diff --git a/tasks/atomistic-materials/atomistic-materials-astra-xhigh-1h100-48h-20261020b/submissions/S023/additional-neb57-cuda-diagnostic/barriers.pdf b/tasks/atomistic-materials/atomistic-materials-astra-xhigh-1h100-48h-20261020b/submissions/S023/additional-neb57-cuda-diagnostic/barriers.pdf new file mode 100644 index 0000000000000000000000000000000000000000..10649a4b7b86abd77844b9ccd6f366ab891858ea Binary files /dev/null and b/tasks/atomistic-materials/atomistic-materials-astra-xhigh-1h100-48h-20261020b/submissions/S023/additional-neb57-cuda-diagnostic/barriers.pdf differ diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/README.md new file mode 100644 index 0000000000000000000000000000000000000000..adb12ebd568d446c5e2a694b5e7cd76d086819b3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/README.md @@ -0,0 +1,64 @@ +# SeqFunc smoke00 + +This is an initial pipeline checkpoint, not a fully trained scientific result. + +The single model has 13,218,756 trainable parameters. A convolutional encoder maps the complete +1,048,576-base DNA input to 1,024 tokens, followed by six 384-wide transformer blocks and shared +heads for all eleven output families. Dense heads predict 128-bp bins; splice sites use per-offset +logits, usage uses predicted positions, junctions use low-rank donor/acceptor factors, and contacts +use symmetric token products. The implementation is derived from the supplied reference code. + +All parameters were initialized locally with seed 17. This checkpoint ran 16 optimizer steps on +the first 16 eligible training intervals, using quantized targets decoded by the supplied reader, +on one H100 80 GB on 2026-10-22. The measured training loop took 4.0 seconds, excluding setup. +This short run checks the pipeline; it does not yet meet the intended full-training coverage. +Subsequent experiments use every one of the 41,020 eligible intervals per epoch. No pretrained +parameters, validation training labels, test labels, or test-derived caches were used. + +Both executable entry points take INPUT_ROOT and OUTPUT_ROOT. They use only public manifests and +the pinned reference FASTA. The package includes its CPython 3.11 environment, PyTorch 2.7.1 CUDA +12.6 runtime, and all Python dependencies in runtime.tar.gz. Entry points extract this environment +into a temporary directory, run offline, and remove it on completion. The host supplies Linux, +bash/tar/gzip and the NVIDIA driver. Model parameters are last.pt; source code and a SHA-256 +manifest are included. Track values are float16; perturbation values are float32. + +Validation reports accompanying publication cover all 200 validation intervals (validation_tracks.json) +and all 2,697 development perturbations across 13 evaluations (validation_variants.json). These are +scored by the unmodified verifier tools. The checkpoint is deliberately undertrained, its splicing +head lacks a local base-resolution encoder, and variant effects should not be used clinically. +The supplied 215M baseline has validation family-mean Pearson 0.302; this checkpoint is a pipeline +milestone, not a claim to improve that baseline. + +## Measured validation results + +Family-mean Pearson: **0.004141** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.004916 | +| dnase | -0.000656 | +| procap | 0.047810 | +| cage | -0.000640 | +| rna_seq | -0.002504 | +| chip_tf | 0.002254 | +| chip_histone | 0.000647 | +| splice_sites | 0.000067 | +| splice_usage | -0.000982 | +| junctions | 0.000000 | +| contact_maps | -0.005367 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4941 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6542 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.0486 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.8434 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.1946 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.1778 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2330 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.2177 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1637 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2077 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4133 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2092 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2001 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..2313fbcc26c62eeb97721fb16cfec21409c78b08 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/package_manifest.json @@ -0,0 +1,26 @@ +{ + "parameters": 13218756, + "source_checkpoint_sha256": "a8d6d7e5a4852d7e869b05d30ed4086a1ed8f2c5711b8a8baaf909f8cccc63f8", + "files": { + "README.md": "325e8e0af474a1a5ed730515b06f61cc57aff5ed6148c951f5287783372e7de6", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "fd99cef3f7bff9a57881cb645a717ae2812f4d5a2229bfd96f288a5cbbd024bc", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "2725592456c55c188b22c376f0cb249e9f46b4ae1370d2226df7143da358a8b2", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "713940fd2b335bf1ce16d1d76241a174f42c3ec15ebdf5c97c1e03f9cd691293", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "2547c9adbdb51ad1acf796a548724cc7770eba90a755f4cb3b37d6f591adf6ec", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "validation_tracks.json": "73a44a7400dce3e367ed4aa3fb4fd4b4d14524acad6aaea26843fc89813d9519", + "validation_variants.json": "74ea2d8e1f2714cd73a6a71bdc5f302139075871e85b7a3100b349c9d9472c84" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..52f3decd4fe2e9d275d08606b74575fafb701434 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/predict.py @@ -0,0 +1,130 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..309576c3d9d03f5dfc5251033b4759306198e7f7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/model.py @@ -0,0 +1,368 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + return self.dense_decoder(target) + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + return self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + return logits.view(len(trunk), self.config.input_length, 4).float() + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self.dense_decoder(self._gather_bins( + trunk, donor.clamp(0, self.config.input_length - 1), self.config.trunk_bin_size + )) + destination = self.dense_decoder(self._gather_bins( + trunk, acceptor.clamp(0, self.config.input_length - 1), self.config.trunk_bin_size + )) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + expanded = trunk[:, None].expand(-1, 2, -1, -1).reshape( + -1, trunk.shape[1], trunk.shape[2] + ) + donor_features = self.dense_decoder(self._gather_bins( + expanded, donor.reshape(-1, donor.shape[-1]), self.config.trunk_bin_size, + )).reshape(len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self.dense_decoder(self._gather_bins( + expanded, acceptor.reshape(-1, acceptor.shape[-1]), self.config.trunk_bin_size, + )).reshape(len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..241fdd4144c4c917109104271fbe80c603075fbf --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,534 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"] + halves = track_halves[tracks] + queries = torch.zeros((len(tracks), 3), dtype=torch.int64, device=device) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = torch.ones(len(tracks), dtype=torch.bool, device=device) + position_rows = positions.detach().cpu().numpy() + halves_cpu = halves.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return queries.unsqueeze(0), valid, missing + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..4412622f650356e2d87e1618d7b942ecf7caae9d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.004915952842900292, + "valid_tracks": 167, + "per_track_pearson": [ + -0.1049344485503758, + -0.04296406868172964, + -0.006648829890009116, + 0.10243842738760528, + -0.03885291385750769, + 0.0358136718673965, + -0.09956069864012582, + -0.018256181739257846, + 0.06338415727185907, + 0.03936396091824245, + -0.07256372746970464, + 0.008759256272046064, + 0.018570667003201917, + -0.06402041552391692, + -0.08752229893403926, + 0.018330830832654545, + -0.018592558419916614, + -0.01650367092600338, + -0.0634570945496311, + 0.026011961648679915, + 0.023012823206842732, + -0.011649342122670719, + 0.008471910448581196, + 0.001422996086190091, + -0.003244479512893527, + 0.02048112471922439, + -0.003969692313483074, + 0.06401915816371793, + -0.003511773651617039, + 0.0009598725108884894, + 0.008447974290155764, + -0.012681669658688801, + -0.005957111354692824, + -0.009743620357138112, + 0.009925950672242634, + 0.03136951398527792, + 0.01706347902882195, + 0.05335933963628765, + -0.01707012021064291, + 0.004372084418689826, + -0.013126457469693415, + 0.008806937382306958, + 0.01031912810969047, + 0.044342589027990376, + 0.014963176755990095, + -0.015679668319426302, + -0.0011391982248855764, + 0.02257780862225278, + -0.013048761188872408, + -0.006224173433243423, + -0.017453573078587723, + -0.006360275100841979, + 0.014740178302500696, + -0.08736847099105884, + -0.02691256607939254, + 0.03815845137572087, + 0.16286080759110336, + -0.08846509037725266, + 0.06448886191478273, + 0.050115487597827114, + 0.046713606751127144, + 0.041577246117223854, + 0.16793916480461193, + 0.004037074608248214, + -0.15068272831161345, + -0.005416279605089634, + -0.05665802066629282, + 0.21138054579320276, + -0.0506201511133995, + 0.03200269734167251, + 0.046288086870654166, + -0.03258841845931451, + -0.0015508880463335878, + 0.03428259873312597, + 0.01132483886846134, + 0.005160109369912111, + -0.02108624857541142, + -0.043902293336634236, + -0.03604729836374465, + -0.0035156868516737377, + 0.04159270903714603, + -0.011253287573098784, + -0.06420320048862509, + -0.024681939401829663, + -0.001895411664483454, + 0.03342008104099099, + -0.007095833569232006, + 0.02988954302635519, + -0.0022632058209565023, + -0.0007655952436115378, + -0.045437004021038896, + -0.004309344615393827, + -0.03558498863375986, + 0.017499397180050412, + 0.02043916240172624, + -0.0021271436312031615, + 0.016524769978415465, + 0.011652604385740058, + 0.014716785429029178, + 0.05812931447766315, + -0.006035382070262623, + 0.03472375412661923, + -0.00926782090974799, + 0.031017728279023868, + -0.01875594023378054, + 0.019756906634058043, + -0.07654176106443154, + -0.011359619781533865, + -0.01289002544449073, + 0.002239230681858112, + 0.025979996384828896, + -0.030339003552382632, + 0.008055341248913528, + -0.005037420551009009, + -0.0019225634424296252, + -0.011357888400880808, + 0.005581319117419201, + -0.00844977316087674, + 0.0016375249147511308, + -0.022846965398750148, + 0.010255194422821249, + 0.014603242796599735, + 0.07190739152087978, + 0.10558134003718543, + -0.13901207237716176, + 0.09197544974918775, + -0.180268456697024, + -0.06223518672827913, + 0.027686705795236053, + -0.02249534623256496, + 0.06751636696588631, + -0.016372963572621965, + 0.03674889015521834, + -0.1252100088863179, + 0.12281523832652269, + -0.053904745431377105, + -0.0027228531803051965, + 0.12710474781930448, + 0.004289328236477226, + 0.21371156234541672, + 0.053067475465147236, + 0.026606471021330968, + 0.1479221709417728, + -0.060288561969015934, + -0.03255935951317716, + 0.028773148221765785, + 0.13519096269791253, + -0.17818708705487088, + 0.1554826220485419, + 0.14386112533848652, + 0.05203550088071346, + -0.011815739399254795, + -0.05488163580538878, + 0.0009288613271203637, + 0.014165692661264481, + 0.0625142610701533, + 0.06756260983337158, + 0.04815468615079113, + -0.022165869538550157, + 0.06331273690212195, + -0.0019459457081554199, + 0.10102841723052028, + -0.10570816509207043, + -0.15661609301687895, + 0.0411942540031637, + -0.009526796844992688, + -0.145658084173542 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": -0.0006558140456984777, + "valid_tracks": 305, + "per_track_pearson": [ + 0.07697626216568408, + -0.22257741165720266, + -0.07518985596813597, + 0.027268604134990743, + 0.05277792052661022, + 0.00815215885399161, + 0.24316652656963705, + 0.19868790391729807, + -0.08787273299034311, + -0.03130813960123382, + -0.22212136299808802, + 0.08894943496513336, + -0.04203933018322692, + -0.04684602580119035, + -0.020090095788601662, + -0.03833064798630419, + 0.06914403570367729, + -0.06223929183788844, + 0.17804050120714998, + 0.07250690420297011, + -0.04140835011735207, + -0.046695778078108645, + 0.1484869783272709, + -0.0056447704539814535, + 0.034422821175133685, + -0.3151625911422362, + -0.019688206634053292, + -0.13613386452812803, + -0.04473471504336859, + -0.05454794405470366, + 0.1296514303199018, + -0.06637543071975366, + -0.11261694830084626, + 0.07283969515914984, + 0.27339805249369303, + -0.1325973742916315, + 0.07850826159701689, + -0.21977230561641578, + 0.06263739676985605, + -0.09022506050324372, + 0.22956801349972797, + 0.08309690790255572, + -0.06942968874692543, + 0.10027298178864963, + -0.09149248887754147, + 0.08749644445757755, + -0.03638011193433431, + -0.037909150213555974, + 0.05673187222427165, + -0.051831669323920235, + 0.00841261962050875, + 0.014793910183275145, + -0.022279641030540514, + -0.1815062652551078, + 0.012282581651553357, + -0.04492634242606018, + 0.035898346304948876, + 0.013717576049550236, + 0.08755228223272632, + -0.020440282817738506, + -0.03777402970377306, + -0.086860022029546, + -0.09319731236560458, + 0.0434233215777856, + 0.025332793671471086, + 0.011660459892857368, + 0.024452039909951845, + 0.04223499640899518, + 0.004203769618092326, + 0.0028838799906964374, + -0.11671003605014062, + -0.04196233768146925, + -0.01044868772777168, + 0.04252688076952047, + 0.008108696118637732, + 0.010953357434078072, + -0.06462136172493746, + 0.011071905277472096, + -0.0031960454375618203, + -0.03480872050272047, + -0.007180975022438096, + 0.0029564088676367695, + 0.033577992779235226, + -0.06908173354437719, + -0.010527951774191082, + -0.018493198834986176, + -0.049038720826727736, + -0.028936320262302372, + 0.08520051032265828, + 0.013318263095672687, + 0.02762449826584863, + 0.011801381134678603, + -0.24343448266967713, + 0.08178174754753062, + -0.09820540610173793, + -0.009312402413445685, + 0.06722767070734526, + 0.14847444902391674, + 0.020470736339288306, + -0.0027533461479582895, + 0.0034973683558494908, + -0.0064920661505787946, + 0.0009300090348513975, + 0.016914629027133483, + 0.003557913983496758, + 0.0076208768511253485, + 0.01445157189321519, + 0.04309623076577374, + -0.3199452646386289, + 0.0775966632911464, + 0.04639544632382553, + 0.02723332015713853, + 0.004595448837429458, + 0.06941643792282866, + 0.04131043495237105, + 0.020838361102808553, + 0.05078274063041354, + -0.1264819080101629, + 0.007326008028336482, + 0.008860895828307598, + -0.08325653546318305, + 0.03447677516246737, + 0.009021745635852573, + 0.01593112533408824, + -0.007868496131900623, + -0.031031246309959126, + 0.020845418825145873, + 0.05688375629152054, + -0.0725800661990507, + 0.025669858014664093, + -0.01869695473750459, + 0.03868385114997932, + 0.012480591636732916, + 0.10298049302060863, + -0.00496288962272397, + -0.009008157009842322, + 0.07121347646690336, + -0.05076560786838299, + -0.054645908387360796, + -0.0970318872247008, + -0.06858026618306232, + -0.035964302839248084, + 0.058227853496658585, + 0.10553170082467672, + -0.018956226910864404, + -0.15248168248612257, + -0.05639755034154108, + 0.03658502074632661, + 0.011000922861511988, + -0.014741420294892142, + 0.0025953129593074232, + -0.01595312645313084, + 0.2734068878615953, + 0.07512255984856997, + -0.014670048398687343, + 0.018489235097899993, + 0.022919337249488914, + 0.0700632548382534, + -0.010691900817225288, + 0.1332144183965775, + 0.12476449620758188, + 0.17130822363638906, + -0.03137323444722649, + 0.022028200525358445, + 0.13367956448306567, + 0.11165121441443657, + -0.0074780179480743395, + -0.11376483881972937, + 0.019130430174785797, + -0.10535225470475836, + -0.0228805102134763, + -0.2805466901400529, + -0.01946923439386228, + 0.045298219093881084, + -0.06653591076788762, + -0.021740818883862884, + 0.017604104793960823, + 0.006102186793438175, + -0.2156942829418701, + -0.008885514681816049, + 0.024305693544180478, + -0.06550074767081629, + 0.0692897685648065, + -0.1460922303602109, + -0.03211829254765735, + -0.04440635740674315, + -0.09807320080620933, + -0.02220555882954823, + -0.023932690820579052, + 0.24756664957896318, + -0.023748787724343464, + -0.08076829198051434, + -0.12865082495697433, + 0.0013979395596948822, + 0.04617334630733081, + -0.010585140392542528, + 0.05368690416877331, + 0.0188092953480458, + -0.008990434532690212, + -0.0458580732878604, + 0.2476051669567738, + 0.09507796360150796, + 0.09879095058364964, + 0.0389812677614839, + -0.09501327442415655, + -0.13121953949451698, + 0.12167690420038299, + 0.04039853742315811, + 0.032616895057698624, + 0.059966183917229667, + 0.030116741872553213, + -0.027453953241604066, + -0.0047299085270210185, + -0.06988782657264819, + -0.031080707115875666, + 0.043580445134904645, + 0.0604377011555602, + 0.034159337268642634, + 0.047131232899287966, + -0.229791503559828, + -0.053748319332713616, + 0.028307484821724373, + 0.2130129685763444, + -0.06486196423559673, + -0.06668311254848422, + -0.0006722431452100054, + 0.11340063432510727, + 0.06291797711453091, + -0.09761094965024499, + -0.033448268205008014, + 0.18945435758607693, + 0.011901570961781266, + -0.0895970610923913, + 0.1536844629070711, + -0.125950407287403, + 0.028078707857129466, + 0.08347009411981851, + -0.10616197329360609, + 0.24402506584925518, + 0.10537870665630723, + -0.2513037588657103, + -0.06297242484008941, + -0.10811414514587964, + -0.0003731007120418199, + -0.04076346300468445, + -0.05197714012831287, + 0.015490277104047753, + -0.08433347319629532, + -0.005434977642498193, + 0.06901604928730085, + -0.1343120860480401, + 0.18197748642952505, + 0.13380259973323297, + -0.06453462623999856, + 0.2577230962815801, + 0.26433446737303573, + -0.24066067921409576, + -0.1586875987962099, + -0.24364278607112283, + 0.17166270477206916, + -0.1421984315190966, + 0.0491475655282753, + 0.16407127097648913, + -0.02031405663644738, + 0.05412489230861275, + -0.13022484968736545, + 0.19764351151353493, + 0.07044394573634356, + -0.02872992992555773, + -0.016238353154272734, + 0.19547681358796176, + 0.009493119461840867, + -0.19634699622538268, + -0.06312114901990394, + -0.04513976858928525, + 0.1088079356251484, + 0.16206453376008595, + -0.03696287857517137, + -0.00921061500726768, + -0.0865328592116344, + 0.008024715236596964, + -0.2414855068031609, + -0.03406730574267633, + -0.16633889233085014, + 0.032132574533178145, + -0.23284058487742637, + 0.0917030787146783, + -0.1475315278212493, + 0.261702529897944, + 0.07099098197643959, + 0.036422301404540704, + 0.03757269811547241, + -0.046768290229103476, + 0.17422104778516542, + 0.09718923353519814, + -0.11409721848732042, + 0.01812496730098392, + -0.0003280215642036123, + 0.005573260212552601, + 0.0031728658018335036, + 0.21881907700297487, + -0.2122051103205003, + 0.023115510694976325, + -0.13563835333245658, + -0.19137373872227761 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.047810251739052685, + "valid_tracks": 12, + "per_track_pearson": [ + 0.09443307441664421, + 0.11218784051844026, + -0.01322803736950871, + -0.07717956623615374, + 0.02203714896596964, + 0.11153216269537765, + 0.01646593338470248, + 0.1079941760593399, + 0.035218808462919665, + 0.09951652202657794, + 0.007299620801328238, + 0.05744533714299469 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": -0.0006398070240639304, + "valid_tracks": 546, + "per_track_pearson": [ + 0.023951453508568487, + 0.013147203080573152, + -0.02661297697059498, + 0.015258716828922407, + 0.005221606522779399, + 0.03317130371015644, + 0.004981419613887956, + -0.003148667784411165, + 0.021423764929732057, + 0.003338117611649091, + -0.028872430874687546, + -0.008103801030973575, + -0.020316200018100715, + 0.007317527805206199, + 0.009716712185722653, + -0.01109396518298901, + -0.002729856957160806, + 0.02532931856382478, + 0.017850754711922635, + 0.005383149593969176, + 0.03304414605194766, + 0.021279458719388352, + 0.02172802568483194, + 0.006809141946143997, + 0.02463518999756073, + 0.00494645304931846, + -0.0062215833358811225, + -0.01881605635377127, + -0.03548327952566213, + -0.052720635003263135, + 0.013872200571324864, + 0.06481949803154924, + -0.004853623033837316, + 0.024111989911952066, + -0.03140002741792722, + 0.024208001968982453, + -0.010561330464074752, + 0.02361053686285138, + -0.022151507054425174, + 0.005770452094281798, + -0.005331039640666828, + -0.030976912688801574, + 0.015511320845390888, + -0.044547607886725765, + 0.006897195806370991, + -0.00709416391517533, + 0.0016368372404678034, + -0.011115260379242913, + 0.009029061088238617, + -0.007341221309513248, + 0.009622849334109024, + -0.04972001641234187, + 0.03901530910470702, + 0.0008447118300125287, + -0.011335888419228514, + -0.019716347810787054, + -0.004664076807786254, + 0.014836933399953666, + 0.02698661289675705, + 0.03282096195012757, + 0.046588256861876044, + 0.0014994961460137912, + -0.005299788186259413, + 0.01657446734272188, + -0.07046875323982864, + 0.015769003354796467, + -0.0566595352584894, + -0.012791927852956332, + -0.015517748670739273, + 0.01822429488371342, + 0.011560376849013754, + -0.008012802658071931, + 0.0003637170912399818, + -0.028174315643875066, + 0.047223358951551705, + -0.02785016397941971, + 0.014702627822712895, + -0.03791643596395592, + -0.00045338906369356546, + -0.007801358492585547, + -0.016435104760926533, + 0.0026348486529321406, + 0.023666082032430584, + -0.00924425169233044, + -0.013676082697295616, + -0.034877569414765525, + -0.03371404284749306, + 0.004418476751476513, + -0.041887083070381304, + -0.0035315537051177773, + 0.0021015922186327207, + 0.037650819181593964, + 0.0025038714614144305, + -0.011823866687841298, + -0.007589377380804431, + 0.028555793610581304, + -0.007902897689333388, + -0.007493765529384091, + -0.0005892278330067791, + -0.004549633530463323, + 0.027274921726729788, + -0.03518216154192506, + 0.03469873242365726, + 0.06162315433281121, + -0.026382418241793656, + -0.045996698271927, + -0.023219950519702243, + -0.012435139419308204, + 0.008052249224274385, + -0.050314901816796154, + -0.026689567102651458, + 0.027707957787181502, + -0.03778421413027301, + 0.011143527359511378, + 0.03193153032236783, + 0.0033575186013222805, + 0.02910779615967355, + 0.017652935744978046, + -0.05240328878466097, + -0.062143441945275726, + 0.038807209815354834, + -0.0440323335394178, + -0.014310492143685353, + -0.013086045691238302, + -0.024329701954426335, + 0.01376010765932079, + 0.029818616489661873, + -0.01177434627989678, + 0.05822472015992072, + -0.015689647417081832, + -0.016951626375636145, + -0.002180498399215762, + 0.004110946844682977, + -0.03266343799596412, + -0.005884367420894671, + -0.04553863316343174, + 0.008476888781473414, + -0.028586772520815762, + -0.0005270844798331344, + -0.009596753781216666, + 0.008485781381104953, + 0.03350173926388806, + 0.022215746671656225, + 0.0031752680847960493, + 0.006543357177334167, + 0.0627041421940358, + -0.009752569068168297, + -0.03501683177834204, + -0.011345841637216695, + -0.009272038426423431, + 0.020868108111308032, + 0.008326934343273723, + 0.037174947539375884, + 0.07128735959670394, + 0.024760644764957183, + 0.001237847096558141, + -0.0015047541064169888, + -0.014651247270556651, + -0.03167745127107714, + -0.0015304000742241515, + -0.031959825149997366, + -0.0014089701390075512, + -0.035178570609791196, + 0.039976557157204315, + -0.019939436042940673, + -0.017143978459730615, + -0.06363650465235687, + 0.014178562501321327, + 0.02811904247408289, + 0.002875222009389363, + 0.037035744510669925, + -0.018032935063340548, + -0.00983383837515716, + 0.011902820518408937, + -0.013308805277181911, + -0.013988967239156292, + 0.002666270897087497, + -0.0114926180974661, + -0.043279316620637116, + -0.04960189198897492, + -0.027112857155824305, + 0.0071580294517893765, + -0.007053694799659126, + 0.02170944194614118, + -0.028473186304842815, + 0.020377184336339282, + 0.0021648857743528783, + -0.005582384764406157, + -0.012563022051444286, + -0.01808516622067034, + -0.01741528673438484, + 0.02807058858602464, + 0.009634160709053867, + -0.009266077023598504, + -0.026916458386306467, + -0.0017199989059081427, + 0.04157810613602076, + 0.03485816433893678, + -0.0007825673237501059, + -0.05434947620597544, + -0.02313071294734707, + 0.011566131085876277, + 0.012269745521251374, + 0.008441139791408595, + 0.014928178236919847, + 0.02183938875078247, + -0.05064774018037036, + -0.002101151879414658, + -0.008988809010869705, + -0.04204808690311966, + 0.015325274505958521, + -0.02778083248364618, + -0.007546532083414761, + -0.023069552521332706, + -0.030473181654530102, + -0.024838053874161613, + -0.03378233645361558, + -0.008856111890538766, + -0.02204975875423906, + 0.04336396461310624, + 0.005262987657646947, + -0.002819983412068066, + 0.016209690608948345, + 0.057945941939955975, + -0.027550171500613667, + -0.02995363119030222, + -0.010667920579891675, + 0.0034241936616129466, + 0.03485471233331008, + 0.03465849488159704, + 0.028667144866893427, + -0.012413171532109836, + -0.03606863107660367, + 0.008296282279342109, + -0.01597696337216923, + 0.014874111628258845, + 0.0507354340681768, + 0.02569546725711454, + 0.0007837283484423875, + -0.032194829944356934, + 0.001943048217036917, + 0.01555705832233775, + 0.008404307978490613, + 0.00693846358609936, + 0.010113023788585843, + -0.02877126742530296, + 0.0015362766554838432, + -0.027193528379893035, + 0.020357548478123568, + 0.03902704008000682, + 0.06142071776027533, + -0.006799521293086881, + 0.05125300984491045, + -0.002127529270690684, + -0.05566556809236316, + 0.010141141111963608, + -0.001510324727185709, + 0.00656274964678346, + -0.014285300110020617, + 0.004230608098493486, + -0.03293773973134276, + 0.036631367720347065, + -0.0049477243722234185, + 0.03628221147320138, + -0.008365243059512268, + -0.01441290480240485, + -0.011454078679521024, + -0.004173216723723627, + -0.008201058190114687, + 0.040204445334735384, + -0.00772313104363278, + -0.00926331961416408, + 0.006330825653595023, + 0.00735214047604246, + 0.003211463621549894, + -0.003933173539976148, + 0.0024399653826470773, + 0.002331428965576291, + -0.015337707132711634, + 0.0007296688405119091, + 0.0234312767784264, + -0.01793297766199079, + -0.01647457666588986, + 0.004086651487719547, + 0.02396278323788307, + -0.015555373785540335, + 0.002877838817195087, + 0.01983315006153555, + -0.01569485666045998, + -0.011031398637332426, + -0.004795621828733936, + -0.005922228743669344, + -0.012547654075865946, + 0.007570480202361614, + -0.03227278375607484, + 0.01991144373967925, + 0.0053039344635379, + -0.004209534955733116, + -0.04106888436741776, + 0.00830525414808586, + 0.022485886211213092, + 0.028305701496766016, + -0.010376783198926853, + 0.00195403089385448, + 0.010051560167983973, + 0.006673887162696256, + -0.0008551172375954153, + 0.012958883224493872, + 0.015983411958570563, + -0.07615433049988411, + -0.014358935256029728, + -0.017361734026657468, + 0.08059725807061524, + -0.025552256881059363, + -0.07399896105113432, + 0.019866508163333938, + 0.042299319453748034, + 0.0077161050661531845, + -0.020890978692265798, + 0.016228018915327994, + -0.004206288204215308, + -0.0033951491006807195, + 0.013146179088637732, + -0.023426375958876755, + -0.0059456013754414444, + 0.041498571279103585, + 0.023656024390273103, + 0.023132394548825302, + -0.02326784356616203, + 0.017379197190277036, + 0.024863085394346243, + -0.03421397017739831, + 0.005014343040859842, + -0.028877602405372956, + -0.03445964113456444, + 0.01014486275547836, + -0.0017326810739511475, + 0.040117756724989816, + -0.01257730964530592, + -0.04369734626014036, + -0.0013865031712099775, + 0.004489696822512935, + -0.002621065793299286, + -0.0003844716057536062, + 0.052681278472670984, + -0.012654489587078918, + 0.029839831877143456, + -0.008835983459200147, + 0.00860193476986726, + 0.010125460566094409, + 0.03077917080862525, + -0.01697667438033614, + 0.022257195843716042, + -0.009813043124758523, + 0.003240088723534801, + -0.03304826751918534, + -0.042678279888200756, + 0.00754210862624653, + 0.007396580714492659, + -0.029726346749011905, + 0.0030999548787240118, + 0.010513631969700484, + -0.04965497202730831, + 0.007708847542226579, + -0.013428030716780556, + 0.037194205104232124, + -0.005109209041236154, + -0.02102865054361144, + 0.009012529912958204, + -0.02107132511024811, + 0.026757053931357747, + 0.008726957319940807, + 0.00374183230121052, + 0.00923752256116834, + -0.019277961709165757, + -0.0047212455631725605, + -0.09781729193317507, + -0.037703226973477055, + 0.07938744050553137, + -0.021084998800209087, + 0.05143318987713172, + 0.0019047351878912033, + -0.03396393722410705, + -0.03413658967992254, + 0.037319365241781643, + 0.02163325307165503, + 0.041704380681115844, + 0.04201553755598295, + 0.0029173913010142414, + 0.039726903685718, + 0.02076011669328164, + -0.021513832641286805, + -0.03328894089396239, + 0.002787293629334359, + -0.00395011377018043, + 0.02633678256149198, + -0.003428485937096404, + 0.017488061517565327, + -0.0010164293110895475, + 0.04390712123469582, + 0.05174952810595704, + -0.044400096293587606, + -0.02054785217251241, + 0.017938688965937616, + -0.006713973031710014, + 0.02295923130041524, + -0.023112590426460914, + 0.023341527295475503, + -0.031568561549582506, + 0.009667884494841735, + -0.005245265947354965, + 0.005189382587426895, + -0.028490304151338568, + 0.025027567506701955, + 0.02289575976371269, + -0.016193581404865848, + -0.022901055573066084, + -0.0175133948661421, + -0.03735853309220987, + 0.012212642258983825, + 0.029316980071847562, + -0.018862056338452116, + -0.029544232470297385, + -0.007636035255266807, + 0.008497947220605907, + 0.018034060248246282, + 0.024851304785182007, + -0.01376421040824361, + -0.029917598694868514, + -0.00045355108408423904, + -0.01326804223307175, + 0.016790437088672883, + -0.010768465019428238, + 0.014062626322005625, + 0.016238732691207965, + 0.0026988051512351215, + -0.03271079541465961, + 0.020082362401301457, + -0.022458446476264063, + 0.007033233249362265, + 0.0007803866878068722, + -0.006518889050691166, + 0.01120347094096168, + -0.004619732967636184, + 0.01867197998782932, + 0.008823128752373578, + -0.03718461151867448, + -0.03836001782709004, + -0.007954877187256603, + 0.015800027134532865, + -0.030204970726117275, + -0.030484291691377022, + -0.04353455760563841, + -0.026707338774731613, + 0.05492483462182612, + 0.0009496508687788455, + -0.02572725852463475, + 0.014854616978170419, + -0.009754481195234739, + -0.0036791659376030224, + 0.012532167983991597, + 0.026762966001986053, + -0.005127300146235178, + 0.02499642299430325, + 0.013433401954564293, + 0.014559381032421954, + 0.008064823263569443, + 0.008725953691517442, + 0.0020854653102250005, + -0.0034905222364758175, + -0.005501198601739235, + 0.006108658215616611, + 0.009951084768341065, + 0.0018223752244599612, + 0.0017933358222632523, + 0.007869538163388888, + -0.0010057844432221326, + -0.006665063334498295, + -0.0033755038934443413, + 0.002850256772292334, + 0.0065696775405561314, + -0.0004440423651646733, + -0.004567871023238729, + 0.025385164471620616, + 0.024763633331539296, + -0.03812827686359909, + 0.02088244626922988, + 0.003869809972219522, + -0.011692420324840544, + -0.010574834851972869, + -0.019724090091202025, + 0.01763518501373722, + 0.043276982282434776, + -0.025119543226104515, + 0.010572990141348581, + 0.045787557487224545, + -0.050658678703714925, + -0.010997050685498364, + -0.015552794507403956, + -0.007081400320933711, + 0.010273775747373546, + -0.02749454215318593, + -0.012333829919406302, + 0.01742922524946678, + -0.01410813418540856, + -0.05261916505368357, + 0.012143843130440255, + 0.013546143655845787, + -0.05401198553791609, + -0.07910894210561485, + -0.02040646234723726, + 0.0033713513673724557, + 0.013129457634188844, + -0.007008710358739695, + 0.007833868094955976, + -0.007295489334138134, + 0.02810874232608222, + 0.008207598295347628, + -0.007159189303529865, + -0.00017996550060169942, + 0.021120292318549515, + 0.016326021160097855, + -0.02803958356302888, + 0.004768534213155511, + 0.029132003100790435, + 0.00794868893391664, + -0.002226612366086485, + 0.009664691603375379, + 0.002972548166686194, + -0.02909457359107026, + 0.0003324002610091622, + -0.01321238682894319, + -0.0194433124380604, + 0.02435847590859949, + 0.0230826925909909, + -0.007274540365589866, + 0.0035748236138019984, + 0.015860309585364395, + -0.030444944165939694, + 0.011655434427993331, + 0.009769551307158028, + -0.005881129171051334, + -0.00829047254416802, + 0.07867722733675334, + 0.01198401531542034, + -0.0112622065195895 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": -0.0025039561785947986, + "valid_tracks": 613, + "per_track_pearson": [ + -0.014197857698690891, + 0.019386175191781336, + 0.011154452921935176, + 0.06818958468559949, + 0.004043823763722735, + -0.024286375358021337, + -0.0434541124067152, + -0.002802668364047326, + 0.0834059585187715, + -0.05831006196245696, + -0.028970916997398843, + 0.016511690710273273, + -0.0171827706178109, + 0.056325848372603585, + -0.07192635887318448, + 0.07954487868833468, + 0.009536349526530605, + -0.065085524534133, + 0.07708961885368963, + 0.04962827436480554, + 0.01701483133186324, + 0.011457316357283252, + -0.028575686486095604, + -0.025161238186099288, + -0.029731005993377843, + -0.0017361437898110747, + -0.013313349309487355, + -0.033829898147099496, + -0.0479221379724388, + -0.1274193437298469, + -0.027490521749838407, + -0.0038073899925854865, + -0.05925260934536202, + -0.004803514933398761, + -0.07806754691579887, + -0.022609515727050427, + -0.04263608196824902, + -0.031378812403992117, + -0.04033542074831431, + -0.011197480397570257, + -0.05901962823681326, + -0.021322534861932183, + 0.04397392369719608, + -0.060362254044504536, + -0.020322833647379985, + -0.008920836871283743, + -0.06032900985174777, + -0.08895091316490772, + 0.0809538669335069, + 0.008379288505956616, + -0.09990709183445325, + -0.012161115345721946, + -0.009909111697053214, + -0.05418602143453866, + 0.032994523131389265, + 0.042258453539632666, + 0.06483228131898229, + 0.04225950908684035, + 0.09323318423050937, + 0.009305017993277978, + -0.02184717108093116, + -0.025559060357210263, + -0.030744678385787573, + -0.01700010252620173, + 0.026491613496057363, + 0.018074527074360965, + -0.07497511874707158, + 0.025581609630398742, + -0.05573859335744578, + 0.047202228430865995, + -0.023908140192305753, + -0.04303535277737864, + -0.044427059693333736, + 0.017778452312933884, + -0.04417356182423982, + -0.07420822046604166, + 0.017406224973011072, + -0.07543843556548194, + -0.06359749939007839, + 0.06082136051939296, + -0.009838405641624635, + 0.014366003388180452, + 0.07203392343288902, + 0.06688761801335841, + 0.022607779455797693, + -0.06945460379361072, + -0.015445489361872004, + -0.11509481282372183, + -0.020426565994108747, + -0.004071209894524824, + -0.042260612987860116, + 0.010708954042356446, + -0.0414686831806857, + -0.06955688498578635, + 0.01315157165319352, + -0.028675917797852838, + -0.08654576190494999, + -0.04681752717297527, + 0.024588505367360913, + 0.02627017832176775, + -0.0012966818648435948, + -0.030886945097200445, + -0.006705041291623368, + 0.03490126996988748, + -0.007002529132936286, + 0.020112967475081362, + -0.06680627649643539, + 0.010984969226475481, + 0.011614906984402455, + 0.010832880454853358, + -0.014042413266046275, + -0.012909746039077123, + -0.0810706458639125, + 0.03468681240891485, + 0.015732595019559772, + -0.01909747626111267, + -0.0887055784731903, + -0.07900577833202779, + -0.011009410117721304, + 0.025849683735806085, + 0.031537819627125756, + 0.04959147783047132, + 0.0031431173407823194, + -0.009359910157433905, + 0.05224889022855513, + 0.07823530676838894, + 0.03331593898205449, + 0.05454303071580894, + -0.017836865475282508, + 0.035444275940298774, + 0.10759889602580669, + 0.058675118825186695, + 0.025480118188895385, + 0.03099525809522552, + -0.013685888883726176, + 0.041045399197106355, + -0.004559913265404951, + -0.046145660299430825, + 0.0661498534152783, + -0.15293946733057964, + 0.024617556344526066, + -0.017920772105483804, + -0.0384292160157512, + 0.038044618625351696, + 0.041381867115485756, + -0.017333746765995472, + -0.047633622984862095, + -0.08160050801296403, + -0.017899661281003454, + -0.020912143572256996, + -0.040247948068919535, + 0.020642463567594922, + 0.029913992936384637, + 0.06179963726365914, + -0.046489096622087105, + -0.05229296552616071, + -0.06944681894416924, + -0.0939757035559443, + 0.08518247589568749, + -0.07947247047647979, + 0.04043914991544687, + 0.02843781331966477, + 0.050688871381682835, + 0.0016361276314581128, + 0.03061629790150897, + -0.029131126425549287, + -0.09593285807057937, + 0.012562588875253838, + 0.025481827626025674, + -1.4773029213342924e-05, + -0.017867220759130532, + 0.021323289240368196, + 0.0047324056151944556, + 0.05113105513684077, + 0.03076139776374642, + 0.07465213991923675, + -0.019052656642673672, + -0.08102739991962785, + -0.04767632400025109, + 0.0011077228668684017, + 0.0590029238552483, + -0.014230616853294752, + -0.0647239331725484, + -0.014793585198730039, + -0.00133408050832219, + 0.0013023824691478462, + -0.07037280052443146, + 0.0023059889136920336, + 0.0021570701587084514, + -0.0922384656147043, + 0.047574674238514976, + -0.012538238653660388, + -0.011964858536693894, + -0.014098551531183718, + 0.014019634779521768, + -0.038803202092018575, + -0.020528234111177615, + 0.00997671514120924, + 0.02946610991641389, + 0.045491305448680984, + -0.0477827002659991, + 0.0479604318610598, + 0.1374332767207188, + -0.004606850180645199, + 0.003023341997247542, + 0.05000902033114424, + 0.005259703295095689, + 0.008847929896732474, + -0.011730728762382193, + -0.023409033531678936, + 0.027630993011940237, + -0.03449739225439637, + -0.033172658405923254, + -0.007398858414045624, + 0.024993512175135173, + -0.017130190205519223, + -0.010104020768376718, + -0.04198994973494816, + 0.02457053459292064, + -0.021040513004338804, + 0.06156990900508954, + 0.0021173929748249, + -0.017445605157616804, + 0.033480918341587944, + 0.03776926658961539, + -0.020482263485292584, + -0.028863623042334783, + 0.0595120965276473, + 0.015551523323995349, + -0.0072319337270980435, + 0.0381337817670249, + 0.040348517887067746, + 0.006197054816303614, + 0.005356455099066764, + -0.033952134331143184, + 0.006356276355217421, + -0.008267242905625133, + -0.014256183918725096, + 0.003867182422352186, + 0.03208838752632602, + -0.05714995489095779, + 0.03170977401992961, + -0.05203512404050423, + 0.02812428501435352, + -0.04773630031571455, + 0.027997051218333145, + -0.14895953552278685, + -0.08545586952546383, + 0.08489339260054, + 0.014986444406327954, + -0.0771571996415923, + 0.05188695266533788, + -0.01105694475357753, + -0.0462494074463006, + 0.10560716661418362, + 0.13197949799021788, + -0.01623697567025835, + 0.035109606581903034, + -0.038438910680037276, + 0.05689643406034124, + -0.011826879592567369, + -0.051878257741342575, + 0.019114508864928725, + -0.036616406939665266, + -0.03621647983406846, + -0.011091938312366665, + -0.057984354520778075, + 0.06772397636568751, + 0.04452835254505995, + 0.001998614860783531, + 0.006900187298923128, + -0.010157664559067055, + 0.03300419926203141, + -0.01163341414898343, + -0.02494814032927799, + 0.019780680395780934, + -0.0004676551906943202, + -0.02496383717450541, + 0.03626467709133821, + 0.014369323941342113, + -0.025374251974891994, + 0.019558884783941473, + -0.025627862616080205, + -0.0021797531886832773, + -0.08305648164983426, + -0.025879381987914674, + 0.05849537880705119, + 0.025836447262148295, + 0.030891263086586428, + -0.008437517537071395, + 0.06074849079183482, + 0.040204048849908094, + 0.019683243805619702, + 0.0026930047291771285, + -0.0063450378795951, + -0.008129643967193697, + 0.038102299140118326, + 0.015208667321611583, + 0.02526313587397228, + -0.024568069983392442, + 0.08929413037218951, + 0.01278963258995123, + -0.05320118359173815, + -0.02567054882990891, + -0.0019826629692126017, + 0.014271941477304439, + -0.029349394745515983, + -0.045938949669885196, + 0.0066558285348730355, + 0.039284270257120155, + -0.032015510974107216, + 0.007872055246821824, + 0.013120686708172385, + 0.025244925194796427, + -0.005149521967056136, + -0.015345372330341394, + 0.003928840433092041, + -0.004203266019754384, + -0.024214451922734605, + -0.0023519441193859277, + -0.0028790268571413165, + 0.01992015529059947, + -0.038065653605800887, + 0.02870662887320823, + 0.0258191391591036, + 0.014615497849708873, + -0.023433952492661086, + 0.029753141068539195, + -0.03125875887823487, + 0.01172323754393271, + 0.030216516741922984, + -0.016301347041635206, + -0.018616256026170848, + 0.018723718875910587, + -0.02882898385541501, + 0.005519523082038747, + 0.03921165809460991, + 0.008241103125095956, + 0.003397543343746435, + -0.039926756836453685, + 0.02471387473929674, + 0.002347686143380721, + 0.0013894354831993254, + 0.01775508452397726, + -0.007176445509117753, + -0.008806100182777846, + 0.05304234723863511, + -0.026326514955585075, + 0.033500883051008866, + 0.0157446599153502, + 0.0010779422273849853, + -0.006915713462422147, + 0.018594687542770393, + 0.03425615699895968, + -0.06269722685141944, + -0.0009369692528241622, + -0.04566946267504032, + 0.0009676623148467784, + 0.03177245210191802, + -0.000719430034743821, + -0.056314000290755103, + -0.031850824048722345, + 0.022505163916660353, + -0.006791215202584326, + -0.012931426295866232, + 0.032251943764979994, + -0.01267747516410351, + -0.024799370388020346, + -0.024228310300823883, + -0.058966508687947096, + 0.014031523003468997, + 0.011509564146299995, + 0.0031101666174188796, + -0.0248458104017321, + 0.014935195552531274, + -0.038445188128432506, + -0.04102389200681308, + -0.009661053112519956, + 0.070933154193509, + -0.023814891204208967, + -0.05689654628615822, + -0.05623278451839211, + -0.0358202971293665, + -0.014597970422980737, + -0.04834296402342536, + 0.06606420134698492, + -0.024735627750960845, + 0.028652490629587274, + 0.05233708721877207, + 0.01661297759846549, + 0.05812685149958814, + 0.02457526330675983, + -0.02341109624563569, + 0.00795766090269611, + -0.0343712160273005, + -0.0391898297474644, + -0.013095129031444204, + -0.025466715057806646, + -0.04255433441302246, + 0.008060229389029642, + -0.013925050950350763, + -0.05043600054561909, + -0.07328937381840081, + -0.06501818273622918, + 0.009120517177961878, + -0.02891424059208122, + -0.02013288968291645, + -0.04489223675013673, + -0.07713474063208038, + 0.06956538641840203, + 0.0017854791317367133, + -0.02025836105872281, + 0.019383127380001797, + 0.0543034557220651, + -0.08904687546828541, + 0.009244222035112552, + 0.021401109282953978, + 0.031190939855686127, + 0.020411169372286538, + -0.006410732459074095, + 0.02313997217273051, + -0.029815976496701504, + 0.056971496336156544, + 0.02941978005691233, + 0.00108203465928737, + 0.008534651880102712, + 0.00923120453949494, + -0.019600912325848537, + 0.038603999786582535, + -0.013156604044325197, + -0.006801720276091837, + -0.03924233860530147, + -0.0054727463133270714, + -0.025808338842241117, + 0.006039692723758612, + -0.01151560025390113, + 0.02764678571222448, + -0.04442238350555685, + -0.0018862964782763376, + 0.06695092738367027, + 0.06147306947141569, + 0.000284816516832969, + 0.03858824877476088, + 0.00022409351718909865, + 0.045655760542304615, + -0.032841845808034634, + 0.04376217452754583, + -0.061935117895924215, + -0.0382338052785009, + -0.0736311123294346, + 0.0044657618861387955, + 0.01667356394874083, + 0.00876071366727856, + -0.015165246153360783, + -0.022775378443396548, + -0.029317605930127954, + -0.013515202972096853, + 0.013190955586687588, + -0.0004782104827800014, + -0.01846822744486034, + 0.011834752780066564, + 0.03826998885264174, + 0.02083138247433435, + 0.0001639818189377368, + 0.018095053151915542, + 0.012845250967195387, + 0.038442624238670925, + -0.028727218709029608, + 0.051210225806823585, + 0.0023314874106067354, + -0.023381876120634082, + -0.03617526584612359, + 0.006993492818640267, + 0.06954332034524324, + -0.017884352241266346, + 0.007550328868656922, + -0.0006784347390395854, + -0.012127883067046982, + -0.021435688176669646, + 0.02846338736557751, + -0.005709557795414471, + 0.0027863752530558573, + 0.002116060516951374, + 0.00350652233604059, + -0.022921750709200207, + 0.0005510087539836948, + -0.006166863638528089, + -0.016552038338372304, + -0.002054154451054814, + -0.01627874415954531, + -0.03140490016198638, + -0.0035068660876416994, + 0.02741214656802954, + 0.013480195292921588, + -0.0048223649634877705, + -0.024872207649011383, + 0.008764333923459672, + -0.019614803500105814, + 0.06230381000505599, + -0.00687075170626403, + 0.012557369690637234, + -0.0007874237468015225, + 0.1110821370642603, + -0.04911341964362374, + 0.009529441101269786, + 0.01725396552654509, + -0.00416677635760508, + -0.01612717802601854, + 0.03538108832783642, + -0.01749963340227716, + 0.017212917459514646, + 0.010053284215531711, + 0.02561384011864482, + 0.013764213783020493, + 0.021679485268586117, + 0.06904704777081921, + 0.01554442339067823, + 0.14134920917546806, + -0.03459790848233182, + 0.01114045613126079, + 0.022696900510398278, + 0.013715329255117968, + -0.0302426759514741, + -0.009240208816852244, + -0.027276576715342876, + -0.05383038530432436, + -0.05397719151292267, + -0.03343479376429211, + 0.010084502076782327, + -0.06176355928829812, + -0.004812635485805059, + -0.0194868609086418, + 0.01796264275226742, + -0.016988288507807165, + -0.01832713554496484, + -0.032100344225583624, + -0.014177451607867167, + 0.01042113315785883, + 0.0025701215522916723, + -0.0049195326398684385, + -0.018533735440142907, + -0.08302549059805814, + -0.06533444416277802, + 0.025671255992910452, + 0.023993431923470077, + -0.07312897087880299, + 0.0611260829098914, + 0.026859243912952677, + 0.01960050885426911, + -0.013806755603888348, + -0.09198709016608124, + -0.029250320479598163, + 0.018870257534895334, + 0.0852649343966724, + 0.002054767505208499, + -0.04518395204254914, + 0.00909305403469003, + 0.05369129799257049, + -0.0738901204631279, + -0.05425301085020579, + -0.06318438235911307, + -0.06596198119974685, + -0.028202252164804112, + 0.0018292396529690793, + 0.07750125353742798, + 0.03403848255698989, + 0.0656950113984721, + null, + 0.03868672756954751, + 0.04314629468710596, + null, + 0.034061425544678864, + -0.10784668370355907, + 0.014779606106848147, + -0.0008405386703803857, + 0.11832651754503583, + -0.07352249509761374, + -0.09887503383030853, + -0.11619309368966223, + 0.11416203339612248, + 0.009069457404361781, + 0.011793416347844792, + 0.03650593982941431, + -0.028863479762882126, + -0.011055206049010476, + -0.0249323487925614, + 0.08148639442297755, + 0.014014737599597963, + null, + -0.05072380726298017, + null, + null, + 0.00482351890502337, + null, + -0.07244548849586396, + 0.013830936701501921, + null, + 0.09922373354516976, + null, + null, + null, + null, + null, + null, + null, + -0.09036858357509164, + null, + null, + null, + null, + 0.06407012173585959, + null, + -0.01649168903102447, + null, + 0.07726721517488946, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.09221096015342198, + 0.027127487880854673, + -0.0157678052388639, + 0.049313262605324176, + -0.020841867098821094, + 0.08623215461318365, + null, + -0.1260306689660422, + null, + -0.05451427387095014, + null, + null, + -0.09429603848622872, + -0.055825625875920874, + 0.05875637287251016, + -0.015487891158949352, + null, + null, + -0.022924035871648474, + null, + null, + 0.012255839267991748, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.01231007405053748, + -0.005833536459765145, + 0.029155859895905373, + -0.011954741689366286, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.00225404725535643, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.0009539704773853841, + 0.006029550984899722, + -0.01363729267044161, + 0.00019776596211022116, + 0.008867702601629812, + -0.03378690243433331, + -0.0017318612374036318, + 0.053039294060862656, + -0.0123695610376042, + -0.0138144730724857, + 0.09111791115533444, + -0.02375288625834209, + -0.017612842661928575, + -0.0556925568567252, + -0.01770239662009208, + 0.007784039010499302, + 0.003364646214857121, + 0.010049707192700423, + -0.014717265803138488, + -0.015765278506175078, + -7.269510169179501e-05, + -0.010249803909810938, + 0.01795365400548753, + -0.014266630952039967, + -0.0011925114725979309, + -0.010719129588888137, + -0.026080535965392795, + 0.009970150894701707, + 0.016544370792099014, + 0.016158678026585844, + -0.0008507975929150507, + 0.0006989474549015676, + 0.026849695467286876, + -0.008957234916426048, + 0.07489104100468562, + -0.014211865596197982, + 0.010045348086389455, + 0.007061905849988031, + -0.0014265921098663835, + 0.005044676658767836, + -0.014505353125900535, + -0.03157425915984459, + 0.0026436310139736985, + -0.018730600345193655, + -0.0008516807637811107, + -0.009119264338170936, + -0.0028345029572325745, + -0.0038069024064364475, + -0.05483043792419051, + -0.021131575492196082, + 0.007466935296420199, + 0.021333793094526893, + 0.08628102870407568, + -0.025279349775130723, + -0.04424091230657647, + 0.010574208829459482, + 0.017611654813377453, + -0.006999408995563961, + -0.007949136931686664, + 0.034258998500689385, + 0.0011840705062844375, + -0.0010828332405734531, + -0.04254431460176746, + -0.008282942808334534, + -0.01403221300809671, + 0.009606361354249693, + 0.012748947626442662, + 0.04335921729931178, + -0.02186640397324256, + -0.006325383966708782, + 0.010201031274609808, + -0.007272238142439797, + -0.0036040298721601734, + 0.013179767751645866, + -0.011096076563594097, + -0.04335435618702535, + -0.002677188596619769, + 0.013681429231839204, + -0.005537184254138615, + 0.008830636323876694, + -0.07546463735663259, + 0.004532944664345853, + 0.0461334425721399, + -0.005901400350028901, + -0.012951148888299771, + 0.006443657663479367, + 0.048373224404605314, + -0.03635476314983661, + 0.070338125605211, + 0.06758883592902189, + 0.004720514802649232, + 0.002516323687579374, + 0.030444779025420945, + -0.005081683263010287, + -0.024981485563325415, + 0.04101373012618042, + -0.01957394003341058, + 0.009963592028629318, + 0.0072634909626781265, + -0.007680144424672355, + 0.01249171866201989, + 0.004742268808986037, + 0.021188816165596362, + -0.07308308922928997, + 0.03858795838630589, + -0.0056174028621161975, + 0.023902462211839514, + 0.0026718305364039885, + 0.09459358563255317, + -0.04040584845170769, + 0.019008078000778132, + 0.01503503398117461, + -0.004677966891549274, + -0.014884906020537691, + 0.014760779598926814, + -0.036620881148984555, + -0.00932782964825477, + 0.02363421154514312, + -0.008705846733249697, + -0.0014073872198896381, + 0.01674185806040279, + -0.039268928743265805, + -0.0025883467068459147, + -0.005870808829255756, + -0.018926465516920524, + -0.006201983546241458, + 0.0007088434748672322, + 0.010343877186827808, + -0.050430338306257955, + 0.10659874237242388, + -0.0036531244762311945, + -0.037529928388750175, + 0.020648990028843454, + -0.00427494356299009, + 0.01081109062599084, + 0.0036031960026143258, + 0.002911613029301325, + 0.040711914219644034, + -0.0037616771920479506, + -0.007105008516692439, + -0.013745859892316262, + 0.03240556818261921, + 0.013280680370175614, + 0.0055000817526197265, + -0.0070216448156599056, + -0.09211524542507622, + 0.024224840043913305, + -0.0019030682169986323, + -0.004398242079587239, + 0.0020679260691615618, + -0.04432754608498302, + 0.016673213958800465, + -0.043272965865646155, + -0.012255532761012499, + -0.059596996833995684, + 0.022785094899416627, + -0.00713601171560928, + 0.015289456789324286, + -0.01040243632085785, + -0.04455147162296024, + 0.008412430044007777, + -0.02156071638132987, + -0.027293318827072366, + 0.011903031653095478, + -0.010601937476890772, + -0.03398646240965596, + -0.034686151094165375, + -0.055889704931235504, + -0.007306715814852207, + 0.00558782288308558, + -0.006521618067317653, + 0.0006621731370778105, + 0.011746895076712383, + 0.0051242360313963784, + 0.022823074974093745, + -0.008045203967973136, + 0.021257117367901146, + 0.00971825106071302, + 0.012622598496463713, + -0.03354039152520356, + -0.004450068467086098, + 0.02952104021044632, + -0.01576841875435523, + -0.027978155297282113, + -0.0065461004364073296, + 0.014818812861923115, + -0.0019835756963309586, + -0.006129292795649084, + -0.0038342849004822724, + -0.0018904339695515282, + -0.012046893448983879, + 0.004465149795414506, + 0.009654121643647902, + 0.004386412661028729, + 0.0013750480231821437, + -0.011202602564111364, + -0.0045280556314050485, + 0.018201833646926523, + 0.0015585942074229225, + 0.008732304287472412, + -0.012316812025940559, + 0.0035099091970397806, + -0.006639981012035324, + 0.002821814903730733, + 0.0491291803742401, + -0.013293833761672086, + 0.01930544107664559, + -0.014740779353314552, + 0.004571334779299883, + -0.02500671894577798, + -0.0008041318479880957, + -0.006596413596649441, + 0.01712490672266806, + -0.06205422668662472, + -0.009780437562461958, + -0.015172956263174695, + -0.019234467003123177, + 0.009472284527287518, + -0.05901376048907442, + -0.09040234905648363, + -0.021093366608751715, + 0.10965624643878273, + -0.03446554585265888, + -0.0009965149376679294, + 0.06110878044543299, + -0.1164559673050239, + 0.009866803619839019, + 0.037219012464543774, + -0.008411308111905751, + -0.032611314663754845, + -0.0015655322486833269, + 0.019448133264532104, + -0.020855871494260554, + 0.007239277268525096, + 0.09326192628849535, + -0.012429876576746294, + 0.020441445084109, + 0.01054342016805152, + 0.01868283325560369, + 0.03509328263997068, + -0.08603808627846875, + 0.023984748199226105, + -0.03675225236517586, + 0.023689362530375263, + -0.020816386602300604, + -0.11605624388767774, + -0.00432177944015241, + -0.04895450277700733, + 0.0033272486564057107, + 0.07987264561453651, + -0.002920411848307053, + -0.09134055440192941, + -0.055050447483841275, + 0.019620709582307324, + 0.009567905085960103, + -0.0029764295754593864, + 0.10062650167875443, + -0.017969852569144962, + 0.036745544349359414, + 0.0021378107951097913, + -0.013750003140968347, + 0.10189008887465083, + 0.06405126228864344, + 0.02138460351985134, + -0.0008571983337241043, + 0.09065826750269451, + 0.017596227404863933, + -0.008252476010647257, + 0.04726997240737258, + -0.004612236778559708, + -0.03274316906511476, + -0.08874762594541233, + 0.015770815233318115, + 0.1847936746414082, + -0.010089426445459265, + -0.004949985781594469, + 0.07764586394612984, + 0.03689241733514609, + 0.017528715129079744, + -0.024468324699421125, + 0.06354832386333083, + 0.015557827253846595, + 0.013020151995803439, + -0.08386871258795596, + 0.009670291154642862, + -0.008876135553084502, + 0.0479000139441141, + 0.0032176002883940397, + 0.03726249095107441, + -0.0528266996696196, + 0.01953877568312605, + -0.01170790377281269, + -0.0007027135837856963, + 0.011339442212105002, + -0.01593827356562924, + -0.013605560415346872, + 0.02121803243018389, + 0.037959448041439546, + 0.03994994227082574, + -0.06401854082499239, + 0.12526329222897503, + -0.014032986412784061, + -0.049169929598779576, + -0.00032043875794023633, + -0.04878984260387179, + 0.03180844479507101, + -0.19482060451241467, + -0.0032932066937548603, + 0.07666829863203822, + -0.18438352709080585, + 0.059548103252593354, + -0.0003140707823409323, + 0.022109484938422, + 0.011666656847441743, + -0.06955437051104621, + 0.026524468674373504, + 0.00253864593584595, + -0.07644471375102557, + 0.013117183803647129, + -0.03563521950692318, + -0.005086741612667944, + 0.008233541276429331, + -0.03330022973327411, + 0.01659982953967945, + 0.025456845879460996, + -0.13697383689547377, + 0.047213787206947536, + -0.0025672385030759593, + 0.017016220446065433, + -0.03940502939080553, + -0.046998534131504364, + -0.013676843808429078, + -0.019252711925717973, + -0.022408803355247014, + -0.04011065782635078, + 0.06175937924278056, + -0.0895859790320709, + 0.07239541518890778, + 0.0037094004191508684, + 0.04451238342993478, + 0.008101514201109246, + 0.006465018858341621, + 0.0037984101251001936, + 0.06834822156852433, + -0.01907595244497371, + 0.007689107261696223, + 0.007806768590263954, + -0.006675460602248968, + -0.027815913335317622, + 0.00019191760654950523, + 0.04545815368066738, + 0.012612004701979734, + 0.0035297188043904664, + 0.11148395317012025, + -0.017304029074634675, + -0.01470954375562919, + 0.01819231710709347, + -0.16239179665925593, + -0.014004639506331913, + -0.05639476838597933, + 0.021887595651049467, + -0.00443650408208847, + 0.03093304402332586, + 0.06463279844435763, + -0.010051489074459175, + -0.01497644618859695, + -0.026981175509528773, + 0.04517677468872343, + -0.02513875103855731, + -0.047164875106181774, + 0.023628763378394108, + -0.006883035312449664, + -0.020565703601089547, + 0.007541298740949813, + 0.0004295842043548402, + -0.04324410977863834, + -0.0019462052548579283, + 0.22607329605932072, + 0.0063571877121754135, + 0.05539194591247467, + 0.06452107087562535, + 0.043456945044754516, + 0.013964926447918203, + -0.02211342233799181, + -0.006914499854696649, + 0.003944579211731077, + 0.014982382662052799, + -0.010074381736244604, + -0.05426978323629371, + 0.03994601597712738, + 0.12130124093663468, + -0.00958067789821128, + 0.023480144649995028, + -0.08355712214050147, + 0.003454429399352426, + 0.007235493077764121, + 0.03816276322114522, + 0.06045932583359566, + 0.003167711502985745, + -0.0434786818393042, + 0.0010268432681905279, + -0.01148077041299634, + 0.04953327224563297, + -0.004618272758440258, + 0.0411346343138987, + -0.12181741549338237, + -0.10271030996588351, + -0.01197212629106024, + -0.013907870532020791, + -0.02354214630015897, + -0.007666871721100027, + 0.011838676180227846, + 0.02506664082277268, + -0.10797656288675099, + -0.1584126750281551, + 0.021898767834865558, + 0.1014195711980356, + -0.01937738580838196, + 0.0793707408335976, + 0.07457313196788654, + 0.021023818295429857, + 0.03688878936233987, + 0.014573402736027176, + 0.05440445646291787, + 0.04192703811428751, + 0.08333334090873756, + 0.08296944137201047, + 0.011869274554184966, + -0.0008891741971584936, + -0.007396897843012902, + 0.011860952473516656, + 0.19043862812093879, + -0.010087639754489763, + 0.05898952052441082, + 0.009589234594773245, + 0.011286648136085972, + -0.052546238607526094, + 0.028093316014552516, + -0.016333984982522778, + -0.08202877750357372, + 0.03476345818105731, + 0.006221341303489859, + -0.07270222357802165, + -0.020326190400005106, + 0.057711279920023624, + 0.016663650221891008, + -0.008727820418941667, + -0.03528356945531694, + 0.06548883119317297, + -0.004450400766918647, + 0.006071585217086368, + 0.002016625933013979, + 0.013968077918386218, + 0.011926411335416142, + 0.17235928416694962, + 0.006734032642043677, + -0.025565013266292732, + 0.02059742786521097, + -0.010713157401686406, + -0.026403591673915504, + 0.02432285118161758, + 0.016797070578884143, + 0.035010310139339446, + -0.09906372422116169, + 0.007566706927415849, + -0.040784778202638435, + -0.0295289652152469, + -0.07513468543509237, + -0.04684521805376673, + -0.10121403824881806, + 0.04811185636164507, + 0.02157406201528984, + -0.1656542965522033, + 0.15623780128854756, + -0.0032526784517657803, + 0.00944327953545381, + -0.0036663326794954564, + 0.009795337681794167, + -0.03089382955321717, + -0.00895904905596154, + 0.0334869674422098, + 0.013187148171676617, + 0.04035942601246851, + -0.004179067752278619, + -0.03790608527353525, + -0.08560919331122638, + 0.0044636160878887155, + -0.024434875434530306, + -0.021786208501578492, + 0.016629518357646415, + 0.028187005757467215, + -0.007782988970670979, + 0.0075747669119597145, + 0.037806104240039565, + -0.00504616919978247, + -0.02595388440000813, + -0.01210831919324255, + -0.05323726913392677, + 0.01677143933068766, + -0.008289174916225555, + -0.010098677688690523, + -0.004200589908606546, + 0.14266506460802153, + 0.03467952303946401, + -0.01710399571519861, + -0.004368748590551028, + -0.015978487228740446, + -0.0031236507277826496, + -0.01210813400053811, + -0.032292656249053454, + -0.037320436124681805, + -0.037805292902644845, + 0.02044748723709047, + 0.06926800514060015, + 0.20819633938028081, + 0.00175234366626547, + 0.007471503549942773, + -0.018242130439865972, + -0.03387703233404757, + 0.06982890296648124, + 0.03642999108696333, + -0.0427004386825584, + 0.010840010324379463, + -0.10864919815375042, + 0.0578912206112586, + -0.07050753574280876, + -0.06358074280987586, + -0.020274155678952042, + -0.0002992813113161203, + -0.012600172691855666, + -0.0623229922502781, + 0.008307551872869974, + -0.03580576763843659, + -0.03291679892423845, + -0.009249836236211059, + 0.01316874987846992, + -0.05632399248264649, + -0.055551009842097404, + 0.00914473676891947, + 0.017386348034467498, + 0.02630616006836141, + 0.0005232288986854532, + -0.08729021353073338, + 0.001957100290433851, + 0.03374299694860174, + -0.022941659478062704, + 0.01771581287983572, + 0.11154296008198795, + 0.004832330993948274, + 0.05366420823444158, + -0.00767034014383477, + 0.0024819435113446913, + -0.03138601954231129, + -0.007133040678005205, + 0.010199450095448917, + 0.009006994001675851, + 0.030987131111272777, + 0.02862342581128748, + 0.04369648951687521, + -0.05564457281644871, + -0.043502142369234935, + 0.010418662502605298, + -0.01896000125347518, + -0.05834625013567422, + -0.01906245177761252, + 0.07526157410551708, + -0.06366551546066794, + 0.047229350184373506, + -0.01687636564045528, + 0.09542818787567617, + 0.07426725601612225, + 0.0006451390505497877, + 0.03215987306456806, + 0.05000506141665795, + 0.08090051192471293, + -0.0007493032242597352, + -0.02486386323223287, + 0.00838920741314772, + 0.006775297430794362, + -0.0003934581508346889, + -0.027548129412495542, + 0.03408208638408395, + 0.01866341513596715, + 0.004978681366124246, + -0.023484681359127547, + 0.02264712098550524, + -0.04153339412132335, + -0.022766127163372188, + 0.08433687750937542, + 0.05719752823665939, + -0.09033877246202265, + 0.012091842936227663, + -0.011707714904066057, + -0.01030792970465594, + -0.000822834621010597, + -0.015728947748824235, + -0.00815786032688837, + -0.08807688284682474, + 0.0030677909906789793, + -0.00018377802200887126, + 0.03348856744797734, + -0.003087847620035145, + -0.005161855618800614, + -0.048875547325394356, + -0.01654701647772676, + 0.016258727386619796, + 0.06509870139572875, + 0.02474502910400294, + 0.06442512558813736, + -0.06733059476285257, + 0.02975754811383242, + -0.0676929593730183, + 0.0702656871268916, + 0.06990705439760007, + -0.004888158462159743, + -0.006664684322810412, + 0.03498962607075079, + -0.009179703088123915, + -0.008216621556210113, + -0.03554741397732838, + -0.007362622273681893, + -0.037499250354474, + -0.03989852863357956, + 0.05048344162318648, + -0.017458067690649275, + -0.10883037117879646, + 0.04672940279036778, + 0.08831231914254638, + 0.08169509491384126, + -0.046131410942112594, + 0.04355735221678086, + -0.041672913391911995, + -0.060556072809133846, + 0.010187872155311497, + -0.008198733067109136, + -0.04409070685618112, + 0.06554069721575576, + -0.1263271751731181, + 0.060324415602723475, + 0.053677328721317266, + 0.03554953480782808, + -0.01768243555183165, + -0.001624187034639991, + 0.025752371608284634, + 0.03424871592937598, + 0.06500784345087696, + -0.10972170532869238, + -0.011558195575098537, + -0.013460706815549277, + -0.06658040202252902, + 0.009110432530898417, + -0.030439194106372653, + 0.15744964488278249, + 0.07305179361457442, + 0.024489100736301794, + 0.004475771303857887, + -0.06189559022350518, + 0.03359955211923804, + 0.14445812010789286, + -0.002319051160800908, + 0.07446771416213992, + -0.04343028664445346, + 0.06561679486654726, + -0.02653092713984609, + -0.008503706088321893, + 0.09959643251990177, + 0.021052964917675478, + 0.054415193641945594, + 0.010650914801789885, + 0.08066290652929785, + 0.11726015015378893, + 0.031135657366856105, + -0.012129393667493863, + -0.05692195382658979, + 0.024331935249258238, + -0.04317313041488552, + -0.0008797611779747122, + 0.014036485640689012, + -0.005032120084124512, + -0.024062168708080645, + -0.028993861680280848, + 0.20983752151889168, + 0.01000240355090062, + -0.007824909251824242, + -0.035601339959814064, + -0.02614281498649709, + -0.019122217341417308, + 0.10461546080357655, + 0.03261648193310947, + -0.12592037297559308, + -0.03667153443005981, + 0.0671762954742247, + -0.01798371793969637, + -0.11437994172930524, + -0.008489521149803453, + 0.03103557072234169, + 0.05008301646876336, + 0.23995555472862193, + 0.09865711804394099, + 0.04687762874620855, + 0.04607160071124781, + 0.11456677620459185, + -0.007280078117875329, + 0.05426469858234697, + -0.05251675543229738, + -0.001752525994693901, + -0.00248084968105575, + -0.0511027466197612, + 0.004060795398949058, + 0.004497767119710275, + -0.0021763438216317985, + -0.09956263747319748, + 0.05969086940461289, + -0.020102144190722107, + 0.01225926578083607, + 0.08791489211036316, + 0.0022810714214485, + -0.018627344656412026, + -0.07433001537458393, + -0.03450084384767686, + -0.06145470357701117, + 0.02600129438204144, + -0.03722466512677332, + 0.02770200410433601, + 0.028885914559150083, + -0.014775585573025374, + -0.03577995594831348, + -0.12983654548083678, + -0.025318055771554986, + 0.029054008292930346, + 0.08767914835335713, + -0.03812433527866241, + 0.03833963023940205, + 0.04869681481605941, + 0.02990924710863083, + -0.0003405376392949652, + -0.0012313999138397127, + -0.005539552679571519, + 0.13509861038440105, + -0.10295149416037032, + 0.17937493862826, + -0.02891069513839572, + -0.007844671165804534, + 0.008397031763202317, + 0.21749361283381136, + 0.017424615326750267, + 0.033818438743951275, + -0.011027408752534744, + -0.0011252662393625218, + -0.00848936208031465, + -0.01104056093452122, + 0.0011368885649381266, + 0.01261710549589346, + -0.04694478006985998, + -0.028942449925948233, + -0.0022475945211896816, + -0.03778738243220315, + 0.0032434792446081272, + 0.003763751969798838, + -0.0027086525807035254, + -0.016462883179486826, + 0.015449845787936895, + -0.004891761609906213, + 0.018047987707868546, + -0.01672849773548078, + 0.012566425647921696, + -0.026495749980125684, + -0.00930236326736551, + -0.009062747608135208, + -0.037106227885365106, + -0.03128718187692695, + 0.010099949120765611, + -0.06212136465886868, + -0.001665935680901308, + -0.002440903837603327, + -0.03142309240368313, + -0.014548352426901017, + 0.005609268184759883, + -0.016756231599820058, + -0.0038169322011095265, + 0.01749009428502831, + -0.009963201735015902, + 0.0010458898075719504, + 0.03606569050673604, + -0.013147482762336111, + -0.013045816107138983, + -0.009850990745119683, + -0.027408904184813336, + 0.04758402129868893, + 0.0340313166712962, + 0.010103416192746599, + 0.03466111848914967, + 0.019710187424662812, + -0.029247306351035715, + -0.03324567952618719, + -0.02231576044113686, + -0.04868001381874836, + 0.003381465560957495, + -0.0882882975656664, + -0.0061043975137996965, + -0.10113238370530415, + 0.035044042661892566, + 0.03098170922425413, + 0.3168022220971996, + -0.003060815205939941, + 0.061394207889080774, + 0.08029926337457412, + -0.02220517926423977, + 0.004301994191315687, + -0.015772992349703873, + 0.06499692891478868, + 0.0386495723888399, + -0.01564217079463393, + -0.030647156563266352, + -0.00337046001315227, + 0.08840465173435541, + -0.06582676559583783, + 0.07455338838417228, + -0.05721906982905636, + -0.0037458584561390054, + 0.030384336298017896, + -0.05519363416395151, + -0.01261312722772055, + 0.034256474499707625, + 0.04559163621264971, + 0.005146141881341269, + 0.056082710009919066, + 0.042453826014614685, + 0.0008788530520471683, + -0.011868785369167006, + 0.01039373199937499, + 0.11622690709433436, + 0.008920339291935267, + -0.00011623829494705951, + 0.006351104577308368, + -0.0033582441738251316, + 0.005329864432122581, + -0.0458632195349269, + 0.009311361467922498, + -0.005676057235601064, + 0.016154013486719524, + -0.0183383752368275, + -0.007163148956307241, + 0.023665701536430103, + -0.005398023408605651, + -0.0036544604424938827, + -0.00469546257113368, + -0.10593456164256058, + -0.004627671466279243, + 0.02680939072456951, + -0.005109192311459868, + 0.07105913173661133, + 0.03089023273394121, + 0.014156708874904056, + -0.0056720202705807955, + 0.007860132516115021, + -0.07715730996814736, + 0.0009425455197648236, + -0.01341946469757934, + -0.011413706952180483, + 0.006452464327043369, + 0.008486684578019484, + 0.04570226733085654, + -0.02939811931450567, + -0.0033005093862148644, + -0.005833889653231259, + -0.026734953116237543, + -0.00180806335956061, + 0.010323098745322758, + 0.008415662145797228, + 0.0008069391245878467, + -0.07850778731698616, + 0.018049386906442928, + -0.0006428541855532708, + 0.00782294296643815, + -0.03360962845644275, + 0.1453154170439575, + -0.0011567778412150918, + -0.008279510667512278, + 0.005825734321243252, + 0.017132792218436332, + 0.007457363538776342, + 0.004932409295799645, + 0.02479155629336912, + -0.007767802664047107, + -0.03032557571346357, + -0.06895140882251284, + -0.019284265273196582, + 0.013275492780925234, + -0.020671157643695197, + 0.012377625867890903, + 0.011318302719158237, + 0.012350848083103579, + 0.01746674342798749, + -0.02622192771707338, + -0.0038361298347607095, + -0.01079090876881264, + 0.04155260030160244, + 0.017391944913023084, + -0.007917579571949523, + 0.014374474978036521, + 0.005707348979268843, + 0.008485027661992032, + 0.010629020827131522, + 0.03859564525560698, + 0.01867214635459831, + 0.055515309451042966, + -0.012020328444648594, + 0.03304487344713102, + -0.03408945078584483, + 0.005093217254816138, + 0.004894655968179055, + 0.033029514204205274, + -0.015929275594955113, + 0.002154740453576688, + 0.11206217170151422, + 0.029155295513059083, + -0.09703255633896846, + 0.008109444169435147, + 0.0041366380869817315, + 0.019031568723345896, + 0.004098720465679381, + -0.010329719842076241, + 0.03699192599279818, + 0.1336113298008907, + -0.008131261460330719, + -0.000822668694317997, + -0.03161812814958835, + 0.009896559964863142, + -0.020444761663486157, + -0.012393238272787589, + -0.0022798747383653358, + -0.016600148701793913, + -0.007350808592256582, + -0.01013745140168988, + 0.027657528961293663, + -0.012734515572948692, + -0.09052191977427516, + -0.006288462954130643, + -0.04747404068317294, + 0.016606020584400662, + 0.012523539651656193, + 0.033268670119050624, + -0.005708067326823296, + 0.001276842270911089, + 0.0033058968569555976, + 0.020924451731002224, + -0.04790917916483593, + 0.029103618224151917, + 0.024179740343410555, + -0.017630339803897135, + -0.007914641048568599, + 0.011366247318372394, + -0.002438296922114339, + -0.009468209221769285, + 0.010485101363404298, + 0.0005086058797110534, + -0.02203266672088143, + -0.006858214489181456, + -0.05442336923191849, + -0.0053823873746578635, + -0.02188670461805658, + -0.025329290001590003, + -0.004104808998477728, + -0.011173537622787399, + 0.00736330845921518, + -0.01214036928284869, + 0.08673128565935938, + 0.07132930462304911, + 0.0003504368387899231, + 0.062947401510466, + -0.015052384844122328, + 0.023870875971184202, + -0.002086083633191807, + 0.020303357978114955, + -0.01155157635899844, + 0.0005060035985987286, + 0.005378896535931288, + -0.057783295924640196, + 0.008666436635458338, + 0.008992002953789503, + 0.00400871534877186, + -0.01893609396812115, + 0.04628659114320212, + 0.009330984604510041, + 0.024252077800788606, + 0.004717226131818511, + -0.07677233068353094, + 0.0042037412269603095, + -0.0012659192737285636, + -0.0027505486481736513, + -0.0032201611935340703, + -0.008489725419755437, + -0.008218993935609235, + 0.026168495871675353, + 0.02248770259029353, + -0.0028702343821439508, + 0.02070375715034489, + -0.010344294906948845, + -0.019064228050390035, + -0.018353375705819967, + 0.011419695873085667, + 0.007026128848676389, + 0.014984044674650601, + 0.0002792797790874949, + 0.025415861868492984, + -0.041253602449563526, + 0.05409991358723948, + -0.0395271683184, + -0.003211294548432664, + -0.030539428500784494, + -0.011065242407091905, + 0.0033373803609486035, + 0.014041737696923929, + -0.04310065881434882, + -0.020178786141962942, + 0.08875019352745811, + -0.04497047858752332, + 0.031649182890940125, + -0.14293214576711433, + 0.09634693260551645, + 0.016248595127285704, + -0.016636505649274498, + -0.008920495073394596, + -0.004013067202622826, + -0.028939568516342905, + -0.025184870340876946, + -0.05516136095829036, + -0.047105858561132086, + 0.003255544698291661, + -0.014911056870648759, + 0.055339223564893604, + -0.03338055048633281, + -0.010821729839898014, + 0.037075239278418354, + -0.030364318405456434, + -0.004002202739361186, + 0.05689714661696101, + -0.0048003125260710185, + -0.0021197938671733505, + -0.004615338901063086, + 0.027109130201960364, + -0.04272926158456554, + -0.013429725374373844, + 0.03173265295797749, + 0.008564052572272687, + 0.008186868659154075, + -0.021148057592925946, + 0.0061724829261772945, + -0.003502368689194772, + -0.034337342271306075, + -0.013340080916451589, + -0.00528615349361893, + -0.04987206405617275, + -0.056979369516535316, + 0.00765974756312395, + 0.07357114585362858, + 0.004890693933160364, + 0.09917098877382599, + -0.0764809642908164, + 0.006317704833484147, + -0.004039683071615888, + -0.01660669210112752, + -0.004764927505040795, + 0.010584010406353388, + -0.0063558108283892755, + 0.02618107965289908, + 0.061006639624155465, + 0.03570848812894633, + 0.004906235589733848, + 0.021009057716663358, + 0.017945217410757237, + 0.0105498443359283, + 0.12493142340709432, + 0.003323498633506866, + 0.0028408397091207894, + 0.011275389092951598, + -0.03712973315387086, + -0.0037422476873991335, + -0.19160769515053547, + 0.03477442531674683, + 0.003925465081318135, + 0.016185503974168614, + 0.08202120230880008, + 0.055409870339151596, + 0.11637892379701532, + -0.00676253664775217, + 0.06342225787578658, + 0.008339110045779054, + 0.03260206536740945, + -0.02990086849685145, + 0.023344237040552447, + 0.00679743115419503, + -0.014886186275895637, + 0.0021510378868727794, + -0.026031794778264117, + -0.030554912208593566, + -0.00535105833671771, + -0.02007372581955451, + 0.040946517155550116, + 0.0011418819528214118, + -0.047189175799015055, + 0.04258820479353744, + -0.005856383263828668, + -0.021828533647664076, + -0.0050105803734486155, + -0.04626014509096166, + -0.03699849502901251, + 0.006640522606878781, + -0.11648456340227355, + 0.01638833655519093, + -0.013282073487568848, + -0.006244284393586261, + -0.0003293001354448259, + 0.00014961792706321426, + -0.004423402327671939, + -0.005054290999827006, + -0.026981024070672505, + -0.06118433757713928, + -0.011657054234503396, + 0.007146124730653984, + 0.004776242200557108, + -0.008337920228585044, + -0.018671246626698064, + 0.004221180187545364, + 0.009735570569143292, + -0.014406407989738995, + 0.024820243768110012, + 0.004065939629267591, + -0.03932855790983218, + -0.007274323007309241, + -0.01404875794450396, + 0.20491949945057175, + -0.05923597110998121, + -0.014013249585460389, + 0.007168169491853553, + -0.0187333028382754, + -0.01692685209016164, + 0.0018811720099772312, + -0.0008492683978308019, + 0.042740548930384176, + -0.04000490937815265, + 0.006695345924715636, + 0.0034800359951776185, + 0.04103888939392856, + 0.007602268228593055, + 0.016520734936213494, + 0.08866063920815799, + 0.03355894763890228, + -0.00936046090483523, + 0.025701340564417013, + -0.006943235381281598, + -0.0005714813935364517, + 0.014306922056054699, + -0.02214400250931349, + 0.012987243643420566, + -0.0024772009911396098, + -0.00382665581834441, + -0.010104624806444742, + -0.04118109884038479, + 0.06255386967289354, + 0.0016994919790084616, + 0.07251110016976689, + -0.16397978336934452, + 0.00673541164652488, + -0.0018249963788032069, + 0.03404774460264576, + 0.009742293317645827, + -0.007409875918815232, + 0.0067129616100115716, + -0.029446345107891545, + -0.044666973513304564, + -0.011682963793288889, + -0.043351049398400655, + 0.042951497541529125, + 0.0011326663941343769, + 0.013977491089459124, + 0.0051591416583362994, + 0.09250775322676007, + 0.01037048938038721, + 0.007415087897514869, + 0.06747130363511211, + 0.0744039181410255, + 0.004028903606279218, + -0.08300035631031698, + 0.009915514435378404, + 0.0766986716998723, + 0.0012420312250255793, + -0.014264967694538272, + 0.0014549772523772364, + 0.004240290021778356, + -0.01969722715326324, + 0.06014733348119845, + 0.013359165745338344, + -0.030021792200983673, + 0.06609412723107574, + -0.013577725974580622, + 0.01656980579437342, + -0.005416621084160895, + 4.666620816464985e-05, + 0.039072317620198437, + -0.05380558739768827, + 0.028250158141382365, + 0.010246073613921405, + 0.00961910617578478, + -0.01929538291582936, + -0.006582789663700562, + -0.004367576902808109, + 0.0059299114067849585, + -0.011145409190303322, + -0.006605181766248444, + -0.010951537889145074, + 0.018691744006808453, + -0.0013831562387030404, + -0.0033912527094629623, + 0.001432620871687277, + 0.004965254523508991, + 0.0043431608474984685, + 0.005391093730688203, + -0.0022839838031612405, + 0.002097515050805787, + 0.036215362080631655, + -0.003661265708452219, + -0.019719990619529167, + -0.12673710547573241, + 0.002252609833906675, + -0.009366905703808853, + 0.028082068479480013, + -0.006165771389473055, + -0.058145938501311605, + 0.0066773881944697175, + 0.03601346866091214, + -0.006690672946521734, + 0.02877978264513884, + -0.008744072342392613, + 0.05166666291034149, + -0.007727677430069986, + -0.011017675967179483, + -0.03302998678016515, + -0.028412499664934695, + -7.784430778041395e-06, + -0.01594549646466865, + 0.033453320099585133, + -0.00693504459024016, + -0.01727582746196611, + 0.07575193677556494, + 0.027684798769607697, + -0.009601864189469775, + 0.011173579482648166, + -0.026875846492715166, + 0.009191826919626686, + -0.043493546856528564, + -0.046219796530293566, + -0.02107595505710498, + -0.025467396573355592, + -0.008410109952492476, + -0.017080870988364805, + 0.0072852489520263035, + -0.025358772970117176, + -0.006260951888494383, + -0.007610695502503453, + 0.007401344466277547, + 0.030092762830979117, + 0.02055701779218166, + -0.008177612629939261, + -0.026615239333544833, + -0.016002733170343963, + 0.0802026473232225, + 0.05656205282406242, + 0.007307881624631041, + -0.020915171555586212, + 0.14116560980319648, + 0.012946673643908757, + 0.09739634516004196, + -0.08625454330877234, + -0.01606636726246228, + 0.003802689451572021, + 0.04118097657365934, + 0.0005879073040352784, + 0.03936841442548986, + 0.05462843904462125, + -0.003314692358511527, + -0.03699848368071327, + -0.021437040100171417, + -0.012873270531871237, + 0.006607957897341898, + -0.09458041473330543, + 0.04938329276489294, + -0.014828827099809332, + 0.005057710869924936, + 0.011602736759119272, + -0.010698654305283642, + 0.017459880630483916, + -0.039635487019547326, + -0.058730172968332904, + 0.020342449191187762, + 0.0028868701490908223, + -0.008818368374200872, + -0.045087570179923, + 0.03070358864532623, + -0.09506352189919343, + -0.10159636925070284, + 0.05097136222575383, + -0.0007193614433931485, + 0.01321383212324294, + -0.031484010126888944, + -0.021980359281979053, + 0.025502115356210612, + 0.0319629102859712, + -0.022882367918866584, + 0.01570161552933152, + -0.015590679363097463, + -0.021048127821341606, + -0.025646640165921095, + 0.018064692677650786, + 0.014579599432631617, + 0.0019393801459999737, + 0.0012775455898624848, + -0.03318736627561767, + -0.0041240855274948356, + -0.02799881973770163, + -0.032008141930856625, + -0.007419208236658135, + -0.012715264560132255, + 0.08155011545209517, + 0.023879362591190324, + 0.004621682268080129, + 0.0006729171086000824, + 0.001947573852735307, + 0.0023762323078296583, + -0.00022340328632540265, + 0.013401392221469842, + 0.017483630027235166, + 0.009736733707537179, + 0.0037239856614588772, + -0.0016562230000442173, + -0.008263373254958213, + -0.011323570213937496, + -0.013788750209587829, + 0.03464185773942037, + -0.01821932514590955, + 0.11117030570662252, + 0.0036507631743279323, + 0.0013171122761175553, + 0.004601411624749906, + -0.021955704081190817, + 0.0005070483135159839, + 0.013255751987194041, + 0.010451664102194927, + -0.01691432092243441, + 0.013533946454788006, + -0.006089917713565101, + -0.00912685744938048, + 0.014837199849971778, + -0.014828349569395959, + 0.008386436605230835, + -0.005913103985838628, + 0.006235858420766108, + -0.08275129620205833, + 0.04491934490598154, + -0.04163657716993833, + 0.00722029438119419, + 0.006763379822438879, + -0.012570843414782417, + -0.05907919471012853, + 0.024164164463679003, + 0.005414903566898486, + 0.00684543789912076, + -0.007605347481243869, + -0.026476409948914106, + -0.007098904793665978, + 0.011871636851910165, + -0.0010008773111054945, + -0.03348039874089914, + 0.12686608650732367, + 0.004593570803664504, + 0.03969857382992203, + 0.0049890080366864755, + 0.005747804676197603, + 0.04358208739345977, + -0.010559948174313999, + 0.04664997445070486, + -0.0007646776962505844, + -0.015042213520404774, + -0.0680764671522179, + -0.0003511195488933776, + 0.027204102885621237, + 0.012746185174039797, + 0.0059167227184961264, + -0.019689030616625153, + 0.05436926081852806, + 0.0027405780041446413, + 0.016627340839859034, + 0.006634794083880188, + -0.0026786207687254514, + 0.017638116475940047, + -0.006825271764802527, + -0.03552116217147041, + 0.010482626643298359, + 0.013937894641479594, + -0.0029499894612478655, + 0.0052988515589485335, + 0.007162049315969803, + -0.013465096413670769, + 0.0004055835748191862, + 0.005918117160316663, + -0.004767361637591459, + -0.006652078509940848, + 0.010297985713335735, + 0.03466377407339687, + 0.014289687427163009, + 0.01604687294582568, + 0.15419245843386306, + 0.0024984508064361866, + 0.021513487011997776, + -0.04949185470918199, + -0.03480805525705886, + 0.014737307357394102, + -0.0017778082364314821, + -0.005515518361284029, + 0.0001263838204815265, + 0.0015746326821325222, + 0.0030683149175348102, + 0.00912836485138345, + 0.11065974297096237, + 0.0382163469640719, + -0.006288616435917406, + 0.010922854978799168, + 0.021978735530624422, + 0.030982713934034064, + 0.09125276153704079, + -0.03419177913565874, + -0.0035468449090303735, + -0.0524710088561228, + -0.0313639727837455, + 0.016853994061963754, + 0.075929477319009, + 0.016694801684535365, + 0.005921036613844192, + -0.05419760602997928, + 0.050864718328907516, + 1.4866630428685097e-05, + -0.007327048139689712, + -0.013334174178390534, + -0.03080078419460417, + -0.0029578126046557783, + 0.06413014870707572, + -0.03524941392938923, + -0.02838643446082251, + 0.04933575780947013, + 0.0027759751837815004, + 0.025917839457845773, + -0.07784034390759427, + 0.02047395392153421, + 0.01608582203431795, + 0.009543926926792647, + 0.031184417419177494, + -0.01786208670737458, + -0.06928464622224725, + -0.0017058848228760804, + -0.09275801079198848, + -0.01829227480276132, + 0.01829707103904051, + 0.05537199467308835, + 0.005145877904866834, + -0.006755047168907059, + 0.013515339924258511, + 0.01798115830871492, + 0.005219090510185338, + -0.0039563169550464895, + -0.019145068484949394, + -0.002521153322093465, + -0.009258626574908767, + 0.0012218727898994951, + 0.07076054196614548, + -0.029407498159147644, + 0.027660601927762308, + -0.015194427776064041, + 0.017893721427707947, + -0.012690346248223912, + -0.025450628653195237, + 0.025637139295748294, + 0.0030350144349562272, + 0.03908127300442319, + 0.041369371093841534, + 0.0010232443007268893, + 0.023830282325938125, + -0.03340086288195211, + -0.1958634500607918, + -0.017937533099513014, + 0.00325766724041044, + 0.009036184230302145, + 0.0360977123698943, + 0.0039941451369887825, + -0.08389209065415226, + 0.04049981064400865, + 0.0243730088094281, + 0.004084590450405779, + -0.027769861289710065, + 0.006596211600604112, + -0.005482050268045051, + -0.10847913983623195, + 0.0059260271299462035, + 0.046469532920403046, + 0.00928966181128989, + 0.03334092119989887, + -0.009154389678609612, + -0.0324686722378556, + -0.1739850676623973, + 0.025463765639887625, + -0.016638248944679197, + -0.003837185636006361, + 0.0024149068411892286, + 0.021456988139691078, + -0.008519308735063708, + 0.018330897943391978, + -0.019742612783767093, + 0.03308904510344503, + -0.028519378611251395, + -0.000999605313644199, + -0.009368777254727732, + 0.046294891492455095, + 0.0018590150729137253, + -0.03964480527732062, + 0.031057058046912178, + -0.03830812820853417, + -0.0350469388475018, + -0.0014180010544426382, + 0.02191761788823747, + 0.015432391074973259, + 0.0055761168269359085, + 0.007810783778721288, + 0.009064801433716813, + -0.016144519154127266, + -0.007173177418821952, + 0.008500674374901457, + 0.01984658265860302, + 0.026451167743212066, + 0.009569824685493195, + 0.0028805281826782946, + -0.0017115392048084016, + 0.012623661749092178, + -0.02425868350460538, + 0.001708393131349829, + 0.03608837109698921, + 0.01854370604704259, + -0.04524294995648177, + 0.005475630571430087, + -0.017278597577779903, + -0.0012863132237692363, + 0.0044201593154407975, + -0.0014174281037447672, + -0.012674859885670318, + 0.03468500303204294, + 0.0510957115483576, + 0.004674068116249783, + 0.0035227586544058272, + -0.04935366251699038, + 0.0008942056793673027, + 0.00986196659505026, + -0.1405521972305515, + 0.013108344577993319, + 0.03321707262327872, + 0.010810494618049673, + -0.006908838714755528, + -0.007049531141418933, + 0.019028206919180068, + -0.0009479058589116499, + -0.004968729490805732, + 0.03532906323117435, + 0.0074054026407587336, + 0.015988193846677316, + 0.014313958128490581, + 0.010989200998490767, + 0.001226257808585977, + 0.0006465456027977014, + 0.03283834646682751, + 0.002560923954896963, + 0.0064989993375466565, + -0.0031353265623696182, + 0.009762531032033213, + 0.010081364060704243, + -0.03089572495871825, + 0.03763521992915897, + -0.04061077973777602, + 0.005954731858239315, + 0.025590546822067337, + 0.01222310841512301, + 0.01091196311010698, + -0.07803226055786394, + 0.004954314436504812, + -0.010922217154514288, + 0.03713606523586364, + -0.004279140223717743, + 0.037185550771938816, + -0.007956176788635614, + 0.01990141673777588, + -0.011591919719548807, + 0.0129690459609473, + -0.045796697387749934, + 0.056168246972859805, + 0.05773957836400108, + 0.040279825725282416, + -0.11600505181055039, + 0.039817798717943734, + 0.0014057253575713038, + -0.015301017782893877, + 0.012247389144350649, + 0.06750655525102228, + -0.0007460952771037302, + 0.005321002003581949, + 0.0093248287087852, + -0.013193374185453096, + 0.0011384649563520568, + -0.02958528054942878, + 0.015017496124824252, + 0.08297557778577487 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.0006473981255731555, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.005840731825184795, + -0.1737268554537944, + 0.061347162033433635, + 0.011862136334992298, + -0.029598588218308193, + 0.07068037122223843, + 0.008171603530320705, + -0.06918709569969698, + -0.10931096122327226, + 0.012652447258253886, + 0.00891034783193683, + 0.009423007496809091, + -0.00397517622295138, + -1.0974790819938784e-05, + -0.003998389973381107, + -0.010443678943621408, + -0.013078084495039785, + -0.017664267306874887, + 0.0090308045065693, + -0.028472312858081706, + 0.01230852342357901, + 0.026656996296269032, + 0.043437154309879385, + -0.140017876235705, + 0.01837876565832592, + 0.050157947922941135, + 0.03279777661048497, + 0.006746096807627687, + 0.0818018333503392, + 0.032378872822936215, + 0.016088160370667345, + 0.0007674339688438293, + -0.0047484927550932214, + -0.021922325193691833, + 0.06051810460407476, + 0.011644619380644285, + -0.00042654991483807213, + -0.015545993410831245, + -0.013843157463906137, + -0.025630480318167746, + 0.060054535004663664, + 0.02155597881316642, + -0.17987588860107134, + 0.011022399807182764, + -0.0018725963589902694, + 0.007782524042611099, + -0.07122377960642738, + -0.018948453107968798, + 0.008322344616397113, + 0.018357626343786654, + -0.008840312530891105, + -0.07709992750724914, + 0.13080946838567079, + -0.0007752247139485339, + -0.002765843445740616, + -0.011269989218410348, + -0.007645984420547694, + 0.031063418915934882, + -0.07931972620292158, + -0.08187800201130586, + -0.017777497469799383, + -0.009525557963907822, + -0.015272754141088725, + 0.07752219965385726, + -0.008160299925370554, + 0.009623572644599327, + 0.04979047175615179, + 0.05409857418723703, + 0.01648177007845472, + -0.03810435715175279, + 0.059205048087814134, + 0.01062525372065717, + 0.06766477381908104, + 0.010220156650168464, + 0.016623082146506228, + 0.007823890153287126, + -0.023754471897699175, + -0.03625859263658984, + 0.024809805056684505, + 0.017882438620715622, + -0.0005425817437069244, + 0.03690950574736948, + -0.07424050383655018, + -0.054807371127507706, + 0.0001319208518875941, + -0.020291543866705813, + 0.012810111316895082, + -0.0004675907762589185, + 0.031080572917375152, + 0.0027751226391266723, + 0.013586244670433194, + 0.011521531839857761, + -0.01037314939472594, + 0.03505404230516366, + -0.02766046565704243, + -0.08279370217381572, + 0.0553794583117373, + 0.012173830143172207, + -0.019478864013313613, + 0.005176934215329381, + -0.019795990783849, + 0.02104058487481733, + -0.00098210544366228, + -0.007521142774541187, + -0.09407276546683385, + -0.050487928725902464, + 0.01269545075057295, + -0.002151433614481569, + -0.045873612905347624, + 0.102448628924616, + -0.1293633961814192, + -0.040080929226000865, + 0.03554673263270133, + 0.07170330762398967, + -0.0651253982039706, + 0.01306633424033492, + 0.12515607121127853, + -0.020232005820850166, + -0.10266616639948783, + -0.0699917594925905, + -0.07440230834105618, + -0.10294459423136519, + 0.20844489292338408, + 0.026691206444869355, + -0.02576303636554636, + -0.02979747663689224, + -0.0068931041332229645, + 0.0639318821152962, + -0.02559729254494554, + 0.0211134244939147, + -0.04124540372194835, + -0.018109321964769005, + 0.02363432812458682, + -0.008847048015310163, + 0.03799242378364725, + -0.0055043039585034636, + -0.006450472531484535, + 0.013276005093407602, + -0.035100414719854, + 0.0005811342039753704, + 0.0003641198624127998, + -0.01944063613219066, + 0.020213876783124796, + 0.11366581002542468, + 0.0019813964413918695, + -0.05546380898983734, + 0.004803940357562276, + -0.03684210140669684, + -0.012137287251290053, + -0.12585934652913625, + 0.0025170857035359297, + 0.013136234010025342, + -0.006613317160694033, + -0.009781050365588356, + 0.006698064247257371, + 0.025915321416700575, + 0.0014561585344462454, + -0.01078857519433396, + 0.07620582027241561, + -0.026438639694926845, + 0.002509368612462139, + 0.02342075525868729, + -0.034600640711577854, + 0.0057402247384842784, + 0.0019691752075094624, + -0.03219207952989973, + -0.02256903300920726, + -0.010944594316748773, + 0.0029829582005419793, + -0.018188870009001868, + -0.012099819527175945, + 0.0035364458469264006, + -0.024874632078286732, + -0.03794189841941028, + 0.010287403049153558, + 0.011240028005401751, + 0.03860249127212227, + 0.013415092018259564, + -0.08815103396045636, + 0.008614105654386455, + -0.013714004164525865, + 0.0010177275354880074, + -0.00537856478872265, + 0.01165004871030711, + -0.00584184870048017, + -0.0017149301292154861, + 0.015676661701703305, + -0.009597502672752542, + 0.03710690787074681, + 0.005845795370424038, + 0.006405133236780898, + -0.00603192983150941, + -0.00877197081372381, + 0.01530763444713737, + 0.008529569508313278, + -0.0288378753065949, + 0.03779904227276576, + -0.013886594782234599, + 0.05632400154718736, + 0.07547740032313806, + 0.033381041104255556, + -0.0030029152981258835, + -0.027947897013980684, + 0.0004898952121060466, + -0.1415002689868419, + 0.010268914286968587, + 0.025731739405551606, + -0.02162079678869491, + 0.030643166445919544, + -0.010124162190406386, + 0.01803708283171844, + 0.03610156632876341, + 0.01113260714672411, + 0.018041572171427118, + -0.02727641214105501, + 0.01779884120233853, + -0.01294586529418544, + -0.04309546393012848, + 0.00182037737768188, + 0.0182948030300125, + -0.0679655630206758, + 0.07672466340637786, + -0.005245061357050354, + 0.0018567406994293599, + -0.010216364784694494, + -0.00230766335167652, + -0.04742367647431051, + -0.02880705576713808, + 0.008231576582095173, + -0.0013808995598119914, + -0.014821028079008645, + 0.002443265263562091, + -0.0494989609338981, + 0.025249887288862977, + 0.010340746695245021, + -0.007682551968221604, + -0.010619519740234194, + -0.005188629533307786, + -0.020319156364333183, + 0.03490797843547351, + 0.02341966121289673, + -0.036267415082429144, + 0.032363969714029064, + -0.0026709152830755993, + 0.0630185435145737, + 0.080295024763266, + 0.04869826677027735, + -0.07325892327889341, + -0.02798532222261154, + -0.05088787252478807, + -0.03370586867094544, + 0.03910483720337209, + 0.03900865366210339, + 0.02865055613381984, + -0.01272101779447848, + -0.012852121819513507, + 0.014527764299475807, + 0.0038290890354439654, + 0.0016962951194192505, + -0.01705848669699822, + 0.01180421983877476, + -0.024762000827116583, + 0.012015290920664737, + 0.03301945868805041, + -0.034877264779630955, + -0.010193065050902781, + 0.020973352401536218, + 0.009588110749078324, + 0.006662625310456605, + -0.011106550072712797, + -0.00803596384474602, + 0.01552825123873622, + 0.008176869537453757, + 0.02912898190768393, + -0.05810059859125237, + 0.00842519746821633, + 0.014688754825244905, + 0.003464320518014883, + -0.045115780962599675, + 0.0026699647442659127, + -0.001380910577475343, + 0.04980355612058871, + -0.017667715556747746, + 0.005086029454781073, + -0.0683367590298116, + -0.0007059650365704177, + 0.015543773137065354, + -0.017694827556040234, + -0.05018712477292458, + -0.01362789496814125, + 0.004751019269912532, + 0.008250568123843961, + 0.001807016027716527, + 0.02255956059162652, + -0.01980617334655508, + 0.026140007812695282, + 0.04265552437896264, + -0.03825087800381365, + 0.03168487645447078, + -0.00651527583786333, + -0.10488964368702462, + 0.15317691445334733, + -0.02006932605085781, + -0.022458909173894554, + -0.00846914826720357, + 0.031020916418805743, + 0.025677463491745985, + 0.15492921920919636, + 0.013542155915878797, + 0.008331557662259962, + -0.0020212641763253042, + -0.04382570372611285, + 0.014924805378541486, + 0.003736750136447758, + 0.02110928835965871, + -0.003434899504646202, + -0.07672421126729183, + 0.03182311596618037, + 0.018064468267640587, + 0.02511231240239908, + -0.05024418856542104, + 0.04082072569966037, + 0.01577221516813872, + -0.007850634256211686, + -0.05072579614576143, + -0.021548720996159114, + -0.008002664475304308, + -0.025039205921487177, + -0.05472766999327841, + 0.011305327148942598, + -0.015931030882510983, + -0.008646809863624368, + -0.013224278049121694, + 0.0976367302200424, + 0.06089291392500004, + -0.015064470839315206, + 0.012451375289736744, + 0.02888121453639633, + 0.001283286409936766, + 0.024676017533210554, + 0.03952817269940006, + 0.08274561322491868, + 0.011113684665720464, + -0.020359746887411705, + -0.0043352769350284365, + 0.021294078854422843, + 0.07824561982013799, + -0.007071118924922797, + 0.10108790662366827, + 0.11576734563442057, + 0.06595895206663158, + -0.055862994596561785, + -0.05258084567683067, + 0.07083982238717657, + 0.0001357055827620114, + 0.04786087532761603, + -0.07206562765523465, + -0.02554150643111586, + 0.10918292272373893, + 0.020620718225511016, + -0.051742352215021784, + -0.056387593547892405, + -0.03770790603102677, + -0.07520422982129303, + -0.07516859487240467, + -0.05803685391604377, + -0.005692127697843187, + 0.08645716661561159, + -0.013221406379930576, + -0.012028701613512717, + -0.02824865873148509, + -0.01060891515046309, + -0.10601238504545282, + 0.011862281222973179, + 0.008556154048758495, + 0.034621683523603994, + -0.002815574153587793, + -0.038321398136630797, + 0.0008463171195039735, + -0.0031795792149819802, + -0.003481654966692188, + 0.01795882371226126, + 0.06105259194766606, + 0.022201003009958042, + -0.001089004781715484, + 0.04343971219957147, + 0.04279932934679246, + 0.012509275337860648, + 0.024313143595921023, + 0.0010779270491984192, + 0.021442096208164365, + -0.009275038052658151, + -0.0016413390038060825, + -0.027837789682520035, + -0.02300910463112975, + 0.07508866329058429, + -0.03724878041808527, + -0.05363675306438362, + 0.003554929948253861, + 0.06244676274787712, + 0.0525961046802472, + 0.04563437217225647, + -0.013218533435542763, + 0.04451615503991878, + -0.008329675578400548, + 0.023422237578789633, + 0.013842603313928229, + -0.008474795219786021, + -0.015108571544484526, + 0.009372134371213273, + 0.041179451843246374, + -0.018332096168358285, + 0.017511325342571608, + 0.014555196548546746, + 0.0024630318121613964, + -0.045186858467019135, + 0.0001021303087621371, + 0.01091408853599182, + 0.00958308099700704, + -0.0028492320980073884, + 0.009075045462227463, + -0.026889797310928193, + 0.024337849670478483, + -0.003958360119607519, + -0.0017626111464521437, + -0.2110126362954674, + -0.02662164952250898, + -0.022935471852165518, + -0.007392524987799536, + 0.020598307495522467, + -0.025539693155859536, + -0.004322493503766473, + -0.10122675038650596, + 0.03793857383555638, + -0.002247447794175654, + -0.025726657500322835, + 0.018399474304596288, + 0.013733597944918994, + 0.053796875882504175, + 0.012784457672695782, + -0.007937486193307374, + 0.09635664681843809, + -0.026100871094372157, + 0.06598431130940614, + 0.01834350378810312, + -0.0037355301980286594, + 0.016520592719020784, + -0.05664189360925566, + 0.02496801772070251, + 0.02539808834050042, + 0.02204793913077675, + -0.02679819690505795, + 0.036937104900174156, + -0.011766172548677387, + 0.0804117825014761, + 0.0004974509222548881, + -0.006056895789110609, + -0.002706091972736995, + 0.0071346257367444965, + 0.051559110012512926, + -0.012272938912431792, + 0.05959940885820423, + 0.006527398577521723, + -0.008112869812584434, + -0.0014349914032666816, + -0.018239713283589993, + -0.004294378242534925, + 0.016862618839861124, + 0.026064536043599734, + 0.001457279694042929, + -0.004540866205755628, + 0.010797060786481462, + -0.00523513410905684, + -0.040423175683094656, + 0.007421147799173903, + 0.006296535560339199, + -0.009010983220383942, + 0.006040231599679927, + -0.023642385033212615, + -0.003234044973363196, + 0.0021806984273004234, + -0.006522470614631411, + 0.004255563531813815, + 0.008219963198753361, + 0.007352072884193249, + 0.0316932370572482, + 0.009484117526148932, + -0.04883152887237508, + 0.001947438570452912, + 0.044131860768051395, + -0.001628810150410327, + 0.05767281584257538, + 0.09218295929106024, + 0.007063634237936356, + 0.0835288242734788, + -0.05542398576955019, + -0.03526993456790506, + 0.043372114530015785, + -0.020675117536302873, + -0.0028315646194945923, + 0.01569626559364684, + -0.013238485279878608, + -0.0011736981516950983, + 0.06751645367901905, + 0.04313242292671671, + -0.07141539584340215, + -0.009840891461930653, + 0.0022894746529667047, + 0.052487978510903664, + 0.010953315819605507, + 0.09086523836116503, + 0.025146588841280793, + 0.01803069729398236, + -0.030923392575711084, + -0.012362641519715886, + -0.0015689667891063245, + 0.0065714849425912315, + 0.010747772243281987, + -0.03655393363446822, + 0.005768153112735357, + -0.006640746282917839, + -0.036077582001314584, + 0.09219959193342701, + 0.010567988727497256, + -0.03441218561580288, + -0.009034223957042087, + 0.0419294903810584, + 0.0726170594818551, + -0.007010720420951402, + -0.0015688181127138196, + 0.015567906374188827, + 0.021921831474703782, + -0.01319179108688728, + -0.0006023102317469585, + -0.00925142854459952, + -0.02112020487364849, + -0.005999598632457355, + -0.0034563426440187626, + -0.011258624763992356, + 0.0012750875698900358, + -0.07490856169081254, + -0.0009201963886464552, + 0.015085760179993819, + -0.004262459386528275, + 0.004440238469726828, + -0.0019879775954790616, + -0.027192993258142048, + -0.015277762780668952, + -0.22379574514220718, + -0.005762341556458254, + 0.008774902639548332, + 0.04906861408416153, + -0.0058038905635840205, + 0.0082812337927407, + 0.003576806523873236, + 0.013550661171697896, + 0.05696680564573766, + -0.0336516011737885, + 0.062613321835691, + -0.006551840143732046, + -0.007786340161420255, + 0.0024772902596219507, + -0.0008465692416998428, + 0.016312600121657004, + 0.019990900720058227, + 0.02685526459890591, + 0.00047367647046816315, + 0.010698658993328855, + -0.034747336058602835, + 0.039291516925361834, + 0.023171216070871994, + 0.03749702743883509, + -0.007279336748248209, + 0.008720299359451029, + 0.01075323990002371, + -0.022946706731045395, + 0.016713628343341242, + -0.013407323774187617, + 0.03522096327082713, + 0.006156588127860086, + -0.006470482440791969, + -0.04154697990853936, + 0.012535792429051543, + 0.006413615447558693, + -0.014835959014638755, + -0.002491620511254521, + 0.014956665535545983, + -0.035351546760171576, + -0.014906069224607247, + -0.07300945219782852, + 0.06868547672818781, + 0.0695356446693482, + -0.02171771634159402, + -0.03847290664878122, + 0.3118351517394946, + -0.02762545776911505, + -0.0037773010167072543, + -0.02082201602824065, + 0.0074126136218375755, + 0.011799856110612319, + 0.012145266552506818, + -0.000450403751834663, + -0.001766993051193298, + 0.025037291520320068, + -0.007541565942436076, + -0.0068286165243218525, + -0.022619385201532678, + -0.019381266856585765, + 0.0235886756603574, + 0.0234307801263229, + 0.012897925682763345, + 0.09054207859482388, + -0.07160137957878482, + 0.0005690872035358878, + 0.02614651866894757, + -0.005509051437378701, + 0.005432871152294835, + -0.004421808578036434, + -0.0019187416962464451, + 0.005440081870346291, + 0.013321960691794147, + 0.02989873151490042, + 0.03799322814547548, + 0.02569055420289016, + -0.00872955077017958, + 0.02543050430398923, + -0.03605684719113331, + -0.0973297856683198, + 0.03740103225211155, + -0.014878188835327339, + 0.02778309566002294, + 0.005194378235447434, + 0.0035028116276102734, + -0.004505597269954404, + -0.0038753972650597485, + -0.022995495382476312, + -0.010310870336998157, + 0.0026870244179896794, + 0.0013189348712512906, + 0.04959813428407917, + -0.01391616028191016, + -0.047278521624639684, + -0.014074237504009764, + 0.11412104244403044, + -0.0315716475273411, + -0.003602566184659866, + 0.004819063232193907, + 0.005338840659545288, + -0.010104190035442067, + -0.0074626179695794475, + -0.02338648198430201, + -0.011916948089302845, + -0.09321640480955806, + -0.08111477915190707, + -0.16494530767659757, + 0.009612108971156727, + -0.040624183153175675, + -0.007581949892200926, + -0.006669469752593108, + -0.0021438608276377505, + -0.0018669702556242018, + -0.010553773594993598, + 0.014022879396598383, + -0.0002631870223163681, + -0.01877480567693743, + 0.0076205593179208796, + -0.009054222607005238, + -0.026396466720441224, + -0.0038853795189068984, + 0.017448359275439522, + -0.09255359113563288, + 0.023899931477528273, + 0.004376668732992772, + 0.00555742119112802, + -0.007765726587576783, + -0.0033676904181250953, + 0.0032312606115758656, + -0.013445689199779894, + 0.028361207945520092, + 0.005437784674267766, + 0.01773532455843062, + -0.004588200213694857, + 9.493641522120231e-05, + 0.006048361807479647, + 0.020629694089078752, + 0.012894131472103788, + -0.06896963635807554, + -0.0509330957841666, + -0.1053434123856957, + 0.04061108636033375, + -0.046706058472522335, + 0.024373574142924976, + 0.003132812336602606, + -0.018045612493394654, + 0.015520360382660114, + 0.005705358833204071, + -0.04921270599930088, + 0.0029457746416955604, + -0.005406855153330236, + -0.007404739273660783, + 0.00964426306783111, + -0.009741674968261225, + -0.01903726422450847, + 0.009155923102921107, + 0.020994255396346806, + -0.016489463227708702, + -0.0021156571039261302, + 0.013080457315255013, + -0.030446145569983145, + -0.06155611818038935, + 0.08083239290872753, + 0.009021492880283857, + 0.0232399209848395, + -0.0013164275611083959, + 0.01012500443823473, + -0.004909571642219045, + -0.0027543294205633812, + -0.03107353988932352, + -0.004807866215518265, + 0.10123531771813932, + 0.06979332006802529, + 0.0006579101036847171, + -0.0014829323950326012, + -0.04304775871454964, + 0.01655718800702585, + 0.011256647691758397, + -0.04183117619715329, + 0.03989607923449993, + -0.09000682302300768, + -0.03571171773382284, + 0.12066266722393065, + 0.13419958328727247, + 0.09310117153440592, + 0.10314250318974633, + -0.09226689774242544, + -0.03226936224755769, + 0.0003251875331525368, + 0.009989757011561648, + -0.023796165171518287, + -0.010053599972468817, + -0.007127587776351647, + 0.0054969521426185665, + 0.02120095154449306, + 0.04235034468825289, + -0.011329281483923048, + -0.06143536189722525, + 0.0077533380458782546, + 0.013028566276477483, + 0.026533765655367143, + 0.09061773121729477, + 0.03819987024653939, + 0.04060609868902014, + -0.0346741471269833, + -0.05036232080092004, + 0.004522601655728859, + 0.016012540116587697, + -0.05557947287013586, + -0.005691865272302183, + 0.02066632725115205, + 0.07978290317145434, + 0.010749377113063396, + -0.0037605714397654934, + -0.019682790236613356, + -0.01750389542867219, + -0.08840013063093496, + -0.09341826841814856, + 0.024215640926227428, + 0.009868178790477891, + -0.026009484810047615, + -0.016178277488082843, + -0.009878121514640768, + -0.07175478297008205, + -0.003831003518794317, + -0.01863526257184426, + -0.02511324063506045, + 0.020413232147518764, + 0.006817714182988461, + -0.013034119459627883, + -0.022853339846545147, + 0.03364715509440663, + -0.04133139381022885, + 0.05573022824407324, + 0.016292681346124993, + 0.013958307708044296, + -0.009106289631685286, + 0.0052009337446087805, + 0.003196783824710953, + -0.03409998719671471, + -0.05764864144028029, + -0.048189452629338304, + -0.02627710230147921, + -0.04054500803484709, + -0.025401333962035543, + -0.0015397909781576017, + 0.04453261738581236, + 0.005547979798651636, + 0.005649974302688528, + -0.010294214586623423, + 0.03705160052944086, + -0.034040123723112055, + -0.02823236165500284, + -0.03654856388797196, + 0.038683060994109245, + -0.057609570123554756, + 0.013233756793726872, + 0.036896366029339496, + -0.08490688681108555, + 0.01203280765360993, + 0.027599761267741323, + 0.02712315612238023, + -0.012634809573969818, + -0.03795734304988689, + -0.0100294641914689, + 0.05394467538428582, + -0.014258417968953722, + 0.03102502365352268, + 0.011163500559132357, + 0.11855163373941409, + 0.01580559471864909, + 0.09891694953405655, + -0.021838945609364172, + 0.08923911642397939, + 0.018309068389168553, + 0.027234260675417787, + 0.016671021210984183, + 0.013303960738270178, + 0.020595890360242643, + -0.002723047565963503, + 0.06218415023973966, + 0.05050886800374143, + 0.0374925716995533, + 0.012785686878825272, + -0.05331108836595363, + -0.005580897193622581, + -0.011767783087348948, + 0.011618664382286733, + 0.01533091590855229, + 0.040843340173901205, + -0.00214065153220437, + 0.028828400045450352, + 0.057653087201464025, + -0.0015204295958419567, + 0.0030483901833782026, + -0.01938709613537287, + -0.011178025958657798, + 0.01167395767806602, + 0.042371136932590515, + -0.003382359452700465, + -0.003270986338866526, + 0.004227554376059858, + -0.002439108906868871, + -0.024635871016762942, + -0.025521511014291672, + 0.04842603227627366, + -0.05786633855932832, + 0.028221058045518957, + 0.010064831074318718, + -0.016587207227404893, + -0.02766270651621954, + 0.02066262402260346, + -0.034364778578612816, + -0.02688232581816887, + -0.0016233107104888025, + -0.006525587224034064, + -0.002276365422771777, + -0.0077622048847796965, + 0.02164380897378288, + -0.010333302761884367, + -0.1060269799002352, + -0.16699629680323683, + -0.23618705457481884, + 0.009462886831971537, + -0.0008480818997942082, + 0.0134496042569966, + -0.015683373617184446, + 0.009127417187024723, + 0.009647419198630963, + -0.05684795378797804, + 0.0068314938157961755, + -0.017138733547140256, + 0.038885123906711526, + -0.020382660443435428, + 0.008204331551714529, + 0.01168518604729156, + 0.014426320383880573, + 0.27681033105945824, + 0.1083106459349685, + 0.17771251582518802, + -5.0354864780693045e-05, + -0.004207762400210057, + -0.05252472037636007, + -0.026306425224143427, + 0.024257997200937124, + -0.027850275952592755, + 0.051804692922189345, + 0.06431960021586554, + -0.01480284172540199, + 0.001610691536857524, + -0.025751153703784058, + -0.04491750102006762, + 6.427955480217669e-05, + 0.007428549162007398, + -0.06173160887686304, + -0.2255771074035816, + 0.003003632666553226, + -0.16467323247749474, + 0.1012573080293696, + -0.045351798978524226, + 0.0015407195943503873, + -0.004847088848968608, + 0.021774976453320827, + -0.0041531798841325665, + 0.012320893635132085, + -0.0013753394719655798, + 0.004334345476256272, + 0.0064215928697872494, + -0.1883511191839441, + -0.00939554910640176, + 0.11319612872915254, + 0.1657107318642522, + 0.11838182739501371, + 0.003288687732045489, + -0.0521757459509922, + 0.15063755893606473, + -0.10505385128325677, + 0.008860147807277476, + -0.0006873968094398789, + -0.024634870299265223, + 0.00658884947430197, + 0.08229142332010694, + -0.0506275528729429, + -0.01603305913797668, + 0.06595960049612094, + 0.019421847442941502, + -0.009233611214323829, + -0.01880049561073924, + -0.00846769360203667, + 0.052807143744959865, + 0.024949269992330932, + -0.0042283658250565235, + 0.03459943032442967, + 0.013973695323800472, + -0.06072224073730545, + -0.12233605058875681, + 0.034331269052741656, + -0.002326697891925133, + -0.07822386916655637, + -0.03305593658923672, + 0.0013986916963114404, + -0.06051926183357169, + -0.09115434604734811, + -0.027115075276688085, + -0.015629734839146373, + -0.05667650841378262, + 0.10172266146129404, + 0.009709578694078139, + 0.00469090792624175, + -0.018465859529235445, + 0.02121190121217327, + 0.00044119902454180953, + -0.0035511109773424007, + -0.028504133430852015, + -0.009165768518565681, + -0.0016668446039544492, + -0.03779708052450676, + 0.03498813085166629, + 0.05641313635602071, + 0.02904792192247014, + 0.002132142656893825, + 0.01799370298984939, + 0.005231013533047167, + 0.024792398680333196, + -0.005378549408485394, + 0.03138344058871791, + 0.0097744756161583, + 0.0006968371246060103, + -0.013244523236849262, + 0.03473281397318444, + -0.06039264125847467, + 0.08299590741111283, + 0.0173945064601899, + -0.08368160548753853, + -0.02645802091986158, + -0.02682404290602596, + -0.02758803974945111, + -0.05994145116226475, + -0.0218842467057962, + -0.0021994013313456767, + 0.0011478614563803053, + 0.006365380709877195, + 0.017337004929242613, + -0.002298771535287835, + -0.0013584674738605907, + 0.008350034041103757, + -0.008925731692622067, + -0.0016195320665487475, + -0.015203357580354382, + -0.03088055345808092, + 0.042255767544596916, + -0.0004081938472802589, + 0.024644231196438778, + 0.11593526419679327, + 0.02204172059241651, + 0.002064271603802536, + -0.003206840140880006, + 0.041304273648573864, + 0.026619783227072445, + 0.03941209626844069, + 0.014134671955047952, + -0.010048180653949943, + -0.015394493658997064, + 0.042330827143000294, + 0.01743929010450256, + 0.05974631926762418, + -0.013208534913180072, + -0.00439760533419415, + -0.031212590153393827, + -0.029934218937118704, + 0.012170947511755572, + -0.05165240155627228, + -0.09027027625502597, + -0.037514402355771616, + -0.011921221163778483, + 0.06481457664708323, + -0.02287478611474895, + -0.005390969819821308, + 0.0075905686036947815, + -0.01493336832007771, + -0.0024307123943666667, + -0.05126638667796887, + -0.0030144932122166774, + -0.009877519814570312, + 0.1121684508891536, + -0.014191447550611542, + 0.009833044987139295, + 0.015576781814639085, + 0.0020947184735690284, + 0.01860277012000895, + -0.0030951136055682445, + 0.015076810582175745, + -0.09184335798497792, + -0.010648080328708485, + 0.005994565602031932, + -0.031371863283356595, + 0.02358047697460353, + -0.007921467296706659, + 0.06358980021883753, + -0.025266704540082573, + 0.05930900983371045, + -0.006563774794674804, + -0.03973079870887307, + 0.0009911962772738178, + -0.009709463440739443, + -0.0351346466875219, + 0.00012676802038769674, + -0.02795656975578312, + 0.008736113845901228, + -0.03147302554917009, + 0.04936419172478635, + -0.02467382985611943, + 0.03139276724768367, + -0.045642106484750596, + -0.10302564362062622, + -0.04092107408061473, + -0.01173503275637971, + -0.022059457211763014, + -0.0012678542413532567, + -0.0034805035941225517, + -0.004997957599545, + 0.030306071505670197, + -0.051244085599729684, + -0.006784402403916307, + -0.0487269757128264, + -0.0020900857595353307, + -0.002915080256696721, + -0.007412050858368195, + 0.0733194997779931, + -0.012455173653767745, + 0.0932158763314371, + -0.047824866354531266, + -0.04211578676469633, + 0.04954597147886757, + 0.018733630865548263, + 0.008686669229388981, + -0.04175916475677617, + 0.019781116129324105, + -0.022192531147115408, + 0.008072650847264672, + -0.005311193276046314, + -0.006096739843267404, + -0.0022647642774542786, + 0.014890820773021769, + -0.07456777777290777, + -0.031974768155648384, + 0.0014246829905922482, + 0.0608563134793802, + 0.009926998540895848, + -0.012321676264595655, + 0.016748509312303955, + 0.0005548908416643846, + 0.07207618414682619, + 0.026056659047669038, + 0.03450977357836057, + -0.005724817602027528, + 0.011779802068050743, + -0.06309809582336738, + 0.01647119951946234, + 0.00786591631946932, + 0.002080484158249768, + -0.009408060559612847, + 0.0840009754199815, + 0.022789488391633212, + -0.0067501964091491165, + -0.026329644540062386, + 0.10952518787260598, + 0.02780205295149016 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 6.691052854160055e-05, + "valid_tracks": 4, + "per_track_pearson": [ + 5.510308005996432e-05, + 0.0003008837997827056, + -8.53419510711466e-05, + -3.0028146051211645e-06 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": -0.0009817962771498862, + "valid_tracks": 626, + "per_track_pearson": [ + 0.011802452822855719, + -0.012108202066191608, + -0.01151229425064781, + 0.010897279649569129, + 0.013912084676257824, + -0.01439004443068802, + -0.008535146979760981, + -0.012290254808855072, + -0.013038980589240311, + -0.01062214064402039, + -0.014247293048065095, + 0.009475172417474795, + 0.013835554020221368, + -0.015314914064516068, + -0.014067758783611484, + -0.014201380013490724, + 0.010535429860192624, + -0.013970037251769556, + -0.01160498633833813, + -0.014701113006060784, + 0.012063725025005937, + -0.013569215850541183, + -0.005210835680956282, + -0.012450581058458511, + -0.009529139062364422, + -0.009027351270318404, + -0.014283974038800959, + -0.010736737930682125, + -0.01317533374050629, + -0.013533661437635108, + -0.013273025368435509, + -0.01431162271022395, + 0.012565582766345973, + 0.0037823000870914124, + 0.012407848182886752, + 0.011314115754553043, + -0.006792470719130332, + -0.009518329148835353, + -0.008985664831666493, + 0.013716091532094944, + -0.014358887681278132, + -0.004531900159885125, + 0.0106917175089717, + 0.010249854489059547, + -0.01408677565757169, + 0.013374208630914916, + 0.012378448991344796, + -0.010911931197866738, + 0.013712442335033595, + 0.009460451580902205, + 0.013689816195660915, + -0.013190844521806721, + -0.008561183313936349, + -0.003968608545233517, + 0.012614695116271205, + 0.013828256092884393, + -0.01335003258472749, + 0.013582579924969957, + -0.013816922906029644, + 0.007478614087000115, + 0.014009377028096383, + -0.009838876360442934, + -0.009226390980596352, + -0.012525887675705515, + -0.013962501728893555, + -0.012221333414187832, + 0.014420383839783182, + 0.011939855885074856, + 0.0032545996612843636, + 0.012982556868733982, + 0.012281838408507616, + -0.014134605867567577, + 0.01434266302343081, + 0.011119582784678585, + -0.00869373639954281, + -0.009000042803784842, + 0.012156649375833406, + -0.013511319167863443, + -0.011285302809148332, + -0.014753024167386119, + 0.011828123240894344, + -0.013027061854429488, + 0.014207049648208722, + 0.013755115433293835, + 0.012364199349297375, + -0.01476872943469646, + -0.0027926722024865878, + 0.01246132608853468, + 0.014462167769451294, + -0.007903294649969255, + -0.013720987001642001, + -0.011309313729309689, + 0.012541603845872018, + 0.008364223081163075, + 0.01085871503965786, + 0.006349495432479432, + -0.013830745679062426, + -0.014296445752458337, + -0.012949178217249042, + -0.014096108770802645, + -0.01315029225028011, + -0.013987748012499067, + 0.009997901408205027, + 0.007783372405691697, + 0.010378299106459698, + -0.011533073945118846, + 0.013389967492221877, + -0.0086653886833264, + -0.01141478137324906, + -0.011918040572683536, + -0.013429500578717191, + -0.012026591514895272, + -0.011552960917530051, + 0.013823329413710206, + 0.011511687169213984, + -0.01393871082971102, + 0.0122722991861677, + -0.013198294994253179, + -0.013661270105691906, + 0.008946586162137343, + 0.010979973602389058, + -0.014034360327759008, + -0.00899378373429166, + -0.004952648256883944, + -0.011779805050570458, + null, + -0.012066731955650977, + -0.00581144569922944, + -0.009635501009819447, + 0.008943462993559345, + -0.012899965898785157, + -0.009088446289077613, + -0.010576271708675072, + 0.0065186578521974885, + 0.010322191434292768, + 0.01090662244650362, + null, + 0.01140273760374512, + 0.012126928113117474, + 0.010274671251149678, + -0.008584015642514306, + -0.006244252675245298, + 0.013026514009803473, + 0.013048092914507776, + 0.010179791293260675, + 0.012867422166676305, + 0.012603258482701538, + 0.006950268991764882, + -0.008913125191438375, + -0.013174859323219082, + 0.007960419243274158, + -0.0060654917174133805, + -0.012449160930217558, + 0.01057866210814017, + 0.005583944448854207, + 0.009378703544828367, + -0.010145762313481941, + 0.00532310880687504, + -0.01392123572207987, + 0.010321297053800284, + 0.010899303149056244, + -0.012060995038873302, + 0.013888496635250084, + 0.01346936089877733, + -0.012950078409625945, + 0.009637175668538422, + -0.011453537907401994, + -0.010566677216295043, + -0.0027143621112382258, + -0.012524769909075334, + 0.007135712340683758, + -0.013875791536604794, + -0.006509720630189059, + 0.010958710211944877, + -0.012948380005834233, + -0.0036949560939148144, + 0.0007705836499329098, + -0.0065022509577167525, + 0.009363342616838756, + 0.007581683927554849, + 0.0135221649148155, + 0.013019869090569425, + 0.010993236432010587, + -0.011462413672083418, + -0.008377169705918704, + -0.007238557527897918, + 0.00989802151946277, + -0.011685218749949525, + -0.010642999489076327, + -0.003177975433130012, + 0.013872590160706688, + 0.009925805210286662, + 0.011130905290807167, + 0.01419670706708923, + -0.014357537720962104, + -0.01188343219322931, + -0.014313678496380548, + 0.011934171047089008, + 0.010554614789245867, + -0.014189635607676769, + null, + 0.013846644964945386, + -0.013206455576322574, + 0.014310906599470437, + -0.015124456032101448, + -0.012924462917637028, + 0.00924216264351324, + -0.007919867261728148, + null, + null, + 0.005290495540979593, + -0.014625552815339418, + null, + -0.01441984584393485, + -0.013451450947276481, + 0.013896034880458088, + -0.01245564898269167, + 0.009525600241028671, + 0.011421867777600786, + -0.012354099442259745, + null, + 0.015408601729869636, + 0.013686263754132797, + null, + -0.012467773764179098, + null, + -0.013659030909648864, + 0.01234258652012512, + 0.014045232013797766, + -0.009605301953698027, + null, + -0.012174738454775432, + 0.011577083050357359, + -0.014679772098485233, + null, + -0.014018653928158566, + null, + -0.015321159570783255, + null, + -0.013593265942047307, + 0.013828596199902283, + null, + -0.01496589741306781, + 0.014965883835916597, + null, + null, + 0.01332187940305893, + 0.015099489543497545, + 0.013953433068653976, + 0.015579608068426423, + null, + -0.014527298301934485, + null, + 0.014958253443945765, + 0.00989021520513557, + null, + -0.011127307170579142, + -0.014275801639566675, + -0.014402202049384713, + null, + -0.014850146070836956, + 0.013026466055883663, + null, + 0.014884838681035645, + 0.015212173588998207, + 0.015190309528726124, + null, + null, + null, + null, + 0.010881912643076691, + null, + null, + -0.010439439665145885, + -0.008681725676466872, + -0.015091783202496366, + null, + 0.010374962215108528, + -0.014992502764789473, + null, + null, + 0.015731366485523302, + -0.012460884642991523, + -0.015073735794574801, + 0.015052578142176072, + -0.015423359096656422, + -0.004713888315073984, + -0.006279626019174607, + 0.011595648488951615, + 0.014266049799569084, + 0.014584536281251755, + -0.014917330199638064, + 0.010592279805464422, + null, + -0.015133015444784532, + -0.014918488531598738, + 0.007963403939418912, + -0.011805837114408444, + 0.013734630339525034, + 0.0142650278514438, + -0.014521239099163644, + 0.013270141896682172, + -0.01195884120114952, + 0.013708551629075593, + 0.014343796267268621, + -0.0034999174855110432, + null, + -0.0066709739163237506, + -0.01297317832199318, + -0.014712433244513294, + null, + -0.011522729007839677, + -0.012277290240958375, + null, + -0.013625039399686787, + null, + 0.006850434180931008, + -0.013857950386720436, + -0.015608324789865869, + -0.013895872334142259, + -0.011135177826338997, + -0.013721417062018028, + 0.009532568042264183, + null, + null, + 0.00034850603683351817, + 0.014319088047987597, + null, + -0.015343193956370033, + null, + -0.011250491185328348, + 0.015477955896996708, + -0.014872350544852203, + -0.01378829282042031, + null, + null, + null, + null, + -0.013844817700475806, + null, + 0.014201895992319611, + -0.012221463651881844, + null, + null, + 0.014891291110692994, + -0.005113852500992758, + 0.0009143323240224532, + null, + 0.01461953059206658, + -0.01316401237622183, + 0.00952552305611452, + null, + -0.012795943700742633, + null, + null, + 0.015434589373844931, + null, + -0.008555551031822858, + null, + null, + -0.013845764804109394, + -0.01250579699619895, + -0.011299870155296071, + 0.01544337978823887, + 0.015185262606570674, + null, + -0.010801189734626404, + -0.0014081695386949278, + 0.010170748058214137, + -0.010159263724369972, + 0.01017485467611971, + 0.004024363040724158, + 0.007671397584744845, + -0.010746339108002882, + -0.005309658019371856, + -0.0081105112395911, + -0.0043383840889091845, + -0.00037215883104171064, + 0.004304811308954538, + -0.004674238909474427, + 0.006665433180878398, + -0.010793654310803556, + 0.006626337024655561, + -0.005793344464299152, + -0.006192727179675716, + 0.00663035271823732, + -0.004366147838145122, + -0.004410166782966186, + 0.00933377782479108, + -0.0068240561175217105, + -0.010592541245229622, + -0.0016288009317258927, + -0.006876025252893777, + 0.0014757778695522726, + 0.0016397179748788894, + -0.009468811674781356, + -0.007514696404131544, + 0.008871146807470061, + 0.008474705147327665, + -0.011503132775774031, + 0.0017186974409634198, + -0.005267345160375514, + -0.00722411417828557, + -0.010397615744506762, + -0.0008703486108718352, + -0.007617015950948592, + -0.0027442065445310243, + 0.003731347862003103, + 0.007694193632905275, + -0.003949039013028466, + 0.0052075458396053645, + -0.003351447138779778, + 0.00703350163071847, + -0.006429310738260295, + 0.00915812977953868, + -0.006855373340931601, + -0.0069914616451645486, + -0.006172739329510172, + -0.0076921143673329626, + -0.005646588308609131, + 0.005480727796598037, + -0.007287668781075393, + 0.008602943263099566, + 0.007883568745657241, + -0.0004212345913587278, + -0.0059005038309288835, + -0.009176056532873534, + 0.0016791266894334547, + -0.00255810077837647, + 0.0022526194931384983, + 0.005489038906996785, + -0.009420461752591262, + -0.008022559086740691, + 0.007200004828297993, + -0.008864040524188953, + -0.003964045208575094, + -0.012862358878612088, + -0.00046426820761937096, + -0.008823108520286742, + 0.009040052497322986, + 0.008780815201805212, + 0.006305498476201856, + -0.006692126072074799, + -0.006309789357188526, + -0.0077074151786071, + -0.007712331999007935, + 0.008268977369278117, + 0.0038356484653256145, + -0.007414895935714051, + 0.0028622521007912838, + 0.005835613463945028, + 0.004379922479294871, + -0.007846513834377873, + 0.008224963124690239, + 0.007755763489253898, + -0.006527064006576679, + 0.006693327127737366, + -0.007521248830163417, + -0.007781396833567705, + 0.006622290404351831, + 0.006480522991906542, + -0.007038990092298333, + -0.008925854715309163, + 0.008797957136674923, + 0.001041350570575357, + -0.0026848739372274445, + 9.088781990446089e-05, + -0.0064501904106703915, + 0.005038980217817558, + -0.004418044869985732, + 0.007389046663751872, + -0.006912995618643174, + 0.006339427845054683, + 0.004835356290763166, + -0.006644059339595112, + -0.00829005055134223, + -0.008211146024040064, + -0.005091018947298831, + -0.007279454560068937, + -0.01063756059528894, + -0.008238630968992857, + -0.008320824977443351, + -0.0062810341435579, + 0.006118905488737654, + -0.0016736471995814938, + 0.00676036686694873, + -0.0062681872691140485, + -0.008482739952783437, + 0.0014208490120616434, + -0.007754760683988525, + 0.009951390428768305, + 0.0037969478833300182, + null, + -0.009468781925788898, + 0.003367567032901477, + -0.006955678441781522, + 0.001347183275764809, + -0.007768955458435039, + 0.006623518823480263, + 0.009411710924009867, + -0.007347108633802, + -0.0003001275225625154, + -0.005258374433498596, + null, + 0.006452177567078205, + 0.005157252029092265, + 0.011080548233386757, + 0.005750689315187936, + 0.00461859683487107, + -0.005259577748780859, + -0.006988058405340167, + 0.004424906019412606, + -0.007358142413853671, + 0.00519551242955263, + -0.004987041354045545, + 0.011606272221682853, + 0.001057385888871273, + 0.004620823261813187, + 0.001215207987948644, + -0.007917368652921766, + -0.014280181925727171, + 0.01417989192981238, + 0.013281640349697121, + 0.014109863122238336, + -0.005556074162159939, + 0.007365922024448537, + -0.006903460199610956, + -0.007483951285768692, + 0.006126594173178926, + 0.003701187286201655, + -0.0025481977857664787, + -0.006718948458651686, + -0.008956235210568798, + 0.001925439790147831, + -0.006093610895024475, + -0.004960579119524452, + -0.00443831048344594, + 0.005676794574816532, + 0.0010480343931529741, + 0.004617308918049308, + -0.006693829166410865, + 0.008882334121717706, + -0.003316321747981575, + 0.0010045231836504059, + -0.0006072730357340979, + 0.006491410521201141, + -0.008961660473181816, + 0.006488590376427751, + -0.0049645098467804615, + -0.010628156736931241, + -0.0070931305661898775, + -0.006785820110556255, + 0.009145028921296856, + -0.0019164962253727731, + -0.001646904589526016, + -0.008703167016171555, + 0.0073098852752749455, + 0.011770942385112745, + -0.005754049049814147, + -0.00028989646327286597, + -0.00800075726339346, + -0.00830155557264369, + -0.008718156366382571, + 0.0018990544039713406, + 0.0049307296229463395, + -0.009124847215345304, + 0.0037594056234180415, + null, + 0.009260775304874744, + -0.005291259595293873, + -0.008666844522159057, + 0.006564529074347591, + -0.0034786630058038816, + -0.001347566195514156, + -0.009562897332054817, + null, + null, + -0.005183033960293, + -0.005180417931488262, + null, + -0.011037427740425008, + -0.006958620951079077, + -0.00666247723629573, + 0.0040975588337830455, + 0.004739204184377045, + -0.009727048532510458, + 0.0020521355957077796, + null, + -0.002284547035514302, + 0.0027558178003503275, + null, + 0.012128850639681131, + null, + -0.005998379648437383, + 0.008618703106213635, + -0.008947485545452973, + 0.008805788459618503, + null, + 0.011967727363329286, + -0.0029580971745276022, + 0.0020779044982257096, + null, + 0.011229738386642343, + null, + -0.009991509223340689, + null, + -0.0066225011512049715, + 0.007632461163194826, + null, + -0.010253866352402599, + -0.0019150461700192516, + null, + null, + 0.008911326843556347, + -0.011650756156487847, + -0.009319481242596876, + 0.012104930896559707, + null, + -0.008200662137949273, + null, + -0.010872847118784532, + -0.010590854957099032, + null, + -0.009760650111798061, + 0.008940885807078745, + -0.006052723608195451, + null, + 0.01065957691872012, + 0.00987712966620771, + null, + -0.006855033935889628, + 0.006547076888406563, + -0.003215082419217991, + null, + null, + null, + null, + -0.002462441209439598, + null, + null, + 0.0036790956951125034, + -0.008089321674298567, + -0.008250314851026374, + null, + -0.0018110207303044979, + 0.0027300600189332034, + null, + null, + 0.004669931068012891, + -0.002774878861998738, + 0.004371552981446182, + -0.010283329719815297, + 0.0025929709447452785, + -0.005766969266834917, + -0.008247977608686596, + -0.0023935207808230587, + 0.014274599399462823, + 0.004375242495980159, + 0.010948971859067702, + 0.006779000599095946, + null, + -0.00868487430504005, + 0.009994203979748942, + 0.008017783123065756, + 4.091767592743407e-06, + 0.006608837543088041, + -6.69400947642287e-05, + 0.0019085706937027045, + 0.0017378308988327143, + -0.002290497948155164, + -0.003639199244760259, + 0.004525564381506737, + 0.0049618267610282134, + null, + -0.010682048017943865, + -0.0033092778867561587, + 0.0018193000768348122, + null, + 0.009935902448031768, + 0.0026460736982552996, + null, + 0.004886018191178686, + null, + -0.009678903222323116, + 0.010235319019042469, + 0.011819406629872777, + -0.007495597607282102, + -0.002912078517124871, + 0.003506501241588347, + 0.01065437346137215, + null, + null, + -0.00974267302757311, + -0.004228699469913919, + null, + -0.0089071268410548, + null, + 0.011729210861503246, + -0.010496340484336397, + 0.006246690117491831, + -0.0027372737762838156, + null, + null, + null, + null, + -0.004073912693980205, + null, + 0.011130513220489746, + -0.002492655396952012, + null, + null, + -0.0053902419708013885, + 0.016381217003387384, + 0.003634636167928362, + null, + -0.00624236885246699, + -0.0051009221129020565, + 0.011433784278040964, + null, + 0.005897881036412052, + null, + null, + 0.00410658550235403, + null, + 0.008669702637493695, + null, + null, + -0.007171697184321225, + -0.00128782162074219, + -0.003630735939290236, + -0.003614891135770355, + -0.002758904860872575, + null, + -0.009265591160590881 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934, + 0.9989429973574934 + ] + }, + "junctions": { + "mean_pearson": 0.0, + "valid_tracks": 626, + "per_track_pearson": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + null, + null, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + null, + null, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + null, + 0.0, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + null, + null, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + null, + null, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + null, + 0.0, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "contact_maps": { + "mean_pearson": -0.00536736727089396, + "valid_tracks": 28, + "per_track_pearson": [ + 0.02389599908072645, + -0.0011864255370531613, + 0.04160082080688651, + -0.00177750708988016, + -0.00784560935054002, + -0.02394990888568218, + -0.06364372390452092, + 0.045185405369988496, + -0.005348991575943296, + 0.025289526301312155, + 0.005888644270057398, + -0.001202508588453693, + 0.049922554098134043, + 0.02457612477129726, + -0.023453947832631585, + -0.03729921173806724, + 0.012249600281251504, + -0.034595713949826655, + -0.01904664485102044, + -0.021814581103723453, + -0.0033709499403530266, + 0.00805430316631191, + -0.007341477905616975, + -0.017913655351214784, + -0.0324298794270259, + -0.027017263761972707, + 0.048657267693981786, + -0.1063685286314522 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.0041405290631839195, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..07d982e384245fae3d41ee635e4541d169a4fa79 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S001/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4941389147398904, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6542302506588221, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.04857778001776154, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.8433797414542621, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.1946354697971151, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.17777922359375958, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.23303249002260132, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.21772140965743444, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.1636819584913644, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.20768655997770985, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.4132730815910868, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.20918637590093828, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.2001169301893302, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/README.md new file mode 100644 index 0000000000000000000000000000000000000000..2c0308999e98778b8cdec00a4c75fb7c98cf0b4e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/README.md @@ -0,0 +1,70 @@ +# SeqFunc local01 — base-resolution residual path + +A single 13,256,456-parameter model reads 1,048,576 bp and predicts all eleven output families. +Its global encoder and six 384-wide transformer blocks operate on 1,024 tokens. A parallel width-32 +convolutional path keeps base resolution, with a 21-bp stem and four residual depthwise blocks at +dilations 1, 4, 16 and 64 (701-bp receptive field). Base logits augment splice predictions; average +and maximum pooling supply aligned 128-bp residual features to dense, usage and junction heads. +Contacts use the shared global trunk. Sparse sites and junctions come from model scores alone. + +All weights originated from local random initialization with seed 17. The parent model trained +for 2,000 updates using a balanced splice objective; the new branch was attached with zero residual +outputs and trained for the next 2,000 updates. Parent parameters and optimizer moments were +preserved exactly at attachment. This checkpoint has seen 4,000 distinct eligible training intervals +(9.75% of the first epoch); complete training coverage is still in progress. All 41,020 intervals +are included in each deterministic epoch permutation. Data are AlphaGenome FOLD_0 train folds 2–7, +with the supplied quantized-target decoding. No pretrained parameters, test labels or test-derived +caches were used. Development labels are used only to evaluate and select models. + +Training uses one H100 80 GB, at most 22 CPUs, eight loader workers, BF16, AdamW, a four-epoch +2e-4-to-2e-5 cosine schedule with 500-step warmup, weight decay 1e-4, and gradient clipping at 5. +The splice loss balances positives and negatives per type; the other family losses follow the +reference. A full-length mechanical benchmark measured about 0.09 seconds per steady training +step and 3.7 GiB peak allocated GPU memory. End-to-end training runs can share this same GPU with +validation inference of a frozen earlier checkpoint; no second GPU is used. + +Both entry points read only public manifests and the pinned reference DNA. The complete Python, +PyTorch and CUDA-library environment is vendored in runtime.tar.gz and is extracted temporarily +offline. The host provides Linux, bash/tar/gzip and the NVIDIA driver. Model weights, source code, +checksums, and complete validation reports are included. Perturbation values use float32; track +values use float16. Both entry points use the same parameters and model functions. + +All 200 validation intervals and all 2,697 development perturbations are scored with the verifier's +unmodified tools; results and published anchors are appended below. This early checkpoint remains +undertrained, has a short local receptive field, and retains a low-capacity contact/junction model. +The matched control continues the same parent on the same additional training intervals without +the local path; it is tracked separately in the research ledger. + +## Measured validation results + +Family-mean Pearson: **0.171715** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.313440 | +| dnase | 0.488170 | +| procap | 0.267572 | +| cage | 0.130995 | +| rna_seq | 0.149987 | +| chip_tf | 0.254508 | +| chip_histone | 0.223019 | +| splice_sites | 0.030428 | +| splice_usage | 0.023365 | +| junctions | 0.004990 | +| contact_maps | 0.002394 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4399 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6804 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.1099 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.6716 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.1962 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2155 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2710 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.1522 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.3468 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2619 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4727 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2523 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1482 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..4de65e82d62cb1e1d03f2a78a0bf5917a7cfc766 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/package_manifest.json @@ -0,0 +1,26 @@ +{ + "parameters": 13256456, + "source_checkpoint_sha256": "2258257f7d405240c62d9f9faafba2c7f87da26041302f387dad790c2f123209", + "files": { + "README.md": "9d2854394e54a769b519e4e60aec257d83439eed1ed0f584f3b58999cbfb5905", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "eb52b09e500dc3089b840c0ac471e4b449090889799d73e0741f61b787536756", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "2725592456c55c188b22c376f0cb249e9f46b4ae1370d2226df7143da358a8b2", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "2871dc8893879c3e14a526dea659701d72c8c4d0e2f835b0bfb857034a2c2e6d", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "2547c9adbdb51ad1acf796a548724cc7770eba90a755f4cb3b37d6f591adf6ec", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "validation_tracks.json": "4cbc1f3246fd93e871c3b198e14d976fa3bca2e1c616160f606a5176d7e4a51e", + "validation_variants.json": "edc97df92a3954c3c57bdc9ed3354b57d5f03aa92057cdb7fcb28e3c89d9209c" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..52f3decd4fe2e9d275d08606b74575fafb701434 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/predict.py @@ -0,0 +1,130 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..241fdd4144c4c917109104271fbe80c603075fbf --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,534 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"] + halves = track_halves[tracks] + queries = torch.zeros((len(tracks), 3), dtype=torch.int64, device=device) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = torch.ones(len(tracks), dtype=torch.bool, device=device) + position_rows = positions.detach().cpu().numpy() + halves_cpu = halves.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return queries.unsqueeze(0), valid, missing + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..23eebf243286ed9d6660220227afb74fc6763990 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.3134396216722404, + "valid_tracks": 167, + "per_track_pearson": [ + 0.5045350901319029, + 0.5326929166313537, + 0.31157864254497103, + 0.4041006146467703, + 0.30027958460719645, + 0.28031197228501403, + 0.2696824466102366, + 0.3405379682442458, + 0.3751103486549035, + 0.42026824207152186, + 0.3851627111386787, + 0.2875855459493913, + 0.1914198544163493, + 0.325496246692045, + 0.44612208572893836, + 0.18515135930992485, + 0.2622014115482546, + 0.2021051938016909, + 0.2545978394166643, + 0.21834591135793208, + 0.22098341277936334, + 0.21109367978164653, + 0.20580126173589086, + 0.25903487460572794, + 0.15359579936896103, + 0.19473608041257626, + 0.17733440749777402, + 0.23205309655581158, + 0.22344963243971616, + 0.17277211904820436, + 0.16900431456084378, + 0.0979095686184638, + 0.17641707086712935, + 0.1037391574796751, + 0.09837180340524986, + 0.25551146279274567, + 0.17786915986040785, + 0.2723757244750177, + 0.1688897417498533, + 0.14000899342675116, + 0.23086620657195692, + 0.1741251916597719, + 0.1627222636902773, + 0.12785847288446675, + 0.31119414527265615, + 0.17757315806569293, + 0.18301032090793878, + 0.1935353604227165, + 0.17126207390881662, + 0.19339578401496976, + 0.2075499247470102, + 0.1376079355449552, + 0.12318650873342561, + 0.33604187255283646, + 0.170061616568071, + 0.4831784734510651, + 0.6580986655671702, + 0.34016470209132027, + 0.4674729727931069, + 0.3892203228091323, + 0.5598255147990482, + 0.30385392758061647, + 0.5498569125070064, + 0.40855221168159056, + 0.5069867455188477, + 0.2242178066518295, + 0.3888147441667532, + 0.4455006208659812, + 0.6394561534303196, + 0.5646974017711864, + 0.3161190140067193, + 0.20043137570403777, + 0.12969974746988686, + 0.22594490699652411, + 0.2623884179505378, + 0.16355570544542594, + 0.26830915597327987, + 0.2676284527841207, + 0.264648731014886, + 0.19256866672353734, + 0.211932473528763, + 0.2888543014218945, + 0.23436613784900703, + 0.205371504592, + 0.20619777565524494, + 0.20867920024097736, + 0.13671995543536647, + 0.17475483395802602, + 0.19807818649475428, + 0.21271862153889365, + 0.2040505489376804, + 0.2010180069467041, + 0.241188353283315, + 0.24360823112571334, + 0.3180242800976669, + 0.11781504289182071, + 0.1690413143598215, + 0.17312398861380676, + 0.20609746174424198, + 0.20700875189503684, + 0.1736681907067968, + 0.2147985487868197, + 0.22157179169401117, + 0.20746706770821965, + 0.1876587718726419, + 0.1962068179817945, + 0.26324459415014895, + 0.22773961154534167, + 0.19532329980943394, + 0.1626401322006259, + 0.23563348027023512, + 0.23358185525494599, + 0.21907353016047276, + 0.18321902112620667, + 0.21943958464880053, + 0.1515561879574205, + 0.17579857852902087, + 0.22722976669736802, + 0.17313777234981942, + 0.1956506874250577, + 0.45231270179197575, + 0.460435851783062, + 0.5563516361643159, + 0.43775921307929966, + 0.4650655960118591, + 0.34806980630778483, + 0.6643117793891666, + 0.43049368878769495, + 0.4875466018902838, + 0.49318907922687427, + 0.5499293068079265, + 0.3864981232865663, + 0.41638865667875563, + 0.3736490621623661, + 0.5361461557006119, + 0.5043997339345597, + 0.5085847638269505, + 0.34494929011550585, + 0.26956877414186214, + 0.5915720246846831, + 0.5802700711778355, + 0.5892113856072527, + 0.5900992063238923, + 0.4396129603670507, + 0.6219515022838781, + 0.45475112693031977, + 0.41196825427596667, + 0.5157587746388596, + 0.6435793145771089, + 0.5491854839168254, + 0.3449596414338582, + 0.40978744125398575, + 0.342516686947317, + 0.4796999155624358, + 0.468554038539967, + 0.35606671541543883, + 0.5539811660535093, + 0.4298207562955967, + 0.3996744988439331, + 0.37730759855623724, + 0.25856434428923625, + 0.4715023126290336, + 0.5763794902390353, + 0.48023088427597677, + 0.37776644725845493, + 0.3273956537862697, + 0.46389150797755363 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.4881702179130813, + "valid_tracks": 305, + "per_track_pearson": [ + 0.3376103323580481, + 0.8801701345944525, + 0.7142122848428154, + 0.2866997947229506, + 0.21295328678392778, + 0.2441854703450732, + 0.8833999257625005, + 0.530628790495039, + 0.21972188091650086, + 0.22508983012348582, + 0.8163248880895634, + 0.5096894449660722, + 0.22814693352450882, + 0.25500103322810586, + 0.3087822480034644, + 0.7297491198723961, + 0.7323140079714009, + 0.8125287040613811, + 0.8071823923488869, + 0.8443909945060006, + 0.43785173932996524, + 0.3557379522346118, + 0.857708503834845, + 0.32439447491502427, + 0.3575819759950697, + 0.7338915985699705, + 0.37346034315023546, + 0.5381586871269092, + 0.38608847343943664, + 0.21815186770983203, + 0.4173426618416259, + 0.23188330501309773, + 0.8073221228498382, + 0.5977095345464783, + 0.8468542715341293, + 0.8539833274262697, + 0.355496844320634, + 0.7889448806181973, + 0.8456467898164565, + 0.8235508289934563, + 0.7875894649419061, + 0.6410307875719491, + 0.6687170761924057, + 0.46104297479911566, + 0.31469652150543653, + 0.603928549533276, + 0.6966908844056595, + 0.32286242305682317, + 0.40754019422556836, + 0.2663014964407721, + 0.1621247606474232, + 0.3910089535803324, + 0.24773044619032747, + 0.7984227937452057, + 0.2651426579250883, + 0.2481767782951468, + 0.2552161580478411, + 0.23175648020429016, + 0.24734721522172123, + 0.3820773702376061, + 0.2498284640272051, + 0.45032960884481454, + 0.2756530482564647, + 0.28783410473031584, + 0.27816780350416265, + 0.28893349931721296, + 0.35505831724709136, + 0.23307073143735932, + 0.2403131860910929, + 0.21684119576662517, + 0.3675257193613631, + 0.43038940289864036, + 0.2251389223299747, + 0.31709920691443755, + 0.17524400860875156, + 0.7694806116825665, + 0.8559088441016705, + 0.47506515453828596, + 0.26667714094031925, + 0.2666430927256703, + 0.20039587799490627, + 0.19231704518211692, + 0.3677854403519792, + 0.2847251199590152, + 0.3430244861451218, + 0.2841859634130694, + 0.3545311294133929, + 0.3232521252213592, + 0.49264262885064286, + 0.20381616775708333, + 0.4322683540171525, + 0.2583025836882163, + 0.6694441687312975, + 0.27277082107786704, + 0.6490977828165242, + 0.28791733232087285, + 0.4497261385325771, + 0.4051655334725461, + 0.18941979418791535, + 0.19321368342246675, + 0.28081034565387436, + 0.21979906763222334, + 0.21004783325370147, + 0.46739471359843204, + 0.3510557380760912, + 0.24789864436161965, + 0.2162419129339311, + 0.3363025165943207, + 0.8064864370751468, + 0.3553124994459673, + 0.24959917022862904, + 0.3360663153636207, + 0.2757460377448972, + 0.29512183384162616, + 0.7304032974979554, + 0.23607563340929102, + 0.31798862113460213, + 0.4012734489641476, + 0.48844232355645534, + 0.21297116451826234, + 0.2683540310106748, + 0.28092326281684316, + 0.2965511302138408, + 0.2723756596876592, + 0.34588803530613976, + 0.4002080732872202, + 0.3600333874797637, + 0.5057035386031795, + 0.6477008222449481, + 0.3351514025472826, + 0.2680496811834203, + 0.4466767469068688, + 0.32745920223820757, + 0.7182879059149568, + 0.2955645025639889, + 0.16694703615339324, + 0.42639684842013237, + 0.3004947405046331, + 0.4514607872679884, + 0.43580385687607764, + 0.40091125911347886, + 0.3941187750782602, + 0.2916037036966637, + 0.7205117059571453, + 0.4143075359541708, + 0.7175985484211111, + 0.20049741299875384, + 0.20089546435983294, + 0.20301173719368165, + 0.22931468351979703, + 0.8589961064203138, + 0.14319244328051636, + 0.8409561878672119, + 0.5053567637334409, + 0.30255677006152665, + 0.15849225671385306, + 0.20598014449687926, + 0.2636488545711264, + 0.4725406576237378, + 0.4539640974721358, + 0.42957417503371303, + 0.44074893982316815, + 0.3974787042638755, + 0.3238808370818829, + 0.586896897872976, + 0.40750471740574606, + 0.21945854808158805, + 0.36676728977317, + 0.3129978736756235, + 0.2970338727728756, + 0.23601318729803586, + 0.7933466562840152, + 0.2384373161883415, + 0.20963690199440443, + 0.42949475569171813, + 0.6082808101233054, + 0.3043457134664245, + 0.11489947750278101, + 0.8461952244470823, + 0.21768839475428325, + 0.6981070229142543, + 0.2214267208080999, + 0.4279677115852942, + 0.46358209141568346, + 0.3242717337150262, + 0.26978575738463606, + 0.3295727824346457, + 0.7808324126287312, + 0.314382973035291, + 0.45014834496909745, + 0.24898307799170358, + 0.4899372938110555, + 0.8577102431190883, + 0.2444187053086582, + 0.4414637859423947, + 0.27582891931547415, + 0.40567353875575374, + 0.43540387168033473, + 0.4686228994213537, + 0.23700319200961018, + 0.8329879846849123, + 0.8285239372481202, + 0.7092653407912551, + 0.7869252893227912, + 0.7999465741033157, + 0.7893455016925334, + 0.8045453760024563, + 0.19534688920530294, + 0.1901396182757391, + 0.3065420897800468, + 0.31611522633292705, + 0.2548825117204278, + 0.3767362615026195, + 0.8049582287761291, + 0.8142732418818026, + 0.446269859163781, + 0.7871475643367781, + 0.7741508080792838, + 0.81017960769012, + 0.8562346181078689, + 0.2551633592721882, + 0.2583427926658179, + 0.5049352823453939, + 0.29277756284211937, + 0.7630332548400894, + 0.2543334641230637, + 0.8698173345464357, + 0.8450380924752939, + 0.81865060755121, + 0.8613667297540414, + 0.8520137702958263, + 0.8279441945557936, + 0.7901977603045547, + 0.8160841630205545, + 0.8536371972183172, + 0.22283643289096308, + 0.8865479189157458, + 0.8303697687247105, + 0.7787110149406706, + 0.40099931488761176, + 0.8291783098711745, + 0.330555245219035, + 0.7113012510402978, + 0.8180725818588499, + 0.30852916148615905, + 0.27244241304665856, + 0.40622527269025555, + 0.3364421589583226, + 0.25810243580033576, + 0.5930195423167831, + 0.8422859094707489, + 0.4033549720048683, + 0.7632339257076742, + 0.45735502468157024, + 0.8210803971677846, + 0.7646882040427299, + 0.8012906287464853, + 0.864745910603474, + 0.7888244522182661, + 0.8495509840225602, + 0.8147382508152201, + 0.28991901167705847, + 0.8315640501805873, + 0.23710891250179786, + 0.31603755708430664, + 0.8244258467757466, + 0.8558356015140925, + 0.8287524771519945, + 0.26030980382515756, + 0.3760986681323619, + 0.5221090339415091, + 0.7538407135200125, + 0.8456015211284004, + 0.42493489231816894, + 0.8522765868263998, + 0.8653022540968192, + 0.44260391871316423, + 0.17759119536088402, + 0.365727724158363, + 0.8555692018336708, + 0.2116450126982466, + 0.4778498124207535, + 0.842256498264301, + 0.7916246003763248, + 0.31758003863509465, + 0.8116330129811421, + 0.850142081854328, + 0.8410431232926059, + 0.8343433931056625, + 0.8293482455178733, + 0.856592853769736, + 0.8145538058343287, + 0.8738743725978946, + 0.7359149586830565, + 0.7945003072609417, + 0.8485139921369242, + 0.3015799676784371, + 0.29640514221193737, + 0.2854396247208527, + 0.2540674443032498, + 0.8644273506321826, + 0.7409242308002697, + 0.7964963896793599, + 0.7390944162885157, + 0.7156466465098618 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.26757215760493364, + "valid_tracks": 12, + "per_track_pearson": [ + 0.26551292649618136, + 0.3097764631633246, + 0.22835037931866317, + 0.2988738560115892, + 0.29666862028958324, + 0.299106761110599, + 0.2345297300353553, + 0.2860799904362394, + 0.2052343366010947, + 0.26493776804988467, + 0.2498163097213669, + 0.27197875002532146 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.13099503352621464, + "valid_tracks": 546, + "per_track_pearson": [ + 0.1230323435494062, + 0.1578243662718878, + 0.1430953128968205, + 0.06016994515507459, + 0.0387395795983475, + 0.12861802870370956, + 0.05108870630661899, + 0.08555742163434597, + 0.14219510333792137, + 0.11403007713906921, + 0.15850264808499145, + 0.061492766598279396, + 0.0876751314185277, + 0.03598264470100404, + 0.047252331092992074, + 0.07498457093984234, + 0.13281033300003253, + 0.16644030592758033, + 0.10553400623861336, + 0.1497353162982746, + 0.1296968008704812, + 0.12789944613850535, + 0.21819955564271495, + 0.08738629372319898, + 0.11666839276070255, + 0.1444963311540077, + 0.14013372539456145, + 0.16288496475955466, + 0.17672951603032905, + 0.17731701543687856, + 0.05904276531784681, + 0.1593644225955877, + 0.0664548802558186, + 0.0694778020582731, + 0.16386870498748438, + 0.16592033493517422, + 0.15702774777836764, + 0.1467711944561051, + 0.14874031556496276, + 0.1976139800402008, + 0.21248643987143043, + 0.1719345371936645, + 0.1623636943551676, + 0.16352321945998946, + 0.06806498194336799, + 0.13694638497646877, + 0.1404981151522526, + 0.13351010064723287, + 0.030869243688379722, + 0.06333369486495748, + 0.1328301549361875, + 0.1718558866402488, + 0.1882913480947577, + 0.11663765762318283, + 0.20190066111428046, + 0.21973854082462818, + 0.10080907166413072, + 0.20379232206523049, + 0.14197605069501348, + 0.15078253870259437, + 0.1958277622599124, + 0.11013973013141608, + 0.17299082583665476, + 0.1244424966361152, + 0.18315303313450554, + 0.1948341957446335, + 0.2042866849911785, + 0.07290798653610367, + 0.07132354285023494, + 0.05876826752369955, + 0.16165543273526423, + 0.17511537755593265, + 0.14805361615641424, + 0.1530837304953303, + 0.1812493203732808, + 0.13549148587466736, + 0.17154163215451676, + 0.15961954563083722, + 0.17812857273462926, + 0.18023277594095014, + 0.14606315434313988, + 0.058171295928235715, + 0.16190445724676888, + 0.15163047674319505, + 0.08455910815644485, + 0.17256079253407597, + 0.13164919376591033, + 0.1874552614233116, + 0.1136339837436957, + 0.1787134582864973, + 0.12223719433405872, + 0.15670713082527496, + 0.1670805464674545, + 0.05992621736472348, + 0.17015253181159964, + 0.1328771199207501, + 0.14833702359937206, + 0.17524884710490826, + 0.14828487195555945, + 0.0622369894838038, + 0.13954873368136717, + 0.16833685452219774, + 0.15132012515749363, + 0.23020015426501803, + 0.19078601239647153, + 0.2116375745233821, + 0.12838364685528517, + 0.17786698250336433, + 0.12127907009776982, + 0.13154544253407566, + 0.13480557071173493, + 0.14466459869832937, + 0.16445025358277918, + 0.16552589284076927, + 0.12084821584632427, + 0.15113466172746146, + 0.1715968426378312, + 0.17216017322828947, + 0.1562966873206244, + 0.1608950926751809, + 0.1862912248975222, + 0.18438485582784708, + 0.10459999665833768, + 0.10280662903749065, + 0.14447701872511476, + 0.1740586276900174, + 0.15817057086061426, + 0.18576436041795089, + 0.2155193372549819, + 0.16329458795029717, + 0.19680107920668582, + 0.1361107767895018, + 0.16848457284215201, + 0.16230119064386467, + 0.21946227672083934, + 0.14517333297269952, + 0.2036752755909474, + 0.11649952814681343, + 0.050807325355354094, + 0.14773798089919496, + 0.19117184205513438, + 0.1689212368917266, + 0.17786226419457715, + 0.121208082586409, + 0.12439235745721407, + 0.14863690219810666, + 0.14954676888206028, + 0.19791882842329275, + 0.1249760029552698, + 0.1251173400662258, + 0.16202075853311512, + 0.09581817185545914, + 0.15202578763936772, + 0.16974842579146737, + 0.11635632249107536, + 0.08725365191578394, + 0.17261764646364325, + 0.1982031602208889, + 0.1324387380521249, + 0.06734541967356272, + 0.08695148147825012, + 0.14691560156118968, + 0.12853509355807785, + 0.15394637360506214, + 0.1315735435228633, + 0.0997479018513507, + 0.14703030039474632, + 0.09751786173738779, + 0.04830559715554881, + 0.1719334435562778, + 0.1350853046678005, + 0.11368408196219489, + 0.1257176059033061, + 0.22432377418027225, + 0.15618529177875287, + 0.04857633100022557, + 0.07946861592407849, + 0.12547811545667883, + 0.1496269724514148, + 0.19333340803027863, + 0.19157566832295392, + 0.18110494354810855, + 0.06845028713314288, + 0.1875039254336694, + 0.1528132391119109, + 0.1133785995218735, + 0.12072010235399115, + 0.08959074065082528, + 0.18597818271772992, + 0.16758481599219419, + 0.12551674951753347, + 0.06569892861858888, + 0.06844271337722121, + 0.14436906509271052, + 0.11906474273857498, + 0.11758516286397941, + 0.15556512780426962, + 0.1472139404206392, + 0.08769650375466768, + 0.1344006101130698, + 0.06863084247057505, + 0.05293717590403379, + 0.05800658120540277, + 0.05150921812132622, + 0.06646958604756395, + 0.14223294207899886, + 0.14575598825850894, + 0.12031714600253773, + 0.040130584024259876, + 0.12720351861488038, + 0.173726372492505, + 0.12796432233400176, + 0.15165805682835576, + 0.13727902813129594, + 0.13875459532653434, + 0.11456536788507324, + 0.1668018265121992, + 0.18032495440894444, + 0.1682420191184007, + 0.15848500709606725, + 0.12816140978404666, + 0.09909007071124311, + 0.14029584044616258, + 0.15836595626935154, + 0.1452951990061161, + 0.18705667573345103, + 0.1464817666508992, + 0.11014376500393429, + 0.12071382413446004, + 0.14483017723261546, + 0.09908019659160226, + 0.10975993657974782, + 0.1942462781700336, + 0.15561433386623336, + 0.11106460610254534, + 0.14428607428517493, + 0.20681599396135816, + 0.08947997428709947, + 0.15285605819891956, + 0.14555461852149312, + 0.04129443226911942, + 0.08899224309010997, + 0.05780865830635497, + 0.12225603553748302, + 0.0857617002854357, + 0.10154589129084997, + 0.05157180208402805, + 0.19295911291629061, + 0.14859119657539693, + 0.16167734019220098, + 0.17985308511517872, + 0.09551204068521547, + 0.18238413894878525, + 0.04713813197826276, + 0.1693844246485327, + 0.0819691973797009, + 0.21600527544673956, + 0.14674100486741157, + 0.149570388669195, + 0.24073153998588137, + 0.10816220823463032, + 0.1474422902605979, + 0.03563214555324313, + 0.16193271567902073, + 0.11833764911184409, + 0.0846304335546926, + 0.043536748248596116, + 0.04323122382832362, + 0.05734312637419927, + 0.14636223836654572, + 0.18607475782819594, + 0.02884149541695436, + 0.03886335959606013, + 0.10854296564110391, + 0.16983308128549493, + 0.12116300837167837, + 0.07906786037792234, + 0.03364663377961001, + 0.13074660120716322, + 0.06648547424547728, + 0.08033094012708404, + 0.14181752400698233, + 0.11760248048588981, + 0.1544712547431679, + 0.061859040392074585, + 0.09913288616917634, + 0.05896171299560517, + 0.05801033796452564, + 0.09233435202015991, + 0.13622434840947403, + 0.1504508755508798, + 0.11029363771993343, + 0.13469934391196522, + 0.11569066696585542, + 0.12805101265661137, + 0.20126991876784622, + 0.09554025702897076, + 0.10277800707332777, + 0.13660429450123607, + 0.14612200293328445, + 0.13426239294659276, + 0.1584803530943814, + 0.15351495813735502, + 0.050296311088443214, + 0.1424575421125422, + 0.08849473890730619, + 0.04466121965573929, + 0.1514603163068678, + 0.17773058678507098, + 0.1419455067310412, + 0.15572150897771464, + 0.1115291439870799, + 0.18755384006827813, + 0.18560818455418018, + 0.17471662630404874, + 0.14295242438987465, + 0.1495097400110478, + 0.07788997619644374, + 0.08961363888507191, + 0.12971088695547647, + 0.15009288469111062, + 0.04826030154919818, + 0.07646282651276999, + 0.11200141261670186, + 0.17186966466281697, + 0.1635056063357191, + 0.12099986434127874, + 0.1760684644562973, + 0.18821616014368314, + 0.10441296814985784, + 0.1749477415634007, + 0.12664222999854147, + 0.13050172501239662, + 0.17690172116728384, + 0.10079997016734928, + 0.1606638216385084, + 0.12213611921495, + 0.180345657768847, + 0.1663095231441376, + 0.19127160680107141, + 0.06310609449566278, + 0.06625404711250112, + 0.05595848341449878, + 0.14748901394797526, + 0.17636344418478941, + 0.1497395642022129, + 0.13883924274953224, + 0.16190542966545463, + 0.1394068292039401, + 0.14220844201184735, + 0.13979366706640695, + 0.16028486856472082, + 0.1758634649147492, + 0.12285571737731922, + 0.0660315506589019, + 0.1562812799640637, + 0.153665960107336, + 0.08883909921119153, + 0.157094117533922, + 0.13136253349718585, + 0.168562332667052, + 0.10820730291749736, + 0.1348069311971132, + 0.1260571204887015, + 0.133145740423921, + 0.16221558480859485, + 0.06830070041739564, + 0.15391218021738123, + 0.13846742467500717, + 0.08282506754910526, + 0.17553283203269868, + 0.1345560139796328, + 0.06691450855441447, + 0.13875128855628532, + 0.1589263268266964, + 0.14861076184928726, + 0.21186370743591781, + 0.16833823682195276, + 0.18352643408142216, + 0.10884330695083053, + 0.14968236968299836, + 0.11998407063689878, + 0.11682382731814216, + 0.1315128586053916, + 0.13002827634348946, + 0.16027869507521827, + 0.15891070599664506, + 0.12336152315820798, + 0.14332474349216182, + 0.1478510209850141, + 0.1470059631763466, + 0.1453630270137508, + 0.17041902106044604, + 0.16182188940265765, + 0.16377402050679635, + 0.10928921558467528, + 0.07819581354839882, + 0.13790604980181193, + 0.16079092216352045, + 0.14350460303403628, + 0.1741503409934693, + 0.2217498465898111, + 0.1358243813026069, + 0.17093183626707903, + 0.12776042384633476, + 0.15321395429699503, + 0.12872332699341968, + 0.19567096577361281, + 0.1345231745259145, + 0.17209027860206713, + 0.11397484013953563, + 0.05077471025136908, + 0.14412878845390506, + 0.18147443659390594, + 0.15190224817987227, + 0.16779861508457028, + 0.12504489772660474, + 0.11613787429324705, + 0.11770337983057456, + 0.13775287227675356, + 0.17133685386985203, + 0.11919379009049184, + 0.12162985296420512, + 0.11168907812969657, + 0.08941543433210715, + 0.10668591420790337, + 0.11665068704310262, + 0.12807733246707034, + 0.06271132569995269, + 0.15133444045406058, + 0.17314932122780313, + 0.09168985377349734, + 0.05126634737334952, + 0.07698802660999766, + 0.11969095852229786, + 0.11875335573852006, + 0.15631235179590772, + 0.09570425312990345, + 0.08486699021775741, + 0.13908331376056474, + 0.09009349859929666, + 0.06759173285742089, + 0.15707777069745538, + 0.1449230456413187, + 0.12278875926449509, + 0.10172399782655367, + 0.21272377459696706, + 0.15503271758763199, + 0.05467997977572482, + 0.07315677143545592, + 0.1350021041420106, + 0.1153005536408496, + 0.20196796250174046, + 0.15811720324754952, + 0.15627244385789146, + 0.05500336353328497, + 0.17669781729380346, + 0.14566329327732913, + 0.11854110272385733, + 0.09966934761386848, + 0.06873296083041676, + 0.1528187256414107, + 0.14719778264834227, + 0.10970244705101086, + 0.0859011141347452, + 0.05801269329173442, + 0.09714584980828009, + 0.1489199567686649, + 0.11573856976245943, + 0.15191881928525178, + 0.14149274142203644, + 0.09822886167467257, + 0.10913870333808283, + 0.05243320844712186, + 0.016344791483317456, + 0.0564242181924847, + 0.06086172671118271, + 0.03232994621776735, + 0.1329791988578451, + 0.09488741921065719, + 0.08617280768674578, + 0.0515318499947532, + 0.10441252807926929, + 0.1732436827305524, + 0.13171768050555346, + 0.11912575904612548, + 0.1508963928769097, + 0.13931961976055438, + 0.1403788086264783, + 0.1581638672796751, + 0.18494907556510642, + 0.172259055077691, + 0.14073217651856562, + 0.14749912213779784, + 0.09437663165291912, + 0.15146351496540292, + 0.15307289726540618, + 0.1311330688119594, + 0.19968730847862518, + 0.12402625901449836, + 0.11551641633634045, + 0.08040982327045645, + 0.136076315424433, + 0.07029434000228256, + 0.07284359357346414, + 0.1943441611733948, + 0.1523706070306197, + 0.11585527340536406, + 0.14318574092900369, + 0.1827375412779595, + 0.10141703432282953, + 0.12620605030318502, + 0.13682565558892879, + 0.049133256110780295, + 0.05963873315962628, + 0.07138985151859878, + 0.12959802557652714, + 0.07902468335454907, + 0.11121096800094013, + 0.04820868720677222, + 0.1920254285442637, + 0.14944389961647112, + 0.144903505566883, + 0.16258835038706174, + 0.10590996045151746, + 0.16475740491666716, + 0.06622213895222812, + 0.14017387106831097, + 0.10737921675040753, + 0.20581859045634504, + 0.16341574689020827, + 0.1582680405112557, + 0.23818010390053398, + 0.12385452592125347, + 0.12716537790999727, + 0.05190262253104799, + 0.14195228895989428, + 0.12638827704243255, + 0.09209958360655483, + 0.049533634224696595, + 0.06375313469450171, + 0.07411659107914127, + 0.1538230004010845, + 0.19364644093247538, + 0.05414726831739456, + 0.044111879564283786 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.14998685477752022, + "valid_tracks": 613, + "per_track_pearson": [ + 0.21906196533689815, + 0.15726015570572083, + 0.1515286613117332, + 0.22271013813060328, + 0.17693277771401963, + 0.1652793752434026, + 0.18877869223585916, + 0.18969975480382734, + 0.18825499522307113, + 0.21601522942770926, + 0.07103921502457983, + 0.1599643633636075, + 0.06731548314796429, + 0.24069172063474084, + 0.22900240356225898, + 0.24190516293020473, + 0.19972923605382292, + 0.2059388296733705, + 0.16653431164734797, + 0.18967193309207672, + 0.15959226901606885, + 0.13655136148730712, + 0.19986305711246363, + 0.20237169894211152, + 0.08708192526185021, + 0.14685733462974573, + 0.10155900402950649, + 0.1771866742189859, + 0.18703824266726635, + 0.19186365045516984, + 0.24043441388165382, + 0.14889044223057168, + 0.23077129790224796, + 0.1508761937705524, + 0.2622537675609644, + 0.09863003623608085, + 0.20121819431021712, + 0.2020527609900752, + 0.15710834879022062, + 0.190615773009778, + 0.23448781830611765, + 0.14108086131573172, + 0.15802873537306122, + 0.17574263320005087, + 0.21300552005411327, + 0.15590248270985288, + 0.1974786569831867, + 0.23747839732016165, + 0.22952475583492316, + 0.13798471352612457, + 0.16670710392967397, + 0.1691985621447913, + 0.20425680142460378, + 0.2072899527682106, + 0.10789549405929963, + 0.20981376041792277, + 0.1981749678606526, + 0.18513796956148032, + 0.1557596994257595, + 0.15857938509775413, + 0.09998434127067521, + 0.14721098789909584, + 0.1469796513254626, + 0.16124110480673545, + 0.16281851284458598, + 0.14786386758554448, + 0.16443187143644514, + 0.15642544058701352, + 0.1653068922708735, + 0.15458687303974153, + 0.14861509854875898, + 0.15464739029485192, + 0.20587086077984415, + 0.14695752245841034, + 0.18407369824138445, + 0.196874214399542, + 0.16717471262129277, + 0.16957001094245788, + 0.2499233398178226, + 0.1713736126529773, + 0.1654237836292928, + 0.16313492944300886, + 0.22968222190951004, + 0.1931947685334054, + 0.10942787432872535, + 0.1437293874956102, + 0.1915430253772897, + 0.23225319538635456, + 0.1843912088117731, + 0.23818954536202516, + 0.1724240435107655, + 0.11737816230228547, + 0.16382329837544757, + 0.1568016730133772, + 0.1748016455040377, + 0.1774058077248006, + 0.15234167422383352, + 0.17725245701317294, + 0.17469964404628577, + 0.15050336719408564, + 0.16982385990792961, + 0.2023645514826708, + 0.18060449791416913, + 0.13923085744058739, + 0.2040030696604529, + 0.2088802182089047, + 0.1615325784830019, + 0.24834251306116958, + 0.21652666499988035, + 0.22626670390974443, + 0.1415195270393388, + 0.2828297756712593, + 0.25427420995078787, + 0.09226684534122863, + 0.1930659959979526, + 0.20147799007380748, + 0.24221140927850096, + 0.2053331907316726, + 0.22195193877175498, + 0.21643010569277338, + 0.16162368719441458, + 0.19405207685856396, + 0.19886102232282676, + 0.2313809196432209, + 0.21692538299962263, + 0.20469768962618393, + 0.20901247529414618, + 0.13864529892208113, + 0.1991962684928731, + 0.15280184544753186, + 0.26175741533636115, + 0.20222960357309583, + 0.1944408094025717, + 0.0761189878766021, + 0.14407336598835618, + 0.16242917272972626, + 0.23823506034002973, + 0.24943795851385586, + 0.2391373110484622, + 0.230318646875809, + 0.21043205267823134, + 0.16303279050592118, + 0.2398924843337022, + 0.2805319659098848, + 0.15870682864540794, + 0.1645320129570112, + 0.11454280315808062, + 0.17978679830765268, + 0.14633032673385243, + 0.21331770388140034, + 0.11207493795263257, + 0.22543650016885866, + 0.13911403394614338, + 0.16086817869447123, + 0.23606622024387003, + 0.16500373111366134, + 0.19745063249113157, + 0.2270953654982966, + 0.13420936198508318, + 0.23255136184740433, + 0.2687146133149385, + 0.20417299044915835, + 0.2049651202834184, + 0.196670290852344, + 0.1975505009749016, + 0.18059951925143547, + 0.2041034424860789, + 0.20340717355481455, + 0.2031825737636463, + 0.22129127400272108, + 0.04668316297763318, + 0.14503853962301813, + 0.12802192076364405, + 0.21928010908824702, + 0.17910233005934986, + 0.1590769770930663, + 0.19184083978017194, + 0.19736725361450164, + 0.2604314955630313, + 0.12201665742954197, + 0.2290423888620727, + 0.10790093316209912, + 0.18135299420895665, + 0.034553551580600715, + 0.16525486384760815, + 0.21504884165832738, + 0.13001858829389293, + 0.07326756823704768, + 0.09977985424822121, + 0.28930265427051943, + 0.3138194708136173, + 0.12327177557458349, + 0.06965406219888082, + 0.20065120302952585, + 0.013837523645834297, + 0.1316844455137958, + 0.11515788432678183, + 0.20627015588940578, + 0.09835443798372058, + 0.169450687894804, + 0.2595272972655076, + 0.01516663223115086, + 0.3229188049322583, + 0.19576944521416625, + 0.1297882852688742, + 0.14170426048695317, + 0.13371858370259448, + 0.15552063781569736, + 0.11884906046991606, + 0.05764164839083074, + 0.053426056060705766, + 0.06149541592611444, + -0.018866200540796987, + 0.07997296563352616, + 0.14159150930859446, + 0.16973444110220406, + 0.012374988729943637, + 0.18440228119461655, + -0.020733798727522176, + 0.09512851845668277, + 0.3093547044451814, + 0.05759664900317944, + 0.2510496817672633, + 0.1930806596214044, + 0.21198955614583537, + 0.2736663694637451, + 0.20577148708931095, + 0.24709567265263327, + 0.16617446090016139, + 0.056253796620996645, + 0.17908453248636402, + 0.30354292527641435, + 0.1088401551949197, + 0.2233237674482011, + 0.12105972623135884, + 0.2227370811096613, + 0.13845839405593957, + 0.17032188588195335, + 0.19342430010578304, + 0.19056784923822714, + 0.18657011231841292, + 0.07076405780877806, + 0.17990387370953428, + 0.018381415329367875, + 0.13647108510840017, + 0.20187482699160164, + 0.28587435547549567, + 0.1544033796321535, + 0.15378039700258356, + 0.1656474807928796, + 0.18806066201417665, + 0.15793395622583128, + 0.09737346242542175, + 0.1401536441717418, + 0.1654085368478134, + 0.18608436187654465, + 0.17237998023119439, + 0.12729170254763444, + 0.18532481684339805, + 0.10531195384230524, + 0.24469768234601622, + 0.18519269772846306, + 0.18600635609470859, + 0.21253805833604783, + 0.18105300622557677, + 0.12714040417958888, + 0.11416484308294327, + 0.12176485212025462, + 0.13508343766816822, + 0.14617835865020246, + 0.205386186115956, + 0.11597724874597724, + 0.06951708279981691, + 0.06571950202933628, + 0.1527831233969745, + 0.11825538593817632, + 0.10090107695259894, + 0.1326244946503305, + 0.11682615204773478, + 0.11870807404561767, + 0.15120521736177228, + 0.017195300214548813, + 0.09206761858038544, + 0.03530987809730821, + 0.1447425154074287, + 0.14869740910464654, + 0.13825886758438058, + 0.10471303305728115, + 0.13624030575438784, + 0.1361959235466598, + 0.10846520392505196, + 0.10776254969026745, + 0.06677080525591704, + 0.13575667080371717, + 0.11901572447120583, + 0.0333132877503882, + 0.07295843126323906, + 0.0820181608832838, + 0.13040537696748758, + 0.11514961213842699, + 0.11635751467422772, + 0.16501954231202504, + 0.1020121495504797, + 0.10139630268175594, + 0.08879381785905076, + 0.2115685417862036, + 0.056810037702653395, + 0.13437205101967908, + 0.12888518160962364, + 0.12535945564029316, + 0.13325004637380514, + 0.15540110904993618, + 0.13600606932195303, + 0.10413867640172098, + 0.12205433974520584, + 0.1329803264468209, + 0.06629177013610144, + 0.1272821746244358, + 0.15744219163678744, + 0.14407058563488753, + 0.06638282318446959, + 0.07633699919822236, + 0.10390790292674629, + 0.12384367539531267, + 0.13924512383074364, + 0.06317776520368337, + 0.14477084157616385, + 0.1220210860960766, + 0.09884089613313685, + 0.05476611277598571, + 0.07269621258721654, + 0.029052044368870697, + 0.06981764956835464, + 0.09636733555531211, + 0.10741985557650614, + 0.11888775778608912, + 0.04394258862428886, + 0.11049556115373928, + 0.08052184926198414, + 0.12050648565154495, + 0.09632390491353297, + 0.07863893162437567, + 0.12266036548438752, + 0.15294786389495152, + 0.09377714896286941, + 0.1276669716328934, + 0.15410945097410314, + 0.10370869972540635, + 0.10230367056923273, + 0.16995648475155492, + 0.10512663818340699, + 0.06587464486565087, + 0.08962735488754034, + 0.14555031840829571, + 0.1254112491911914, + 0.03799216397176685, + 0.09824950432910882, + 0.11180818091405699, + 0.13565428396489332, + 0.06746943545371768, + 0.11412590474666456, + 0.09116415168964564, + 0.019400068734075286, + 0.09594139592747977, + 0.09743567745486922, + 0.1201357974302932, + 0.09942986922263497, + 0.09712621540506135, + 0.11510442123287554, + 0.12950592660483184, + 0.1041271340265215, + 0.1174634766342985, + 0.13645390678588742, + 0.13050012382749074, + 0.06233939407047659, + 0.12858878316004554, + 0.16455286160872676, + 0.08397615365162966, + 0.16756661794926905, + 0.12615524697805455, + 0.1388819413469636, + 0.08838880213759066, + 0.19811283480533484, + 0.1785974206012568, + 0.05848590757927179, + 0.1314990687977439, + 0.12075799341167408, + 0.1545307962504375, + 0.14331431717213647, + 0.1416256241335428, + 0.13196171478611843, + 0.06103030961583059, + 0.12731083843593471, + 0.08372639207696507, + 0.15023959970651596, + 0.11935193643263763, + 0.16024109800976183, + 0.12999784124745098, + 0.10557445399255476, + 0.12128419268904675, + 0.08886181516306485, + 0.1904912184517107, + 0.1523695228852971, + 0.09270236973711386, + 0.050968325310632515, + 0.08305342927624505, + 0.12340333929452658, + 0.13590085885160982, + 0.16288432667983552, + 0.15816686866994978, + 0.13715314274245916, + 0.12097103603997404, + 0.06245929801292622, + 0.15453558253797448, + 0.19962526814502773, + 0.10801160626482341, + 0.0944590631185765, + 0.08263861716911533, + 0.1068784735521281, + 0.0442519618411097, + 0.12702001007453437, + 0.10541236140558892, + 0.15605014845990645, + 0.03109589040707457, + 0.08725873095770979, + 0.13141975167095005, + 0.10562340349472273, + 0.1259354735407758, + 0.14757343113941981, + 0.03716735963992385, + 0.13110975441690095, + 0.1537002602858403, + 0.12322555342659473, + 0.1413077444033803, + 0.10086102806675876, + 0.10327666663342652, + 0.1065073030159841, + 0.0926664223600661, + 0.12088014457658275, + 0.13370132131403176, + 0.17403408762255446, + -0.0009972153422592102, + 0.08107015416549591, + 0.09725130301504264, + 0.1675458321947019, + 0.11686094400614916, + 0.09059447920763769, + 0.15680932517099241, + 0.16888794755622338, + 0.20645974909242626, + 0.07305683805536238, + 0.1905288438293658, + 0.11094430671156982, + 0.10235281060661042, + -0.03189820050003649, + 0.12016933087699987, + 0.1678108628337677, + 0.06229154861679437, + 0.031439789908298585, + 0.03806339558125709, + 0.24824100205849162, + 0.23832575545203585, + 0.09401755744593408, + 0.03784125580613981, + 0.15367903541395844, + -0.0055924457987622704, + 0.05387762024770059, + 0.08589625838049122, + 0.1858043539757933, + -0.0014117159278441002, + 0.11716510871052072, + 0.22598904837115624, + 0.007945153393532068, + 0.29065938141599323, + 0.12903822629208017, + 0.10020754164584755, + 0.09184920838047846, + 0.10155939854052878, + 0.10280348188343941, + 0.1203119224812976, + 0.007993212170891106, + 0.02701165602951395, + 0.008568523722992317, + -0.028262860356609444, + 0.013252672928741216, + 0.11684438207756308, + 0.12378804079137307, + 0.02617098397763072, + 0.12751574429914928, + 0.008531603268299987, + 0.0051075964611658675, + 0.22344646627468814, + -0.020457633761740108, + 0.19482780751697212, + 0.10400264256165404, + 0.11963650367317434, + 0.20555116704993964, + 0.11501115503718067, + 0.18452837202902445, + 0.048483418011051056, + -0.011078724949001846, + 0.09571522490036724, + 0.25729192665634065, + 0.07264827923357758, + 0.19815102323164818, + 0.05292948820819136, + 0.17975628977123756, + 0.08184040130722132, + 0.13541837799496573, + 0.10935386175805034, + 0.11031701299900468, + 0.09605847531879934, + 0.01684852545283635, + 0.1241135204100823, + 0.02159583141055517, + 0.1507012948764154, + 0.13296654280501047, + 0.25443684653581244, + 0.10700649234178393, + 0.10329100097181668, + 0.09339078739967023, + 0.11501814958570268, + 0.09197455813177696, + 0.04703498031624621, + 0.0678374966532965, + 0.11868943480970018, + 0.13948319383819946, + 0.08698180260410077, + 0.05842476842469226, + 0.12060835277507999, + 0.02925054744675957, + 0.15281005353828936, + 0.08797402326974722, + 0.0786778808113503, + 0.11670015815963292, + 0.12428821371875853, + 0.12167670553254231, + 0.09361132942042584, + 0.04951763825162597, + 0.09519922909748837, + 0.11486564463129173, + 0.1636822759123758, + 0.17589274705519722, + 0.06270905532327577, + 0.23389499741881212, + 0.28478595823645103, + 0.2381393766015344, + 0.22943946771034943, + 0.23005960191898236, + 0.17808808812467322, + 0.1902355291501989, + 0.21219073725342405, + 0.16505451592310755, + 0.17566236157787313, + 0.1943894215544732, + 0.17320731346466972, + 0.1756540444194282, + 0.17705071578226983, + 0.18336903840677884, + 0.1738077021030659, + 0.16892164624847483, + 0.16654713297584284, + 0.2365969523397261, + 0.22842205554041928, + 0.23646398314600947, + 0.23743938805181894, + 0.2721753506127896, + null, + 0.2539488598318288, + 0.2905879819640072, + null, + 0.030225191156727792, + 0.35916668461728546, + 0.04469175315745085, + 0.2473434348746846, + 0.219263583332358, + 0.29133414003516195, + 0.35059692979579965, + 0.33421480865623504, + 0.3445246691759555, + 0.3237546716757768, + 0.3055770065651484, + 0.2712481347221715, + 0.19894420986709288, + 0.34891218227921134, + 0.1472232971061043, + 0.25723351383697834, + 0.23932951031791524, + null, + 0.23400643570645718, + null, + null, + 0.2207489558735324, + null, + 0.1664933596262848, + 0.1841616764723154, + null, + 0.26328075771650616, + null, + null, + null, + null, + null, + null, + null, + 0.22269271638225124, + null, + null, + null, + null, + 0.19213942833668107, + null, + 0.23419394150866873, + null, + 0.17041321005682358, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.24866797343788483, + 0.13851968666662787, + 0.2283738077870903, + 0.22576513748914154, + 0.15625992068022543, + 0.21282704682081433, + null, + 0.27093241471826884, + null, + 0.20888314119211257, + null, + null, + 0.22123369230502551, + 0.25181834306516904, + 0.20238735718675793, + 0.2101041250836267, + null, + null, + 0.16796117283130882, + null, + null, + 0.2040428377625844, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.20985828819198943, + 0.19892015795006968, + 0.21524468510197386, + 0.1899179370109519, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.2545079702799696, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.05084704138610699, + 0.043110670888422895, + 0.14982942878960442, + 0.19692283751287232, + 0.18669968016242178, + 0.07274554935024957, + 0.13372939134116932, + 0.1851422048669583, + 0.0740719733394928, + 0.10505346883240152, + 0.3444121563861709, + 0.1830400447560529, + 0.1274808613489319, + 0.15848521550366543, + 0.1284582136168474, + 0.15637483041765646, + 0.07772035562882924, + 0.07906583877533865, + 0.14658066785646423, + 0.08813997696742984, + 0.10162019582023801, + 0.1253779339919102, + 0.09288419624461443, + 0.1428782623813654, + 0.08741298090449719, + 0.0748751923927756, + 0.0791786019118735, + 0.22806111435323453, + 0.23770883487672564, + 0.26604988867939955, + 0.13695019362497532, + 0.14125976597008003, + 0.38730973547186665, + 0.11733714346448341, + 0.5123647843060434, + 0.23091036797639344, + 0.08041663618096594, + 0.07580330956678753, + 0.12203288932032252, + 0.07868513539332528, + 0.08605287235601584, + 0.3831225179134526, + 0.12714003597596452, + 0.15838975384624746, + 0.13849304686210864, + 0.22815862737648032, + 0.11435642567903292, + 0.09896930907535008, + 0.2788564871730102, + 0.222541428373054, + 0.19116064406311972, + 0.2691622159677444, + 0.3186529274447129, + 0.1613667242736824, + 0.2552546717795553, + 0.17203976309243366, + 0.037358808922965385, + 0.10081539100334032, + 0.1281291454431001, + 0.15830251074736232, + 0.07791533640893346, + 0.09103166775088835, + 0.2977992893085028, + 0.16055782013428702, + 0.14592427142123934, + 0.10628058251965461, + 0.16278854270361554, + 0.2319777994029014, + 0.18376811640606175, + 0.09123658636987886, + 0.1613252978195307, + 0.5846452237602066, + 0.13686830873886674, + 0.1392832924701176, + 0.07545786064231025, + 0.2616434384258846, + 0.03205921813156432, + 0.47964922897257556, + 0.12262536510818288, + 0.20262780105664074, + 0.22598506795030415, + 0.13841162832405868, + 0.2757372720178809, + 0.07767894282213558, + 0.1063300064378251, + 0.08132056088190115, + 0.30743085582817975, + 0.11880183629642777, + 0.7426448652793892, + 0.3067290027680109, + 0.14097742429945262, + 0.1607268164685405, + 0.14605793160255412, + 0.10301488246942797, + 0.20215131218527113, + 0.19978322154351877, + 0.17526782865562754, + 0.07873770932439461, + 0.1199199470255824, + 0.09971057627710068, + 0.17097915001730268, + 0.24235300682796002, + 0.23615740190317117, + 0.26203361925454416, + 0.3089762489222026, + 0.004452071273085462, + 0.050824582333366126, + 0.06425315609923142, + 0.32963827273982593, + 0.14799460902395606, + 0.12742049503721276, + 0.0683866359232012, + 0.13197615425782508, + 0.21610385451806827, + 0.038640379076043826, + 0.37513794789428573, + 0.18010751608711587, + 0.17721704687768666, + 0.14786507366799112, + 0.12066786668454489, + 0.09376065638294716, + 0.25258017832174207, + 0.16979775420872587, + 0.03239010956484154, + 0.1959905109606266, + 0.12890081852501942, + 0.09013549676221821, + 0.09526537422206456, + 0.16121493424898797, + 0.4601956953150506, + 0.12358379929156016, + 0.0938035040180751, + 0.20707725185647824, + 0.15209733509988435, + 0.012757205153953623, + 0.05276459755163321, + 0.01881663470085609, + 0.13868710340012896, + 0.05611353529990442, + 0.038871401504734095, + 0.015666754935767035, + 0.17404886999683367, + 0.19784183181261916, + 0.05047024189488492, + 0.1419568000811117, + 0.2899746260745536, + 0.24541160913505625, + 0.02821559714615093, + 0.14627388225161592, + 0.15801596994494366, + 0.176720985002033, + 0.08563371706963503, + 0.13271089808128098, + 0.13135856802174178, + 0.23775229645994655, + 0.1310388783462526, + 0.04051049875049887, + 0.0669268954555226, + 0.06086590317413636, + 0.21708897434261953, + 0.029742180011149515, + 0.16794150022099041, + 0.16879383929584293, + 0.10558626621285455, + -0.014178984344197853, + 0.19191002181234787, + 0.16311552826743317, + 0.29604880323200294, + -0.015800612494565908, + 0.07980714721450406, + 0.04383745434829563, + 0.012703533422110608, + 0.30271884817806605, + 0.047389628838025, + 0.028577257783542847, + 0.2232960873620172, + 0.05397210403395838, + -0.010558589664545535, + 0.014361385179388076, + 0.1417892963758799, + 0.02162917785105717, + 0.08989065054169414, + 0.09094259560799475, + 0.15118673219720466, + 0.05931057347676792, + 0.03956237279921387, + 0.15425882519707354, + 0.06807437141813175, + 0.0923818666909765, + 0.12802861320418024, + 0.061430107068477464, + 0.08013256316275905, + 0.24834299444490335, + 0.22166342878669693, + 0.05940877128219688, + -0.0037169406012192052, + 0.011046134191133998, + 0.036568246965561935, + 0.10132799083759969, + 0.14653597728066464, + 0.06370664025403912, + 0.12510857647726825, + 0.07535754524518565, + 0.08136651292945696, + 0.17339775401997537, + 0.03212518312033895, + 0.016931410440930043, + 0.16222049556976492, + 0.0013861521922778034, + 0.1553772267191901, + 0.02943962869023323, + 0.16774971056255805, + 0.11679139284236595, + 0.16064855530447472, + 0.7144327302109976, + 0.14664743611880224, + 0.27249215814452804, + 0.2091598143247379, + 0.3202801076106802, + 0.4722460850286655, + 0.5551488065899076, + 0.5449782001406953, + 0.15862244938587325, + 0.1559722343613761, + 0.3579814487701598, + 0.6810192426170315, + 0.1493768547965205, + 0.17291642542206806, + 0.4685298159081824, + 0.4338860346523018, + 0.17516313876112685, + 0.18476821088169623, + 0.13563066733861434, + 0.17107451863217715, + 0.5230702392885445, + 0.1971660290617748, + 0.11875807193772425, + 0.557799184384483, + 0.473476717362428, + 0.33143864494998504, + 0.3832168067466219, + 0.6536390206701078, + 0.3178436616984529, + 0.27538385930961073, + 0.2287080111700239, + 0.7155089134013622, + 0.1382834691344265, + 0.5989474590637022, + 0.22333730516558137, + 0.47489994335704094, + 0.3181411661015359, + 0.3391607157302382, + 0.6638273029850029, + 0.3995634515672343, + 0.09008781287943238, + 0.18211620287431773, + 0.5991800792826427, + 0.17361333425301734, + 0.279602185947826, + 0.07730254607477358, + 0.18202056719763657, + 0.23189455329067996, + 0.1575201912647297, + 0.20658463698971374, + 0.233955428122076, + 0.4120422658202932, + 0.16355568406341253, + 0.1555707664665573, + 0.5803961359791993, + 0.18906234975057487, + 0.3676236745399778, + 0.4401514325482301, + 0.11889614029551425, + 0.6111353896930584, + 0.08579923338845093, + 0.13170538829544362, + 0.49664038631463464, + 0.3527678344697545, + 0.5281269519613544, + 0.6837876164678458, + 0.6018090552644997, + 0.0846262685690847, + 0.3371144880853249, + 0.4408084099313786, + 0.11196476218112557, + 0.09850240609099822, + 0.5065996134261794, + 0.1718932011749074, + 0.6316266131954167, + 0.354832763789495, + 0.21914021031045366, + 0.18108956057954617, + 0.2092781659972032, + 0.37921189296776375, + 0.6603139837116875, + 0.5621869498382869, + 0.1959334381594193, + 0.4181398898171065, + 0.3702020386399975, + 0.41329857682717014, + 0.6895901070998957, + 0.13022469748041202, + 0.3556404079271355, + 0.12322514594658965, + 0.5258247934969591, + 0.14555534062601413, + 0.6837644228930162, + 0.21632098603714944, + 0.3391878712338362, + 0.6411304291290124, + 0.4121455577200399, + 0.12766309168695336, + 0.5059669751088987, + 0.23721918545492582, + 0.35424218263026297, + 0.43588494225929336, + 0.16445011411959837, + 0.43974611898008253, + 0.2131179673603661, + 0.34010068425000184, + 0.17135365814708559, + 0.07039510534329686, + 0.22793093064635608, + 0.3811325253970466, + 0.4601090866220907, + 0.606464176141893, + 0.2035675957930142, + 0.1274478337104415, + 0.2855724141689701, + 0.31189031608476164, + 0.3468130265359933, + 0.3114333271247678, + 0.0961398052709026, + 0.20526247766096267, + 0.1363483077683781, + 0.3927949387036982, + 0.7307236432114462, + 0.44815044922784514, + 0.5239453933268269, + 0.5427619273717499, + 0.6909400848090086, + 0.3752420224285509, + 0.13143037662921953, + 0.5139386037942213, + 0.1966078280960118, + 0.4105806857199478, + 0.26778433066336804, + 0.1615789481424455, + 0.40780018855835454, + 0.7313307827930198, + 0.14959892028824753, + 0.36007108284352984, + 0.28211032230859506, + 0.5627134356538522, + 0.227839449738793, + 0.11231136156884877, + 0.26956170379857486, + 0.6895804096080341, + 0.2429196169853523, + 0.3109705914647614, + 0.4378068716364741, + 0.37052350744864554, + 0.31067818379273077, + 0.3637800553959942, + 0.5503144745635331, + 0.37951208068298337, + 0.2677771698665436, + 0.4123766551501368, + 0.2572427771322414, + 0.49676406326865263, + 0.3759018545683469, + 0.29689534500026116, + 0.43058482629008077, + 0.16305492380795186, + 0.04829007960507925, + 0.18826440385788154, + 0.20574711581109348, + 0.6473524475322343, + 0.15521909946155169, + 0.4996990607310576, + 0.21858741603349632, + 0.6334366071557942, + 0.15239295138865108, + 0.12515222889013006, + 0.1336895508711412, + 0.07537931816305644, + 0.1397548885328285, + 0.09818321463533433, + 0.5565610599026629, + 0.1725196152393873, + 0.6430902234832201, + 0.5182678030005513, + 0.26739080924553404, + 0.4295254733252911, + 0.18186379262196622, + 0.20052738206265286, + 0.27842607374248257, + 0.3962798328848615, + 0.13813701004092713, + 0.3724927500977956, + 0.2345634698727093, + 0.4338261034741855, + 0.27196097762567, + 0.09515624748244897, + 0.5135433547048429, + 0.35733412682512256, + 0.6179900357261685, + 0.5004546180256326, + 0.39984700987859806, + 0.1664533237463089, + 0.11876857600359846, + 0.1804845944444345, + 0.2536472704860158, + 0.6958522284078528, + 0.3886327208549877, + 0.2287071089682327, + 0.6936585006981248, + 0.24821965923704556, + 0.5659371997787199, + 0.4517022423020179, + 0.205443177613115, + 0.23663931297404817, + 0.16775334090590463, + 0.29769664823525854, + 0.5395901545549946, + 0.42242864684721165, + 0.40931370288980196, + 0.20672460236491189, + 0.2983526224471084, + 0.1232307541959039, + 0.13919365133665218, + 0.7259795260337324, + 0.21838128882404895, + 0.360727768688288, + 0.14953045703728668, + 0.13020260340639192, + 0.6282445539942354, + 0.3169476733426411, + 0.1522165021310808, + 0.4446912804970491, + 0.19257773018343008, + 0.1116095095813731, + 0.3658380109900661, + 0.29190452785540416, + 0.7364311672103228, + 0.16257399908228656, + 0.20912813972392816, + 0.11081538039256558, + 0.2855744347814785, + 0.1254737651992422, + 0.1492784462269034, + 0.15643590918464276, + 0.08173657660578089, + 0.2523030672753781, + 0.6569797438210315, + 0.11680130206554569, + 0.30246049207482656, + 0.4128134008172396, + 0.3033980728160437, + 0.26259745218846076, + 0.37089421717104043, + 0.1817747905269477, + 0.4221095686309367, + 0.3846998877292168, + 0.10008095388708438, + 0.5594299270449142, + 0.6104928152718813, + 0.39351235283018454, + 0.36116146269519067, + 0.4192166182062234, + 0.16571410907425696, + 0.16033856733663207, + 0.3759765485722295, + 0.7263525982840838, + 0.1053533593109874, + 0.5528799653296627, + 0.1140393698699308, + 0.37220018579954633, + 0.3109778551433768, + 0.17857120817513567, + 0.24968685650359843, + 0.08916327781860281, + 0.16942909212842824, + 0.13607196087697795, + 0.4182262347580313, + 0.30314254992734485, + 0.05608325689736973, + 0.2974408675328614, + 0.35456577487026647, + 0.13407874691207722, + 0.38764429761684227, + 0.16989798087650093, + 0.13649394217427643, + 0.22542002635771627, + 0.10123023334710718, + 0.320495590849411, + 0.21833058297037553, + 0.514299090145719, + 0.33201479666875194, + 0.7265805087501882, + 0.09813677809249684, + 0.5012283283646499, + 0.6943644394463118, + 0.5827161905268644, + 0.08588693814597564, + 0.3351387645267361, + 0.2784471229024159, + 0.12658988358786574, + 0.1501880882029534, + 0.1838898958335371, + 0.32477893838344096, + 0.42472966008780366, + 0.1966978582992998, + 0.5528073516836371, + 0.6953447231001301, + 0.17605127931311107, + 0.1412484470119516, + 0.2855145172484729, + 0.30732722100141713, + 0.35890248635917854, + 0.41495459679469343, + 0.3938159396153111, + 0.12994672439047636, + 0.5682409878688466, + 0.3742300376051519, + 0.46108199915945697, + 0.6535876692371542, + 0.17419193589016096, + 0.539924962663655, + 0.22598736647253675, + 0.3082110348087097, + 0.2044298863866653, + 0.415576673140035, + 0.31536275554314763, + 0.1740770792945469, + 0.1779577140898505, + 0.1968372619047179, + 0.48242843990687473, + 0.4975007007032868, + 0.24733052805805503, + 0.15357718273376222, + 0.20588498560829638, + 0.683017813573798, + 0.21778392814351924, + 0.28609456354007057, + 0.3515028990109031, + 0.36197885169916677, + 0.5069695649481022, + 0.06785303351533427, + 0.5043305056918836, + 0.1457882803776601, + 0.4275773204977407, + 0.2897531257475052, + 0.11748073554201202, + 0.14143508762732798, + 0.29124427482058124, + 0.14497465104745016, + 0.4228615422198264, + 0.6696458549805191, + 0.6502946694071158, + 0.19268915281227011, + 0.10753567636135017, + 0.6567925659583617, + 0.3535567610805231, + 0.2511211703984571, + 0.3666036107457111, + 0.19545933763065584, + 0.533904495632226, + 0.30558474043658845, + 0.4879251278364789, + 0.4330694912772203, + 0.14281975807197783, + 0.4044353313256126, + 0.592500330811909, + 0.40387704674254776, + 0.17497713009031363, + 0.2349630626663554, + 0.10947822976894932, + 0.4494847721896429, + 0.2123103341351472, + 0.40275099327517966, + 0.39682051813393604, + 0.15362421413656252, + 0.6801003184070036, + 0.17475853113628514, + 0.6810007698911446, + 0.24367207618972203, + 0.43115581335030456, + 0.7022673583091581, + 0.3334850426227637, + 0.4720167609920973, + 0.14592309481680768, + 0.6472831962016007, + 0.20977231012787315, + 0.08531744737345481, + 0.43261193874864673, + 0.4255999411112035, + 0.5756581783167404, + 0.49082884061526116, + 0.2058862198478074, + 0.36988153433509663, + 0.10886800563307271, + 0.21401224356747153, + 0.514937308695903, + 0.4071314603509284, + 0.16025898016691567, + 0.4290093158998881, + 0.5080859465171583, + 0.5254748970919905, + 0.3484283173747729, + 0.1034187381665236, + 0.43682759838933083, + 0.3318618141806005, + 0.4904181508616569, + 0.18458724631277676, + 0.16676837061728308, + 0.2465676942483891, + 0.39627780887975844, + 0.22554199308213838, + 0.6849161519295803, + 0.4183199842903278, + 0.34788358666419983, + 0.3892879985496876, + 0.2373163485261356, + 0.3336654598203551, + 0.47440528839839613, + 0.48224517021460195, + 0.6718815999832426, + 0.44919123033456354, + 0.5211205581805582, + 0.425385635014667, + 0.34231658579839097, + 0.5139652627735506, + 0.19199854364689783, + 0.5421078364900701, + 0.5328974903723217, + 0.6804455307941136, + 0.41438584180208504, + 0.47658169130396133, + 0.47451533324227896, + 0.169238877302566, + 0.4821631162821325, + 0.11735327247006491, + 0.6183776735246992, + 0.3573506589839247, + 0.6901339987973516, + 0.6808114091066617, + 0.2242096565041413, + 0.329855074664873, + 0.41038685948503933, + 0.10086051043584705, + 0.34989569540763427, + 0.7093098104064759, + 0.42148318787190825, + 0.38682389019073654, + 0.5194741181591421, + 0.3786992938948817, + 0.18780874162392816, + 0.5480212482625947, + 0.3143767105265697, + 0.23551224630031878, + 0.6697868908256758, + 0.26789282670274767, + 0.15468060339291737, + 0.22009220824467507, + 0.4101046905183594, + 0.4949876939780349, + 0.5357955441303504, + 0.6315510656621263, + 0.567332585903788, + 0.531919076704571, + 0.3217634971895032, + 0.5545905078050696, + 0.46009304047288446, + 0.7321825206897232, + 0.6469790447872227, + 0.08768774374235908, + 0.17298443434555655, + 0.09625726886816852, + 0.40174645426406586, + 0.2936529090414529, + 0.7301249873099067, + 0.23301179919651366, + 0.2986101214963185, + 0.4735636525196489, + 0.3748250738308703, + 0.42793835879061515, + 0.5629991035345489, + 0.31122669920621754, + 0.5602137998128569, + 0.23878305200962305, + 0.613533664007132, + 0.7240510079114683, + 0.7171239530102117, + 0.5348876706951206, + 0.31035938390118073, + 0.4441771123953721, + 0.6726686679270387, + 0.6370193646172587, + 0.5616394711200465, + 0.3479364905653101, + 0.3356713525006146, + 0.40962449090536596, + 0.6827706442421299, + 0.2917424233007329, + 0.2896134735919098, + 0.16939592262011652, + 0.5208505987767995, + 0.10056220717419408, + 0.18594187216950073, + 0.6037962541650049, + 0.485785445033138, + 0.6408591451940234, + 0.417736585364937, + 0.4288518325274875, + 0.6145426440946827, + 0.5811372328770962, + 0.5050950395140991, + 0.2701072260648935, + 0.5858611429746461, + 0.4241569991046108, + 0.2804602656535604, + 0.20174428072411169, + 0.40922492592535487, + 0.5935170520915795, + 0.38316898138843675, + 0.5396200615473488, + 0.48582109901298914, + 0.321563705465512, + 0.2633746836885393, + 0.6350441445622256, + 0.3824995174031165, + 0.49261728650221454, + 0.6866407136954291, + 0.3101214647116654, + 0.1852592500122701, + 0.17782193480608, + 0.1299665729792429, + 0.6647463066145797, + 0.6960780211153514, + 0.5064101462901752, + 0.3154681377226837, + 0.29361249093323, + 0.22211822594480068, + 0.7095885716380609, + 0.19039378658426187, + 0.3843309081296315, + 0.1799435154936269, + 0.4041144415584761, + 0.19246772078492122, + 0.4839239157503238, + 0.3110616764459964, + 0.10723102904957704, + 0.6338296926015875, + 0.4066477187377388, + 0.14114071672665546, + 0.1435940248203668, + 0.12421148279592223, + 0.15183054283018307, + 0.15157323601017098, + 0.1587295399192725, + 0.16581127748163263, + 0.2011743164020818, + 0.13595508192104275, + 0.163664618265796, + 0.10351200359930679, + 0.19762108217704694, + 0.12373774738011709, + 0.1069595602819089, + 0.15804602186587474, + 0.28718428229572507, + 0.08518110032695037, + 0.24500311320499765, + 0.062953978801538, + 0.1096651502067939, + 0.15324715306380401, + 0.4817543278964999, + 0.11573817005390921, + 0.10277987227495745, + 0.667001185682077, + 0.20906718101847677, + 0.20204763181652127, + 0.1683867057935742, + 0.3491697712632864, + 0.14261984687748772, + 0.21531682159190355, + 0.20310991235037068, + 0.22445258042433375, + 0.12759267702191535, + 0.29406213468741454, + 0.28826488888731794, + 0.3020431662422909, + 0.17414469886269449, + 0.38621734885384074, + 0.15538255116821284, + 0.10993519699249617, + 0.15256987757995139, + 0.18480287779901927, + 0.38664206684110536, + 0.1670876967559176, + 0.37673805047066783, + 0.1990779105461216, + 0.19621586181230818, + 0.7658668734719578, + 0.06003269293780069, + 0.25579143495394663, + 0.24808597214151637, + 0.24492362922764924, + 0.11550272037813164, + 0.20416793814554382, + 0.3454306197583314, + 0.4418329569823516, + 0.2549867092038159, + 0.26043956554999814, + 0.06937917567757278, + 0.3051025777687316, + 0.2479535703804657, + 0.24188453113868152, + 0.28846819066793666, + 0.21802495332527494, + 0.18764374396230513, + 0.34042604749329924, + 0.25189090436819417, + 0.29834752468465287, + 0.09634896480716008, + 0.17823650034401814, + 0.18758496712191416, + 0.3539928130278301, + 0.23285419672235663, + 0.15268777877129774, + 0.11339621640719114, + 0.7344927020103006, + 0.07934605654708973, + 0.13908688241390268, + 0.19845084201304553, + 0.1361581594816257, + 0.03886901933922529, + 0.06332401176955921, + 0.049625794901562936, + 0.14592790917781956, + 0.2457447716370768, + 0.08229242496365265, + 0.12065327794760698, + 0.28361608853774795, + 0.30743615979841765, + 0.01397737902864462, + 0.1220042122524761, + 0.4714395270307464, + 0.1210662081930135, + 0.16498805575167166, + 0.20074532586273136, + 0.33789321087063356, + 0.06344516225369683, + 0.1280857384470161, + 0.06431475249830321, + 0.1899110657522074, + 0.3554608097534971, + 0.08969344071723338, + 0.18961371922932335, + 0.14925152573978964, + 0.11893626680119855, + 0.11167614166849461, + 0.10160399050430653, + 0.22975235272484326, + 0.12897765703722786, + 0.07765099423150323, + 0.3200581022416775, + 0.07553843127263364, + 0.1542647804468934, + 0.12404829084968991, + 0.7367841653794199, + 0.34333623176223277, + 0.18818171823906293, + 0.04167797183290499, + 0.12278104540949614, + 0.10408314825100362, + 0.636486139955835, + 0.19026620582669898, + 0.10469725116243785, + 0.2984099745087677, + 0.2019187924575005, + 0.21361310806253786, + 0.14647457274368733, + 0.36712856028733204, + 0.11854194595392802, + 0.11966373502077177, + 0.3311158429655475, + 0.22900388590787124, + 0.13491177198717158, + 0.09792005714158022, + 0.07711912391332124, + 0.09306958252116233, + 0.08772294485526909, + 0.1527445064580476, + 0.16682894590032044, + 0.5180584178012202, + 0.004630019836031502, + 0.14190606556672505, + 0.13065780099569885, + 0.3360500045232327, + 0.05740910863703443, + 0.05146478486120637, + 0.06742629332980803, + 0.16696070627911813, + 0.23526250881018412, + 0.0760851810138564, + 0.4226685465882746, + 0.12177057789052803, + 0.14361057034573152, + 0.23206818092026965, + 0.12367523138221076, + 0.1432602342656839, + 0.1962544710245432, + 0.3780085877929026, + 0.14329225474161056, + 0.5873634233564333, + 0.21561707392626614, + 0.38008711610624196, + 0.0741090095630991, + 0.19080559738154687, + 0.12151308331902434, + 0.1197788400511211, + 0.08357563532832277, + 0.2848244077527789, + 0.518304062418376, + 0.07911067178841684, + 0.09073684526535918, + 0.3988253404086629, + 0.17715002705021574, + 0.22920089050447756, + 0.4839643463139573, + 0.16827104421925507, + 0.18047483261012934, + 0.05983578807535564, + 0.19320273789614364, + 0.14407809310473746, + 0.11240195625239664, + 0.32766496602232653, + 0.15568384530091042, + 0.267785946472636, + 0.07226849579387733, + 0.18022385464638982, + 0.13225345772652317, + 0.09056104469881483, + 0.116102740434082, + -0.013122432128587908, + 0.2657627160687447, + 0.4056038369656076, + 0.31093826015396175, + 0.22659768839029004, + 0.2704643275978704, + 0.11102500301810933, + 0.13019195575901932, + 0.15284647382171507, + 0.14242057473050368, + 0.15609777256012372, + 0.18606167222962325, + 0.1975950767618286, + 0.15393347866911716, + 0.42929817227573625, + 0.12385162205529171, + 0.09039309083254458, + 0.3983703817440798, + 0.1407634524784882, + 0.18059266439304544, + 0.106299751651337, + 0.1850646653831906, + 0.5427926804547445, + 0.7405440711173623, + 0.19721230781548232, + 0.213578111220431, + 0.2703968946110471, + 0.23044847737057342, + 0.12568486093013162, + 0.22222213094347215, + 0.17536230429323635, + 0.09459064914479369, + 0.057088170145456904, + 0.2965546845163136, + 0.12217059283065868, + 0.1137505492660963, + 0.11700019415191068, + 0.076889710700808, + 0.23700771788223995, + 0.12718663388631476, + 0.12738212526093512, + 0.20123267931904304, + 0.24959645460335836, + 0.051731524085920036, + 0.02651079995011497, + 0.0966496827365664, + 0.2662196895268104, + 0.09319669687925758, + 0.21716558176088277, + 0.18319965800527285, + 0.04238825315941604, + 0.0570165278028179, + 0.3507053013846224, + 0.10571915664469107, + 0.14065479313784648, + 0.18697558076151738, + 0.1477504378223727, + 0.11103608762166976, + 0.5120977082331277, + 0.06725116478759856, + 0.0400876345258565, + 0.238372941591097, + 0.16063855097409183, + 0.20831513133700214, + 0.04519989462914627, + 0.12654159141763216, + 0.16795968484368742, + 0.12324709799719677, + 0.025229286843373784, + 0.1504312192501685, + 0.10135204797457245, + 0.1455055633762256, + 0.21984795595507042, + 0.12026877464725276, + 0.30944020751119705, + 0.2880275929454476, + 0.129024467882378, + 0.10864479524545564, + 0.13897339384093324, + 0.06428635328026619, + 0.2260966598446269, + 0.2628742629911438, + 0.3448732105640458, + 0.24884598410305125, + 0.15060359075696456, + 0.2841685579976886, + 0.20209920216782884, + 0.2434560971717521, + 0.22200769372539617, + 0.35764623731358963, + 0.17001483737497722, + 0.1338781677363543, + 0.37717928417804486, + 0.12816999925080336, + 0.20925857012342122, + 0.1960704371989009, + 0.06571160084788459, + 0.36553669393353905, + 0.0658205538377491, + -0.03268638015348395, + 0.09649886046166253, + 0.03863741003321039, + 0.20296576176490272, + 0.1564693388015163, + 0.07012343533339953, + 0.240192065695797, + 0.45694685154559705, + 0.13433732531801118, + 0.30806543654665175, + 0.735343473217233, + 0.21145836732273762, + 0.0849597701584015, + 0.2895116742731238, + 0.3210759008907088, + 0.223162007281131, + 0.07495212882549378, + 0.47018641842403386, + 0.13758953393460444, + 0.1016583529162348, + 0.313707387850061, + 0.15649541436034348, + 0.22938051548553032, + 0.42495657187661534, + 0.22524968173877757, + 0.43796145798455044, + 0.1132655277010874, + 0.15126182868129623, + 0.36944818192651224, + 0.3639000727264108, + -0.02791910987725607, + 0.022356945963551, + 0.033137620056395076, + 0.462869849057855, + 0.2344781854093198, + 0.7745157554496966, + 0.3743013069502334, + 0.09433112762127618, + 0.1491514067144256, + 0.3529216451727987, + 0.3583092355607623, + 0.5422723289909912, + 0.10243362427751089, + 0.402942759382233, + 0.09166706669435289, + 0.15388126680160089, + 0.021565809455565684, + 0.08244670945796828, + 0.18804669774832758, + 0.020640347606940154, + 0.06833411329316137, + 0.3636174902430816, + 0.20142153251983905, + 0.03055116392583074, + 0.07370478717893464, + 0.1648431813108761, + 0.033382203013259955, + 0.2224255272977705, + 0.2190468985708073, + 0.0888186699536904, + 0.15138030724497734, + 0.08452502204746364, + 0.46792399473870666, + 0.30401207160599547, + 0.11026083115185344, + 0.5207293863994127, + 0.17773292572915822, + 0.11116067447569633, + 0.06354102148788876, + 0.026849159486773265, + 0.05016301066083, + 0.10205759910502389, + 0.24146366785190748, + 0.1643181684387267, + 0.6397967352236696, + 0.10641578345311989, + 0.05229388675686789, + 0.07344910935016888, + 0.06213939857380882, + 0.3928715143446278, + 0.11924720515505131, + 0.11608779203396799, + 0.12134869356904762, + 0.07394183077101968, + 0.042371665406509065, + 0.2673865222734536, + 0.10712080866494572, + 0.28239703775938085, + 0.7128019992277634, + 0.21704341133595598, + 0.09096831251585366, + 0.06079430897743229, + 0.15125418682276517, + 0.010328913137581988, + 0.06585222943458273, + 0.09880765999486829, + 0.7056669134943446, + 0.4552981467094128, + 0.10635210429068057, + 0.09466474109456556, + 0.1520866083896378, + 0.08022698245925929, + 0.07641535499034145, + 0.22629887985347918, + 0.20880276629897113, + 0.0031916069303544415, + 0.1187907894419551, + 0.09159940470133782, + 0.30652493116167584, + 0.2651472248693612, + 0.4327758509600161, + 0.2502682606832532, + 0.07957446459454859, + 0.3118817438941115, + 0.18786822073515067, + 0.42309144645747115, + 0.3106781870436742, + 0.10602977374550518, + 0.6666292119575901, + 0.6634239983744149, + 0.09378590620122326, + 0.10843585415241853, + 0.16376312998337172, + 0.1597865696381572, + 0.14870500011554041, + 0.1949391605204298, + 0.3428798797640975, + 0.3622630355212553, + 0.22856379616356576, + 0.2337225636152208, + 0.1741260292104065, + 0.12214202226273731, + 0.1695115477472056, + 0.18935068307398703, + 0.2435636601465135, + 0.20007504373091162, + 0.19467618660757896, + 0.24831530624197506, + 0.45442530340885295, + 0.2833193537677215, + 0.5482991493076516, + 0.33667371395143486, + 0.2174656851425974, + 0.20496197851623188, + 0.18373198071561422, + 0.13926499004019544, + 0.5304496793259929, + 0.5072042593234452, + 0.23842953637984413, + 0.150748788476306, + 0.2410845594769941, + 0.3198637896293493, + 0.17522268974045133, + 0.22161381091491397, + 0.21662689346165592, + 0.33726913311735757, + 0.23592257935460673, + 0.1820424739428784, + 0.19332863638122866, + 0.3580834403619291, + 0.21211974604914052, + 0.18944815646463328, + 0.18998223018288357, + 0.23015964615206605, + 0.07244242976592563, + 0.02061171193687005, + 0.07897635721400723, + 0.3152480814155235, + 0.18902823169308092, + 0.015550217673520272, + 0.17502099927973558, + 0.10930031528427901, + 0.1028464467532318, + 0.14188957276109362, + 0.20803166607762408, + 0.11931969508274982, + 0.19655195681430296, + 0.5239368406881416, + 0.15828451986463554, + 0.16604033713739588, + 0.4672359939294409, + 0.16685189239090756, + 0.34871401688364867, + 0.4384718121497041, + 0.19613030910083504, + 0.35002463125093014, + 0.21548775869126385, + 0.23043862698336687, + 0.11752871899635998, + 0.24095397667360216, + 0.17967461297659323, + 0.37080768336587366, + 0.21389161660901057, + 0.17184049750662475, + 0.31285999414293925, + 0.27619332104035216, + 0.2184997149709228, + 0.31787164332600093, + 0.36653223944652313, + 0.3014713498593121, + 0.260930372107945, + 0.2091771120873415, + 0.26530742700807514, + 0.1252313929444823, + 0.1815934481137461, + 0.2610210758520015, + 0.17704865558910066, + 0.21838761501183387, + 0.3686467631466575, + 0.15705134539693966, + 0.16412301417980066, + 0.3054188619585025, + 0.4272637453489143, + 0.20517130348599905, + 0.29224830208689145, + 0.13094761922412973, + 0.1633133493832384, + 0.16068927312856807, + 0.18929367969761493, + 0.1897551284004165, + 0.16718761669359566, + 0.1878993094595046, + 0.31480356566990225, + 0.32501828002209415, + 0.6706523648764166, + 0.46875297860530174, + 0.3833619469767086, + 0.46783785181266035, + 0.36128131094637167, + 0.34351629409124407, + 0.4631941097738808, + 0.19322100966653186, + 0.1302804677235915, + 0.3287094450233492, + 0.14741601163546475, + 0.31158104770307016, + 0.40409234506219105, + 0.3024070478977723, + 0.3549203560147541, + 0.21658574940576913, + 0.15778944153131327, + 0.20884852204222315, + 0.40497598999958306, + 0.16077265325770376, + 0.20189184052187525, + 0.1719235330062389, + 0.14811670664514023, + 0.18793788206277676, + 0.20943947058927004, + 0.19305338534956853, + 0.25070560468137587, + 0.18361588773443024, + 0.2900127802523154, + 0.11240140887668393, + 0.14186625416432982, + 0.3209324630205665, + 0.36031861293289436, + 0.31059566842840997, + 0.31385230739212405, + 0.16860565948024003, + 0.165590009965591, + 0.22843635244977145, + 0.18075108605962212, + 0.31677313582309663, + 0.2587953087131293, + 0.20404366270695418, + 0.22438962286915676, + 0.24462169876908219, + 0.22211074119205054, + 0.22059009486458034, + 0.3471118534875376, + 0.18438843693405754, + 0.08689838975780143, + 0.12734736672734492, + 0.3442161182296788, + 0.16463053244495265, + 0.19251374522951756, + 0.23198172970169662, + 0.2224007470373135, + 0.20108964917314628, + 0.23426914109139102, + 0.2395969479760665, + 0.1289359356460006, + 0.1707807623767553, + 0.04926867012330044, + 0.0372321827893411, + 0.030148563791151904, + 0.048646479017442554, + 0.05692458203978886, + 0.0873497236908959, + 0.08211349328349979, + 0.03552927748548861, + 0.0785803429867203, + 0.16057755404882532, + 0.09989520126237823, + 0.16078451797926374, + 0.18397122284007492, + 0.3045314807340291, + 0.1316122452996656, + 0.22987182213861315, + 0.1787918369231893, + 0.15148226686583777, + 0.15163681713182967, + 0.2925063210794814, + 0.20636396862719464, + 0.4623566144284156, + 0.10453736332950926, + 0.14984970716500814, + 0.21341027636011667, + 0.2471782335640134, + 0.2583658923795859, + 0.1827448846261616, + 0.1365505386676025, + 0.16328285677208154, + 0.278276634702001, + 0.26510640340276365, + 0.44423646292808283, + 0.21030743283468595, + 0.16162576072858534, + 0.12415903880563668, + 0.22411244943231223, + 0.16230276043501435, + 0.18869836525141565, + 0.2184338187249879, + 0.17358232914047092, + 0.32485808773710895, + 0.5096809306618894, + 0.17073385628068993, + 0.1393748790808871, + 0.3631829699104957, + 0.4829562932909644, + 0.32787782637416135, + 0.29623812738186434, + 0.16794287696249793, + 0.24412277095469134, + 0.18906316821770527, + 0.15520079231984316, + 0.19351199210427217, + 0.16385644685634845, + 0.22857286864109339, + 0.41635012941891497, + 0.15030966146470962, + 0.22764874831821033, + 0.32923637124845767, + 0.2632993380043742, + 0.2551210802976008, + 0.33098509914263236, + 0.06740399139160584, + 0.19442529797205196, + 0.1307187811925928, + 0.20479506245753654, + 0.17675288375197606, + 0.0984799869575092, + 0.24933715506975132, + 0.0934056535666371, + 0.1241305722604726, + 0.18769756799862197, + 0.1426030816079482, + 0.16102201493227966, + 0.13878425646284762, + 0.11150119513363965, + 0.1359874276467512, + 0.16450024810931949, + 0.1523928352418132, + 0.2605788061950057, + 0.13579917691776452, + 0.1842901333412722, + 0.20233090886796057, + 0.4933980355422876, + 0.164945395986994, + 0.10853221484742162, + 0.35005952463792894, + 0.24208709982907128, + 0.17452374582778227, + 0.1439206799365174, + 0.0840023322761849, + 0.09531200051025074, + 0.04169060696780488, + 0.20820017649765107, + 0.11004422727779736, + 0.5938889168079489, + 0.3138619613535375, + 0.14879352801277113, + 0.1852297237918476, + 0.25185185935682664, + 0.20821741348593575, + 0.26800613604835904, + 0.22823690849650086, + 0.2160246875646056, + 0.1684523642108589, + 0.29964181288090014, + 0.04556484745746279, + 0.3491291850563468, + 0.21082687998260105, + 0.11995998891096364, + 0.3815661359428986, + 0.3758900209487385, + 0.13297196935624378, + 0.22141704875734503, + 0.15248428030965727, + 0.2117760004075333, + 0.09675891075721411, + 0.4216374446530753, + 0.3288657471594154, + 0.4993686849894295, + 0.29065673946888615, + 0.2859085840221695, + 0.34436324964517095, + 0.3400032158479153, + 0.24705559841522434, + 0.16937278061488584, + 0.26879424597571044, + 0.18146007084784807, + 0.11171180150141506, + 0.2349145842759616, + 0.13901062215852156, + 0.29119326411266183, + 0.6537763214392746, + 0.1584889452832279, + 0.22747568498831355, + 0.13272964705632867, + 0.17256943513352127, + 0.04494168344461615, + 0.029362852639646744, + 0.13603101645356594, + 0.062118528257116064, + 0.2907030312857335, + 0.06949622854653491, + 0.16968268924468238, + 0.1825420919822085, + 0.28811353142253, + 0.2292274835861476, + 0.3138621183016801, + 0.1633708258255194, + 0.13596920229989665, + 0.39433517695131043, + 0.40092819340278507, + 0.15440966040761278, + 0.3095336730031909, + 0.20854587877147446, + 0.6522956672434652, + 0.12492297100514146, + 0.4788352944345264, + 0.6947495680236186, + 0.6719809996683761, + 0.3054886715933287, + 0.15872495436278192, + 0.025459516300435536, + 0.27430038817871844, + 0.30997974954705165, + 0.39024168079114363, + 0.4351445041449062, + 0.2922456631976607, + 0.46391230971710595, + 0.15220896027446407, + 0.17378844335278235, + 0.6314999011974336, + 0.6330931650022049, + 0.21768202912743437, + 0.2605381397868236, + 0.2855814367089099, + 0.2587273435447783, + 0.27489486158841586, + 0.5911966070506813, + 0.5729932052332096, + 0.2788379116012287, + 0.28020028296868477, + 0.293847432900617, + 0.07612639990501899, + 0.34137502022088706, + 0.079503363184966, + 0.2391303771737109, + 0.12541721001786463, + 0.20467250844283982, + 0.2955809642535557, + 0.16915048745379072, + 0.14459664473974948, + 0.37359770437811946, + 0.05441092519230993, + 0.2308665567431291, + 0.22510730182143698, + 0.34050200281975096, + 0.24656229341643118, + 0.16473785935934843, + 0.23120303386072655, + 0.10457415494098798, + 0.09897116446132903, + 0.11609574966239833, + 0.23077391327250513, + 0.2278986485259225, + 0.11141546837716582, + 0.2510220351163227, + 0.09978663535190235, + 0.25181503396918725, + 0.2059247223789632, + 0.3104251131510594, + 0.10879323484096, + 0.16484663181334658, + 0.30257286525936, + 0.09805169500696992, + 0.281373589524411, + 0.1715479498907439, + 0.24357099016861292, + 0.27340711938187096, + 0.17767525521213814, + 0.05451790588485433, + 0.07611748832880666, + 0.0931595733521663, + 0.3232415716926337, + 0.2548931812461559, + 0.34101020949594146, + 0.07935058796373447, + 0.1281926381063794, + 0.2368319280968488, + 0.18008890472467456, + 0.16406612431727, + 0.6108473524227103, + 0.19685904954998412, + 0.2575843889614724, + 0.12489892224149421, + 0.23926833959045224, + 0.23220347693838914, + 0.157603175663373, + 0.12522799614243374, + 0.19032166027931383, + 0.16475256897980645, + 0.2547633296658752, + 0.14889183491266023, + 0.1837344585957289, + 0.1426684728895754, + 0.1747497615863183, + 0.1353861949002277, + 0.23244291576638804, + 0.13270035440287495, + 0.21473568204076607, + 0.05630023863361924, + 0.043082389697125904, + 0.039452890087395664, + 0.20612338650237125, + 0.27456307361452165, + 0.22704269562568857, + 0.18719329651049832, + 0.2694482801590195, + 0.2729806449197155, + 0.3585360028174801, + 0.2889646113751727, + 0.1731781480679381, + 0.2651546574220181, + 0.14107702152850585, + 0.24722094237141784, + 0.10882688554255626, + 0.5938662000464753, + 0.19966927045665012, + 0.11379026351120981, + 0.37703085031950245, + 0.30013101704471745, + 0.16614046058482715, + 0.21457635849677525, + 0.1787175551795019, + 0.7267122635287422, + 0.14166196425481337, + 0.29324488810516297, + 0.34961747832754875, + 0.3566469132641749, + 0.5921038016877517, + 0.04181950368482305, + 0.2528985436680851, + 0.29365218360844575, + 0.13967558351382922, + 0.10436589744007109, + 0.31069784839709724, + 0.13906500844350397, + 0.6527448936560426 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.22301927877624753, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.2211625544215715, + 0.6822995886989826, + 0.6638095489120472, + 0.7684182918204894, + 0.616365386429294, + 0.4952985676923958, + 0.4916042852190903, + 0.5208006621126912, + 0.3515087805272557, + 0.5786327893682152, + 0.17268656775598062, + 0.2076201116165631, + 0.10578830617280381, + 0.09696300350289176, + 0.056494344600641025, + 0.07631858921113578, + 0.12372976234356757, + 0.24395280472808986, + 0.1880528766635591, + 0.15785939721685804, + 0.08076973796656932, + 0.5854988428647373, + 0.1952726171659962, + 0.5371030279340587, + 0.1973513755919944, + 0.07316378260406345, + 0.1423443269919236, + 0.1519805054913642, + 0.19848102911480553, + 0.2491911230646518, + 0.1292097449854571, + 0.21421938332320795, + 0.13631835667608588, + 0.1814826423657201, + 0.23314734889943642, + 0.17650618355040323, + 0.11642968712823945, + 0.26672544344407023, + 0.2678941646529064, + 0.5758944439060646, + 0.7372519533825374, + 0.7286230556079346, + 0.7162157569730638, + 0.5325275243982024, + 0.11347863711351623, + 0.10324268124321447, + 0.7240110295320782, + 0.09922535049281998, + 0.14645926815132643, + 0.12232907073219439, + 0.21932408496708927, + 0.7242262113862505, + 0.744501816042314, + 0.16225431922611794, + 0.08111600980192643, + 0.10742620236742959, + 0.12349191826616952, + 0.14383957312572848, + 0.22898049785312197, + 0.2635003436522213, + 0.11829783453942566, + 0.12907837081852427, + 0.11295263220499673, + 0.332917345984357, + 0.10963974140458306, + 0.10071282695654249, + 0.24926835323760538, + 0.2766378402851419, + 0.13278144138750556, + 0.1765429191178496, + 0.17650877303306886, + 0.12731214882753378, + 0.23251057873585046, + 0.13393509618649238, + 0.11499798390513426, + 0.06224685699196827, + 0.10505493252997712, + 0.19980047006291599, + 0.25822465627474955, + 0.1893438157589516, + 0.19750731266295848, + 0.229172065832601, + 0.23031185283908182, + 0.17739103245150567, + 0.2819240147217147, + 0.21678674058578776, + 0.16970834038033292, + 0.0636773200887856, + 0.15969804302336477, + 0.16538571056838106, + 0.22071832641601877, + 0.139013483092927, + 0.11162282778866384, + 0.3561508589961902, + 0.1323471495219962, + 0.32058630671675553, + 0.16471813992326081, + 0.23641664652147643, + 0.17052849819137184, + 0.14062502927327655, + 0.313481485701119, + 0.17345118879656973, + 0.10668411300800677, + 0.28729295017070733, + 0.17276733499602814, + 0.19379586399542786, + 0.12257130483806358, + 0.2073402912812897, + 0.21862279616304228, + 0.7546749230827063, + 0.7435971514204759, + 0.7518188855456149, + 0.762731889352742, + 0.5201177375846014, + 0.6214515448357005, + 0.6560290101793792, + 0.6940611239752504, + 0.6156267571097294, + 0.5161602674219583, + 0.6074543696852264, + 0.5089209119406539, + 0.7174421141166274, + 0.6064489807021574, + 0.23124498153783088, + 0.21487390860750313, + 0.11700069372890114, + 0.10878846078843837, + 0.238782313038938, + 0.1191399465851539, + 0.08956026558737361, + 0.17257284044480403, + 0.24632488623396315, + 0.16675978159326155, + 0.19930560067949368, + 0.2026755800602642, + 0.1790520864754067, + 0.12874835907327514, + 0.11825774942994963, + 0.1607707311373255, + 0.15687796652989142, + 0.10855150625490484, + 0.17962426253263883, + 0.08760115602022124, + 0.5348633490029218, + 0.10536314164393146, + 0.47101941462832664, + 0.08906601310841379, + 0.3593390019781144, + 0.11256208674302733, + 0.5026987922800171, + 0.11309797782103935, + 0.40054248346873356, + 0.049110279168119025, + 0.16063076923421357, + 0.1048819973365816, + 0.20814750568186774, + 0.0911319030225526, + 0.045013536725642025, + 0.2248129738255197, + 0.19810455568243904, + 0.15238578434989797, + 0.1028000759538518, + 0.21007865956777885, + 0.15427482748383367, + 0.23363272580774577, + 0.196948482222041, + 0.16752806506979562, + 0.1364139087012814, + 0.0936881907737778, + 0.1287002247812244, + 0.1646089234771587, + 0.07455971834788822, + 0.23714204890493584, + 0.11175036451906963, + 0.09713255025186736, + 0.11248525651174102, + 0.25490803174342624, + 0.26225832885867656, + 0.47647624024643603, + 0.18360468737331037, + 0.12088351464664675, + 0.13820203097600287, + 0.15234043428657057, + 0.18565763241457067, + 0.124270313048505, + 0.17202041663677872, + 0.1342111907205966, + 0.1867592433358336, + 0.09982681471135162, + 0.0780358539632695, + 0.10425037147048972, + 0.05637243128415236, + 0.1779130031124555, + 0.11801668081619741, + 0.19974324421728448, + 0.12937455109590051, + 0.5132755439624703, + 0.14597470112561986, + 0.1123350061602566, + 0.32201659449315007, + 0.1688910679061752, + 0.09999307495081139, + 0.09954287236829831, + 0.2049231939157952, + 0.5288417715967372, + 0.12378195284912212, + 0.11419521648972687, + 0.2680030670201499, + 0.1694934198721951, + 0.22438386708991925, + 0.2517567861575864, + 0.2377421929100411, + 0.10575437505964237, + 0.15718906730601673, + 0.19989724348582438, + 0.15672934267360006, + 0.14680389978267946, + 0.24805923055740126, + 0.14557382105605468, + 0.200581352312687, + 0.26794335933176183, + 0.3352937521420119, + 0.18749866822024802, + 0.21630379378054923, + 0.18951833663773654, + 0.09982501033119752, + 0.37496547557914256, + 0.2561019445087846, + 0.181429615805911, + 0.3537329306020395, + 0.09083871854096555, + 0.13354369942418956, + 0.2682606738828322, + 0.26620325604767864, + 0.12376643695603787, + 0.2832483109502599, + 0.2269736598894032, + 0.16122204342521132, + 0.18311615139533152, + 0.3210233247959207, + 0.18359396474425405, + 0.30365976412535883, + 0.19278491709774667, + 0.16074015301024072, + 0.2538122016667077, + 0.23759509516157926, + 0.17575669342696823, + 0.20723274279058687, + 0.1954586911885643, + 0.17761132111601816, + 0.3334570253201773, + 0.17387164365779537, + 0.26012383350165474, + 0.2071921408238441, + 0.09093193554904092, + 0.11817867282756449, + 0.1723491615456937, + 0.13738188984314162, + 0.17599989149640297, + 0.08124281344555431, + 0.09214058982298261, + 0.2121187475364702, + 0.1480590516336121, + 0.2071534311888849, + 0.16262298972208492, + 0.0895318199435581, + 0.13549136925296257, + 0.1720248748759704, + 0.11826791131746056, + 0.11452540812826982, + 0.11849765645897156, + 0.16387350640242715, + 0.13858284799504428, + 0.2706474172078684, + 0.2542522050471107, + 0.2276467818721916, + 0.1107728276514458, + 0.2278329036308931, + 0.3794685301895339, + 0.08529051974125089, + 0.07533760208777575, + 0.24312134637450988, + 0.12053718268131212, + 0.09883740698235101, + 0.20992175435432667, + 0.09825144104378612, + 0.1660828820592115, + 0.1105459903913791, + 0.15961618761483232, + 0.1399470725556955, + 0.19023280010326285, + 0.07391935166845728, + 0.0839363923027399, + 0.25850118260683025, + 0.2205558075975373, + 0.32316748019745944, + 0.33865407844437934, + 0.35373108003193177, + 0.5852041380953259, + 0.22886543756181604, + 0.45778040056223174, + 0.5740973372134411, + 0.5741608213311975, + 0.5492655844670737, + 0.38147619069543615, + 0.39737017899052074, + 0.21464597270212613, + 0.6014341678523566, + 0.3443838306025957, + 0.3631770764135608, + 0.14970024050185035, + 0.298503146210551, + 0.290284797511708, + 0.15558583824422045, + 0.21897170014328346, + 0.18739296003209177, + 0.32222771181763976, + 0.2219720791467671, + 0.25841143680363693, + 0.2394470736732632, + 0.28225701358611716, + 0.2954687726602382, + 0.20402739925861121, + 0.17213215462770468, + 0.20069579790921363, + 0.28703297910942405, + 0.1948061465788665, + 0.21938142193141363, + 0.36953181319654166, + 0.04153294026568016, + 0.14199258393929773, + 0.10703268108121684, + 0.16279287177406046, + 0.19201737398082686, + 0.21593589663510276, + -0.007572727241227233, + 0.07729423329747233, + 0.1667239649600017, + 0.11217750362778225, + 0.2152643285663823, + 0.24428388677317223, + 0.25289562107455527, + 0.198374530212173, + 0.17373237368060845, + 0.08511871481457557, + 0.1791548997868367, + 0.266098877278779, + 0.07017236961311098, + 0.7653363344964383, + 0.4326425232344463, + 0.3112219461752401, + 0.2897392165868991, + 0.32378270312831176, + 0.35058027072522585, + 0.25250112435593103, + 0.3761973369066009, + 0.23406127281678935, + 0.35268105932912114, + 0.3003324834776663, + 0.3471663060324272, + 0.3699451532565887, + 0.3413944949510497, + 0.31139366921161266, + 0.20542728574557012, + 0.24578650944983516, + 0.3771751176906512, + 0.43028769987434723, + 0.31094055824014877, + 0.44568587524225745, + 0.3233356069339178, + 0.42997796862260174, + 0.2767290501953954, + 0.29453907786647937, + 0.3433198845870119, + 0.17842053900995677, + 0.1780739433711323, + 0.10593384318179147, + 0.17179063508686818, + 0.1825012721153695, + 0.2448410500836054, + 0.09196962811902173, + 0.09474583155621051, + 0.21345039362243065, + 0.07658432306502604, + 0.2145524123980612, + 0.2070077423029543, + 0.15518271689881535, + 0.10989629950525191, + 0.165611862030093, + 0.13761329528306945, + 0.17939584713626622, + 0.13856148240269375, + 0.12077126199943711, + 0.3055995524609017, + 0.26507262633243595, + 0.2902106928076062, + 0.32769063301869683, + 0.2707365570055821, + 0.1256872588857034, + 0.5349828984413562, + 0.2532742281973791, + 0.20804924805792369, + 0.13221684092541866, + 0.27631722155564226, + 0.19443963180310758, + 0.1663808298963323, + 0.13398630089496896, + 0.17877283498868438, + 0.1414217661494178, + 0.19924562586106986, + 0.016081210738205728, + 0.06891547583332341, + 0.08116322372112748, + 0.1253121105519205, + 0.2252797408138312, + 0.2396827926776194, + 0.07595522995351203, + 0.24698817036743875, + 0.113626349968062, + 0.15213568302345024, + 0.12540638032643947, + 0.1746269017306165, + 0.16504813339136085, + 0.05354497559777376, + 0.11249786310708648, + 0.7310064664672357, + 0.1939823972578179, + 0.13890601221192342, + 0.2092334827218565, + 0.16021467099030132, + 0.1606118296888482, + 0.16776809315230667, + 0.27618764341731267, + 0.19771984028459713, + 0.07466748491595128, + 0.1327140037865216, + 0.21417374735484826, + 0.18108077484567975, + 0.1869129485863801, + 0.24429352674097982, + 0.3367667500345849, + 0.34731326313540106, + 0.17004443207896341, + 0.34652171295122264, + 0.18595255294740626, + 0.17167954002019695, + 0.11648458647545529, + 0.3166760725477639, + 0.1436499002435682, + 0.18036584636076694, + 0.2133452619240538, + 0.09875451394428329, + 0.11627456684926045, + 0.28595215136156193, + 0.3810985685109827, + 0.16634948441965403, + 0.11550167446200911, + 0.18187031886021657, + 0.19468903971683293, + 0.3136204788475361, + 0.23957540183332576, + 0.46348023245366965, + 0.1379621424437251, + 0.17406715826590627, + 0.07700617009333924, + 0.24794265261403992, + 0.22741938283471125, + 0.13111664995527528, + 0.1396689963778281, + 0.16972761307895318, + 0.1813400867317059, + 0.09144519721341655, + 0.17164677640533754, + 0.19999868217147934, + 0.01673076900754847, + 0.08369777104652901, + 0.14843793125077823, + 0.12235217250742422, + 0.09928146146817049, + 0.11907557491217778, + 0.092047538947921, + 0.09754053412774111, + 0.3142320136142525, + 0.2537508398552454, + 0.10842334045092586, + 0.18674662171180967, + 0.15506233210624523, + 0.1637331892689122, + 0.16833218107871395, + 0.19216598135212015, + 0.26411012016912755, + 0.5865369850709524, + 0.5652236031810406, + 0.3118230401256234, + 0.4050793050543224, + 0.24325282653564193, + 0.37325695893508226, + 0.2594382697591829, + 0.14072189699606596, + 0.13285140665317263, + 0.42423822081110585, + 0.25656896526863177, + 0.2547706338634867, + 0.5357513098253165, + 0.5251438356613638, + 0.49532873004999356, + 0.5295395141513647, + 0.4338617782844239, + 0.3905042532564866, + 0.3307987418210864, + 0.2598602116845868, + 0.13576388835178269, + 0.14107527336042858, + 0.20875100733862392, + 0.1283734918471969, + 0.09448841427903332, + 0.13157767262651318, + 0.08103423324510531, + 0.18036224636833995, + 0.1458216371341595, + 0.18653007668682126, + 0.16889348585304076, + 0.37966235513007024, + 0.2023576675319486, + 0.2930021187358969, + 0.1809859977155715, + 0.09823267282107764, + 0.36877060829277397, + 0.10432299794968, + 0.10814029281270414, + 0.18671585926515488, + 0.12019162189805654, + 0.14111773826945442, + 0.08690749360814883, + 0.1467230026942393, + 0.2327962742931613, + 0.12310980510476358, + 0.1139109745390722, + 0.1541260439289241, + 0.05489975897207848, + 0.25005581030481205, + 0.212246930601874, + 0.18167806858558114, + 0.12851217473746526, + 0.12743802164777965, + 0.1760295267818885, + 0.11196214448787081, + 0.18348497007414563, + 0.3415661837766944, + 0.13108357779464547, + 0.20956588892028477, + 0.21905639547367803, + 0.12559685313438015, + 0.22001095417625416, + 0.18152604961148097, + 0.15839044330799218, + 0.2525542551911841, + 0.2729343477370414, + 0.22650928101980608, + 0.15332239865337055, + 0.19499578224224567, + 0.04043994935972145, + 0.11590453326191522, + 0.2058506829234867, + 0.15814375235141234, + 0.18009235678252786, + 0.16342443649147484, + 0.21728912613703513, + 0.1646184734531446, + 0.18668779823457327, + 0.12920479500542337, + 0.17061138130584755, + 0.15828970839497558, + 0.0829621390509537, + 0.1547300633001313, + 0.08462083326177654, + 0.14348846211319546, + 0.0823875015841896, + 0.1582639879242575, + 0.20538316735900072, + 0.07983201999787332, + 0.12094679151121182, + 0.05415393954945022, + 0.10366556996381372, + 0.1548827922088058, + 0.2941749865131873, + 0.1632933980314527, + 0.4470137344221425, + 0.16057228564253703, + 0.2417370441670535, + 0.5104296252618723, + 0.19931526181252412, + 0.1584669525420385, + 0.14760507920541643, + 0.718350972511512, + 0.24770822604603487, + 0.15479064623122815, + 0.23961969319502432, + 0.192597800238443, + 0.18182547669733698, + 0.15477279705532812, + 0.18349561053464916, + 0.19427092112055028, + 0.17271465146547255, + 0.19796387060096415, + 0.19834705571394717, + 0.009851540114260336, + 0.039941127625312205, + 0.08232642460424466, + 0.08213365237329674, + 0.0745165411392614, + 0.24501231041362928, + 0.18310642517731232, + 0.09256730726804302, + 0.23402041244445168, + 0.17996377850787995, + 0.09520293736064857, + 0.15221875178141464, + 0.09657658535055373, + 0.21820467740162414, + 0.26958048233523496, + 0.18916823755481002, + 0.18228642744974666, + 0.15425017480369316, + 0.17677010270449098, + 0.12477918945344013, + 0.17140237905983463, + 0.3217975947825578, + 0.29134265524599356, + 0.19955324369237915, + 0.14098581550699232, + 0.13055012756524412, + 0.1469935966347677, + 0.14323296521651055, + 0.163344691033622, + 0.1592388179902929, + 0.15977188155533248, + 0.14555042842136104, + 0.16491621308958593, + 0.23097276130266894, + 0.23613634209275355, + 0.24665869339006036, + 0.25896678886145486, + 0.3666993419181821, + 0.23819299867084906, + 0.17458380880155674, + 0.1077325457742412, + 0.11216701118162706, + 0.1339451447017967, + 0.17347344134524031, + 0.16219746961084053, + 0.10950768116838029, + 0.6089297695158388, + 0.6201325638051082, + 0.5110669880106038, + 0.5122070280140596, + 0.5202391581581605, + 0.5342818230991466, + 0.12750194540490506, + 0.16324540743993854, + 0.23253262739958994, + 0.163308950076431, + 0.1994582025810201, + 0.16479788342593263, + 0.15141390075811637, + 0.16541191439832492, + 0.15624715313328955, + 0.15654154679920537, + 0.10545686546814116, + 0.19005121530252864, + 0.3830314848138822, + 0.22337301034411008, + 0.17578940313438823, + 0.18019448406138214, + 0.1452920767080415, + 0.197890389113298, + 0.1691403858105726, + 0.16685695487893962, + 0.15486344804579283, + 0.20511487599722297, + 0.12715572990696797, + 0.14201203753950314, + 0.10180033761446775, + 0.11378320509722446, + 0.11128354894263005, + 0.08953530426845473, + 0.45268956822504297, + 0.4040924488871007, + 0.44261360091292207, + 0.2449902923509381, + 0.35770864498386234, + 0.17610390660676692, + 0.1269902215967462, + 0.2622240645579949, + 0.09347970002686509, + 0.029148580931946076, + 0.2373004477497235, + 0.12337649794783904, + 0.18539381903996371, + 0.2750538229589148, + 0.11955611218556789, + 0.11334984089326046, + 0.16425876424145083, + 0.14179675460077215, + 0.14672818903048407, + 0.09065530320099616, + 0.08760067229132089, + 0.12059087693046545, + 0.19997480307913124, + 0.45700436323292865, + 0.43000029042429794, + 0.3561064283032245, + 0.4251985300768936, + 0.304978142588715, + 0.30841468072904443, + 0.4151248115848732, + 0.1582060370557371, + 0.26111125938397606, + 0.3048037974101514, + 0.37032329324646795, + 0.4724711822999238, + 0.04974173612222475, + 0.23150343139003549, + 0.15318584616891545, + 0.049933940672113014, + 0.048792935874927994, + 0.17160269574905154, + 0.16641757417351927, + 0.2830833533635608, + 0.6627100338661097, + 0.5541559711810012, + 0.6571352284564584, + 0.6905005987116555, + 0.5703722857742369, + 0.6273937201624076, + 0.625209654584257, + 0.12528998781784853, + 0.16119272159178324, + 0.1511340452236186, + 0.15610305430936577, + 0.08548462605685306, + 0.16767061513579415, + 0.3088195281032765, + 0.272207350100223, + 0.20230089543225294, + 0.2502039567741763, + 0.18859257982288138, + 0.36921706694853557, + 0.1621612260177656, + 0.27975323014398157, + 0.30179315583952426, + 0.15911712945650502, + 0.2279615820944684, + 0.26396527138433856, + 0.19091319112452312, + 0.14889453263684282, + 0.26242322618388636, + 0.07126014506482253, + 0.3934359241019338, + 0.3671413727913437, + 0.1838066913120205, + 0.04656505271470767, + 0.13025854140700788, + 0.42772218530883926, + 0.501853636677344, + 0.5843526261565153, + 0.21802526727900542, + 0.25265404232303745, + 0.07216932399318712, + 0.21431531685199828, + 0.23176944421003595, + 0.20628561821296443, + 0.28774994345203814, + 0.24331355132904756, + 0.27680584414250886, + 0.16273495339751096, + 0.1782790538194199, + 0.21571321307702462, + 0.19411227816324547, + 0.22296823217867995, + 0.18222075154625247, + 0.23240165467284746, + 0.09177701545694693, + 0.08034393425089918, + 0.1443548928633493, + 0.1584046789405146, + 0.17721303491826604, + 0.15987796654944683, + 0.2104240919942733, + 0.3607927135197067, + 0.09935226399039503, + 0.4545781493647073, + 0.6519446839355906, + 0.3482764904219031, + 0.2455043559114264, + 0.18174850300376585, + 0.16098946938019124, + 0.1619589111235625, + 0.24836886756688545, + 0.3256653558731501, + 0.2479468083617667, + 0.2126372268474428, + 0.19674618122871018, + 0.2104273680311819, + 0.37964743839767884, + 0.49022510987046397, + 0.545365739457803, + 0.14604001384323012, + 0.4624260542277198, + 0.08142076330867917, + 0.12456477943189015, + 0.11486894887898012, + 0.00676409222954309, + 0.12292389587687036, + 0.14209981820401885, + 0.24773180708297338, + 0.16655134762707388, + 0.5464985463714652, + 0.6074625076475383, + 0.6929965330033179, + 0.5115787465119233, + 0.7013375645181309, + 0.05821771252409292, + 0.16151390758320627, + 0.13716871178270632, + 0.14694621953733816, + 0.17457761485609896, + 0.22697593176632355, + 0.20593401949448872, + 0.3201966583389772, + 0.2586129044672076, + 0.2042641887060165, + 0.2164964442358624, + 0.16905723038928025, + 0.10405814837884093, + 0.10283746593977813, + 0.011647638248939626, + 0.1392004885157868, + 0.06521680349401777, + 0.16804817823784668, + 0.28617545786784393, + 0.03544266740376331, + 0.036938662548779075, + 0.09217492413375811, + 0.13850969656823725, + 0.2242208434871685, + 0.19779333560145934, + 0.1829286604373828, + 0.15503834077802459, + 0.2793485874553614, + 0.055168181051325, + 0.10497410587861773, + 0.1665045392691882, + 0.24753439364906427, + 0.1996569265490436, + 0.3635425751479579, + 0.07036360090535221, + 0.15020650470574026, + 0.2892471331983903, + 0.17469098702747063, + 0.6802890684145646, + 0.09837010069027011, + 0.00952076139914548, + 0.26265646839052015, + 0.05352910516893445, + 0.1002292536509734, + 0.11682059638963789, + 0.12326606826580333, + 0.6052104763236695, + 0.49178078890217386, + 0.5711063396311298, + 0.12538553287123705, + 0.2306943094321812, + 0.12242411902812758, + 0.023642945825615724, + 0.03645325290386999, + 0.051585266346245204, + 0.17623993845668204, + 0.15721525976470135, + 0.1900143930121513, + 0.09993233454161528, + 0.09009626273864954, + 0.30160116490787786, + 0.1921074916831702, + 0.13169952972183815, + 0.6140744870450721, + 0.3853425661739021, + 0.6178899903130545, + 0.4966111980155956, + 0.14389099956532173, + 0.3164620124696712, + 0.2605510956338743, + 0.24905330355215125, + 0.24235650594633096, + 0.19419828455126562, + 0.25096454291555775, + 0.2373516213530793, + 0.20086615626769902, + 0.2434928429682178, + 0.12578221165446773, + 0.07189387664006171, + 0.007207885011578531, + 0.14857948888753736, + 0.5771465817792849, + 0.11685796848734734, + 0.5086588184199663, + 0.43452567704203443, + 0.27018181894903986, + 0.19198901316270925, + 0.16621724897737722, + 0.23032388815612706, + 0.17173986444396389, + 0.07051391633949088, + 0.04673082149040732, + 0.05634693402169852, + 0.056823910331567425, + 0.7008448934679093, + 0.1868039588782107, + 0.5220565045178063, + 0.6625032955260162, + 0.6312023158874699, + 0.13614273250714737, + 0.7350793505041844, + 0.4142969842818467, + 0.5972365397225733, + 0.2271504715935099, + 0.1883905047237257, + 0.06051883322669733, + -0.012718530498897245, + 0.14727671429741934, + 0.17262477744180893, + 0.307379967918767, + 0.13816201571614367, + 0.03871208944977386, + 0.045726354858812414, + 0.054699589244480346, + 0.17114325334354294, + 0.149768697227355, + 0.2696476997078776, + 0.10910219384039803, + 0.140207231985308, + 0.1474852490501283, + 0.28300777027501156, + 0.20158298833828725, + 0.23068603130698478, + 0.2702582659311795, + 0.3477681051024965, + 0.2655269904403309, + 0.11343960126154791, + 0.17388027120075292, + 0.4047404768524301, + 0.46362022459654173, + 0.4511881362301822, + 0.2540382964896509, + 0.4885904616541688, + 0.5365407453527292, + 0.1830012917727413, + 0.19502792959143203, + 0.23463548007336185, + 0.21733808968913368, + 0.18000294195549565, + 0.06563934237036963, + 0.03999376244774705, + 0.07580492658119868, + 0.15184466958269355, + 0.09068060519634376, + 0.20139225102514108, + 0.038039431365218555, + 0.047218509160301125, + 0.026151174520273445, + 0.07982602645779238, + 0.10829313443724152, + 0.03760768025267205, + 0.031899399250958284, + 0.09086202867558928, + 0.10686690658798507, + 0.09324495842111483, + 0.295450222106408, + 0.43095315991410665, + 0.566765764458591, + 0.11285153773512398, + 0.3526188306197413, + 0.28437012319781685, + 0.2638491852622632, + 0.2956535989554082, + 0.22728682125730915, + 0.08902733010363446, + 0.04473040680421728, + 0.1263179896500036, + 0.11413571367472646, + 0.5607745596878057, + 0.12872162211298335, + 0.15857524953130006, + 0.09430781257518175, + 0.09034030598669529, + 0.12053558753275316, + 0.2816993564591201, + 0.23706668172130174, + 0.4948685034005591, + 0.10927322987160586, + 0.5486964261711008, + 0.2679543227186297, + 0.26013612491680027, + 0.19824202546416275, + 0.2266783064446413, + 0.16636339461749053, + 0.28558200556926555, + 0.2664868955782097, + 0.22743097255366854, + 0.2690734419005951, + 0.2370127480505138, + 0.22177011296593427, + 0.14690113612950864, + 0.26518093824586014, + 0.08165192425803273, + 0.21258525598379419, + 0.18049583886719148, + 0.1738900087967063, + 0.20559019187705468, + 0.1722922317635893, + 0.24311120599901612, + 0.2554379968288769, + 0.24529257571447755, + 0.22067503549573703, + 0.26986801140780625, + 0.21042025438256357, + 0.22035853458951718, + 0.10808103454322839, + 0.23085362935637174, + 0.16645907493953807, + 0.1572006804390257, + 0.14947749301153243, + 0.22976653208742695, + 0.12956615334133437, + 0.20530775688534178, + 0.20761935943663595, + 0.24139884712915946, + 0.10803682956059046, + 0.07681228529110506, + 0.19834985189667387, + 0.17910645969611064, + 0.10207500023953563, + 0.06912030864219129, + 0.1319942881696275, + 0.1190175311838141, + 0.25173873554437787, + 0.30653089179904325, + 0.18504023199207464, + 0.1873658112407764, + 0.03623411808208444, + 0.18619470537735214, + 0.0708070456728378, + 0.35794784312350164, + 0.11862023286883104, + 0.0391907291595434, + 0.13736998150368354, + 0.13162618796779926, + 0.15100965989246223, + 0.11213495344080585, + 0.2609695418351309, + 0.06235489876279117, + 0.2699737049027246, + 0.26784345453815583, + 0.25843760356619944, + 0.15757693507560597, + 0.21731974561785913, + 0.03634151333909997, + 0.05501548245105846, + 0.12585793790689256, + 0.19466987984868428, + 0.10018933776944225, + 0.18845452307837307, + 0.21022731632143218, + 0.12427912505100946, + 0.031191992709670775, + 0.01603536643602482, + 0.11664817567871978, + 0.13526025818665702, + 0.25032834791072517, + 0.19004861438918225, + 0.06054606000671197, + 0.28975882603013514, + 0.2774846163165658, + 0.1854015679302633, + 0.23974723584134391, + 0.12671435441100135, + 0.10092100699237773, + 0.0519118746340991, + 0.09825296875141108, + 0.10399020943585943, + 0.17583649506376153, + 0.20660621513376515, + 0.18601759430948947, + 0.20925664151431517, + 0.20396231791847327, + 0.1351030674165099, + 0.18920793006731665, + 0.17029365454219028, + 0.2361978494932404, + 0.1657216713620176, + 0.14454262712607135, + 0.2780198556471326, + 0.22998126178756623, + 0.2673067182186612, + 0.3197648558221101, + 0.191912280642525, + 0.2444845691459216, + 0.15209257763160913, + 0.23967385212742484, + 0.13666582198437915, + 0.24546431820383918, + 0.1237838047647418, + 0.12723992001096407, + 0.16007493753972957, + 0.21007405572792567, + 0.13636815471212888 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.03042751969145321, + "valid_tracks": 4, + "per_track_pearson": [ + 0.033762409192988056, + 0.03092138727932287, + 0.03261121424520768, + 0.024415068048294228 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.02336527924451008, + "valid_tracks": 626, + "per_track_pearson": [ + 0.048146182108500156, + 0.04746576278650127, + 0.05000675276365983, + -0.0028559742323312205, + 0.04628251827723593, + 0.04694433728868593, + 0.027937118287237953, + 0.015187724971233413, + 0.03461361946039991, + 0.0406394066339295, + 0.04217449030561564, + 0.04678797372589702, + 0.042781741479506086, + 0.04590942264969394, + 0.05542360535257157, + 0.043384372293566995, + 0.04444764556074964, + -0.013138368940860866, + 0.049642359013530826, + 0.047829477433659354, + 0.049419340817523506, + 0.045845820808836285, + 0.03704299307997113, + -0.034850341954938704, + 0.0075253537299023445, + 0.017069383413656825, + 0.047002021653120764, + 0.04327762526694491, + 0.04660085766145259, + 0.05268975637782946, + 0.03761966876134176, + -0.015008002304943783, + -0.01757146002224514, + 0.049579276910473535, + -0.011989793888071502, + 0.044629716193592565, + -0.026215311489636038, + 0.05416185987128262, + 0.05597456468577934, + 0.05573820775507629, + 0.058474739527465104, + -0.002705749818109475, + -0.0063957171756475195, + -0.0029336943683969715, + 0.048901080760881316, + 0.020547970094504975, + -0.003569791472910438, + -0.013864362894131163, + -0.026064820873659036, + -0.017672401014946004, + 0.01793649662179397, + -0.014731094097178973, + -0.02147749907593442, + -0.009588496900275972, + -0.01669800019293046, + 0.04251811822218662, + 0.03937683932813061, + 0.04168335813855867, + 0.045526721132588405, + 0.04430975951022862, + 0.04272518599525705, + 0.04924891968151751, + 0.05084653144326443, + 0.05417210721476336, + 0.04236219342944881, + -0.015506121026844913, + 0.047997987643882724, + 0.008946702194567429, + -0.008562193310812252, + 0.035151418753093674, + 0.04511903967927455, + 0.045513564713039134, + 0.04067955259561364, + 0.04867683220780214, + 0.008226725812073789, + 0.04331711061805249, + 0.049901753573892066, + 0.04780887941107965, + 0.042072718449830765, + 0.048450166693675835, + 0.05103178036095408, + 0.04160474568131189, + 0.04567338196779828, + 0.05324942880611444, + 0.04870984003799136, + 0.039628355484642275, + 0.045042135371372666, + 0.05036839061391507, + 0.04639160037099352, + 0.04355188019995987, + 0.04293333841205862, + 0.04940596288905492, + 0.042776075569253294, + 0.05745464247896094, + 0.03552348767360728, + 0.04405102279753709, + 0.03407266917672546, + 0.04864510390592425, + 0.0409092526950236, + 0.045590530603080874, + -0.001352400206165588, + 0.0474337186508581, + 0.04417996533607974, + 0.05056678426422298, + 0.04527846630522941, + 0.0545774025400969, + -0.021036438516188912, + 0.043223187723619924, + 0.043979291306871525, + 0.04745049885417466, + 0.04709362293204566, + 0.0582497139960076, + 0.043165676556926935, + 0.011000828746155484, + 0.0349984561149331, + 0.05306062383213299, + 0.054468056716071585, + 0.04436239064267971, + 0.04342621258749901, + 0.055528075964588974, + 0.04014994456112384, + 0.007754335874519911, + 0.03712503262841865, + 0.04608296472149844, + -0.018589257662808074, + null, + 0.047515217324712436, + -0.02840913545855262, + 0.03315473323079459, + 0.04467634034629135, + -0.002853535942952351, + 0.0547248043653858, + 0.04228113596796158, + 0.022930738375951575, + 0.05875987379143195, + -0.0015812695165482358, + null, + 0.03362921533667275, + -0.013633714339506367, + 0.05289452526912782, + 0.04179516635006676, + 0.021564974940165533, + 0.047753605023600654, + 0.04045928418947319, + -0.02609006070958031, + 0.04543047520761463, + 0.01320133288596152, + 0.037816641267550646, + 0.0226530278800443, + 0.054855220489533646, + 0.009560653097090513, + -0.012097489258468357, + 0.04457184710027413, + 0.056903074313182686, + -0.02493481792909766, + 0.0013375823731816853, + -0.031197385328761467, + 0.053449032205457984, + 0.02378547974640151, + 0.033507823405587736, + 0.005742920357560683, + 0.04409790980102901, + 0.04804710905399783, + 0.05100830874986678, + 0.051289351377978645, + 0.052301438702102736, + 0.04486760463143886, + 0.047176255562683074, + 0.0064949467383391525, + 0.06305436751911152, + 0.0014466693817503907, + 0.037357569126744976, + 0.055424817781893704, + 0.05038942990661936, + 0.033618065036448305, + -0.021757587296406294, + 0.045785259668133724, + 0.039032484135096406, + -0.04646810430886781, + 0.059114850896133664, + 0.022319164894788815, + 0.01905624815445073, + 0.01590461318236849, + 0.04199912223523865, + 0.019641280362110974, + 0.00040201031283973985, + 0.0516866329892824, + 0.03747678287693635, + 0.04986457154783273, + 0.049288587508207866, + 0.0455796707711401, + 0.05068045347822049, + 0.04184868640986169, + 0.048481155964806885, + 0.05467139852396344, + 0.027826863424842967, + 0.04975801589893408, + 0.05740233659221924, + 0.04542383448685583, + 0.045722855479256815, + null, + 0.05016451556674603, + 0.040691311341504784, + 0.03668985057602165, + 0.053631506326281324, + 0.04435517698937344, + 0.04092344206865844, + 0.043105302847164524, + null, + null, + 0.03992210153488475, + 0.036956433755316834, + null, + 0.05634962408406299, + 0.052158139667815046, + -0.016178100064945352, + 0.054113459627144186, + 0.0437894473900699, + 0.04766163807271123, + 0.049029994175102176, + null, + 0.04390421586228071, + 0.04335795414143591, + null, + 0.018050414991615464, + null, + 0.043027882839537075, + 0.049998597271378245, + 0.04712468001425299, + -0.004266199878990128, + null, + 0.034542939436250614, + 0.047050279491503415, + 0.050829268663172895, + null, + 0.04386435823943586, + null, + 0.04577456319969724, + null, + 0.04464464438292199, + 0.04981775539710777, + null, + 0.033646379746823774, + 0.0532072301049399, + null, + null, + 0.055205145522833975, + 0.05202175001888527, + 0.045594992492066236, + 0.04510249149515024, + null, + 0.04446554387523822, + null, + 0.047456692367762704, + 0.051176139634638074, + null, + -0.017789122979381795, + 0.05128664714847972, + 0.05154981544440914, + null, + 0.05022241458036604, + 0.050339846957078944, + null, + 0.048235310353821316, + 0.05018394747082342, + 0.04855975748673812, + null, + null, + null, + null, + 0.04903638548620729, + null, + null, + 0.05091214632302724, + 0.03876010583570809, + 0.041935155282134616, + null, + 0.057065523514747925, + 0.048547047715617686, + null, + null, + 0.04658460485820869, + 0.03384427802742639, + 0.045198533401730974, + 0.056311128716723724, + 0.03665215942188137, + 0.05568663666901655, + 0.0016838649763201861, + 0.044746590688121896, + 0.030582634242245603, + 0.05741971525736631, + 0.04656452277491098, + 0.04160152125277026, + null, + 0.05728325164498816, + 0.03294935359996643, + 0.04458666600851369, + 0.03472165470187371, + 0.0021689416499286233, + 0.044878904475700046, + 0.052714163495235715, + 0.05088314890079558, + 0.05332205086526868, + 0.059383115288088074, + 0.04201247976849534, + 0.0526018760029425, + null, + 0.047642106949319814, + 0.04541935933512286, + 0.03901027821042002, + null, + 0.05006299440272899, + 0.018115268776133873, + null, + 0.04752042760165849, + null, + 0.05069588734309044, + 0.04735746734509157, + 0.0475213038404529, + 0.03880869677928892, + 0.04870258621666429, + 0.024412505856424935, + 0.05632469103344385, + null, + null, + 0.05125594967771574, + 0.04617399047065662, + null, + 0.055772733297067244, + null, + 0.04032596578581674, + 0.04577684891167427, + 0.05243625698235143, + 0.054120706833899954, + null, + null, + null, + null, + 0.05287053197666678, + null, + 0.042880515851779336, + 0.04721351801809108, + null, + null, + 0.04920267155638118, + 0.03857431124512648, + 0.047931614200286105, + null, + 0.0455800329578712, + 0.03543976132638858, + 0.05517389162664342, + null, + 0.04044377452319824, + null, + null, + 0.05006248836371626, + null, + 0.04717164385243796, + null, + null, + -0.0020050221093413398, + 0.054492470417379146, + 0.04861447970162655, + 0.05637042602923066, + 0.043798094489803846, + null, + 0.04665239026992378, + 0.016683550640952026, + 0.016246224291593427, + -0.0038706737523971756, + 0.0015628731885378886, + 0.018096395289315163, + 0.015780542178384874, + 0.0032084894001108443, + 0.008097080892896468, + 0.018960678710717462, + 0.014010586669514984, + 0.01841901973660258, + 0.01906681684346297, + 0.016385251570646985, + 0.012210088604762139, + -0.004255313422953219, + -0.0030126887811398593, + 0.017905134028515474, + 0.017891521971618588, + 0.02343679431493681, + 0.016742288469506334, + 0.012538546711127106, + 0.009495788365591588, + 0.0037912371061727696, + -0.010300516517563687, + 0.011423361933698983, + 0.017752746250910643, + 0.011487702642969135, + 0.017847247500930554, + -0.001891365738299464, + -0.0024321919191173415, + 0.002494944841373788, + 0.00012014910970856525, + -0.011331675977859045, + 0.005850219763628527, + -0.015129667685163842, + 0.005247209648459202, + -0.011754174729594363, + 0.01622892515411183, + 0.02016795257691773, + 0.008413978003372362, + -0.0006064732825266767, + -0.003106520249155104, + -0.009770089477749343, + 0.003105732517253737, + 0.009045065346487404, + -0.015749208510917125, + -0.005371150763027873, + -0.005174217618214321, + -0.004105231148760727, + -0.008789861025438982, + -0.007388753421287746, + -0.00980703451033066, + -0.002599856524909785, + 0.0013935847748886173, + -0.0058384395667889204, + 0.0011467182196933253, + 0.0056260281495744075, + 0.014399272359133946, + 0.0056132722410062825, + 0.014739965334538186, + 0.005433792959644639, + 0.013940756630818117, + 0.01587603931960202, + 0.022129733600469936, + 0.013861806502053011, + 0.020072069998221226, + 0.016576064238991037, + 0.005154272077027856, + 0.016687346414523543, + 0.00021479308261915803, + 0.016676979788472327, + 0.021239702548186284, + 0.006716248197622217, + 0.010542339603472445, + 0.00651302292977218, + 0.008010472457945955, + 0.01239527866098556, + 0.01926578894797101, + 0.010610632561265971, + 0.019784350210759524, + 0.008395635725815828, + 0.015914146691521712, + 0.023480391281635055, + 0.014044995257440702, + 0.013129735523474538, + 0.015901364839917054, + 0.01690506779894653, + 0.01719507771649352, + 0.010501673756153062, + 0.018359167751682146, + 0.009133640897548603, + 0.013366808079784881, + 0.01584304765792646, + 0.008155775968107687, + 0.009048221693775116, + 0.0092920340070641, + 0.013901638952483471, + 0.016466193025485007, + 0.010648364867158023, + 0.014035684710507038, + 0.01200511634094655, + 0.015015633099520113, + 0.018753922222578594, + 0.024023468840526332, + 0.02594060492872645, + 0.02110487227889515, + -0.002828283991141543, + 0.012356770260309412, + 0.006988006688133597, + 0.01726558815586281, + 0.010574313798600538, + 0.009615619474258317, + 0.010728805824373147, + 0.008883216336164955, + 0.014832801972634566, + 0.01837823328728379, + 0.009150749272911348, + 0.009824632630004883, + 0.007004270087602604, + 0.006403562097072523, + 0.015815364957316023, + 0.015836769976909967, + 0.01137763674541365, + 0.02186490254500329, + 0.012854470858569124, + null, + 0.020546887874031337, + 0.006929672151811714, + 0.008438621448997628, + 0.007868182980452266, + 0.016186833944901717, + 0.01202110303484519, + 0.012942500128830792, + 0.02351321565726362, + 0.016246191340326576, + 0.01921309161736683, + null, + 0.01348962557418284, + 0.016425383491232404, + 0.0072167457783961185, + 0.005155005427764476, + 0.008058489598185784, + 0.007906253012360516, + 0.015023943154799105, + 0.016985703730283897, + 0.014246153474802332, + 0.011059029984701771, + 0.015338902867705026, + 0.014146315205437857, + 0.012732201681164441, + 0.018600465905643168, + 0.011804556218962143, + 0.015292346269358941, + 0.012034630108775344, + 0.0014288075399540203, + 0.00019956288904560636, + 0.004192760940086264, + 0.024979581943660227, + 0.0165632071327803, + 0.015988098941618064, + 0.01141241540387974, + 0.011806159099244576, + 0.0192702222206944, + 0.008513515284138053, + 0.0019435011327831528, + 0.01776422968553407, + 0.020322583831894447, + 0.01232884806106916, + 0.010312394013517733, + 0.009049363439544248, + 0.008011877436322137, + 0.015659860066020204, + 0.016260549187802082, + 0.010939164115446624, + 0.002627821473832453, + 0.01542583080367139, + 0.01309039224004983, + 0.01400783493684817, + 0.014710035517781158, + 0.008924925334085626, + 0.008540423557488632, + 0.013117258315620661, + 0.018444853024440804, + 0.019606938070295385, + 0.016251897410711563, + 0.0030782884721211414, + 0.01898383262482527, + -0.0011767432911136064, + 0.013776357805657434, + 0.012653361752883638, + 0.013071438551452175, + 0.016151552969482313, + 0.017795686168359588, + 0.001230212007374856, + 0.00526602893668417, + 6.867677015134754e-06, + -0.0002734881268310755, + 0.009456854172517883, + 0.02401451924988305, + 0.020291585136857637, + null, + 0.030008814905795034, + 0.012709783158485359, + 0.00772448831891947, + 0.01255495560773047, + 0.02273275908592407, + 0.01558495367446462, + 0.006238949173942178, + null, + null, + 0.019786202166122487, + 0.011273378588815839, + null, + 0.014423866296440968, + 0.00865843850418622, + 0.023240945314740743, + 0.005733777039347393, + 0.004520104874199321, + 0.0017074743319127794, + 0.0187186830979776, + null, + 0.015173881639914902, + 0.006702478758741909, + null, + -0.0018883826399029215, + null, + 0.008284240263624187, + 0.02381089649827275, + 0.01292892595056521, + 0.0028190319463969405, + null, + 0.005944060584287783, + 0.010944868270477984, + 0.011604860868996328, + null, + -0.0019610883087889516, + null, + 0.008644060744096672, + null, + 0.013613402590232975, + 0.007703127902232808, + null, + 0.0057262071125145746, + 0.017624571338469175, + null, + null, + 0.02175750579396963, + 0.010876446071330748, + 0.013376422653432531, + 0.009942324343894485, + null, + -0.0013090608720531761, + null, + 0.011132224308955928, + 0.00658299804332931, + null, + 0.00961890221241189, + 0.008133306518496053, + 0.006404160585392851, + null, + 0.010692102932793797, + 0.007335991852209738, + null, + 0.018495249835753527, + 0.01490573706291157, + 0.01850285700212392, + null, + null, + null, + null, + 0.021382638818809484, + null, + null, + 0.009490692129056105, + 0.010722935265785735, + 0.01811681848076573, + null, + 0.024377938940344606, + 0.019854777530956644, + null, + null, + 0.01108638472245725, + 0.0011506239204072807, + 0.0095205336261315, + 0.009912102332037513, + -0.0005908407883359353, + 0.00537268344302339, + -0.005910752345179244, + 0.004243719031316344, + -0.0029704656817621253, + 0.000686699439766711, + 0.014115585439234956, + 0.00770635048639683, + null, + 0.005734053797364578, + 0.004719574614636144, + 0.015046877550264941, + 0.021645654384505747, + 0.011049338334564687, + 0.014753334486712774, + 0.00538447913918792, + 0.017365244479034845, + 0.011703612150179005, + -0.0015601206551855281, + 0.0018752968066564468, + -0.006451207842537733, + null, + 0.0039583161187323745, + 0.015777128609398908, + 0.020884302492585535, + null, + 0.013815293887625925, + 0.01782271092696214, + null, + 0.01122117159750491, + null, + 0.023192770416603524, + 0.007377751282453416, + 0.017788893195847075, + 0.008606373893665007, + 0.024256449830918913, + 0.004652176499439466, + -0.001265707565418493, + null, + null, + 0.016788419594878926, + 0.017990432994399085, + null, + 0.008954768381248963, + null, + 0.011201243922689879, + 0.01486162632576653, + 0.0016185222513620495, + 0.020850372795574435, + null, + null, + null, + null, + 0.0008639484130665537, + null, + 0.015265625556863918, + 0.01736744356918303, + null, + null, + 0.013359675128452715, + 0.007786172718312255, + 0.0015261955941957717, + null, + -0.006683988075605172, + -0.0006986640179038942, + 0.009795518346848426, + null, + 0.008567760077969222, + null, + null, + 0.005312534068011951, + null, + 0.006746159660650655, + null, + null, + 0.0035704524777388958, + 0.017230383458898206, + 0.02013509719230057, + 0.02336306820871809, + 0.02316547901314906, + null, + 0.016132751231935655 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852, + 0.9096262740656852 + ] + }, + "junctions": { + "mean_pearson": 0.004990454880980991, + "valid_tracks": 626, + "per_track_pearson": [ + -0.01652067236346904, + -0.012815731955812176, + -0.01945226416572215, + 0.01436786600211135, + 0.1372718691391953, + 0.053097708941068154, + -0.017237964294451347, + 0.1646605626656335, + -0.01301435563347865, + -0.016997608448087433, + 0.018380062261911963, + -0.01649535658604801, + -0.015279156208987718, + 0.0, + -0.008083164177663959, + 0.027446733398123816, + 0.09334258820015856, + 0.11260845094768794, + -0.007433102313053847, + -0.01830249805742475, + -0.028652949112032183, + 0.0699416172827279, + -0.015941413309981826, + -0.02490690772707213, + 0.0, + -0.01103657657537428, + 0.0, + 0.0, + -0.01620439117565145, + -0.004223436252753155, + 0.0, + -0.012701817814766607, + 0.0, + -0.009539120712262176, + -0.01975656135662994, + -0.010176757066687223, + -0.013081253440732356, + -0.017290442570960716, + 0.012075195626618275, + -0.0005950170331399062, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0027265882272793952, + 0.014641204969851482, + -0.01593720324324935, + -0.003094398623847711, + -0.014617229195506763, + -0.01515939918353802, + -0.015592173755293806, + -0.016315189720158867, + -0.014233688891429631, + 0.0, + 0.0, + -0.022517307693926902, + -0.0038148282772060976, + 0.0, + 0.0791809942468345, + 0.10365208949370468, + 0.021004999057849912, + 0.0, + 0.0, + -0.016420528218481625, + 0.0, + 0.0, + 0.022680807629283845, + 0.0, + -0.019463822987914377, + -0.020938934603610602, + -0.01309372186693302, + 0.0, + 0.0, + 0.0, + 0.0, + 0.2646147967958422, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.1292271575205834, + 0.0, + 0.09768202170352419, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.01887978455279389, + 0.0, + -0.009377542635705434, + -0.004150184837816546, + 0.0, + 0.0, + -0.017013743158729392, + -0.016452583521287995, + 0.0, + -0.006535812372352379, + 0.0030402730095198532, + 0.02576125509896979, + 0.013542743616827713, + -0.015615767077477994, + 0.0, + -0.015061483030342016, + 0.0, + 0.04290618709445581, + -0.015576100601485678, + 0.0, + 0.0, + 0.0, + 0.02880023839701447, + 0.016610391214233175, + -0.008056120563722582, + 0.0, + 0.16895169531518564, + 0.0, + 0.02865831563851586, + -0.019192728391767628, + 0.0, + 0.0, + -0.015813440380606544, + -0.001513008915441771, + null, + -0.015897391667885202, + 0.02491937576043207, + -0.011050846500295195, + -0.01640995760471947, + 0.02919656287897649, + -0.014876046265169904, + 0.0, + 0.0, + -0.019840418312454002, + 0.0, + null, + 0.0, + 0.0, + -0.0174365668051194, + -0.011417171897838831, + -0.020089393812830175, + 0.02820000902235233, + 0.25262051440221045, + 0.0, + 0.09121215401900837, + -0.005635435110044339, + 0.0, + 0.00850985990233798, + 0.0, + 0.17695405836402692, + 0.15248156858011944, + -0.01387973378916834, + -0.008373044254086392, + -0.009555064254286984, + -0.013091570651986035, + 0.005226290512553896, + 0.0, + 0.0670974166860918, + -0.014666423133760795, + 0.0, + 0.21098523831991078, + -0.011855724209967975, + 0.0, + -0.021134444235463276, + -0.022930661591501957, + 0.06544183631391601, + -0.013949392990219432, + 0.01953052891133132, + 0.0036224711783980285, + -0.01839240666478127, + 0.11644148903874854, + 0.0, + 0.013874004751624158, + 0.05060619001077736, + 0.0, + 0.09714600056464502, + -0.00728666351160128, + 0.040725326273878336, + -0.020693279403031166, + 0.01032530103349078, + 0.02325811343126411, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.043085468807976084, + 0.0, + 0.0, + 0.0, + 0.01717024144006456, + -0.01824530871189142, + -0.01997498315577223, + -0.02293662501494735, + -0.0011056389432563265, + -0.021737113280535045, + 0.0, + 0.0, + -0.015630680658334825, + null, + -0.01635140728973566, + 0.0026605131065248985, + 0.01387079020579909, + -0.009487260210033372, + -0.01280061972993607, + 0.0, + -0.009276094807117009, + null, + null, + 0.013863364463572529, + 0.06660831705962475, + null, + 0.003197427512420009, + -0.021096980913934946, + -0.006083603828494443, + -0.01972771628612021, + -0.01682721764424989, + -0.013991056953117139, + 0.022653057425447636, + null, + -0.015075014490087708, + 0.052367534638954875, + null, + 0.0337186374746007, + null, + 0.0, + 0.0, + -0.010525115338628202, + 0.13647774366334828, + null, + -0.016799107801874328, + -0.007176221888766036, + -0.008188425686967121, + null, + 0.023793698542174287, + null, + 0.037559289312991385, + null, + -0.015126121891152228, + 0.020682302700556165, + null, + 0.04842153514786641, + 0.013683763327727522, + null, + null, + -0.010701424306565854, + 0.027058606939642585, + -0.017778580274735816, + -0.0010256374262188452, + null, + 0.05968284247479773, + null, + 0.03488883607278179, + 0.009300152360023713, + null, + 0.0008788073668111802, + 0.02464630318961765, + 0.0, + null, + -0.012365049308060969, + -0.0611875437311227, + null, + 0.12309756831482198, + 0.054861229410913814, + 0.1412475681815802, + null, + null, + null, + null, + 0.040572374890089746, + null, + null, + 0.08405451114322877, + -0.06460426752787053, + 0.18608956330581836, + null, + 0.018825903070578208, + 0.06466548349004077, + null, + null, + -0.10589683584090022, + 0.027526484112786846, + 0.011883696907185659, + -0.015163202414489272, + -0.009832094092233115, + 0.021087247591896712, + -0.009473609866344478, + -0.01005105924659567, + -0.013993954479753178, + 0.0069426776174429565, + -0.0025465197964884326, + 0.005963963385308019, + null, + -0.00523856033227725, + 0.058510150256957466, + -0.015524458633021293, + 0.11325636391757848, + -0.015154142719926243, + -0.019360880031491987, + 0.0, + -0.024057386676942148, + -0.018170952086816433, + 0.0018349869727789451, + 0.027659323699823546, + -0.016537215020012475, + null, + 0.026261219235076825, + 0.014595891167006964, + 0.0060535316365530785, + null, + -0.03529859764630315, + 0.0, + null, + -0.02465367520005302, + null, + 0.011984465661048938, + -0.020498994114463927, + -0.03176350512626534, + 0.0, + 0.0, + -0.0021779635299021163, + 0.011186871057531229, + null, + null, + 0.1003003023654155, + -0.017828435804618504, + null, + -0.022731733191690185, + null, + -0.019737744291657335, + -0.008930657205141702, + 0.008677764816159783, + -0.0027080437163114093, + null, + null, + null, + null, + 0.03444858052316482, + null, + 0.0, + -0.01909714288457197, + null, + null, + 0.0, + -0.015098774714062067, + -0.014184254523909168, + null, + 0.04785605853397923, + 0.018717080082870707, + -0.011229727906003913, + null, + 0.22828155055710186, + null, + null, + 0.0, + null, + 0.009079674101051679, + null, + null, + 0.0, + 0.009188438646260792, + 0.0033515606383916164, + -0.0040358548093692975, + -0.00617099884960239, + null, + 0.08372939916134498, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.026045066780216793, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.026069879606733466, + -0.023605424832186174, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.02447048241159396, + 0.0, + 0.0, + 0.0, + -0.01557771812639211, + 0.0, + 0.0, + 0.0, + 0.019561792851122053, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.026639265013115473, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.021000380839769007, + 0.0, + 0.0, + 0.0, + -0.017325454642228785, + 0.0, + 0.0, + 0.0, + 0.0, + 0.02161781657239668, + 0.0, + -0.01866150486245094, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.02159178261845416, + -0.016271413376449273, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.019915870649302187, + 0.0, + 0.0, + 0.0, + -0.01986947369152813, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + -0.02224594366293003, + 0.0, + 0.0, + 0.0, + 0.0, + -0.01564973882299356, + 0.0, + -0.01817030457547575, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.018679820042246997, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.013995465936506413, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + -0.019849728205900093, + 0.0, + 0.0, + 0.0, + -0.02496703894806138, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + -0.025090097308382246, + 0.0, + 0.0, + -0.02294151067643491, + null, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + -0.02942081343283792, + null, + 0.0, + -0.0219266629824458, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + -0.02039975394857462, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + null, + 0.12180002387912846, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + 0.0, + -0.026706783184364223, + 0.0, + null, + null, + null, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + -0.019655696065620235, + 0.0, + 0.0, + 0.0, + -0.022736772757702223, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + -0.025315197757498558, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + null, + -0.027410471019220766, + null, + 0.0, + 0.0, + -0.01902202283503447, + 0.0, + 0.0, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + null, + 0.0, + null, + 0.0, + 0.0, + 0.028432134624542714, + 0.0, + null, + null, + null, + null, + 0.0, + null, + 0.0, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + 0.0, + 0.0, + null, + 0.0, + null, + null, + 0.0, + null, + 0.0, + null, + null, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + null, + 0.0 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9966887417218543, + 0.9967105263157895, + 0.9928825622775801, + 0.9936708860759493, + 0.9966887417218543, + 0.9966101694915255, + 0.9929577464788732, + 0.9861111111111112, + 0.9967948717948718, + 0.9964539007092199, + 0.9966329966329966, + 0.9966101694915255, + 0.9967532467532467, + 1.0, + 0.9933333333333333, + 0.993421052631579, + 0.9965034965034965, + 0.9967948717948718, + 0.9968152866242038, + 0.9967320261437909, + 0.9928057553956835, + 0.9935483870967742, + 0.9935064935064936, + 0.9925650557620818, + 1.0, + 0.9967105263157895, + 1.0, + 1.0, + 0.9932432432432432, + 0.993103448275862, + 1.0, + 0.9967637540453075, + 1.0, + 0.993920972644377, + 0.9926739926739927, + 0.9968454258675079, + 0.9961832061068703, + 0.9968652037617555, + 0.993006993006993, + 0.9932885906040269, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9934426229508196, + 0.9966216216216216, + 0.996309963099631, + 0.996415770609319, + 0.9964539007092199, + 0.9965986394557823, + 0.9963503649635036, + 0.9926470588235294, + 0.9962121212121212, + 1.0, + 1.0, + 0.9931972789115646, + 0.9930795847750865, + 1.0, + 0.9965753424657534, + 0.9964788732394366, + 0.996551724137931, + 1.0, + 1.0, + 0.9965870307167235, + 1.0, + 1.0, + 0.9966216216216216, + 1.0, + 0.9965870307167235, + 0.9967320261437909, + 0.99644128113879, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9932203389830508, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9967637540453075, + 1.0, + 0.9969040247678018, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9965635738831615, + 1.0, + 0.9931972789115646, + 0.9966216216216216, + 1.0, + 1.0, + 0.9965397923875432, + 0.9966887417218543, + 1.0, + 0.9966666666666667, + 0.9967213114754099, + 0.99644128113879, + 0.9968051118210862, + 0.9965870307167235, + 1.0, + 0.9962546816479401, + 1.0, + 0.9966777408637874, + 0.9966216216216216, + 1.0, + 1.0, + 1.0, + 0.9966101694915255, + 0.9965986394557823, + 0.9932432432432432, + 1.0, + 0.9966216216216216, + 1.0, + 0.9965870307167235, + 0.9964285714285714, + 1.0, + 1.0, + 0.9967105263157895, + 0.9959514170040485, + 1.0, + 0.9965397923875432, + 0.9930313588850174, + 0.9962121212121212, + 0.996268656716418, + 0.9963768115942029, + 0.996551724137931, + 1.0, + 1.0, + 0.9929824561403509, + 1.0, + 1.0, + 1.0, + 1.0, + 0.993421052631579, + 0.9927007299270073, + 0.996268656716418, + 0.9963235294117647, + 0.9896551724137931, + 1.0, + 0.9961832061068703, + 0.9960629921259843, + 1.0, + 0.9926470588235294, + 1.0, + 0.9962962962962963, + 0.9964539007092199, + 0.9963369963369964, + 0.9967532467532467, + 0.996551724137931, + 0.9965635738831615, + 0.9962121212121212, + 1.0, + 0.9964664310954063, + 0.9963503649635036, + 1.0, + 0.9930555555555556, + 0.9968652037617555, + 1.0, + 0.9962264150943396, + 0.9928057553956835, + 0.9965397923875432, + 0.9966887417218543, + 0.9962264150943396, + 0.9964912280701754, + 0.9963369963369964, + 0.9965870307167235, + 1.0, + 0.9931506849315068, + 0.9923076923076923, + 1.0, + 0.9963768115942029, + 0.9930313588850174, + 0.9926470588235294, + 0.9965034965034965, + 0.9965156794425087, + 0.996268656716418, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9966329966329966, + 1.0, + 1.0, + 1.0, + 0.9963235294117647, + 0.9964285714285714, + 0.9929577464788732, + 0.9930313588850174, + 0.9964664310954063, + 0.9926470588235294, + 1.0, + 1.0, + 0.9966666666666667, + 1.0, + 0.9967105263157895, + 0.9966101694915255, + 0.9933554817275747, + 0.9966216216216216, + 0.9967845659163987, + 1.0, + 0.993421052631579, + 1.0, + 0.9924242424242424, + 0.9933993399339934, + 0.9935064935064936, + 1.0, + 0.993103448275862, + 0.9963369963369964, + 0.9927007299270073, + 0.9965870307167235, + 0.9933774834437086, + 0.993103448275862, + 0.9932885906040269, + 1.0, + 0.9968944099378882, + 0.993127147766323, + 1.0, + 0.99609375, + 1.0, + 1.0, + 1.0, + 0.9964788732394366, + 0.9932659932659933, + 1.0, + 0.9963768115942029, + 0.9964788732394366, + 0.9929824561403509, + 1.0, + 0.9932659932659933, + 1.0, + 0.9935064935064936, + 1.0, + 0.9933333333333333, + 0.9966442953020134, + 1.0, + 0.9962825278810409, + 0.9934426229508196, + 1.0, + 0.9921875, + 0.9964539007092199, + 0.9967320261437909, + 0.9934640522875817, + 0.9932659932659933, + 1.0, + 0.9929328621908127, + 1.0, + 0.9935064935064936, + 0.9965635738831615, + 1.0, + 0.9961538461538462, + 0.9935691318327974, + 1.0, + 0.9927536231884058, + 0.9928571428571429, + 0.9966555183946488, + 1.0, + 0.9967213114754099, + 0.9929824561403509, + 0.9932203389830508, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9966777408637874, + 1.0, + 1.0, + 0.9965986394557823, + 0.9964539007092199, + 0.9967948717948718, + 1.0, + 0.9931740614334471, + 0.9965753424657534, + 1.0, + 1.0, + 0.9930313588850174, + 0.9963768115942029, + 0.9966555183946488, + 0.9932432432432432, + 0.9931740614334471, + 0.9965986394557823, + 0.9931740614334471, + 0.9930795847750865, + 0.9966101694915255, + 0.99644128113879, + 0.9968152866242038, + 0.996309963099631, + 1.0, + 0.9929824561403509, + 0.9932432432432432, + 0.9928571428571429, + 0.98989898989899, + 0.9965986394557823, + 0.9930555555555556, + 1.0, + 0.993103448275862, + 0.9934426229508196, + 0.9966216216216216, + 0.9963768115942029, + 0.9965870307167235, + 1.0, + 0.9967845659163987, + 0.9966216216216216, + 0.9967948717948718, + 1.0, + 0.9936102236421726, + 1.0, + 0.9917355371900827, + 0.9965635738831615, + 1.0, + 0.9967948717948718, + 0.9965277777777778, + 0.9933993399339934, + 1.0, + 1.0, + 0.9965753424657534, + 0.993006993006993, + 1.0, + 1.0, + 0.9934640522875817, + 0.9965986394557823, + 1.0, + 0.9963898916967509, + 1.0, + 0.9966777408637874, + 0.9899328859060402, + 0.9927272727272727, + 0.9933554817275747, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9931506849315068, + 1.0, + 1.0, + 0.9936507936507937, + 1.0, + 1.0, + 1.0, + 0.9900332225913622, + 0.9962962962962963, + 1.0, + 0.9965753424657534, + 0.9966329966329966, + 0.9929078014184397, + 1.0, + 0.9968152866242038, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9935275080906149, + 1.0, + 1.0, + 1.0, + 0.9967532467532467, + 0.9965635738831615, + 0.9965635738831615, + 0.9929577464788732, + 0.9922480620155039, + 0.9933774834437086, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.99644128113879, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9962406015037594, + 0.9964285714285714, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9965034965034965, + 1.0, + 1.0, + 1.0, + 0.9964664310954063, + 1.0, + 1.0, + 1.0, + 0.9962406015037594, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9964285714285714, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.996415770609319, + 1.0, + 1.0, + 1.0, + 0.9964664310954063, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9961977186311787, + 1.0, + 0.9964539007092199, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.99644128113879, + 0.996551724137931, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9964788732394366, + 1.0, + 1.0, + 1.0, + 0.996415770609319, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9964285714285714, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9964664310954063, + 1.0, + 0.9965156794425087, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.996415770609319, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9962962962962963, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9965034965034965, + 1.0, + 1.0, + 1.0, + 0.9963235294117647, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9914529914529915, + 1.0, + 1.0, + 1.0, + 0.9963369963369964, + 1.0, + 1.0, + 0.9964664310954063, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9921875, + 1.0, + 1.0, + 1.0, + 0.996268656716418, + 1.0, + 1.0, + 0.99644128113879, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.992, + 0.996551724137931, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.996309963099631, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9966555183946488, + 1.0, + 1.0, + 1.0, + 0.9920634920634921, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9966996699669967, + 1.0, + 1.0, + 1.0, + 0.9965986394557823, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.996415770609319, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9918032786885246, + 0.9967213114754099, + 1.0, + 1.0, + 1.0, + 0.9966216216216216, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9965034965034965, + 1.0, + 1.0, + 1.0, + 0.9915254237288136, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9917355371900827, + 1.0, + 0.9916666666666667, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ] + }, + "contact_maps": { + "mean_pearson": 0.0023942596311031295, + "valid_tracks": 28, + "per_track_pearson": [ + -0.005046063564099311, + 0.004014919997497826, + 0.011985718571993039, + 0.007626877751262779, + -0.004144083889837532, + 0.0011857805658123893, + 0.004587249293142865, + 0.0034925063189514684, + 0.01301076892846299, + 0.00453102936168211, + -0.01315944441917175, + 0.026218789504695854, + -0.02710731688841972, + 0.0016589796506984644, + -0.0077081741016192945, + 0.00453046477642573, + 0.0032809488886019946, + -0.03405841518399383, + -0.03680074781352524, + 0.06256720146919631, + 0.012323511211964356, + -0.026645680752843307, + 0.0007307280345499732, + 0.03476390529277079, + 0.023839040493714863, + 0.007299548078943807, + -0.004021008282731535, + -0.0019177636232384835 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.17171533163620495, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..a1f9b4860845edda89ece84512c41a2596127bd5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S003/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4398601030917572, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6803609732181162, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.1098670364215773, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.6716153991373395, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.19618952824625902, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.21548331119454472, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2709509046987323, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.15220958146626407, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.3468004299934636, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2618951286619208, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.4727432591017893, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.2523424968266569, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.14815618473839376, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/README.md new file mode 100644 index 0000000000000000000000000000000000000000..b4295d3b375db2b6ca257e8a05d1c53035af195a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/README.md @@ -0,0 +1,77 @@ +# SeqFunc local01 at step 14,000 + +One model with **13,256,456 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and four dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The global parent trained for the first 2,000 updates. The local path was attached with zero residual outputs at step 2,000, retaining every parent parameter and optimizer moment, then training continued. This is the matched control for the uniform splice-context fork. + +This checkpoint records **14,000 updates**, 0 complete training permutations and +14,000 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 6 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup updates. The splice objective is `balanced_splice`; +other family losses follow the reference. Recorded accumulated training-loop wall time is +35.3 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +## Measured validation results + +Family-mean Pearson: **0.255588** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.449029 | +| dnase | 0.581962 | +| procap | 0.331749 | +| cage | 0.180663 | +| rna_seq | 0.200977 | +| chip_tf | 0.366987 | +| chip_histone | 0.352559 | +| splice_sites | 0.039608 | +| splice_usage | 0.130403 | +| junctions | 0.158050 | +| contact_maps | 0.019487 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.3964 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6965 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.1779 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7404 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2576 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2616 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2667 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.1634 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.2635 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.1972 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5022 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1633 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1851 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..26648c6db0c00608abb429da526d4941f96c55f8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/package_manifest.json @@ -0,0 +1,30 @@ +{ + "parameters": 13256456, + "source_checkpoint_sha256": "af8060229a79fbcb3e5ce34ee559af3724268ea7dc8e321bd12ce9d5f2d27e3a", + "files": { + "README.md": "101f591999fec21fdb5f09cabdab1e58f3d50ac366bc30458b67c4ce27c6b5f2", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "fdef062e6c8de5ac6f73c764d638d6cd994fd001370628e19bf4365db4d57d86", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "2725592456c55c188b22c376f0cb249e9f46b4ae1370d2226df7143da358a8b2", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "2871dc8893879c3e14a526dea659701d72c8c4d0e2f835b0bfb857034a2c2e6d", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "a6653431e850d448c414f48cc65cbaaec8c17a430c08376b281441a6f6bce987", + "training_config.yaml": "deab298b32f0aef33e66ddf538541ec470eee0c05a6ebc983684f2575287d791", + "training_provenance.json": "1a721e47432865cf50c2ff1a14ac3cc97b2926da28dff75023e7d008c55a9793", + "validation_tracks.json": "c8a25d419cc2df7c7d834d49d89cf7fe9f2975374c0048dca7035b0c20dc4420", + "validation_variants.json": "41cd2d3e0548c520def53c102fbb8e99c5c7c473d807d62b0d5bc0376f7adbd4" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..52f3decd4fe2e9d275d08606b74575fafb701434 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/predict.py @@ -0,0 +1,130 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..45bdf927f635baa77ab98cc8fcefbb60196aa0ff --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/train_single.py @@ -0,0 +1,155 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f9a3298d69b138a02fed00880182d4004b23327 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_config.yaml @@ -0,0 +1,28 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..6c451660c4de8331481dbbcfae4140a9747cbd2d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/training_provenance.json @@ -0,0 +1,66 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 0, + "offset": 14000, + "global_step": 14000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/local01", + "config": "/workspace/research/experiments/local01/config.yaml", + "resume": "/workspace/research/experiments/local01/step-012000.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 82040, + "workers": 6, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 2118.901816125028, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..991411c0430b6faed3f332637dacbdc5a9471f3c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.4490286571976123, + "valid_tracks": 167, + "per_track_pearson": [ + 0.5787180877972778, + 0.5991515013347578, + 0.35903155111506063, + 0.5193434419972531, + 0.35044756915873515, + 0.3437029574320618, + 0.34756048117451777, + 0.42540369245724163, + 0.47264460256463814, + 0.5209497485899214, + 0.4566865328180873, + 0.3897322480759855, + 0.2561485999623867, + 0.392830801227994, + 0.5730450863697303, + 0.3510127947274521, + 0.4247439334333387, + 0.3651042845681107, + 0.41584548725817033, + 0.40511820355764716, + 0.3720545185131241, + 0.3325212433700937, + 0.3794904461265274, + 0.4044846289545441, + 0.34088904670038417, + 0.3603870422288395, + 0.34918863670743455, + 0.3768702600974143, + 0.37958507518094714, + 0.3583781273707457, + 0.36833146617752877, + 0.3617025207634615, + 0.35761873715031967, + 0.2741533776590353, + 0.24876364328297623, + 0.42769623410994234, + 0.3562879171581113, + 0.47614307417969715, + 0.376639689062023, + 0.34804106233208476, + 0.4277149421346815, + 0.3856446880541281, + 0.36778459764982724, + 0.32680388208399097, + 0.4280376208326986, + 0.3693404429328073, + 0.2926603388401998, + 0.3390525379984045, + 0.3304521292534768, + 0.3358901517979463, + 0.3713851031989818, + 0.291369195709947, + 0.3348811477661533, + 0.4803003333422318, + 0.31847388698793383, + 0.5912899866101075, + 0.7050618993818616, + 0.41812667456178626, + 0.5093735640672955, + 0.4923613400628523, + 0.6138636281722541, + 0.3802104692385244, + 0.5855885756970621, + 0.5077466577288013, + 0.5412959212081566, + 0.37189575851373, + 0.45222198294493865, + 0.5367696541026322, + 0.6912142911857361, + 0.6188357722216369, + 0.40233711547805434, + 0.3287030980669896, + 0.28094874045698837, + 0.40550597303509167, + 0.39073632248756107, + 0.32863185404207995, + 0.3811561431503719, + 0.4066075827817357, + 0.4452432749912775, + 0.34815329539110534, + 0.40084296529664154, + 0.4339697572272894, + 0.3873856762228455, + 0.3626044497999108, + 0.35597436467198007, + 0.35691025801573173, + 0.31157023766608916, + 0.3363073722433857, + 0.32725185680472013, + 0.38117448203272436, + 0.3785619052543757, + 0.36997174134220223, + 0.36887085848978596, + 0.3917450291685058, + 0.450417009706941, + 0.27823992557334126, + 0.3155222023942281, + 0.2865205523520331, + 0.3557071844095252, + 0.38369098043498384, + 0.3167693743076178, + 0.34850457431170995, + 0.3738786695139217, + 0.3476567489255353, + 0.3787980146067932, + 0.39027981093851505, + 0.41761359232677997, + 0.38930952598736385, + 0.3896811573729035, + 0.3995248451092658, + 0.4218735532286697, + 0.3907429704980889, + 0.37784360755086716, + 0.3291786675279102, + 0.3368474362051795, + 0.3253297906209802, + 0.34052251784212867, + 0.36649769866383974, + 0.36129299757781, + 0.34810792744995317, + 0.5597513165982163, + 0.5460021078014532, + 0.6166491144926074, + 0.6154810007630878, + 0.5936062740268738, + 0.4845877936460461, + 0.7594485368541238, + 0.5844817709147023, + 0.6462148166221638, + 0.5812911817194366, + 0.632850306600059, + 0.46883697426637144, + 0.5015798861716791, + 0.47046089043070954, + 0.6080698047092125, + 0.6036079099576245, + 0.594101353709923, + 0.4372612090895903, + 0.3486512844607223, + 0.7029645986118448, + 0.6724949108336272, + 0.6841615662215763, + 0.6807315847140819, + 0.5054787172408788, + 0.7486562053843839, + 0.6547517935104803, + 0.5915206499866476, + 0.614528586838432, + 0.751781939765598, + 0.6909652024203696, + 0.5131597884359982, + 0.5198862050104495, + 0.5892171495087771, + 0.6319374980887114, + 0.6325310639208899, + 0.5042657847525656, + 0.635891209770898, + 0.4969548421046694, + 0.5086896230933758, + 0.5687633347115012, + 0.4548732898190214, + 0.6459349885441865, + 0.6807243888551492, + 0.5374119316844675, + 0.5081594689823072, + 0.4644304163979905, + 0.6303742633770874 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5819615351028306, + "valid_tracks": 305, + "per_track_pearson": [ + 0.4825282257095405, + 0.8974002730271095, + 0.749502436550507, + 0.4376828905737947, + 0.3830108847807131, + 0.4011917044597347, + 0.9208451261320248, + 0.6605305955440451, + 0.3416811029927286, + 0.28836991163471576, + 0.8485526763001894, + 0.6038023779910493, + 0.36602324670046693, + 0.4265459626010102, + 0.40900693942762534, + 0.7651063811057393, + 0.7519769154488417, + 0.8332204280345484, + 0.8263847144749069, + 0.8625045914197149, + 0.5794703075127715, + 0.5264584426256417, + 0.8796988997172452, + 0.430024316985493, + 0.45169426677341956, + 0.7492616418313209, + 0.39794117721266475, + 0.5897319446644322, + 0.42695689272464155, + 0.41687136352670573, + 0.5378633995590119, + 0.36494340162206135, + 0.8193912303093721, + 0.6037896913063845, + 0.8621877852173455, + 0.868006048865225, + 0.38031008727732524, + 0.804974467551571, + 0.8595415256782285, + 0.839636078560268, + 0.7990619046330576, + 0.6486011012064055, + 0.6831959616030377, + 0.5724135856366416, + 0.3980411929821742, + 0.6125831260513943, + 0.709101914891812, + 0.4611053985979649, + 0.528908445444118, + 0.3534388752025239, + 0.3042490440347367, + 0.49085115944150476, + 0.42180052531311213, + 0.8115408180272357, + 0.3608402898220272, + 0.3530011233629133, + 0.37393230536441274, + 0.39943654147910485, + 0.36187222643460853, + 0.4756765347985436, + 0.35078859081876285, + 0.557778983641464, + 0.4079158899136556, + 0.3957195007289911, + 0.4205409972673051, + 0.4151902354997426, + 0.4809152303109889, + 0.39405658851441255, + 0.32139719319031956, + 0.3497692637241536, + 0.4626645268430315, + 0.5608185232442054, + 0.4009653811055363, + 0.5352429552841572, + 0.2697163684900562, + 0.7853876031667651, + 0.8763889964084774, + 0.5902369183294703, + 0.3703770391580457, + 0.3791189988896404, + 0.41262372144998855, + 0.39215867160166024, + 0.43542258121652205, + 0.3969605438950681, + 0.4500823139519885, + 0.4315752366134483, + 0.5073280798299201, + 0.4391698230072379, + 0.5766697665139416, + 0.36362211736165195, + 0.5554945690275738, + 0.3638244602266766, + 0.6882167564369275, + 0.40154776318336677, + 0.6762109002226742, + 0.4358105492636843, + 0.6669945785646737, + 0.6237110081208828, + 0.2772826710923412, + 0.2854085998453261, + 0.3953352527091865, + 0.31780930334875246, + 0.2992469187811878, + 0.6401247760649295, + 0.6106326455090128, + 0.3703223424316646, + 0.3746139243969257, + 0.5323749109225467, + 0.8207545387896096, + 0.48938704396874655, + 0.4617391091371984, + 0.5115634303575717, + 0.42531684570827305, + 0.39197477736864733, + 0.7470606562233543, + 0.44617492430176564, + 0.4310197547683894, + 0.4879071631069926, + 0.5521642783226393, + 0.3847794750278131, + 0.4367224035791354, + 0.44074015116077486, + 0.4678391920179805, + 0.43959001649687257, + 0.543224747178906, + 0.4692161697771991, + 0.5136852623687543, + 0.6226287706822446, + 0.6616290819631337, + 0.5096670887301237, + 0.40680118671622156, + 0.5364894233236762, + 0.561318510830251, + 0.7521441734790806, + 0.38484285026662457, + 0.2609455591640401, + 0.520143220972116, + 0.4212361548541905, + 0.5544291280405642, + 0.5307509435489854, + 0.5362868969200196, + 0.4926008174758058, + 0.3964291406120515, + 0.7453802516590226, + 0.5418364081035901, + 0.7332123650951285, + 0.33263267709281763, + 0.3087638978465614, + 0.354238155705568, + 0.2905529374974485, + 0.8758223823514306, + 0.2702505238802112, + 0.8712975666663519, + 0.6447457970927334, + 0.42922127550273714, + 0.2859311974467197, + 0.301149201263264, + 0.32434863233154293, + 0.588028515617278, + 0.6306189406015921, + 0.6518511899872177, + 0.5771761935321705, + 0.4716700758270473, + 0.40115692995463303, + 0.6298680345324424, + 0.5157496581086833, + 0.43262276231009755, + 0.5433599481220323, + 0.4058489321164665, + 0.40872752538001755, + 0.32190405500458374, + 0.8107064703661099, + 0.36266813151500615, + 0.39262460568425667, + 0.5780010116922265, + 0.6544291899886129, + 0.40553652602947965, + 0.28510648619361983, + 0.8604114902647961, + 0.32309484184841447, + 0.7294956874886884, + 0.35185610713116644, + 0.5407675980391072, + 0.6561697949847166, + 0.4183078817725991, + 0.48810770312054585, + 0.4954281216013944, + 0.7956235347477538, + 0.3825445094412282, + 0.5502851252617066, + 0.46418876074977494, + 0.6426401908661429, + 0.8782246187000978, + 0.3349289045601117, + 0.5172354668515198, + 0.389005516439018, + 0.5570787232650555, + 0.6525185865503367, + 0.5171434063550875, + 0.3776519671427676, + 0.8503010075723495, + 0.8545136561051346, + 0.7282160815159273, + 0.8075321896239172, + 0.820137102973549, + 0.8042696857715371, + 0.8213672588735322, + 0.30301355242251055, + 0.3116441232576851, + 0.442641904914012, + 0.39955458568704627, + 0.3766432104080838, + 0.5306345878864145, + 0.8370611724324138, + 0.8297941854618072, + 0.6042414085122075, + 0.8025407944605704, + 0.797648994719154, + 0.8356447185591828, + 0.8841053121711021, + 0.4125645341654173, + 0.46532936810218234, + 0.5841951533867318, + 0.478759072466552, + 0.7911849149370651, + 0.4805132183989691, + 0.8856755203216273, + 0.8619167431281788, + 0.853539195244854, + 0.8815102852278999, + 0.8748030691599222, + 0.8464547148955549, + 0.8074472402483761, + 0.840626002614182, + 0.8834613897983155, + 0.3636450372010752, + 0.9031581825763413, + 0.8586240778783243, + 0.795770890104613, + 0.5261554811942016, + 0.8543772291026758, + 0.4472505693047547, + 0.7309171508575679, + 0.8532853235119024, + 0.4901464990907652, + 0.48215313147945527, + 0.5907646053354426, + 0.4616415284885569, + 0.4495362997208625, + 0.6834057229015857, + 0.860884200016329, + 0.636378716389505, + 0.7801336905815686, + 0.5425612695873796, + 0.8353631579388088, + 0.7815268747828015, + 0.8140703014268283, + 0.8827472520583913, + 0.8161716676025444, + 0.8766063853136333, + 0.8428897115327111, + 0.44988177710939536, + 0.8543887884464569, + 0.45526122562019705, + 0.45170442614839257, + 0.8481124905579103, + 0.884790807778116, + 0.8537046836638795, + 0.47666552175511717, + 0.5351547259734077, + 0.6063543064136611, + 0.7712425755636855, + 0.8652601151236203, + 0.622833354906427, + 0.8791653472412693, + 0.8944876067151276, + 0.5268880471773332, + 0.2638787245601644, + 0.45831750335841154, + 0.8813107845146603, + 0.3469979093849518, + 0.5361003543908566, + 0.857029629195837, + 0.8067780331110332, + 0.5241892271435185, + 0.8454454838175337, + 0.8760246454438088, + 0.8698598676537322, + 0.8526629412852625, + 0.8442254703182639, + 0.8700521812933121, + 0.8321992161964104, + 0.9058244817144404, + 0.7615346757542177, + 0.8125793735792926, + 0.8926746094592726, + 0.4688173121030755, + 0.47393266168426157, + 0.44579911722408755, + 0.4369128952742375, + 0.8901378837216493, + 0.7710907290579766, + 0.8272097396737335, + 0.7647090047806934, + 0.7376401636751615 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.3317488363965557, + "valid_tracks": 12, + "per_track_pearson": [ + 0.3261400809112019, + 0.3664624313873139, + 0.2863933619906205, + 0.36805408643131776, + 0.3608426609211243, + 0.36114425738366096, + 0.3011709810607287, + 0.3458949865051905, + 0.24822084547387102, + 0.3413506196714335, + 0.3301116730000494, + 0.3452000520221563 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.18066326957284914, + "valid_tracks": 546, + "per_track_pearson": [ + 0.1726681761944187, + 0.20444998335759548, + 0.17469811260136006, + 0.10567393240201972, + 0.05553939589000169, + 0.18464254978983816, + 0.08906861816787504, + 0.1187930161472794, + 0.18910829587781872, + 0.15817858944837998, + 0.20932229940045124, + 0.1108015382956947, + 0.12545456244359876, + 0.10245816956487758, + 0.07807499832925704, + 0.14246649974704498, + 0.17910295772034357, + 0.21745483045547478, + 0.2019907426311949, + 0.20112589595978936, + 0.1718483670231368, + 0.1638043704196551, + 0.2713768177157652, + 0.13911661244362694, + 0.14889946509213833, + 0.18176556213447814, + 0.20504643935946387, + 0.21091067842304698, + 0.22720398762974886, + 0.21532012535314043, + 0.13391510891746758, + 0.22241296833270585, + 0.14729891235799655, + 0.11244986290465893, + 0.22924430639147284, + 0.222388108317822, + 0.20465821098343373, + 0.228724647222783, + 0.19803238418417551, + 0.26356390664803986, + 0.267648463816309, + 0.2338322670423025, + 0.20063604341193128, + 0.19590235635205708, + 0.12081994582664772, + 0.16692893413487392, + 0.17334272460472994, + 0.1796396597483245, + 0.10720083684977567, + 0.14027107776606867, + 0.18957231182253786, + 0.2281242963729715, + 0.23776135359270728, + 0.16048390469452387, + 0.2501381969368553, + 0.2746704965410695, + 0.14031098966781305, + 0.254479736779293, + 0.1911134659539861, + 0.2150192947457678, + 0.25443727033121866, + 0.1510732877822669, + 0.2301584734131527, + 0.1652721061815186, + 0.23216159855690477, + 0.248767280510286, + 0.26351900722930643, + 0.15536653070987863, + 0.0981757400278823, + 0.09299616153957774, + 0.20112210421360754, + 0.23909783013540822, + 0.1948022801867839, + 0.20512518544039582, + 0.2268108124598396, + 0.18695213061348942, + 0.2211006543680953, + 0.2044632293061416, + 0.22944708237540298, + 0.23614595398899174, + 0.19062339331242262, + 0.11029991513081877, + 0.20832341144854052, + 0.24216025182511716, + 0.11600922227985422, + 0.21643616073428212, + 0.17267195595902757, + 0.24483004503559788, + 0.1495255688461516, + 0.23147289061257242, + 0.16863611940297848, + 0.19256557430441448, + 0.21494019030072098, + 0.09480512946935642, + 0.22189149522596996, + 0.1905208635938321, + 0.20249422086676658, + 0.22100818438266914, + 0.2018523157962885, + 0.11814627813006566, + 0.18768812434402699, + 0.2114962025872281, + 0.21041579015844433, + 0.29296404931725983, + 0.24210438186636368, + 0.26181575676695407, + 0.16964523998017725, + 0.21888367556315494, + 0.1657612641926429, + 0.17450322701733734, + 0.18980610215834084, + 0.18837985227418108, + 0.21258985905906125, + 0.20915026171950346, + 0.16376921320381532, + 0.20662066662648504, + 0.21251200323126385, + 0.2166758783460428, + 0.20404850663419993, + 0.23487772869264267, + 0.24054962855503567, + 0.2341081289941437, + 0.14145648597364865, + 0.15149263951657624, + 0.1854377349325125, + 0.2273134651343163, + 0.20422868107086678, + 0.22496203563102715, + 0.2977503918962978, + 0.21366613634644718, + 0.24053021590886567, + 0.16303805734427315, + 0.22830354622852866, + 0.22595065358263638, + 0.27607549755716726, + 0.19176269477788205, + 0.24650652866431433, + 0.1710395928588895, + 0.10297150937129589, + 0.21297696018833237, + 0.25235521093190494, + 0.2148448015305251, + 0.22824722061128347, + 0.18106439695606177, + 0.1698115140842062, + 0.19170138553198293, + 0.20145958753761836, + 0.24695477872719643, + 0.1742375481845308, + 0.17207633277833653, + 0.2084972623045976, + 0.1502848787476804, + 0.1923547095033159, + 0.21637004456949896, + 0.1892928711126083, + 0.1336066810922874, + 0.2244874791565176, + 0.2596680279126595, + 0.16145378288958803, + 0.12884095037550414, + 0.12305415125448384, + 0.1800682387190481, + 0.1799020365279142, + 0.2038416924578705, + 0.17703179419725185, + 0.14265785610612125, + 0.19984496047213454, + 0.1319037560156313, + 0.12076894069261535, + 0.22861251006387387, + 0.20688051322923748, + 0.1482125546245078, + 0.17953985123986904, + 0.2797165663757242, + 0.20579170457178841, + 0.11257708914116707, + 0.12380791068715653, + 0.1614434537494647, + 0.21453438281345855, + 0.2520460799346758, + 0.2388420840800608, + 0.23685789968958137, + 0.11371822242791063, + 0.23860534251747453, + 0.19283147346767035, + 0.15205409928502017, + 0.1698720308721409, + 0.1359331117711021, + 0.2550679289210337, + 0.21495991851383064, + 0.16651144439082724, + 0.1088197894701279, + 0.11139070620238169, + 0.18399266195756875, + 0.1830411816324703, + 0.1646064205994545, + 0.19516173708959275, + 0.19432190241734312, + 0.13819916055071238, + 0.18986074811920736, + 0.10251331756253702, + 0.08317966232997884, + 0.09474595832577153, + 0.08434448589617333, + 0.11314182843480543, + 0.1880300798051844, + 0.18641075384734396, + 0.15869868932482467, + 0.08743375239779505, + 0.17175742984010184, + 0.2358769158044101, + 0.20802181677180126, + 0.21183982359923145, + 0.20900236928133434, + 0.1949664902928648, + 0.19529771961007772, + 0.22708667162960314, + 0.25072953994130476, + 0.2249005559384223, + 0.21999461075856774, + 0.19948322291878562, + 0.14122339001408793, + 0.2013278431992787, + 0.2001494073418792, + 0.21154346844872096, + 0.2597763742553705, + 0.19565669493266866, + 0.16545134991305102, + 0.1918472354647725, + 0.2023471681735993, + 0.14463825553001297, + 0.14403360765874634, + 0.24781207775855307, + 0.1970955825386554, + 0.16277650016395043, + 0.19331984193626714, + 0.25733692165147337, + 0.13280934546918305, + 0.20106968012706583, + 0.19767522465928025, + 0.08048073555073869, + 0.13327375134234956, + 0.10883691754907344, + 0.19086821542330074, + 0.14247800522578166, + 0.17649202993575844, + 0.07082547438995754, + 0.24257872412014012, + 0.20280361318660253, + 0.21257643989846356, + 0.23115255834752035, + 0.13113235349912158, + 0.25749008627738484, + 0.10831443207021092, + 0.22676896407418734, + 0.15682032522436268, + 0.289928412754713, + 0.21536495398159894, + 0.2169553753529255, + 0.3148731386036633, + 0.17810100578001178, + 0.20342739002117718, + 0.09874402609377977, + 0.21172120834802, + 0.2031786436896366, + 0.13858068146065217, + 0.10309795750166087, + 0.10567153406602292, + 0.11861437764971648, + 0.2060168053369955, + 0.253808615028564, + 0.10518921847786952, + 0.09089768305691556, + 0.15750913493818378, + 0.2124690316687272, + 0.16853390710633856, + 0.11666712461078022, + 0.05766426988850531, + 0.17644941318847304, + 0.10752407009406169, + 0.11431911091084199, + 0.18625505094387754, + 0.15145961621982468, + 0.18936135778249769, + 0.11520013076093037, + 0.12309528183524364, + 0.10576427323055707, + 0.09798030896049242, + 0.13439113075981404, + 0.1675018275527015, + 0.19846812396858357, + 0.1879218509724187, + 0.18276942529591544, + 0.16243654743844127, + 0.15694591608202843, + 0.25230878580723903, + 0.13022418689057302, + 0.14011078878824265, + 0.18741894314806093, + 0.18765977686962004, + 0.19029835312358598, + 0.20581468541467712, + 0.2050018231567081, + 0.12038459022400438, + 0.19308526574827165, + 0.1408809094870893, + 0.1051586187682737, + 0.20815738883668586, + 0.21703992692330573, + 0.19940396446339892, + 0.2044106119086027, + 0.17637028157208004, + 0.24207677265298863, + 0.24226818294565894, + 0.21175774072525297, + 0.1954886015351272, + 0.18328358945245227, + 0.11370045141588091, + 0.16182644229565116, + 0.17301110038180004, + 0.1881488612585868, + 0.09985391392815608, + 0.13402049871526991, + 0.17001619083001798, + 0.2201358752936261, + 0.221888296849636, + 0.1479053535708754, + 0.22442196973055745, + 0.24560825524837168, + 0.13893991331668873, + 0.2255281870711701, + 0.1740070982918138, + 0.1876226323895866, + 0.23069259282905702, + 0.1445080822507166, + 0.21331761328068177, + 0.16280480082469706, + 0.22439630635288846, + 0.21506201212771314, + 0.23874621446393443, + 0.14143920439069615, + 0.09628627385100876, + 0.08804288801867405, + 0.1866328363793594, + 0.22151308484236615, + 0.18884997865964184, + 0.18899946977476123, + 0.21483472940005388, + 0.17598262368951428, + 0.19374678793172806, + 0.18725631479414112, + 0.20944877964129652, + 0.22550601742091925, + 0.18173968050461028, + 0.10797162780165785, + 0.19788410273175433, + 0.22034466157616212, + 0.11806439982615885, + 0.2019302223305541, + 0.19424194244137302, + 0.2305622471467936, + 0.1380465222434513, + 0.2270650706345365, + 0.16366362995387615, + 0.17403322278630304, + 0.20121279940363176, + 0.09581243422846143, + 0.2017457081362786, + 0.1745732470683879, + 0.18001862184859194, + 0.22027077443831855, + 0.18524203207545697, + 0.10775184892432976, + 0.1771181894506629, + 0.20934149854443773, + 0.1912602798972745, + 0.2689275343008004, + 0.2189673189778695, + 0.24830982054293213, + 0.1552323864537385, + 0.2014904145637359, + 0.16111424705589653, + 0.15676114774632727, + 0.17419040542573772, + 0.1747984564019772, + 0.19604269444650702, + 0.2000805206018612, + 0.15228266903225512, + 0.18961590522938812, + 0.19208466219678175, + 0.20365951243105512, + 0.18449929608645577, + 0.22346966460968334, + 0.2170039173026218, + 0.21236337895620253, + 0.13884972947256377, + 0.15359850404105427, + 0.1796028184643768, + 0.20479482263019322, + 0.184823551938803, + 0.21630136042555928, + 0.2766047397636958, + 0.1953245320621593, + 0.2229237397269222, + 0.15617976803205377, + 0.19902111933386676, + 0.18959509132488153, + 0.24884871537941772, + 0.18527935395538406, + 0.2308850666820824, + 0.15782683516980525, + 0.100107765473016, + 0.19682921489348512, + 0.22823962793588112, + 0.1918636609066415, + 0.22297177602680554, + 0.1632791388251168, + 0.16027864105780268, + 0.1706573258402939, + 0.17708946240568935, + 0.21017315467506187, + 0.1731453394831159, + 0.1512747523126322, + 0.17609147269353476, + 0.1279472258191604, + 0.16309532875745072, + 0.16476320795289692, + 0.16690553314071052, + 0.11224482526219258, + 0.19376898598377834, + 0.2232652735967115, + 0.14386801892542755, + 0.12286155005994216, + 0.10980724689643084, + 0.15671758492575905, + 0.158779530670779, + 0.2015739058675396, + 0.16905003440192978, + 0.12562086501288408, + 0.18595731524560358, + 0.14158263071873367, + 0.11064938323537822, + 0.2136336133405868, + 0.18935615871068795, + 0.16157892705899618, + 0.14851857468113497, + 0.2632791282399886, + 0.19109862603655162, + 0.10262103460831143, + 0.12232813157543616, + 0.16997554959430625, + 0.16900868204577751, + 0.252114334519964, + 0.21642040209252442, + 0.2276083714146563, + 0.09724542995918524, + 0.22281024126564852, + 0.18775944201511716, + 0.16567257626064089, + 0.1421602090836331, + 0.13046175178959546, + 0.21381895546203403, + 0.19134241279051822, + 0.16482929105349658, + 0.12440918541188073, + 0.0875319100022925, + 0.13164246249343295, + 0.18867456040625893, + 0.15608467623546024, + 0.19165890975487, + 0.18313351333769315, + 0.14218967715230216, + 0.15243337190109288, + 0.08103569323165048, + 0.06049685858308272, + 0.07652148919090973, + 0.07602518573459809, + 0.07681566732608199, + 0.18054072963445178, + 0.14529705767118148, + 0.1270397605273858, + 0.07966990189708109, + 0.17112120232875858, + 0.23631363821495005, + 0.2130679659584696, + 0.20782827266474244, + 0.2096949744023833, + 0.195806686641128, + 0.19428168567992884, + 0.2201823059445729, + 0.25134138922498284, + 0.22236651703673596, + 0.21243259058304503, + 0.2005846034331201, + 0.1359860318820037, + 0.20244329997386426, + 0.19232228038118268, + 0.21179980906689733, + 0.25391361932810186, + 0.17547800328402038, + 0.16126264088930017, + 0.15980583573418583, + 0.1837911722606348, + 0.10841985235022841, + 0.09571375836029448, + 0.23432571408604796, + 0.18660823756504746, + 0.16225636672256885, + 0.18515132070326978, + 0.23614130542168466, + 0.1361436540039818, + 0.1851872720840193, + 0.1776215263892009, + 0.07218544517090508, + 0.10253564055250439, + 0.10024280525057849, + 0.18970674409649899, + 0.12039549296252203, + 0.16621003217280853, + 0.07051372035005142, + 0.2367497584386492, + 0.20453387154416466, + 0.1991309411811481, + 0.22517509948269473, + 0.14777539920215754, + 0.22458541921470285, + 0.10311445517755129, + 0.1998282468597243, + 0.1550222435308118, + 0.2764768838082069, + 0.20868571857130666, + 0.21663348534066318, + 0.3032002464954743, + 0.17614226927872598, + 0.17801480093795252, + 0.09391615427579954, + 0.19293072564403826, + 0.18796644696013684, + 0.14812103079095326, + 0.09524191332724692, + 0.09674778007262756, + 0.11308968693620508, + 0.20872037873151622, + 0.2472456344877532, + 0.09828835650843112, + 0.07919245931528236 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.2009765256073382, + "valid_tracks": 613, + "per_track_pearson": [ + 0.23554563044125829, + 0.20602623740225573, + 0.1883331301503943, + 0.25136592596511803, + 0.22914660518105964, + 0.18920218052659182, + 0.21900527982267845, + 0.22861692731647065, + 0.23319050369613065, + 0.25698948682981515, + 0.17560412564620556, + 0.22134465946152646, + 0.15077690018693182, + 0.27154688091465407, + 0.2644948435447308, + 0.27450692619580663, + 0.2511793847914716, + 0.2339828631875139, + 0.2102832477666289, + 0.23205557054154707, + 0.1992294578223936, + 0.21391621846043554, + 0.23229731052051916, + 0.21880217509751418, + 0.1606532242491078, + 0.20576264581687667, + 0.1648091940216792, + 0.2213844549681071, + 0.22662451716845003, + 0.23729295668934347, + 0.2746224643238935, + 0.20933660208460883, + 0.2888372005112373, + 0.1894702501312026, + 0.2970942910723345, + 0.16293027812410554, + 0.23148293494058791, + 0.22869475427911026, + 0.206944157883126, + 0.22706237919849467, + 0.2586085722098654, + 0.20637398591967085, + 0.22345800203460484, + 0.21994117298548724, + 0.24609755653956217, + 0.19451390755499517, + 0.23057268312509857, + 0.2711130237442279, + 0.24786098932866021, + 0.2034697484478525, + 0.21871686367497092, + 0.2290880461540137, + 0.22626803778309026, + 0.2479169276475476, + 0.17079665208914102, + 0.24749026653087286, + 0.23140770886889298, + 0.22783103799052273, + 0.23067073106560645, + 0.20580263210245694, + 0.17967828110980022, + 0.20475961087407987, + 0.2076485308059104, + 0.18776059087215877, + 0.22702761051148754, + 0.2009357603854917, + 0.2289343815284083, + 0.21912425596621607, + 0.2312723444705509, + 0.21574040704532968, + 0.20997983666339892, + 0.22834850119910752, + 0.24374536474526542, + 0.21333262562210248, + 0.2274501768204644, + 0.24168704396494384, + 0.2272217065299782, + 0.22673864410285113, + 0.27560658411365446, + 0.2206211323733586, + 0.2070960389973582, + 0.20279271041906752, + 0.2646159527828368, + 0.24294813959986863, + 0.1726239287337746, + 0.19571753483010917, + 0.2320551039650409, + 0.25334343040954554, + 0.2299364837803433, + 0.2466035361009463, + 0.1831284655146108, + 0.1885157798601589, + 0.21217585278936898, + 0.21830165636531093, + 0.2288703166395457, + 0.2224307295265582, + 0.20693904553917164, + 0.22672619768381658, + 0.2365016864539691, + 0.2002693742942307, + 0.2077201681062996, + 0.24301939699826702, + 0.21988743023369015, + 0.21423838252320831, + 0.23567712173056304, + 0.2618699261030163, + 0.2124905243053831, + 0.27487091106127515, + 0.2508810373604454, + 0.23367820725996127, + 0.20406565747837965, + 0.31649988844810034, + 0.288175126324399, + 0.17294459504711507, + 0.2218854851910514, + 0.22221268662034316, + 0.2814530768662763, + 0.249183417459261, + 0.2456253164998071, + 0.22507660887246628, + 0.17288548022691902, + 0.23216253643728324, + 0.2170494486403018, + 0.24783529505340957, + 0.24753028256224005, + 0.2340076441057989, + 0.24992720811907557, + 0.16615114293440023, + 0.21514862787742792, + 0.20261939034247184, + 0.29630280443484996, + 0.24364849492877305, + 0.2098810132102792, + 0.16428792196316513, + 0.17845764008088685, + 0.22026437730549198, + 0.2627170954964987, + 0.2714755651383875, + 0.2704503683951755, + 0.2561114519774726, + 0.24381503279818617, + 0.19524044946675947, + 0.2705536460878809, + 0.2924338869215419, + 0.20277247875895602, + 0.20397619452300902, + 0.1782049166485818, + 0.22017737039931182, + 0.17943334291477864, + 0.24354059369968795, + 0.17960926203297453, + 0.2529038668980252, + 0.19822476226953978, + 0.20353988467648543, + 0.25887166357344527, + 0.1889595665058616, + 0.22819419497656812, + 0.263480466460366, + 0.2006702241564689, + 0.259003746712166, + 0.2913445980570118, + 0.2212169027036543, + 0.2475585773054586, + 0.2400574503983318, + 0.2360783431598858, + 0.23454767565484136, + 0.239222520835071, + 0.21456737620057367, + 0.22389394809807325, + 0.25925065795301355, + 0.18581402235967026, + 0.19212927046820824, + 0.2383910954524104, + 0.24879634894254918, + 0.21408764836549118, + 0.19801870736252403, + 0.2142681349720007, + 0.2635302523343798, + 0.2745406646545422, + 0.17787971898143173, + 0.23901434630271143, + 0.1888244988956557, + 0.19509502069753337, + 0.17938613013344476, + 0.19910267476920632, + 0.24295896027688027, + 0.20632968793364526, + 0.12331613198823657, + 0.17163993180660322, + 0.3037030219367529, + 0.32543884478723395, + 0.18392409074746896, + 0.16623592813426682, + 0.22649431334949927, + 0.17324545117660212, + 0.18991528639292937, + 0.19842823266899492, + 0.2109933622249976, + 0.16574585159045732, + 0.2180928191459438, + 0.29993583065343987, + 0.16639440477895084, + 0.33919401307224656, + 0.24589177216092717, + 0.190598321409746, + 0.2047131303877718, + 0.20183227062318004, + 0.1771475588395764, + 0.18919151656630856, + 0.16646853708287224, + 0.16542370979334486, + 0.16454047378128422, + 0.16211520976196686, + 0.17813196548204138, + 0.18104570273632342, + 0.21284097006171807, + 0.16548361031998254, + 0.20501599154726452, + 0.1744984741779595, + 0.18382367900103894, + 0.31850012829863333, + 0.18753284945288895, + 0.2831682271175704, + 0.24683127057130286, + 0.24938227059215454, + 0.28185123729766737, + 0.25138949288681034, + 0.271979184150333, + 0.20980988487906954, + 0.17631954950362097, + 0.22044026835889344, + 0.31715966972560217, + 0.19889426753031603, + 0.25036593942025354, + 0.2095467891118419, + 0.2303845026395889, + 0.14998774512406202, + 0.23152118340107225, + 0.23251908343364586, + 0.2321256358690178, + 0.23319247562978015, + 0.17244835625004284, + 0.22776095463295437, + 0.1838255166275759, + 0.18451119132614044, + 0.2469226864788969, + 0.30675667545963026, + 0.21973360567883943, + 0.2211702740562431, + 0.19346770820141837, + 0.20719099140674277, + 0.22049006298740867, + 0.16974168479839236, + 0.1927992779982286, + 0.22119139182147923, + 0.25199941442630164, + 0.21051631997357192, + 0.19918537631089656, + 0.20681664674148362, + 0.18168494691213663, + 0.2526201968325216, + 0.2253091340210493, + 0.22372544076558257, + 0.253382603291522, + 0.2241637712880344, + 0.2087687729877054, + 0.1918052567047429, + 0.1814010480736744, + 0.19995634819307662, + 0.22190992927579428, + 0.22647094479190774, + 0.13305579694929046, + 0.16393101089239168, + 0.14756275773108615, + 0.196239325367566, + 0.1725914935662401, + 0.1246863597952131, + 0.13933874868743154, + 0.17721070046180043, + 0.17775596566860188, + 0.19996988297844842, + 0.15999760969963245, + 0.1360747064570286, + 0.12742445822851614, + 0.17197598120156776, + 0.17163589988956496, + 0.1680150636856594, + 0.180238490753293, + 0.17564121421103773, + 0.17732143252067875, + 0.17049565185668145, + 0.14281779355438445, + 0.17971115841805307, + 0.15326195148509192, + 0.13759409615721283, + 0.10243244474968186, + 0.15296354485571703, + 0.14125111451437264, + 0.17625050668394104, + 0.16167504914601166, + 0.16396742744107645, + 0.19936266923590149, + 0.17972546105524598, + 0.15665791260935027, + 0.14674048217171545, + 0.24844393135620896, + 0.1267579066816808, + 0.17728140354764377, + 0.1734149855370335, + 0.16240143937380178, + 0.16954359609895955, + 0.1867071134022365, + 0.16909203156042096, + 0.17134411724904985, + 0.16534725612949974, + 0.17144088933259385, + 0.1379664639241948, + 0.1527519499030133, + 0.19003331008720267, + 0.16499546343478994, + 0.16863143445852216, + 0.17444570622430575, + 0.17719729326020112, + 0.15212944044613277, + 0.17463552550135808, + 0.1036688129714865, + 0.1638722810408192, + 0.17024304775118873, + 0.1628143819395391, + 0.17670708345256447, + 0.15523786613044746, + 0.14933614455793032, + 0.1671897883929436, + 0.17551866315128004, + 0.12910197183740685, + 0.1802950626800305, + 0.1630626126412974, + 0.19149487616542704, + 0.17746724511993325, + 0.1912593109155336, + 0.17333116941577864, + 0.16475972477118256, + 0.19116038738640798, + 0.1896966292697144, + 0.18007526501981153, + 0.18332792888326682, + 0.19927339126279517, + 0.1843076568522989, + 0.1858387332638462, + 0.20318690207183024, + 0.1652034479415218, + 0.15542296282567158, + 0.14174745243281392, + 0.19950576880505572, + 0.20052151365145512, + 0.14582845366722288, + 0.16286183320499253, + 0.16792686306082538, + 0.16851529435561577, + 0.16092783903016286, + 0.15054776432706898, + 0.11808297308984765, + 0.14519204367689778, + 0.16043937856796592, + 0.15727837336170078, + 0.16552584589864835, + 0.16864800643191322, + 0.16926913608857735, + 0.16819895390892076, + 0.16962616043129178, + 0.13919912389023506, + 0.1413248883675485, + 0.16034968873776556, + 0.15315073320521272, + 0.17629495781701165, + 0.16466122627284985, + 0.18401548566538173, + 0.17184005221148949, + 0.19170419950008424, + 0.19861917978415145, + 0.1678396070815798, + 0.14277321048472968, + 0.23980742054259915, + 0.20290162206911583, + 0.13889880138309854, + 0.18470981188189478, + 0.14857537219785105, + 0.202323794209049, + 0.1729557056565432, + 0.17907578569969404, + 0.15617681815178505, + 0.13099249951143443, + 0.1759103904500507, + 0.148768073123103, + 0.19579598800573506, + 0.17856597664001583, + 0.1781403487164693, + 0.164247299236794, + 0.126457422489266, + 0.1391904870287869, + 0.13891863497671195, + 0.2224790118539089, + 0.18067883658565254, + 0.1366383852305969, + 0.1181012175182539, + 0.13594383794281129, + 0.19988097760208956, + 0.19140543683068623, + 0.2107278643198629, + 0.18866834111692327, + 0.18407319580355883, + 0.1555804583057045, + 0.13204604364371736, + 0.20194443866478554, + 0.22310110583536044, + 0.14343530746365857, + 0.13935238482722917, + 0.14040300177634557, + 0.1656695849619853, + 0.14754397788514353, + 0.1728263652862161, + 0.1419897091707307, + 0.19271663895893593, + 0.14487214458629308, + 0.13492226598445695, + 0.1794140044863306, + 0.13980250407903913, + 0.18121185397923947, + 0.16222133339692835, + 0.1812288869942675, + 0.17413214270369673, + 0.21765640889578414, + 0.14065012719328304, + 0.1656850735589843, + 0.13060689640983122, + 0.1300604694697789, + 0.12680600436260112, + 0.12637373529324009, + 0.12759713122474453, + 0.1488418746680989, + 0.18593832513823208, + 0.17549502412093065, + 0.146306786994217, + 0.21032361471044814, + 0.16666061082053288, + 0.1384598913018664, + 0.1428951917406091, + 0.16701705783409235, + 0.22104515137196706, + 0.22264899789314785, + 0.132259903458532, + 0.20043851826367337, + 0.16761997539991638, + 0.11629123688418772, + 0.12571916484700857, + 0.13553819215824234, + 0.18227783480953383, + 0.16798932812717285, + 0.1031059396794889, + 0.13926259643944064, + 0.25985630573232993, + 0.26269590152953093, + 0.15229100056622005, + 0.1322341230828651, + 0.17942293647726423, + 0.13880403486828882, + 0.15957782405092416, + 0.16291855555934046, + 0.18946256302495323, + 0.145497633000943, + 0.1795178902785303, + 0.25602260656610176, + 0.15225374285471452, + 0.3023935131958923, + 0.18820421884246105, + 0.16219708669977584, + 0.16503304669123411, + 0.1481130535667473, + 0.11302159018264711, + 0.17592990347663842, + 0.12130513110709604, + 0.161137886929316, + 0.1536574821166151, + 0.1731249680761389, + 0.16889380044227448, + 0.15331563248895738, + 0.18133439936323137, + 0.17173865760906748, + 0.14619747457104357, + 0.17647974469593108, + 0.16645460315723365, + 0.24444344298503087, + 0.16643192272872476, + 0.21392172161490502, + 0.1542131472637783, + 0.15851303182601859, + 0.21377379760817358, + 0.14282093804098694, + 0.2053097781755193, + 0.10659910337920485, + 0.1599908344454838, + 0.12983946532173754, + 0.2757402576803516, + 0.1759180565099438, + 0.20296741128088985, + 0.17863815901093233, + 0.18838561332950668, + 0.10194044423377785, + 0.19198150370584136, + 0.18235870516460814, + 0.16133914491005033, + 0.16926521813659762, + 0.14692408153849013, + 0.2038821211556549, + 0.17683039665922215, + 0.18896648511090774, + 0.18822568110177074, + 0.26438740727495397, + 0.1747926804015978, + 0.17008669719557823, + 0.1414016065853467, + 0.13163361424641587, + 0.17126671999169785, + 0.13674784022473208, + 0.15151468794930006, + 0.17894064759091496, + 0.18439242809994458, + 0.11669631775313835, + 0.1666679056977758, + 0.13457117764247958, + 0.16345235264194677, + 0.18516502492504455, + 0.11270713635904142, + 0.1612214039369453, + 0.18869895494068387, + 0.17861570923994102, + 0.1885401561994078, + 0.17171722785898164, + 0.094133990277567, + 0.16954493310314647, + 0.18335104362207716, + 0.17539560723916423, + 0.19714972514048257, + 0.09879724269311678, + 0.26649338310416987, + 0.3041512266400287, + 0.2787026311202462, + 0.2564078803636943, + 0.2613092196052215, + 0.2171637621003818, + 0.2182132721090965, + 0.24965213245900372, + 0.19372498472732852, + 0.20946000709591236, + 0.21540639761738542, + 0.20134585491740928, + 0.20279622286989804, + 0.2085449251204644, + 0.2197893379255217, + 0.20206552677208878, + 0.21104901624346098, + 0.18817541547452094, + 0.2689607015697996, + 0.2676331625412983, + 0.2674801792369937, + 0.26619171253482227, + 0.30090973534611476, + null, + 0.28375492533037916, + 0.3370664695479783, + null, + 0.06323932219192019, + 0.4004253839180688, + 0.06944196669536425, + 0.2627819166823951, + 0.24289229955822064, + 0.30968395358415796, + 0.37011016676625563, + 0.36179691553501897, + 0.384191413460685, + 0.37732973012871573, + 0.3419001241572681, + 0.2946711320573942, + 0.2246294095690373, + 0.3827715338856736, + 0.18074980712931787, + 0.29700984684246884, + 0.263030593856712, + null, + 0.2565775203001909, + null, + null, + 0.26161210209410696, + null, + 0.24201560686165188, + 0.21151712095741593, + null, + 0.30272723382147604, + null, + null, + null, + null, + null, + null, + null, + 0.25706931138646205, + null, + null, + null, + null, + 0.2092049766131927, + null, + 0.2636095244878663, + null, + 0.21775112353422602, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.27678922056956456, + 0.1528666849431263, + 0.24992463367497467, + 0.25835807949740996, + 0.1706957944392102, + 0.23523990099544087, + null, + 0.29886180208842295, + null, + 0.24113795847737807, + null, + null, + 0.23887940425428458, + 0.2742931202766058, + 0.23170485704004978, + 0.237628979205647, + null, + null, + 0.20252211793131386, + null, + null, + 0.24471031041219002, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.23771348656357738, + 0.23682050268311297, + 0.25007183977269376, + 0.2232892200439769, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.36698738437050543, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.18605359449331416, + 0.25289971569315023, + 0.27528881363331564, + 0.39607796874979206, + 0.32310346969566484, + 0.2028468283413249, + 0.29450672040407516, + 0.34113895712910475, + 0.2818715157714315, + 0.27800257787958804, + 0.38293113176652477, + 0.3383346168403477, + 0.2568633752079149, + 0.33498679624750705, + 0.29160096038069105, + 0.2965596948938549, + 0.3172760936735992, + 0.2636134187775228, + 0.39261310088978724, + 0.2826942605203927, + 0.28766243645707235, + 0.2984055896807618, + 0.2737384251572345, + 0.3964741228010041, + 0.28340048057595635, + 0.2867119401062806, + 0.21008821249492896, + 0.3424714762632227, + 0.32117498722338916, + 0.36692848367903547, + 0.30307393959027074, + 0.34727887503096305, + 0.46163829027898795, + 0.25336545317487796, + 0.6170323487787197, + 0.267509047436743, + 0.18063229536533285, + 0.1510263847688611, + 0.29372613078305526, + 0.20057641614071678, + 0.24366026213310105, + 0.43918033119191824, + 0.2761619748662937, + 0.35796144304115934, + 0.2642792247886365, + 0.3601536662661621, + 0.26780710405146707, + 0.18908754661296057, + 0.4253668409782453, + 0.39452466279638104, + 0.35705882805241235, + 0.3469826410575407, + 0.36087548784345547, + 0.31329377371448186, + 0.30327533032225834, + 0.3132410034979766, + 0.23813747226455245, + 0.3328559030971655, + 0.3150044854205327, + 0.3370494331901398, + 0.2236554917343509, + 0.18264503019018463, + 0.37574598083267147, + 0.37241939334199675, + 0.2529441133976179, + 0.3029198358685377, + 0.31226509555744053, + 0.3857406712938147, + 0.3938055896042359, + 0.321514720699943, + 0.2965602140982005, + 0.6624054587566737, + 0.2913593490506823, + 0.2796184355649715, + 0.2706501000741514, + 0.36502862270352265, + 0.20558751056657265, + 0.5783069171824576, + 0.3235237330332286, + 0.32644372218209616, + 0.33149915165928745, + 0.355269194535597, + 0.4318767514872335, + 0.3400559874505129, + 0.25768543104231256, + 0.22102077866603675, + 0.40313967778352666, + 0.3098741798869462, + 0.7850087890216524, + 0.4159697300960074, + 0.2894078956990322, + 0.29350695698063317, + 0.3227396388517881, + 0.26050609285351983, + 0.30444188658420906, + 0.43538698869284026, + 0.28404555623104544, + 0.2637318842716572, + 0.2839226961581419, + 0.2345711945130131, + 0.22188384637084047, + 0.3500931646861991, + 0.31507632969463634, + 0.3114653888599944, + 0.35843843077123905, + 0.13020785978170227, + 0.14954200132466758, + 0.16574761154206943, + 0.3674521817198699, + 0.23600849920135047, + 0.14119385602334592, + 0.20150169497651382, + 0.17503485131790297, + 0.2578680204412213, + 0.22144154141417047, + 0.41357205254300766, + 0.30722526408259054, + 0.2570225214039928, + 0.20667415541622638, + 0.30316410421116363, + 0.19402012910355965, + 0.30822227074907965, + 0.2574745528932893, + 0.13354205799476218, + 0.24984750273656087, + 0.18872037061792316, + 0.22203559100704906, + 0.18510153648150457, + 0.2101577237941112, + 0.50403146045214, + 0.24345332338155123, + 0.15871182181488058, + 0.29006717352868755, + 0.5014125133577018, + 0.16005296892324591, + 0.2118907088291903, + 0.20418685240877887, + 0.24379408112785672, + 0.15714790731177383, + 0.16691152740782594, + 0.15733368883464502, + 0.39501816336538215, + 0.4238640612511172, + 0.1883956082897433, + 0.2872066102917414, + 0.3519493233294043, + 0.30885826696121976, + 0.13847045341842296, + 0.3120755813618935, + 0.23018399943055082, + 0.25072577553796005, + 0.19859596666756252, + 0.21735839815053284, + 0.36492605766271274, + 0.2948353796444221, + 0.28989160451487234, + 0.14627055118015567, + 0.18945711270424925, + 0.17782678686425216, + 0.27625615194108294, + 0.23991634598882844, + 0.22679543282947337, + 0.2485455263724565, + 0.2745577808678634, + 0.22470528688683714, + 0.24717644282868872, + 0.22545251649952497, + 0.3526481094299012, + 0.14894664004096606, + 0.2524683852060254, + 0.18688198512663437, + 0.17542911726329044, + 0.42248310881267925, + 0.1760334661025144, + 0.13138913317646553, + 0.3349106606617415, + 0.15496920175228268, + 0.2330260618916088, + 0.2252609855854983, + 0.19847514760861823, + 0.23732394651951372, + 0.21629303345496595, + 0.19378560859888683, + 0.30130839511275825, + 0.20244048823737604, + 0.2348680728001292, + 0.32602906019000993, + 0.17692699638566128, + 0.19409884307873784, + 0.3001900825279414, + 0.15223888904167626, + 0.2057899971250178, + 0.3577645076408821, + 0.275295598180099, + 0.18416591384880085, + 0.18171882488712168, + 0.16894055650566414, + 0.1503868308055996, + 0.1844688616553421, + 0.3503823793465296, + 0.19681828235071513, + 0.22293524645493068, + 0.19562543145471584, + 0.17529010735737327, + 0.3286345035875313, + 0.1948390912089053, + 0.2074330698458699, + 0.28872799656349224, + 0.15567061404404245, + 0.37619975287064794, + 0.1673084312434885, + 0.3070909436939991, + 0.23386048118941166, + 0.28758691896945027, + 0.832648611714524, + 0.29649676442360123, + 0.3391644506902616, + 0.34871329297455084, + 0.3972185524852638, + 0.5126235533047324, + 0.5837682465576233, + 0.592767705234659, + 0.31429896665357027, + 0.3292747035278923, + 0.41899705808202586, + 0.7086633371659438, + 0.2419101193716395, + 0.27613941976993633, + 0.509091225082812, + 0.48186766691030714, + 0.4432853926925452, + 0.358825563635336, + 0.2309653640418933, + 0.301842333881495, + 0.567983872561656, + 0.34592420972467014, + 0.21957770770115242, + 0.5910889980186711, + 0.5150465898994514, + 0.3999979021091964, + 0.43767584532356285, + 0.6845016352476812, + 0.4103188271979532, + 0.4314570762943625, + 0.3924218435059037, + 0.7470077102351553, + 0.2516716427195521, + 0.6217666666664271, + 0.32285177824481553, + 0.5156090813903257, + 0.37920910342237946, + 0.3700076886297759, + 0.6856223154004268, + 0.4629856071417176, + 0.23204097601830972, + 0.3273346251599295, + 0.6235206082687014, + 0.28493982981786004, + 0.4064667568461697, + 0.3550309017242522, + 0.2665953518242508, + 0.3955017979615849, + 0.3802816635871203, + 0.3455086372461645, + 0.3569818784087158, + 0.44222359529574173, + 0.2779348911657768, + 0.2550038566436117, + 0.6054909239973406, + 0.3133017598024955, + 0.4148983059591454, + 0.5128154635588197, + 0.2577274568757254, + 0.6402238076963658, + 0.24668454640413293, + 0.25838997867458274, + 0.5318036407814287, + 0.4300752673875781, + 0.5591826938299979, + 0.7143260444799578, + 0.6375762135294426, + 0.32702704148978234, + 0.4323630053341845, + 0.5021742144651438, + 0.22495767179175621, + 0.24547509934907855, + 0.5402249689412176, + 0.4407196866998587, + 0.6875639260089609, + 0.4054324872622091, + 0.29741266897102403, + 0.39227734297260014, + 0.31807457537730444, + 0.43473963806939253, + 0.6835559875266891, + 0.5990820596700522, + 0.38776207596523143, + 0.4826640605503489, + 0.44744967969748417, + 0.4880181880293719, + 0.7192265613447215, + 0.26078210408349095, + 0.4039921607692692, + 0.18750609393255394, + 0.5682967081251813, + 0.2564691569404414, + 0.7172599914847065, + 0.3374739732847154, + 0.3791486996494276, + 0.6707074029043528, + 0.4729186507572058, + 0.2569204548542268, + 0.5386313057412859, + 0.33689746452229646, + 0.48198243533911467, + 0.4847878325493918, + 0.2796042877073579, + 0.46989980401821885, + 0.3257287628059703, + 0.4196345648466626, + 0.28670997595296643, + 0.20094074714214497, + 0.3256679538045994, + 0.46198874494264647, + 0.5281138890864272, + 0.6354104838984955, + 0.31612367149019927, + 0.23623315077483645, + 0.3874478637033172, + 0.3987993080512689, + 0.3960760823583088, + 0.4407264016647176, + 0.37496154634626266, + 0.2982861915246932, + 0.27302855323158687, + 0.45813268588094236, + 0.7514720562330949, + 0.49986398087145095, + 0.569442312345614, + 0.5690465618581382, + 0.7544852546631636, + 0.505647990301987, + 0.2499671204133375, + 0.5544976813624711, + 0.3162460553686853, + 0.47016977361121687, + 0.3698908056675137, + 0.3259693570697146, + 0.4538288944502046, + 0.7604024927403812, + 0.20736913366385648, + 0.42307565994165763, + 0.4213363706857411, + 0.5971548593378089, + 0.35531753416314854, + 0.20234902245525524, + 0.35572691575174487, + 0.7068824375153333, + 0.32979043362537436, + 0.3858384983698464, + 0.4786085498632732, + 0.4101266549772203, + 0.3576450977265701, + 0.4357092476739072, + 0.5842836850884253, + 0.42588912620521374, + 0.3465696384788862, + 0.48931752731449274, + 0.41379948009140444, + 0.5376701656288129, + 0.45955181260475564, + 0.47075842106524174, + 0.5002150681596358, + 0.23624325527721657, + 0.2075963261006962, + 0.3904759294085451, + 0.2578412912628354, + 0.6782525599244604, + 0.237401199482277, + 0.5373255912321055, + 0.46313588463048194, + 0.6660246485595159, + 0.24396696712453833, + 0.2586862084849346, + 0.25861035537711424, + 0.2421993354414269, + 0.23078544588860606, + 0.21430109513269896, + 0.5878364299901442, + 0.2749730651119398, + 0.6642244521576793, + 0.5522128685273249, + 0.3657315779966556, + 0.5083359074986078, + 0.34152485144914246, + 0.46084988378651004, + 0.4646409629954072, + 0.4315700132413203, + 0.3295150190948383, + 0.42174751858562975, + 0.3893920713771147, + 0.4850011290109179, + 0.34283820869088044, + 0.24606856924552503, + 0.5736250405264319, + 0.41966575479218526, + 0.6400547958521636, + 0.5790557094246808, + 0.45890136501181594, + 0.2508966833100447, + 0.25762472107371626, + 0.2820533300549754, + 0.3495820277295755, + 0.733574445725315, + 0.42690954472826936, + 0.31048828754262653, + 0.7196561997957187, + 0.3588036371896697, + 0.5980612151929305, + 0.5004241466152364, + 0.33942494981785065, + 0.4421792435054773, + 0.38495168606147734, + 0.3870403192221798, + 0.5595403189841985, + 0.4852004885351695, + 0.4757112893516709, + 0.34371484312854, + 0.39034032982011996, + 0.33666383971422187, + 0.3950571487079336, + 0.7514633999489825, + 0.4179521296325508, + 0.4197896605311326, + 0.37411037522494195, + 0.21532851636269892, + 0.6506010574261583, + 0.3579911506020503, + 0.2441179261080642, + 0.4805947396880102, + 0.33430197472831547, + 0.24885968425853477, + 0.4620079386628156, + 0.3676440488867963, + 0.7607797897381351, + 0.2843393299485738, + 0.31062535316884377, + 0.24626398067172453, + 0.38923365443695307, + 0.31541579370060996, + 0.23304364678618794, + 0.30941882177021146, + 0.21854086695140165, + 0.34389011046025975, + 0.7199685400620145, + 0.23170990928957885, + 0.3608937794733225, + 0.46020316582353715, + 0.3670400570833297, + 0.3295514514203539, + 0.43550088302279194, + 0.33218022074805487, + 0.47715178077541764, + 0.4595915580883572, + 0.1547355529826961, + 0.5990702961162322, + 0.6523326991451418, + 0.45215882434752197, + 0.4217616401396936, + 0.4889580779335189, + 0.32419135015954914, + 0.3017504317150771, + 0.3981294472301202, + 0.7610997029572355, + 0.27166553500114243, + 0.5783186664899526, + 0.25344077822220246, + 0.41793007702569224, + 0.40710435108229986, + 0.41402683655538347, + 0.3189497031329361, + 0.20368803593368798, + 0.25025134649871494, + 0.2560702484636093, + 0.4684246593418133, + 0.3741499550525201, + 0.19142978592050489, + 0.35616682411873024, + 0.40778697807354614, + 0.23133198442359507, + 0.5094917675193085, + 0.36806040980985516, + 0.2991553283614306, + 0.39571687752650714, + 0.2192801638070743, + 0.37523639179734386, + 0.3820225848956433, + 0.5484075742323696, + 0.41218865438192337, + 0.765216282499321, + 0.2311157875820788, + 0.5286883856055992, + 0.7222838732046093, + 0.6112119195847519, + 0.1741007340193953, + 0.4247009948181881, + 0.38388024299749385, + 0.25968621124849417, + 0.28416694666887077, + 0.2759020683047752, + 0.40226499966868867, + 0.46347635630331474, + 0.285293349156511, + 0.5835431309594394, + 0.7433662003574371, + 0.3724414857018654, + 0.32138549092516483, + 0.3723686810272882, + 0.4130213240250631, + 0.443876100891244, + 0.46234197379894604, + 0.44764182366243005, + 0.23449025619232303, + 0.6140656830428848, + 0.43178240647279736, + 0.5121381836625698, + 0.6857141959281503, + 0.2516950496311891, + 0.5666743918572801, + 0.34375686704693575, + 0.38636610829295653, + 0.31196817571200425, + 0.46887833762671166, + 0.3996808087864588, + 0.2601712833188833, + 0.3199731879941164, + 0.38161970956248603, + 0.5223763599464406, + 0.5496776997722321, + 0.36438188578419367, + 0.3900348311317513, + 0.35778372946801096, + 0.7150912072340079, + 0.33500792399409574, + 0.4281104302829602, + 0.4102391118572346, + 0.42864921495093344, + 0.5645593845434204, + 0.20806184438896239, + 0.5376324839453901, + 0.3082230755437918, + 0.48858497789814687, + 0.3608441608547429, + 0.251434862929209, + 0.25489017758351296, + 0.4063934010314237, + 0.24142957259307096, + 0.4660519511223867, + 0.7052017773574277, + 0.7105676436361289, + 0.3104967641185964, + 0.24407823863376207, + 0.6793420023081937, + 0.4029954263595172, + 0.3225208809947289, + 0.45037014448743345, + 0.31657007036517, + 0.5649372019506219, + 0.40409822413047897, + 0.5298557091609908, + 0.5452610340998364, + 0.3249933482278543, + 0.46284065163220417, + 0.6312661709640282, + 0.45569019501353086, + 0.27942362793556286, + 0.33653945982295647, + 0.23130910611083072, + 0.4988735969297002, + 0.2883530453284474, + 0.4660774060414859, + 0.4536481803083875, + 0.2727691518039384, + 0.7012528844750182, + 0.25854974920555474, + 0.7052337324223124, + 0.3570195094747355, + 0.4942889237492147, + 0.7265869632363433, + 0.41355826054887296, + 0.5154656175536263, + 0.2627687417538505, + 0.6823772504120978, + 0.3267268096183758, + 0.2300569369654212, + 0.5052357111150394, + 0.47524265223803486, + 0.6109954476183926, + 0.5203560640538167, + 0.31072049480631664, + 0.434482705719762, + 0.20706948265053754, + 0.35971949456119945, + 0.5411317240805825, + 0.46114975628704746, + 0.27103967189152256, + 0.487188104407859, + 0.5478978606189799, + 0.5443870819752106, + 0.3882845944629728, + 0.2440768121000574, + 0.47493620394135067, + 0.4170168830085711, + 0.5268662454337044, + 0.4040878766929246, + 0.2633000983279371, + 0.3403347837448227, + 0.4499710851642015, + 0.3365077157443439, + 0.7098526315794417, + 0.46842740798108584, + 0.40539490054584604, + 0.4465153745220625, + 0.3250658591470602, + 0.3835162516535359, + 0.5101378389267005, + 0.5193059922466093, + 0.6942847159152924, + 0.49574629762626377, + 0.5513947487763665, + 0.4813662632016046, + 0.4164555697211996, + 0.5456554717334713, + 0.31590911993452153, + 0.5925378915363857, + 0.5691036650586155, + 0.7122853432418919, + 0.4655937986981408, + 0.5144813787874076, + 0.510090122889373, + 0.3173798175557842, + 0.5101815842224237, + 0.2525756556002707, + 0.6507719030112672, + 0.4241318353341681, + 0.7184387470622334, + 0.7393650820848614, + 0.3631309199757274, + 0.3978445593334518, + 0.45374253677209825, + 0.26474842124131254, + 0.4311501415207122, + 0.7405638702177568, + 0.45754951284959117, + 0.4546467719494046, + 0.555733871859977, + 0.44444640740921465, + 0.27661219966253325, + 0.5798613690285646, + 0.37687176579626475, + 0.3219335612233297, + 0.7188768506095619, + 0.3419050599758866, + 0.3548419055728725, + 0.3244054042200332, + 0.44982055353728473, + 0.5314856111343385, + 0.58216585245083, + 0.6563227911844007, + 0.5994255358654501, + 0.5628518133860972, + 0.3927969151512086, + 0.5826011682446383, + 0.5128143879371558, + 0.7658491005487835, + 0.7196901070141587, + 0.24751137163891193, + 0.28684278278687586, + 0.234858961359655, + 0.46598424565809127, + 0.3649800433489822, + 0.7724054668147958, + 0.313800440001297, + 0.365138548984094, + 0.5186135281167668, + 0.43571671451410404, + 0.5019909032088141, + 0.5955146147163812, + 0.4034181484021365, + 0.6015923909879987, + 0.3182417521639567, + 0.6522672381817181, + 0.7641134801083025, + 0.749030652377317, + 0.5561304506925238, + 0.3758158621323106, + 0.49661134571807547, + 0.7195626596822822, + 0.6658638518643839, + 0.5975099266441531, + 0.4227204675809448, + 0.3867180600734786, + 0.4571855167977552, + 0.7199587374606786, + 0.3622340343463004, + 0.36530510001560146, + 0.2540983052695302, + 0.5667301179957732, + 0.22673957092205468, + 0.32129254906912014, + 0.6282985831910676, + 0.5151962342403341, + 0.6652007228319367, + 0.4867211189949589, + 0.47175374196882863, + 0.6453777550605095, + 0.6097486118369962, + 0.5380493251939654, + 0.33553798142529784, + 0.6203184496363455, + 0.4799771789848933, + 0.35094179295847033, + 0.28272134579003055, + 0.4472465439554263, + 0.6140904003499805, + 0.4511161398953027, + 0.5822641266387751, + 0.533416675863123, + 0.3768249882123123, + 0.3493168843098487, + 0.6655174515378048, + 0.44406203897433866, + 0.5534175329931768, + 0.7078214522146168, + 0.37069065919799155, + 0.2896713215497781, + 0.28302531133743525, + 0.27803148224952295, + 0.6951575830764959, + 0.7197993588189363, + 0.5348266522724829, + 0.375518861971335, + 0.3952452219598125, + 0.3227291553629372, + 0.767484400949806, + 0.28738648035533143, + 0.4386137201859171, + 0.27177518441549037, + 0.4572901302320857, + 0.29583519233928185, + 0.5304399853574819, + 0.4121431717515912, + 0.22774730001766133, + 0.6708275094254347, + 0.4674918660128971, + 0.35577011716613816, + 0.3001935381693491, + 0.3092947863919248, + 0.3532393027928146, + 0.3660389074129678, + 0.32243957885145624, + 0.32002243558209986, + 0.3590491905037309, + 0.3251567561732045, + 0.35773083490209684, + 0.2825116509329533, + 0.38560605854737123, + 0.1819843368032579, + 0.19247291496177082, + 0.22514710879168823, + 0.3457856073964063, + 0.34887190293438225, + 0.3429814123030942, + 0.22416012856371578, + 0.2198537702401259, + 0.21667985755942676, + 0.6048950153712426, + 0.23962162594767308, + 0.20507131321825847, + 0.7595602574535322, + 0.30460278658596734, + 0.33388933844753466, + 0.26414996189320256, + 0.40844676731713103, + 0.22944866122440402, + 0.3257367486336916, + 0.299981738876954, + 0.39061317595646394, + 0.18327978661098138, + 0.36596446686017914, + 0.35822326970791074, + 0.4508113751166983, + 0.292988702253214, + 0.43872530041689406, + 0.2552844848178471, + 0.3448646036249608, + 0.3167312868897407, + 0.2836584221478152, + 0.43236309376270204, + 0.27374665046826246, + 0.4384517095897985, + 0.3005085627220206, + 0.2866858508716812, + 0.7972503361583261, + 0.20337577156504516, + 0.3283510052462013, + 0.31351171303798164, + 0.33068049356121204, + 0.23775839029821344, + 0.2745036603095118, + 0.4234708790601137, + 0.5124345885252063, + 0.3207162523254542, + 0.3385859287707956, + 0.19182982435494778, + 0.3684095056166137, + 0.3234359496713354, + 0.3464007856708311, + 0.3759028591546345, + 0.31060667486376803, + 0.2710254151601476, + 0.4271220215736897, + 0.3080246141413608, + 0.4035477875631576, + 0.18514087840664076, + 0.2765172043882844, + 0.4074626333926181, + 0.42428737774187947, + 0.337451315526277, + 0.2848926378696535, + 0.2701015661031775, + 0.8087137356035442, + 0.265107112854556, + 0.2460256962054871, + 0.29404881313098924, + 0.19589793273606187, + 0.23525435336614198, + 0.2440680095263587, + 0.17375972344388702, + 0.307660962800022, + 0.3262730027431138, + 0.28963758525529065, + 0.2430583992083702, + 0.3696547757610292, + 0.45357491194422067, + 0.20325081066570816, + 0.31225220011746124, + 0.5157111325974654, + 0.226888978267082, + 0.26865047417978094, + 0.4554359056297132, + 0.4004894045462279, + 0.22965616756055715, + 0.1977910010971038, + 0.23783400056598436, + 0.34799828118766724, + 0.40866639191537085, + 0.1896090237910511, + 0.2842244380767044, + 0.2620934956439851, + 0.34967200268236853, + 0.25860590284519197, + 0.17866786056274597, + 0.4160561992553546, + 0.4184428434548725, + 0.16862914108956592, + 0.3987933620840432, + 0.17853018151692865, + 0.24150251688662014, + 0.2189576551337448, + 0.7669798699722239, + 0.38722280247309876, + 0.3767228715134438, + 0.16977186968144029, + 0.3011833638382645, + 0.2732774268451316, + 0.6723352630672181, + 0.32531120326067375, + 0.257658810949493, + 0.40621983549865825, + 0.3542424182809436, + 0.33235856698746813, + 0.3806658031095602, + 0.43282774518841083, + 0.20341162402594823, + 0.22524795593695648, + 0.39547211341943417, + 0.31176013978505057, + 0.33391427029239673, + 0.30917561065415894, + 0.24093725917059577, + 0.1938436958132043, + 0.19479394645092168, + 0.3301528081205292, + 0.2895486720741964, + 0.5716409002992088, + 0.228250803390257, + 0.1955663280638364, + 0.21231143600848842, + 0.45345401770678506, + 0.2615791992119083, + 0.1607334166077325, + 0.21120837625377717, + 0.26441583143600833, + 0.28993684736838754, + 0.18532111190582054, + 0.49082025175409694, + 0.34440464072127835, + 0.20997310335890523, + 0.28919997891228716, + 0.3581979657450814, + 0.2681333159617006, + 0.2755729026269338, + 0.4401553986964813, + 0.2429784634322976, + 0.6240762041860991, + 0.3233477455508727, + 0.4287138809914076, + 0.3194423583149055, + 0.3129909698344175, + 0.3194887395919418, + 0.22780311031343525, + 0.1881529055345112, + 0.33838335925140745, + 0.5591330430852578, + 0.2563371605858244, + 0.16465920660162117, + 0.4504550799934871, + 0.3645223076012088, + 0.3636970628583224, + 0.5583565923067165, + 0.2548840876212227, + 0.3353785550172912, + 0.2016464734307399, + 0.3039252494885436, + 0.352729302433324, + 0.2101015497100214, + 0.4006999138728002, + 0.3373298139617666, + 0.3224488196053553, + 0.14577204539613545, + 0.31774322145253714, + 0.19462880792355702, + 0.21130797895187947, + 0.21908930060984452, + 0.13256148963491599, + 0.36802763200639454, + 0.4719804928583583, + 0.38239746292300053, + 0.41795462156901947, + 0.35469524395694335, + 0.1702817505989871, + 0.25598608683923507, + 0.3920131412034653, + 0.29813096999081284, + 0.32330109692789366, + 0.38936559955458094, + 0.30110276463792224, + 0.27613279785559747, + 0.46951067800512414, + 0.32137051082062673, + 0.23724077106493244, + 0.4623548454277142, + 0.24906477272518457, + 0.36729918485939966, + 0.2188527972720173, + 0.4110019928658122, + 0.5708103330812774, + 0.7749303591886911, + 0.35806906604028604, + 0.30135333555263666, + 0.3667611656153252, + 0.29750891041370864, + 0.29269843550126745, + 0.2796995642991855, + 0.3245081554351843, + 0.24565390798041067, + 0.18793768002729633, + 0.39290689671883894, + 0.23820520106459503, + 0.2594651723510979, + 0.19310625108124277, + 0.15593235076261514, + 0.33070115180132276, + 0.3408043420356443, + 0.3072979290278456, + 0.3039946662508196, + 0.3347865650086959, + 0.20227136674307913, + 0.13267408721789223, + 0.24607925737078093, + 0.3858502967764921, + 0.2086315505642643, + 0.33220180598876625, + 0.2408955335216578, + 0.2065777237055741, + 0.13637905248416715, + 0.451201428951066, + 0.18827646784094793, + 0.2239027030359999, + 0.3192057219979729, + 0.2640927403373323, + 0.2362931608672019, + 0.5589804048801202, + 0.3534753301673819, + 0.15787502807067474, + 0.3779363997356756, + 0.324259823619128, + 0.30285134832107047, + 0.2308766872978665, + 0.2039012747738416, + 0.24358191369031626, + 0.4046403061059753, + 0.1489764242733816, + 0.33793139463789845, + 0.14195457588322596, + 0.23993981105123235, + 0.26484518556276737, + 0.17080381342687936, + 0.3522596912752966, + 0.34859988692808713, + 0.23750013658318325, + 0.30673899939056987, + 0.3079727970313154, + 0.18633724231345544, + 0.30922955475638114, + 0.3366536384952241, + 0.4602814127187056, + 0.43453220901517586, + 0.31463903419373757, + 0.3418660373511292, + 0.31167463218118097, + 0.365399093886907, + 0.4086371121471241, + 0.4950168390081574, + 0.3434036564362416, + 0.2189251341089122, + 0.4680101078204892, + 0.2595988689588098, + 0.38413738502572226, + 0.37517014419927003, + 0.24615496140945933, + 0.4420307184991694, + 0.1519012629930153, + 0.15470521565742706, + 0.18763341306320838, + 0.14293840583153222, + 0.3069198548729355, + 0.28422509709841715, + 0.22312484140224695, + 0.3320202827566742, + 0.5064034351649201, + 0.29920884753776916, + 0.46124415193798274, + 0.761419416398278, + 0.4722514639905688, + 0.141262038271857, + 0.3793645257457848, + 0.3785548769430308, + 0.3286131075943226, + 0.17073038900528228, + 0.5388350159721846, + 0.3810742433817045, + 0.23035796763117547, + 0.3947417085775712, + 0.2829984342156634, + 0.39218327792762536, + 0.46176457640099994, + 0.31666652650673727, + 0.48350153140149643, + 0.3576753289902164, + 0.27177725036856726, + 0.42990494765275433, + 0.430935067577077, + 0.17016176855499807, + 0.18186357847637377, + 0.15908614909430252, + 0.5464333342913333, + 0.44344120334534987, + 0.8027271453107567, + 0.4670864288203019, + 0.3114404793642851, + 0.25336394819264574, + 0.43353329062540097, + 0.4142493696575435, + 0.5756681473592533, + 0.2663790161908105, + 0.4665294108010287, + 0.23549830126738236, + 0.2653529164486089, + 0.13480645113920278, + 0.1469598432731199, + 0.3210218569194839, + 0.12446330900601381, + 0.30590634015176676, + 0.4208390931873941, + 0.30042240480707505, + 0.14915388093917936, + 0.1745659385615523, + 0.3471539624190287, + 0.16492699026269572, + 0.40622041569786604, + 0.366404340589939, + 0.35411543646190635, + 0.33995427368024883, + 0.24822448400531766, + 0.5179409823478082, + 0.3981004441860417, + 0.2664734760421621, + 0.5572714240353875, + 0.2453921203371935, + 0.3312952743516471, + 0.25297434351966375, + 0.14473772550129904, + 0.2082338046546444, + 0.2642724312108993, + 0.36661752896227323, + 0.22696727047338258, + 0.6662441565420716, + 0.38284084341799046, + 0.19983126301613025, + 0.25248114658110915, + 0.1692107972480479, + 0.48812095589572807, + 0.19959946054589095, + 0.2584433367791381, + 0.2824346988210421, + 0.2952497056027492, + 0.15995464749805482, + 0.3371147222101234, + 0.22135424543509943, + 0.3679280251505258, + 0.7507399567491453, + 0.32967640590026126, + 0.27203709603824244, + 0.2767021082967714, + 0.28388235773910186, + 0.10792913108040819, + 0.25835574144101603, + 0.3289229152692079, + 0.7307111874364293, + 0.5045571860319207, + 0.3411919484349804, + 0.273159477338689, + 0.2784174090270741, + 0.2937775494185288, + 0.1760995119734915, + 0.35004789954986737, + 0.42641480425798445, + 0.21952939466539448, + 0.32505358411349006, + 0.30172703944485646, + 0.39135881434180836, + 0.31442347321857145, + 0.48367005926131335, + 0.32122541364759216, + 0.23508281060251793, + 0.3778672074129413, + 0.36068445279602906, + 0.49290350448428305, + 0.47153781756154717, + 0.20946202834615413, + 0.6988543702647709, + 0.6924891981204229, + 0.1790833439035082, + 0.22281966168945763, + 0.2962464079072148, + 0.38303908020118455, + 0.1958445286854277, + 0.2944821417474599, + 0.4298884994890377, + 0.4442123203226628, + 0.3949701366988139, + 0.30347537874539476, + 0.27584795849850025, + 0.2873767321443108, + 0.2768837808498725, + 0.3877941399858318, + 0.3370691409375647, + 0.3257280482302164, + 0.4327869315118264, + 0.34278070445337955, + 0.493465026814079, + 0.40151663637439095, + 0.5806562897373565, + 0.4057690675269979, + 0.34756914786608284, + 0.4334200550764225, + 0.3955078168154765, + 0.29019546111127403, + 0.5934450250813368, + 0.5618602938086247, + 0.3860901914294042, + 0.3244629686324915, + 0.3180132571920619, + 0.46947232024071683, + 0.23522366119742313, + 0.32963115825903283, + 0.31908738286896476, + 0.434735287108827, + 0.37460159274539134, + 0.33325766179397937, + 0.28376922890952927, + 0.463086571804689, + 0.43401639615748466, + 0.41114366071131003, + 0.423826860466582, + 0.3155283167755431, + 0.20559632716663778, + 0.12381791493612909, + 0.2232421827087729, + 0.4667092396524798, + 0.4042512751333207, + 0.13003142974138943, + 0.37845981581236543, + 0.2960932747070228, + 0.30651758164615456, + 0.27282213779574976, + 0.32140095535019014, + 0.2651897480716006, + 0.39627729975890014, + 0.5934571818791856, + 0.23726487903441437, + 0.366368426781588, + 0.5050402992133249, + 0.3074875330912043, + 0.3944542812722244, + 0.5450092435420117, + 0.4063551351040068, + 0.4069583978053078, + 0.3366267556873539, + 0.3235649073415287, + 0.18623131634346793, + 0.32995051084923543, + 0.4349381422248452, + 0.45530107503957457, + 0.42183966730000716, + 0.24211420702434555, + 0.4164612479456033, + 0.38212093759030513, + 0.44703534573330533, + 0.378708355756685, + 0.43930626870999057, + 0.3600635702143233, + 0.4111183329724241, + 0.4011607479564389, + 0.3746428678394253, + 0.20008877925477545, + 0.40257433732062287, + 0.4263602135272406, + 0.21433441530928818, + 0.31068198154804955, + 0.4181163597765165, + 0.33233984838623204, + 0.28598152579797564, + 0.45437799358191344, + 0.46676533607826126, + 0.37309841133227206, + 0.39280285368316553, + 0.26578198905053685, + 0.3826817890537515, + 0.30121043401435094, + 0.3121702632522353, + 0.3295149343857568, + 0.3918289848440541, + 0.32354440262963374, + 0.38726269088008713, + 0.37065122696936414, + 0.7028832989400071, + 0.5244045729089474, + 0.47474156613706203, + 0.511592871244369, + 0.4429050195346732, + 0.395406934952951, + 0.517257152941879, + 0.3151838927632771, + 0.3484461243151327, + 0.4425733767802326, + 0.29653401595325585, + 0.41275287092161705, + 0.4690251010491613, + 0.41637777945289106, + 0.4253723363184862, + 0.4174049793827201, + 0.2879816940275261, + 0.45908251496174357, + 0.46879566295464176, + 0.3340549991844644, + 0.38097589580453184, + 0.34522502104359004, + 0.31304414999539626, + 0.3871320159740237, + 0.3941898010500427, + 0.37755463095923525, + 0.4159272649390872, + 0.34713789309835735, + 0.4301624457197151, + 0.34001944279701496, + 0.3645313144219601, + 0.40798202223902785, + 0.4090206423776473, + 0.3841199678154216, + 0.4442106990174219, + 0.36613167614155223, + 0.3182680111070414, + 0.32381510426884813, + 0.3902916156453149, + 0.38018186139026433, + 0.4157643188897808, + 0.4254579749835348, + 0.3920457410299545, + 0.3806719491988647, + 0.32372602118303284, + 0.37350175640063066, + 0.4659285797706179, + 0.3475881154531543, + 0.18581815703218205, + 0.2930913771510924, + 0.3885171364312127, + 0.2710166939606777, + 0.2771082926527546, + 0.4186725303824527, + 0.4197063623870332, + 0.2527244353222533, + 0.2829827212853183, + 0.3085645453928915, + 0.2866166975862248, + 0.4269086196993659, + 0.22329909038687526, + 0.22311656112160028, + 0.2363614240323497, + 0.24233676740689564, + 0.17500392277593985, + 0.23271393475025445, + 0.21057056237202515, + 0.23221410714530255, + 0.20879938853412291, + 0.33336085613910776, + 0.23376642827205102, + 0.28047849714431383, + 0.2885963400057917, + 0.36249915799037735, + 0.24878642599830078, + 0.3471959634416379, + 0.29257548607804357, + 0.2927115627793225, + 0.2941800049687774, + 0.36336522596309007, + 0.32506288602725686, + 0.5240382875911606, + 0.2534889268792121, + 0.33462159525704654, + 0.43815584809750524, + 0.4236544816004041, + 0.3416309135725086, + 0.39062099185150423, + 0.35341585059299685, + 0.3880754681740758, + 0.4291447781157108, + 0.33777220687386067, + 0.541046671254246, + 0.38109588395686095, + 0.3642861124744676, + 0.3110940681899487, + 0.38634136899178184, + 0.3207279391855444, + 0.39142885838512254, + 0.3779520548648185, + 0.3916867492915712, + 0.42870793189766226, + 0.5604863585541684, + 0.37874516297456506, + 0.3415840550158766, + 0.44294195254778956, + 0.5122660697198569, + 0.47124229609004326, + 0.4110675938226276, + 0.456250872568822, + 0.4739292665455846, + 0.3438505082860616, + 0.3397570612647006, + 0.3773552281501464, + 0.3096933347099818, + 0.5207458397543216, + 0.4657699800656456, + 0.3545729853620521, + 0.3236024507330876, + 0.4343268801158952, + 0.370424776459385, + 0.4537669163187805, + 0.4051968342275225, + 0.3138870401557942, + 0.46726550377048004, + 0.23850321658847656, + 0.3557106609095587, + 0.44517447961352846, + 0.2811713696555483, + 0.353949176585692, + 0.26517877863671313, + 0.29727946800498267, + 0.31636951487380155, + 0.330239246384545, + 0.38372275164111125, + 0.2838196168029728, + 0.2700480237059723, + 0.3642500590122027, + 0.33071829384060974, + 0.3284079542495342, + 0.39412020480137167, + 0.30376212549209675, + 0.37307968658182317, + 0.4625807741157777, + 0.5337788899577366, + 0.3919455694504519, + 0.2974455896966319, + 0.43267566199587487, + 0.39344187103276784, + 0.33973267455663286, + 0.3555978511838084, + 0.26210515202861284, + 0.2666193208592378, + 0.23721925238537472, + 0.32193312463027535, + 0.24699480388445183, + 0.6238661423746091, + 0.3955227697239285, + 0.351750798029465, + 0.27545037309424447, + 0.2819640473184799, + 0.2954173398872383, + 0.30425861513745867, + 0.27515206309345885, + 0.2794030085278408, + 0.2692587985996827, + 0.3411710819137018, + 0.30568308227661817, + 0.40593805793361254, + 0.34858034544926386, + 0.30454456386187784, + 0.46006721190542305, + 0.45613269042925986, + 0.31260795197505425, + 0.3591657709648099, + 0.3525906205392218, + 0.3362945183217051, + 0.2949312139055417, + 0.48000978559692664, + 0.4078323342195921, + 0.5515377669894799, + 0.39757458246712063, + 0.3897739387507764, + 0.40863080735913093, + 0.41463088201900755, + 0.29078685843245244, + 0.48404854908112066, + 0.37596555497164397, + 0.21301359778450255, + 0.38351944084702094, + 0.47874116647105547, + 0.313376569671995, + 0.3695725085173783, + 0.7152305483311189, + 0.38159242249141917, + 0.4311616155421005, + 0.2428930084524839, + 0.40183214507394516, + 0.20350354481692493, + 0.19975357677368333, + 0.24393869871390464, + 0.1967654623300893, + 0.40102922223296694, + 0.1396367252028199, + 0.24921976535043672, + 0.3768703040285714, + 0.36108022377365484, + 0.3544900931203441, + 0.3683560742236908, + 0.26599751088550455, + 0.3058199370577074, + 0.4664600879756473, + 0.4801218875542452, + 0.28146017573527, + 0.3889357562311251, + 0.3248580275314019, + 0.7004447055686296, + 0.2997438715347221, + 0.5249727137669752, + 0.7197435513257489, + 0.6957726864152782, + 0.376757776563167, + 0.29542362332221667, + 0.2385173791725022, + 0.3859510148677174, + 0.4265159239022005, + 0.4503718556668207, + 0.5041324240993744, + 0.3648855156833504, + 0.5119272658996364, + 0.3175889321321663, + 0.30778638910730266, + 0.6736034414183364, + 0.6602439018807097, + 0.3671017830972141, + 0.35160706786643736, + 0.3897697289765821, + 0.34984488702547106, + 0.3900148985128119, + 0.6190767034742352, + 0.6270146641790615, + 0.3622185640467356, + 0.37694145996854467, + 0.3787999972870505, + 0.2783043747679959, + 0.4386595589003717, + 0.2485701029827418, + 0.39609364843451267, + 0.37427907058160936, + 0.2947475237470404, + 0.37839682832802224, + 0.29571148197559577, + 0.23358196928351777, + 0.4608397742791646, + 0.33212867177774, + 0.33487603873973615, + 0.30724816396971333, + 0.4166701971509831, + 0.3423092042567635, + 0.2672406766872706, + 0.2927171164279, + 0.20796182645609543, + 0.19096434295751327, + 0.21078745990993739, + 0.33097528515530716, + 0.349727746227323, + 0.2276205614461423, + 0.2957369093032005, + 0.21843677919296092, + 0.35960550720359735, + 0.33241454392950365, + 0.4230666735448857, + 0.25403656423862536, + 0.3307514347429417, + 0.42886365664916637, + 0.26097151655235384, + 0.42215500395881606, + 0.3605578652536481, + 0.3385361033475377, + 0.35410275682584275, + 0.27612813888714693, + 0.17094043608595885, + 0.21553953907739853, + 0.2183311940001192, + 0.4314538714126609, + 0.34158336906507736, + 0.37908323745001893, + 0.33426783400318183, + 0.35970137013667475, + 0.2755974109082173, + 0.2973599590733853, + 0.28862965396488616, + 0.629195225702457, + 0.3668175785022197, + 0.4059336363685931, + 0.2933537210790512, + 0.3848450483247846, + 0.4110093621295449, + 0.34743168469882085, + 0.32501026693859114, + 0.39408712363246085, + 0.36425789267680697, + 0.4574836261962677, + 0.3340803466222192, + 0.39196755841622233, + 0.3563562790627538, + 0.3455723092117635, + 0.3143244876443409, + 0.39129570642781186, + 0.33028848161340185, + 0.39883508097744086, + 0.2507185449138907, + 0.23885580733577846, + 0.24109333961324306, + 0.31314533225693, + 0.3680291798405156, + 0.30718371465758376, + 0.3194497529085334, + 0.3775580214339566, + 0.34705182939234397, + 0.4001006644034775, + 0.3676830538869277, + 0.23983534797640593, + 0.3381588412729586, + 0.20463345128642207, + 0.34317662773206375, + 0.17104548432732222, + 0.6262539620841252, + 0.26592805931159186, + 0.3286842413532772, + 0.46505274532321944, + 0.3850716554039118, + 0.21503971497805893, + 0.3220040434124037, + 0.26324181697694665, + 0.7531972339458506, + 0.22799436369064174, + 0.417539885677871, + 0.44968224699706044, + 0.433698905100008, + 0.6128020147006036, + 0.3004938030237645, + 0.4328474340604189, + 0.38348429598190564, + 0.28472410575138024, + 0.22841156688859243, + 0.38582324985156663, + 0.22648578564062574, + 0.6903184860477919 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.352559395518743, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.36629062864173684, + 0.7316013908668074, + 0.7179632002613122, + 0.7970644700487808, + 0.6927585554625931, + 0.5944043897206508, + 0.561421567844021, + 0.6204717787770606, + 0.3769664508310355, + 0.647338926147443, + 0.250298819301485, + 0.40384974358243597, + 0.2671500054650273, + 0.2632789550966217, + 0.23743755476391043, + 0.23851787513496311, + 0.2783376644113127, + 0.4420937740936593, + 0.3117803181316562, + 0.3239360247116592, + 0.2754492859504329, + 0.6448374708631298, + 0.2543844031844192, + 0.5681177940168167, + 0.31441523335963695, + 0.11448843123639596, + 0.2927674381867137, + 0.3586802245677119, + 0.35321312658837833, + 0.4225322375239587, + 0.24089848731647143, + 0.34324686990591396, + 0.2856943580994963, + 0.30341769302555965, + 0.35908053731825323, + 0.2930579112624999, + 0.247538337111022, + 0.4276942730936061, + 0.4318697153536675, + 0.6046589104546188, + 0.7646929968347111, + 0.7584400232960389, + 0.7469198737839043, + 0.618950257390443, + 0.32805488774265423, + 0.2804893768306813, + 0.7561030214470476, + 0.2731499380517031, + 0.25028584366250767, + 0.3831451380862664, + 0.28222253826591287, + 0.7541964618608558, + 0.7734961034363035, + 0.3288165123507215, + 0.25080151585481114, + 0.2984674031818475, + 0.257355209690155, + 0.32029781223811105, + 0.30402342337740307, + 0.3952048805548804, + 0.33716327892917647, + 0.34809494118253287, + 0.2974479205622664, + 0.39888462293202426, + 0.3232133340418648, + 0.28496425580356316, + 0.30153406302042846, + 0.4203898811302008, + 0.28900082379186876, + 0.3473945168785042, + 0.38330208739007066, + 0.32525083131359567, + 0.3049258014620504, + 0.29455376662687127, + 0.3218281446558546, + 0.2937961260825502, + 0.21969209620865163, + 0.22569344382067014, + 0.3388177316451495, + 0.27938848730954213, + 0.3660016173558387, + 0.26251273484782534, + 0.30474166647689244, + 0.23668708758334658, + 0.3388765329051456, + 0.4115062951865901, + 0.3285576857899993, + 0.21208718841775445, + 0.30755027263752577, + 0.2578044043621502, + 0.26521894344325064, + 0.2704521041580359, + 0.3785221460271985, + 0.4475041350316156, + 0.2083737169004958, + 0.46203456192390074, + 0.2694970623958501, + 0.41238347459457414, + 0.22404612989885603, + 0.34812380751335503, + 0.3548605563264817, + 0.42262951950118216, + 0.19023609532455746, + 0.34370178006231783, + 0.25255953481598437, + 0.2755307978459698, + 0.35307272607678, + 0.26723433474883634, + 0.2814382781019253, + 0.7856965955310375, + 0.779341007972849, + 0.7848712711211581, + 0.8010301221017408, + 0.5625241430262851, + 0.667749586725831, + 0.6841838304661298, + 0.7347160660125367, + 0.6703330021227172, + 0.5666403041454635, + 0.6349397354176173, + 0.557035676794621, + 0.7487522253997871, + 0.6406738832181199, + 0.37738118397105774, + 0.2941728644664008, + 0.2371472267740975, + 0.27729128701137795, + 0.44274704202333226, + 0.2506890225061245, + 0.23982816294449427, + 0.2900999699566197, + 0.3418320092158728, + 0.23772198001187206, + 0.3262035467419456, + 0.34019009262883304, + 0.3537521602558632, + 0.301277204623485, + 0.26737004516974416, + 0.29757246664855375, + 0.3128887708521389, + 0.25591600350466936, + 0.2731753833264643, + 0.2140823719703151, + 0.5720364391580425, + 0.25455833193822214, + 0.4990340713084567, + 0.3015452647103513, + 0.39723330924271094, + 0.20029695769251296, + 0.5346611076979312, + 0.23976675125574687, + 0.4394045233664376, + 0.2639218241598074, + 0.20562590960168298, + 0.2778087848205708, + 0.25938352911624135, + 0.2155813289132319, + 0.1303535184071216, + 0.2659784142071175, + 0.2785854961201844, + 0.40052126255398324, + 0.3477030851027322, + 0.4046974795241169, + 0.3840693533820968, + 0.29035773526304, + 0.36664072016789023, + 0.31494228564777554, + 0.3450435056288105, + 0.3281498661813067, + 0.2962801941363404, + 0.21143052701634085, + 0.13891525942094876, + 0.33906875295508443, + 0.2349114962809757, + 0.20006217202106466, + 0.2087377999105369, + 0.3479669131098725, + 0.39853422842043484, + 0.529823960777625, + 0.3349566636201588, + 0.19483330779776056, + 0.37169216910249653, + 0.374332238518509, + 0.4008470031962531, + 0.23433012464900846, + 0.41490309829996397, + 0.3331697021542826, + 0.4001613832763909, + 0.19666420428287387, + 0.29397500371906005, + 0.27346858672090446, + 0.20596153211178547, + 0.40084150053363776, + 0.3355217929758427, + 0.429378744615427, + 0.24957504356862859, + 0.5739816104389409, + 0.25185471419284877, + 0.17081281099898357, + 0.3637452885230618, + 0.2383853851062424, + 0.22197563621701105, + 0.16861281050819976, + 0.28848451954882415, + 0.579889328061551, + 0.19775541072363706, + 0.20653246510560075, + 0.3696793280197371, + 0.2755182182093906, + 0.3350759026060387, + 0.38168371628160763, + 0.28081013864361054, + 0.201740433125383, + 0.283954344897424, + 0.2793210606558612, + 0.41799667559657505, + 0.41250771501836736, + 0.4694883164911268, + 0.42562512733220287, + 0.23272006156066388, + 0.2891103663897236, + 0.38720389739985084, + 0.22143254070167975, + 0.29385035823541966, + 0.25099347414919654, + 0.13207367547500418, + 0.4030177980790768, + 0.2855125006103151, + 0.21268962406340636, + 0.472572697595076, + 0.202456400772721, + 0.14101740417669248, + 0.34488859224318585, + 0.34383356251632025, + 0.24101648963900796, + 0.49875839948877393, + 0.2783765752214343, + 0.262988926336143, + 0.20960303853474824, + 0.3845589863492821, + 0.2480724083726478, + 0.40802375322441964, + 0.24166499914463252, + 0.32334659046184344, + 0.3260640063468774, + 0.3307614912766334, + 0.2859584187907189, + 0.2771394713809101, + 0.2904422205074027, + 0.2871761553460043, + 0.40247991777964454, + 0.23855768077759312, + 0.3351114504836701, + 0.2545374106241418, + 0.2798860334553774, + 0.31772155456187057, + 0.33403144985235, + 0.29841079128793424, + 0.3617486168812424, + 0.2249082113416926, + 0.25770544067988, + 0.4323226843437381, + 0.226669519306973, + 0.2867153591860957, + 0.2636003368925685, + 0.20698815328013126, + 0.2549277226876691, + 0.23491230395232612, + 0.22687710372335204, + 0.23510447445859878, + 0.21327281878128823, + 0.26385800014418176, + 0.2630664207342478, + 0.312701810342033, + 0.32120043257563113, + 0.3975780039213577, + 0.1976219000345652, + 0.3451667562696652, + 0.42936000666595364, + 0.22293465803302615, + 0.21925211033201128, + 0.4644957007693398, + 0.19871046336479403, + 0.20490191661109056, + 0.32976999599614837, + 0.31210426046713596, + 0.34732126087461634, + 0.26379476298288856, + 0.39757622936758824, + 0.3205911762580085, + 0.3072221610127578, + 0.26009735712687326, + 0.33206774110247195, + 0.3398828659549598, + 0.3119807048289953, + 0.3481821760143691, + 0.41766169317604285, + 0.38051755543837573, + 0.6165100639220954, + 0.2526477714493638, + 0.5132394651007749, + 0.6059286616155873, + 0.6061767514989326, + 0.6045107043127795, + 0.4261194200885856, + 0.41552564685551757, + 0.3219870117919307, + 0.6454533903085214, + 0.3812543250747423, + 0.42403525664055325, + 0.24108654273481578, + 0.4052685457755035, + 0.3363789478915128, + 0.2915744798730479, + 0.33213154607249307, + 0.26291448482501706, + 0.37896260986440256, + 0.2376676952850595, + 0.27626446884985634, + 0.29614755069758253, + 0.3072045449245335, + 0.34833855565594307, + 0.2286328193234524, + 0.27628742843571114, + 0.3017950057191177, + 0.31961117358604885, + 0.26651515712526275, + 0.310495736907426, + 0.47224365374918414, + 0.34716286941584446, + 0.3001601541850633, + 0.32263594935977596, + 0.35213046437944034, + 0.24534273090590691, + 0.2735577755327546, + 0.24872502704217667, + 0.33333090016693784, + 0.2652826282695996, + 0.3525701391480226, + 0.43817449335689335, + 0.4280374010610559, + 0.31319543678637457, + 0.2578391020018306, + 0.3777986040481426, + 0.22249684050618793, + 0.275818389145443, + 0.3758666038400189, + 0.2653551825728106, + 0.7981153338886705, + 0.46242432769065456, + 0.36329028039525174, + 0.33944442907624517, + 0.39784803454195034, + 0.4076739187354931, + 0.3334899107739403, + 0.44500850515624374, + 0.2982364498438842, + 0.43406484236630544, + 0.3389118598639568, + 0.4135259463924025, + 0.4056167986094416, + 0.3918488594613867, + 0.3701063187527096, + 0.28916075711363975, + 0.29412635509629487, + 0.4452672447237841, + 0.47484644438891654, + 0.35740265173300373, + 0.5311219171115301, + 0.3631142885909491, + 0.4937159754267374, + 0.33821008245679457, + 0.36643749406369813, + 0.38692328937293985, + 0.4576983209334986, + 0.445006346198674, + 0.389412722190785, + 0.43223275653532217, + 0.40989685955563965, + 0.41906928269655314, + 0.3052317440844491, + 0.2999991110125969, + 0.28495333697628944, + 0.19396195467243219, + 0.29611915479621803, + 0.281219836024802, + 0.2471612000117066, + 0.24772452004368184, + 0.344489558210405, + 0.25789083946745733, + 0.24912001670084769, + 0.2729124908415793, + 0.4330496661617938, + 0.4825117401187068, + 0.42464542750408024, + 0.4568949745619066, + 0.36732985708601534, + 0.45544750464833395, + 0.29235127666144384, + 0.5785574522756761, + 0.37200648101503186, + 0.4365901843072475, + 0.34919221711678233, + 0.44981817815536546, + 0.43215644225390104, + 0.39682585450922864, + 0.3560930530535173, + 0.38335294964818795, + 0.38993989806707696, + 0.4096873151042715, + 0.22200782281201326, + 0.2846617280528562, + 0.2572902821637186, + 0.27325481765420295, + 0.41107827082058707, + 0.4565615579607734, + 0.23695626133642136, + 0.4642154756226303, + 0.37469661600925175, + 0.3581579540043524, + 0.2969076998216083, + 0.35949760444645446, + 0.3589091000941917, + 0.2523140266732751, + 0.26907398184861414, + 0.8239337957443074, + 0.3382743705769631, + 0.33646537701753726, + 0.37954547683277806, + 0.36529069731415564, + 0.3677834500872214, + 0.3359545108409, + 0.48923524551177394, + 0.3776679031427387, + 0.2841263795769821, + 0.33389559250860745, + 0.358224256588027, + 0.36801648928939446, + 0.34905828521199705, + 0.3500889235453269, + 0.4099299670768068, + 0.4016480707946923, + 0.23798019904785167, + 0.3949788239898603, + 0.24415781304787207, + 0.3232543289249381, + 0.18668145607380499, + 0.3853328625830226, + 0.2964994803756505, + 0.23601820690098652, + 0.36594492680628216, + 0.16642799807541997, + 0.21880257712865742, + 0.47720812045181127, + 0.43720086927142204, + 0.37465468486999515, + 0.3704611471934616, + 0.3352832246893677, + 0.349935600130519, + 0.4892790235560481, + 0.2985299184187114, + 0.5148592302650287, + 0.27019338690476075, + 0.22190734339479395, + 0.17208906595000978, + 0.3249060998479363, + 0.4012609819975288, + 0.4095484202009503, + 0.44226771113538726, + 0.3467792712335919, + 0.3846198096301066, + 0.2803109056443852, + 0.37129550001544354, + 0.3085777062620411, + 0.21682077803779062, + 0.2676200032438107, + 0.28070640959911003, + 0.3443549680140643, + 0.26153166568272296, + 0.28018360122458263, + 0.2957987940084932, + 0.30853040822053723, + 0.40630800345922546, + 0.32889865743184427, + 0.2574828456459174, + 0.4081162169685708, + 0.3493222127738676, + 0.4458800906201543, + 0.37863925763883693, + 0.47403998136241277, + 0.4874583911916353, + 0.6374070619663859, + 0.6202813676772101, + 0.43716681257354634, + 0.48316099547838354, + 0.36035044050488957, + 0.484597870231657, + 0.3552802709751935, + 0.3620719336089774, + 0.3196156907882365, + 0.5351055009528223, + 0.3303715759900732, + 0.47202745140568797, + 0.6155145066683075, + 0.5951508317410314, + 0.5323175715940559, + 0.6153342327451891, + 0.5021548404125155, + 0.42507372259450993, + 0.3669871598346392, + 0.32582077318057534, + 0.22764575229801584, + 0.28164067239932206, + 0.3875608496148256, + 0.4408333641244913, + 0.2947385800317466, + 0.2322487884129826, + 0.1591526865872648, + 0.3455795555070984, + 0.32754096446973713, + 0.3504746134173553, + 0.25263536593113284, + 0.4666856502086036, + 0.3366419310405578, + 0.38792182394210206, + 0.27529950674650705, + 0.3379610459290783, + 0.4221718920442834, + 0.260967861725663, + 0.28322209306522805, + 0.3371815315390847, + 0.28244900555858043, + 0.2783040666432296, + 0.26447849127794637, + 0.2773743961280985, + 0.27567056011461843, + 0.29665085282223547, + 0.30135825582921744, + 0.3254642004394761, + 0.20497451939481845, + 0.47407518944515803, + 0.3355996273471617, + 0.3993747674847591, + 0.41446084885930085, + 0.29235950240356134, + 0.37666193256812086, + 0.37042012195154916, + 0.3925067397334782, + 0.49062864477930473, + 0.31207420346728787, + 0.3492385129758697, + 0.38912556837001333, + 0.2592575486463234, + 0.3631653825806738, + 0.3329369357544677, + 0.2521268189539668, + 0.3671235950682948, + 0.3750096634706674, + 0.3391395262535734, + 0.22652124711908994, + 0.3069908286817063, + 0.17836529890197828, + 0.2830460117842632, + 0.5052292631444385, + 0.4298621236683947, + 0.45570401758535084, + 0.4392818615968832, + 0.5263489147985608, + 0.4377689979332688, + 0.4750408084491828, + 0.4295889360731599, + 0.4782160792976924, + 0.31189485291608693, + 0.28974235035839196, + 0.3284893981770256, + 0.25882958395475797, + 0.22581677019024549, + 0.3009689972287804, + 0.29604111952354634, + 0.26419859739105367, + 0.2191322097891563, + 0.29486101466653564, + 0.22835332761026492, + 0.3603919432897884, + 0.2566759780660859, + 0.4560295161182336, + 0.4656282092490403, + 0.481962894077874, + 0.23748189043202952, + 0.31671299296380306, + 0.5423030195425095, + 0.251388056775152, + 0.19614334658218877, + 0.21347375156621134, + 0.807191667136356, + 0.4447562985253446, + 0.3513918321789796, + 0.4621737356461, + 0.409119338062679, + 0.4148433007757541, + 0.34710233596882756, + 0.3819130072161786, + 0.41536391738028045, + 0.37273536278839775, + 0.4589453436292879, + 0.42609125897838757, + 0.19242556212037787, + 0.28224553452178813, + 0.2069938261620358, + 0.2766762680610375, + 0.22014230430023424, + 0.41223890985979045, + 0.3455471230484393, + 0.26990893805281635, + 0.37376537487838274, + 0.40729375489650094, + 0.2761412304050554, + 0.3663652901187348, + 0.22047141520860655, + 0.36998402027329924, + 0.41752090394425484, + 0.26794310183637393, + 0.27332089469096965, + 0.21665861580931145, + 0.4336736155628054, + 0.21228921685784077, + 0.27085723773718057, + 0.44140227286090156, + 0.35244914696857965, + 0.3703233711904024, + 0.2632459743567765, + 0.2893690248691715, + 0.3341231785623575, + 0.30469912145355055, + 0.3859699174299929, + 0.25956707107569577, + 0.40666308609991786, + 0.20666594385913026, + 0.4092649851848102, + 0.35964020981048067, + 0.32053949296768686, + 0.4181073428470443, + 0.3294924477460568, + 0.43737243308989554, + 0.27535706638720686, + 0.4236225808864425, + 0.3132549283911886, + 0.23965287402950303, + 0.37287397911262254, + 0.24931114225081383, + 0.42131776071560245, + 0.17094882722601681, + 0.6489087244460938, + 0.6562170280758005, + 0.5437666742989197, + 0.5443843647816765, + 0.5838953593698442, + 0.5818793030457606, + 0.22586810050652142, + 0.3081301719215782, + 0.2801448471928764, + 0.33160153457923475, + 0.2779545681794383, + 0.2952978843199669, + 0.2991062226515471, + 0.3171350966805998, + 0.3599155126027414, + 0.26520650872590396, + 0.296362530890156, + 0.25114889569471316, + 0.43580368210461845, + 0.23567957974043002, + 0.3527958514413518, + 0.35369371416261586, + 0.28722158288287214, + 0.4175933374306093, + 0.3130037299018572, + 0.42631708855242634, + 0.20754825560177562, + 0.39495251479431176, + 0.2998597595543761, + 0.29287691998513304, + 0.2532213017447966, + 0.2948515347589505, + 0.2685479047153871, + 0.2923582392338897, + 0.4690902758613812, + 0.41900156062460225, + 0.499897999790351, + 0.29291596158315686, + 0.3684189101746629, + 0.31021515748092976, + 0.2989128448331175, + 0.39811499993272975, + 0.26418375164161007, + 0.1784123108406575, + 0.3531380370559176, + 0.2732970156229332, + 0.29034673191074645, + 0.35940459821305865, + 0.2733142420343855, + 0.21401618147550053, + 0.2181140673425142, + 0.28200087235320564, + 0.2849518073621364, + 0.22941687462509075, + 0.16930127033342143, + 0.2509458049467125, + 0.4018854359155703, + 0.5441154641000091, + 0.4780719316753888, + 0.43851025059633786, + 0.48315955211908523, + 0.3913475475096094, + 0.45181469122505125, + 0.4503787865747798, + 0.38508374766466974, + 0.30807670220318006, + 0.3361015223140859, + 0.4052026151814398, + 0.5508254691907915, + 0.1266663848492591, + 0.2633359523952902, + 0.24758791058555962, + 0.20013867752765988, + 0.20761149937170237, + 0.3302953113537942, + 0.24487701509332233, + 0.32495036479710965, + 0.7054883259095126, + 0.5851665437828616, + 0.7366567262681761, + 0.7498544723470413, + 0.6210193860330003, + 0.697758768577576, + 0.7133454325394444, + 0.283783635192975, + 0.28049255395172523, + 0.3057637948007119, + 0.4055221156569354, + 0.26783304288359167, + 0.2879807760678996, + 0.32030522018273283, + 0.3080942048937178, + 0.32035776129361093, + 0.3029647494080191, + 0.23134987675255012, + 0.3989946271043395, + 0.3105925073750519, + 0.30947349974473815, + 0.3730393648401722, + 0.24757236297379678, + 0.27972151428479186, + 0.29674463626821146, + 0.2859600913770266, + 0.20595544213784558, + 0.3076726038607235, + 0.22063343015583256, + 0.4736763184533773, + 0.4364047217426452, + 0.30785290979050295, + 0.23434395468379693, + 0.14448454788335696, + 0.47418965803261687, + 0.5427930337319139, + 0.6229100546106322, + 0.2623439127688385, + 0.31963272699803524, + 0.1932541189427909, + 0.37333756926573813, + 0.2949495226238156, + 0.2560989690471945, + 0.42609975311633147, + 0.4036117384152573, + 0.4090772778221968, + 0.2740813133609591, + 0.38174588307140883, + 0.3631024845047149, + 0.413937908408567, + 0.3759468022504969, + 0.40742288160627643, + 0.3150148530414007, + 0.3141973806376536, + 0.3424684915939185, + 0.3187115105879245, + 0.22295999045489662, + 0.28229730631074, + 0.31289621657325223, + 0.3407588614002902, + 0.46105728125309725, + 0.14536778873932513, + 0.5195850953574693, + 0.6802236903365266, + 0.4951539911249801, + 0.33995751190838586, + 0.26844221878130436, + 0.28900997689661195, + 0.22471613145740513, + 0.3234670810990333, + 0.39362641842021795, + 0.37891862852816127, + 0.28127126253391455, + 0.23764211076068556, + 0.3723642106325479, + 0.4657978540076005, + 0.5083521304139414, + 0.5725832339229106, + 0.3371304253102388, + 0.4852995845341839, + 0.2667824515565486, + 0.3232850056444431, + 0.21382224719534934, + 0.19521564442604028, + 0.3161617252095432, + 0.2851898654002957, + 0.41589091353431457, + 0.3554062688778982, + 0.5793963594017759, + 0.6347860183057848, + 0.7298192992249799, + 0.5633178112589066, + 0.7391237802927201, + 0.3132745601987258, + 0.3092101165510909, + 0.3090420151193816, + 0.32804585093803185, + 0.26043237647717626, + 0.41611344567032854, + 0.36824967146648724, + 0.43676169237019963, + 0.3956017523268914, + 0.34494206930106197, + 0.34890936218852747, + 0.33316404941177385, + 0.26484139613604835, + 0.3183338988552016, + 0.2813197495808255, + 0.3206855276241134, + 0.37077942500314715, + 0.3485158026702594, + 0.40638134486521565, + 0.2735784464929686, + 0.3117281807658968, + 0.36955261322869287, + 0.2493068144194148, + 0.35898113468421444, + 0.3339028635633698, + 0.3399517372072442, + 0.3121704270804666, + 0.37714288353023995, + 0.26827598227822713, + 0.3311940090409064, + 0.33826804466617677, + 0.37862832287016257, + 0.284480756071729, + 0.4380399752037412, + 0.3272434929464557, + 0.3032048597842392, + 0.43736816198823664, + 0.35408062653439043, + 0.7177236047495661, + 0.18924659627964344, + 0.22455736601785714, + 0.2993449494174327, + 0.26967566254155523, + 0.1971923856712707, + 0.18378603689507803, + 0.2501714147868223, + 0.6526445676517776, + 0.5309827847183478, + 0.6151612279821856, + 0.27304447153064, + 0.2697133899873886, + 0.21569452134568945, + 0.27052249112450466, + 0.3103335079474996, + 0.15765040285631457, + 0.20150812674482627, + 0.30612500614185595, + 0.28402700103323625, + 0.33993225192808946, + 0.25623436498892876, + 0.3649337960836414, + 0.2949550986104996, + 0.25789686716916393, + 0.6680540246730998, + 0.43299234418027716, + 0.6651872079772576, + 0.5458570527405103, + 0.1828341013239223, + 0.4257845248829182, + 0.3927524327254155, + 0.42860413794373153, + 0.38223944115302744, + 0.32614501951149905, + 0.39557097964523646, + 0.38282276186843833, + 0.425905840869346, + 0.37634339495532715, + 0.28353406168542306, + 0.3108127422322657, + 0.2051331794131004, + 0.23250055738089345, + 0.6168949305137404, + 0.3419740850056944, + 0.5486941308100461, + 0.4454999098222382, + 0.38842561809728626, + 0.352529798282486, + 0.3461055978863772, + 0.3998664389953461, + 0.34350277427156684, + 0.18201626327348336, + 0.27263987428448727, + 0.3485034174361426, + 0.189193500026144, + 0.7367120819738502, + 0.3673904011301564, + 0.5466881832567049, + 0.7070670292944197, + 0.6576859656579236, + 0.31891193539390916, + 0.7845446795517137, + 0.45554421914948157, + 0.6350966569505302, + 0.33673614976402266, + 0.35991619527461677, + 0.27027071398317737, + 0.25209145132514593, + 0.37689641349506114, + 0.3658907408069636, + 0.36435169553145014, + 0.2706261127617317, + 0.2779561015371315, + 0.27903104895016684, + 0.30761914929499184, + 0.3683116436603128, + 0.18936278789501385, + 0.3974711805924283, + 0.3089514787267539, + 0.30835487900570313, + 0.3861866417227474, + 0.4263427059841198, + 0.24165082953416436, + 0.3284781983285937, + 0.3550762709314625, + 0.4611363129304021, + 0.402143947242335, + 0.3328607149256244, + 0.2245447887285828, + 0.44955154071425574, + 0.5048656758753655, + 0.49614878981430843, + 0.3705113705285348, + 0.5458544874088519, + 0.556583949986255, + 0.34952675137177874, + 0.399837932708222, + 0.4451173035351351, + 0.3985794969464103, + 0.3632249120683343, + 0.25499432170278236, + 0.20197595343802394, + 0.22949092022156792, + 0.28875090317908625, + 0.2158570016671487, + 0.22999661544503225, + 0.23645388513177823, + 0.2614153128609619, + 0.28105668252552446, + 0.21610760487679415, + 0.24863740612479432, + 0.23506976114586503, + 0.2506048861907752, + 0.3134684481080359, + 0.23494981611269863, + 0.24041233329919626, + 0.3664659789064584, + 0.45741860680269536, + 0.6124369936010535, + 0.2336764860837915, + 0.44309198717518383, + 0.43456726548122665, + 0.45710382568974367, + 0.440294608077155, + 0.3580440869187052, + 0.31880873874893856, + 0.2760918189915821, + 0.3236865158910961, + 0.3025750982958274, + 0.6020923853033392, + 0.3459300954283519, + 0.3831200003470225, + 0.2028590912836639, + 0.17680694274383968, + 0.33212679444816084, + 0.40443026497961426, + 0.3143914551080501, + 0.5333552414397893, + 0.2697173047744512, + 0.5875045758869077, + 0.4188880899376035, + 0.41629048978521155, + 0.43012171141609734, + 0.40128359685327847, + 0.35032819444357377, + 0.4404997284046873, + 0.42215295241715567, + 0.44110502779968447, + 0.42126677646947264, + 0.36928503551928854, + 0.28584430237552166, + 0.2101783333559105, + 0.3203680764309048, + 0.17906775668100322, + 0.3306661735769798, + 0.34908696321694554, + 0.36073358161181196, + 0.34384166758198353, + 0.36360938757138683, + 0.3479559932432104, + 0.4273146565908288, + 0.37834715797332064, + 0.29411830406499445, + 0.389950247654848, + 0.4449352124845602, + 0.2978997761141327, + 0.22204926213058, + 0.2825911005972334, + 0.3187736870046754, + 0.3440157563888024, + 0.19525198280926462, + 0.3516642751784331, + 0.2020702656176219, + 0.32383321872429927, + 0.3871062341335743, + 0.37501212768947373, + 0.16158159652663634, + 0.23315821412009877, + 0.2582328350451903, + 0.18703913769825248, + 0.18165665866810304, + 0.24560669465841928, + 0.35180744501520983, + 0.20547945276431995, + 0.27539315303848455, + 0.3866749824571506, + 0.2518893130987635, + 0.25872603040943754, + 0.3318957046974225, + 0.36324395368930157, + 0.12919755260610802, + 0.4039156121642507, + 0.28857012066344695, + 0.24492873887548136, + 0.2289049124419141, + 0.35860849272520906, + 0.29501030502983433, + 0.15430021119883974, + 0.3958542502958197, + 0.3315545424738874, + 0.318773976104654, + 0.39013089898586323, + 0.28718268265761765, + 0.2423733801434745, + 0.28654029184883767, + 0.18754701406571425, + 0.29152475116525745, + 0.2595077396705987, + 0.2959012227628453, + 0.30036113524062824, + 0.3243448399098133, + 0.3589828773461198, + 0.3169371609748325, + 0.24385642548298905, + 0.27719813520850894, + 0.3396712504367918, + 0.29312556662120387, + 0.3043128023615442, + 0.3797575237633639, + 0.3219737615576263, + 0.4286670598743424, + 0.47144141954466984, + 0.38787604333198833, + 0.3862442383945503, + 0.268333171940694, + 0.19431608697510278, + 0.28331703361264016, + 0.3591427512105243, + 0.2696383627433714, + 0.37903025231562226, + 0.35983710731503715, + 0.2631731239821132, + 0.365172756236146, + 0.26344707855113286, + 0.2244471773197006, + 0.31834314817707265, + 0.34823992455814, + 0.3595248163942301, + 0.3702621870343647, + 0.23670752536695444, + 0.48793532134512374, + 0.38749676526153326, + 0.4138240691700965, + 0.4214976250814173, + 0.4929637075430526, + 0.3419686654078747, + 0.2618359552399033, + 0.3339552338064686, + 0.33508082307842324, + 0.3328998049766994, + 0.2717477319549358, + 0.24970691779542292, + 0.2625906433175087, + 0.3320717852042375, + 0.21564668531752926 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.03960787550567231, + "valid_tracks": 4, + "per_track_pearson": [ + 0.04239889311807491, + 0.03786319848354249, + 0.04060818056968738, + 0.03756122985138447 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.13040267237774725, + "valid_tracks": 626, + "per_track_pearson": [ + 0.1306572653069055, + 0.12391256623662203, + 0.10047069411595314, + 0.11713822768626947, + 0.14189368388467746, + 0.13475586892667407, + 0.09249953170271356, + 0.09598874941602943, + 0.1396250270921677, + 0.12586638345822249, + 0.12691738008830802, + 0.1275479792989979, + 0.13022004528692166, + 0.1204568706596265, + 0.11692679599376118, + 0.12339367484888586, + 0.11450071270790364, + 0.10686755943422512, + 0.13865555736963492, + 0.1174245749433282, + 0.13865024852380323, + 0.11906008599065414, + 0.12956642104182586, + 0.08448950476430843, + 0.1175417342756212, + 0.12248983986985391, + 0.10583448599092242, + 0.13213293968772005, + 0.12790494725875284, + 0.11290098751119558, + 0.10895162351070331, + 0.11861969451123194, + 0.0871969434347343, + 0.12421715721299852, + 0.0962166597564298, + 0.13500473424608403, + 0.09061227948147034, + 0.1300091247926734, + 0.12952163175279027, + 0.13088767809677596, + 0.12664542103870513, + 0.11318076484020956, + 0.10997931112829473, + 0.10282808600561763, + 0.11030207861707121, + 0.11052514873062244, + 0.09195922975511528, + 0.11700023209534408, + 0.10204362452729622, + 0.08628205177766088, + 0.09939907390068904, + 0.09074176872461526, + 0.09818400868648026, + 0.1131656206877781, + 0.10959212977844329, + 0.10268101568538289, + 0.12645122261755296, + 0.12079953357318989, + 0.12300053551782425, + 0.12802350423988237, + 0.11014055177236125, + 0.13269237397190795, + 0.13520538765963255, + 0.12776720740526376, + 0.13259512396214, + 0.10863060574220637, + 0.1326593002420469, + 0.10451156327994245, + 0.11452609048061106, + 0.1118893144290243, + 0.13199009399090073, + 0.13320200302297464, + 0.11951985397377778, + 0.11379698540011071, + 0.1262917051276519, + 0.1265758430062536, + 0.11635550410769163, + 0.12958066879496444, + 0.12548389195097442, + 0.1305253775140466, + 0.11714507394910939, + 0.1223546963851826, + 0.1333692746772458, + 0.11072996291427406, + 0.11242702644512335, + 0.1253823352132722, + 0.12245677455391156, + 0.12622337029230826, + 0.11634951327467467, + 0.133613164642556, + 0.13002553345300236, + 0.12104010437629972, + 0.12355027964793908, + 0.12399238281996403, + 0.11052221093202809, + 0.12097743358110702, + 0.1217309750532783, + 0.12811847637383064, + 0.1288529862027313, + 0.1214812254545832, + 0.12445634239363154, + 0.12375583507985466, + 0.11919944980435211, + 0.12384051095325989, + 0.11936268210601153, + 0.11721650538049355, + 0.10142375332272083, + 0.12928393841657856, + 0.125515746549636, + 0.12120719549575729, + 0.11792224805721005, + 0.1218092096128794, + 0.13464197981621542, + 0.12441829380582743, + 0.1161090438197562, + 0.12342073159306781, + 0.11456389689879633, + 0.12246634931947892, + 0.11553626654710604, + 0.1089762570747978, + 0.11281679991406894, + 0.12288347098431038, + 0.11784812807798388, + 0.12601842246640219, + 0.10310839829141223, + null, + 0.12339466374869978, + 0.11138492135712333, + 0.11273929054507266, + 0.10905845938066623, + 0.10098615737071362, + 0.11310510307503505, + 0.12265716840339543, + 0.12121475931550937, + 0.13211286940573191, + 0.12138939118243601, + null, + 0.11040787830798585, + 0.11545932179703862, + 0.1042584924771297, + 0.09002058093189039, + 0.1195250676881928, + 0.09873113894400405, + 0.12156614170549235, + 0.11634464856099255, + 0.10433704972466626, + 0.08781850095490239, + 0.09859777880959578, + 0.08703432648901635, + 0.11192416307188839, + 0.10435677097064092, + 0.1142999428510802, + 0.12227293805604568, + 0.12700832027982895, + 0.10482360680125023, + 0.10904484335833974, + 0.10968658791568907, + 0.09747476011204723, + 0.12141584254556945, + 0.1068366873751187, + 0.11194209203349963, + 0.11118832993945217, + 0.1291932923071497, + 0.13226856994726738, + 0.10627700954846743, + 0.1070538244737696, + 0.11476614773747702, + 0.11146147727184495, + 0.11471386223721312, + 0.10951257418857561, + 0.10308125420228664, + 0.12120721730520027, + 0.1101346709484009, + 0.11069560871663346, + 0.10698728840869653, + 0.08611733360542742, + 0.1183258323308509, + 0.1139704999411507, + 0.11299911425818443, + 0.10938555851792994, + 0.1116947815315508, + 0.09837293948476399, + 0.09664112941160846, + 0.1262499888357524, + 0.11881646918082496, + 0.09002380706616245, + 0.1428404608943993, + 0.1260434138300435, + 0.12733909358010798, + 0.1206417954948336, + 0.1341260228349753, + 0.1127739867460164, + 0.1426148220216415, + 0.12647178472502735, + 0.1345922727926258, + 0.1336499125918867, + 0.13214235241040273, + 0.12350160501901981, + 0.13823475125754517, + 0.12654846640941936, + null, + 0.1397890464598405, + 0.13185695642047496, + 0.13432649958393525, + 0.13355693992051423, + 0.12637799098690344, + 0.12267257573067562, + 0.13279638764517984, + null, + null, + 0.14030053131717354, + 0.14367770403982374, + null, + 0.13813173949272214, + 0.1238582290830926, + 0.135592986510638, + 0.1262056373138844, + 0.13254879700406225, + 0.1303212023090997, + 0.1322921667620799, + null, + 0.14235054505223954, + 0.12280691840703022, + null, + 0.11562720682636088, + null, + 0.13069393996045758, + 0.14125892119125924, + 0.14427849767190368, + 0.14551637130271897, + null, + 0.1319151520865191, + 0.1312765763247291, + 0.12199778702125795, + null, + 0.11430184522586694, + null, + 0.13618746485060054, + null, + 0.11308979976320734, + 0.1355687655010785, + null, + 0.12992428411001908, + 0.12541739875448818, + null, + null, + 0.14586804236667217, + 0.14217155116597896, + 0.12644906599257097, + 0.1270009921630402, + null, + 0.13374989266690174, + null, + 0.13988011249036006, + 0.13918913535600463, + null, + 0.09682437092189741, + 0.11955164817753496, + 0.13530758825201725, + null, + 0.1254028439612634, + 0.1365144214329779, + null, + 0.13501290421817794, + 0.1317688862790525, + 0.13850037453149874, + null, + null, + null, + null, + 0.13873825765638148, + null, + null, + 0.12540076998555166, + 0.12674318625123998, + 0.13301260929451278, + null, + 0.1358442935456877, + 0.13575424362030142, + null, + null, + 0.13532794795193104, + 0.13903776064489684, + 0.12524376416343694, + 0.1352144050311425, + 0.14957834232940012, + 0.1363906290317537, + 0.11909172693089409, + 0.14224157216248315, + 0.11926015338757429, + 0.13639118366707326, + 0.13315251180386145, + 0.13115420346852066, + null, + 0.137482894049703, + 0.13427087169131274, + 0.1373668094463253, + 0.13918234532094528, + 0.11193800567946484, + 0.11717709309529914, + 0.1359554762476004, + 0.14214711118870874, + 0.13958883060947083, + 0.13202213032267024, + 0.12456223213738726, + 0.131721051287213, + null, + 0.1417991681857732, + 0.1424785360617939, + 0.13550093531774063, + null, + 0.13035820010906293, + 0.09701121478853343, + null, + 0.15008420130516895, + null, + 0.14977552041575928, + 0.14014688157185284, + 0.13988041652302652, + 0.12447964353446746, + 0.12220718452775751, + 0.12991331239084047, + 0.11986525684916573, + null, + null, + 0.13926218000656485, + 0.12302256838662115, + null, + 0.14346096514736859, + null, + 0.15218036302560015, + 0.13407248865969013, + 0.13970040848894996, + 0.1282184367566151, + null, + null, + null, + null, + 0.1413882417156448, + null, + 0.137907406589677, + 0.13041238370185024, + null, + null, + 0.13543711027672897, + 0.1489433434141096, + 0.13706670736676407, + null, + 0.13623101267491103, + 0.13854910180159874, + 0.13238917219852966, + null, + 0.14723458929281236, + null, + null, + 0.1442181296737932, + null, + 0.13233447991792766, + null, + null, + 0.13547386385389473, + 0.12449098955746415, + 0.12910045773187737, + 0.13054734812438004, + 0.12351158068039518, + null, + 0.14013190411230364, + 0.14577917467660978, + 0.1239436163432087, + 0.11040714531151402, + 0.1205010423142396, + 0.14992426919014892, + 0.12102157083170019, + 0.14752255521393198, + 0.1703975451104044, + 0.16958571750289955, + 0.17364352466303545, + 0.12636224019592315, + 0.1307601609874645, + 0.13725244518627283, + 0.13414495040981708, + 0.16246574074886544, + 0.1610646681864375, + 0.132294653859764, + 0.1698025553825608, + 0.1499313570702144, + 0.14476747804747622, + 0.1344195881282265, + 0.12657523284406025, + 0.12425208371871224, + 0.08246626788771964, + 0.18230105410021336, + 0.17102927508742446, + 0.12424823642976827, + 0.17730299894247886, + 0.15261673299336725, + 0.10287436984046654, + 0.14179937423077726, + 0.12386664177130507, + 0.11733788954568253, + 0.13848000246869877, + 0.10750930444207954, + 0.13389289459068085, + 0.1234646692666095, + 0.12377964426187654, + 0.1323000748996905, + 0.13080449914656478, + 0.11461217836638182, + 0.1490312727685537, + 0.1419146247641103, + 0.11954266510807127, + 0.1264910142996332, + 0.12145379831222759, + 0.09855257279115454, + 0.13048604447415196, + 0.12198975011248034, + 0.10023328529112846, + 0.10295281244522234, + 0.08798515315713962, + 0.13048718970549417, + 0.1098499227805957, + 0.09658241220013435, + 0.1606977243886449, + 0.13780592932052382, + 0.13163106333955846, + 0.17536721780160147, + 0.1237401576882913, + 0.10751597526449659, + 0.14127783599916052, + 0.15266984172891637, + 0.12414986141012839, + 0.1354213739834041, + 0.19150392689193235, + 0.14480652905726502, + 0.16438091155054438, + 0.17763035389344445, + 0.1570405576501629, + 0.16436379446370322, + 0.12704800900588023, + 0.1286237456940058, + 0.1300700839778146, + 0.12307160544512386, + 0.14929333559656002, + 0.13738201652889295, + 0.15085678222141208, + 0.13092072300897806, + 0.12624176075148524, + 0.14666498465751032, + 0.13325333394331096, + 0.16458815630167153, + 0.14236993872335083, + 0.16073286647098087, + 0.12464134148337729, + 0.127151492847172, + 0.12237586058321556, + 0.11655022151432011, + 0.1339716220236615, + 0.1338212116934364, + 0.1284331964749938, + 0.15710415144372522, + 0.1266437941123702, + 0.15868800380182235, + 0.12206468250299267, + 0.13691435717689085, + 0.13194289285642666, + 0.1314763211698808, + 0.1430423084012762, + 0.10926970596207537, + 0.12871084508242506, + 0.1349624386085073, + 0.1625391466346545, + 0.1638586015414857, + 0.1312634857420479, + 0.09803214111727636, + 0.14807631898858842, + 0.12779725676981074, + 0.1226450405901268, + 0.15545150145596306, + 0.11796314469360261, + 0.12225408729997082, + 0.11560323427043105, + 0.12054551819268572, + 0.17512471868722665, + 0.15454614867927438, + 0.1509424375327832, + 0.1542068033521314, + 0.11390336317356724, + 0.1369646504724708, + 0.16721895737726883, + 0.12356281539691212, + 0.18273680439324003, + 0.13375602105099485, + null, + 0.14306729676650468, + 0.16243991824013457, + 0.11580899121753603, + 0.11323332627149829, + 0.1323239250661116, + 0.1533511245582069, + 0.16120884630018967, + 0.15300326272371848, + 0.13589219250799156, + 0.11404402940874184, + null, + 0.1523941162091157, + 0.16384108592844107, + 0.15725947995657805, + 0.16648393400089062, + 0.11379680354758527, + 0.16478886874258933, + 0.12085032467083476, + 0.1331207361544298, + 0.10674221606995826, + 0.12225154135433655, + 0.16867566862587152, + 0.15948636857550735, + 0.1660451279196766, + 0.1297349421189938, + 0.12983812915897402, + 0.16778005164372717, + 0.15903552595311685, + 0.14236141272537844, + 0.1395522220938938, + 0.13696272742030352, + 0.12348931596061792, + 0.17989660930091983, + 0.13019305162480407, + 0.15237615255578635, + 0.1041583772840063, + 0.15282491496765263, + 0.11101579882417677, + 0.10766073394641516, + 0.10950671499089647, + 0.14894808451121824, + 0.1639379821179455, + 0.07363980272479505, + 0.13620440233825626, + 0.16646691610022257, + 0.1359657382624257, + 0.18006583799293316, + 0.17338479020415465, + 0.11422131617698018, + 0.134020400830209, + 0.13496917410406759, + 0.14285474637550943, + 0.1259661382257175, + 0.10665105850554711, + 0.15064269597357, + 0.15935399612012538, + 0.06183617536444805, + 0.13556431673293895, + 0.1270982797915753, + 0.14133847968316368, + 0.12905022941605276, + 0.11224621157904052, + 0.11951849512565944, + 0.15466743597231475, + 0.13796011936274108, + 0.12552304785584278, + 0.13666603461870455, + 0.17181381337938412, + 0.17194379833872078, + 0.16437138428866632, + 0.17709942938324477, + 0.17138081755839493, + 0.13148698605641268, + 0.13915219246953525, + null, + 0.1439273332835575, + 0.1324755023922641, + 0.11334766579617837, + 0.13425043453185326, + 0.14457880218456834, + 0.1479532100831199, + 0.12179144902530555, + null, + null, + 0.15910197596078574, + 0.1543097718425359, + null, + 0.1745883724730241, + 0.10482454592483047, + 0.1783299149266471, + 0.10489443129828532, + 0.16437356272655568, + 0.14219675774335572, + 0.13966353852059413, + null, + 0.1376207452072303, + 0.13373811301062627, + null, + 0.1676147837890567, + null, + 0.16839285860573697, + 0.1385509074341739, + 0.13157467024237746, + 0.15429158118072503, + null, + 0.16374808698393037, + 0.1561115640416497, + 0.12206288904093572, + null, + 0.1445887179439889, + null, + 0.1378929451988219, + null, + 0.11199040841908721, + 0.11768607214701542, + null, + 0.12527866868014514, + 0.1349246629334848, + null, + null, + 0.1418970141835635, + 0.1249691988348386, + 0.11667104966360141, + 0.13764312702293136, + null, + 0.10873593090289344, + null, + 0.13801238891205367, + 0.11705362134521444, + null, + 0.08437715567965681, + 0.1188788538479036, + 0.11790532910424058, + null, + 0.11624505413900581, + 0.130288757886464, + null, + 0.14513389141405278, + 0.14301480313761034, + 0.1529196573348648, + null, + null, + null, + null, + 0.1526881809874079, + null, + null, + 0.1107739546664221, + 0.1423240292158516, + 0.14891982452447394, + null, + 0.16238619761288609, + 0.15619448460871685, + null, + null, + 0.13183195626088912, + 0.12658584093179578, + 0.15100110485656762, + 0.13748061318473478, + 0.1284152322508227, + 0.17472513038968238, + 0.16381933064055365, + 0.1357452355182313, + 0.11136647117411391, + 0.1698503903715403, + 0.1464110519129561, + 0.12339280382533856, + null, + 0.11119468325901684, + 0.12393357287982676, + 0.1285546379988379, + 0.14391311772503962, + 0.1654141920563374, + 0.18782260843457424, + 0.11795564721802156, + 0.13879848489695593, + 0.14916859469902713, + 0.11322839717755087, + 0.16004896506070113, + 0.09887351475237946, + null, + 0.15966726693543587, + 0.15021474341335217, + 0.1499410228748348, + null, + 0.14933657081219004, + 0.15338618775694765, + null, + 0.1480927607772005, + null, + 0.13508262069600663, + 0.1206917614438717, + 0.19071905751410703, + 0.16511130665152793, + 0.15020308263166227, + 0.16341208651057054, + 0.13278681549425642, + null, + null, + 0.14491788237752018, + 0.12514862426072146, + null, + 0.11868215603520291, + null, + 0.1301513359026871, + 0.13655940306297387, + 0.12088401590184668, + 0.13522278313066527, + null, + null, + null, + null, + 0.12505397737789065, + null, + 0.13449023686136719, + 0.13573148588042896, + null, + null, + 0.1349250992361932, + 0.11439733075359701, + 0.10526022310876255, + null, + 0.11155080113624319, + 0.1256269935052763, + 0.12368942995047574, + null, + 0.14431900965062622, + null, + null, + 0.12602523912969776, + null, + 0.11091006370276686, + null, + null, + 0.15848263595466536, + 0.14303529234706538, + 0.14659695308915843, + 0.15313663068087677, + 0.1353088845647616, + null, + 0.12686860119081478 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508 + ] + }, + "junctions": { + "mean_pearson": 0.1580498031293085, + "valid_tracks": 626, + "per_track_pearson": [ + 0.26237488489565214, + 0.2830819189579111, + 0.19062496128626233, + 0.1278161428665119, + 0.29965691153676594, + 0.31891553345038004, + 0.02097992772778701, + 0.18551031078679392, + 0.21395000617390916, + 0.17672887424110809, + 0.20006795334142566, + 0.12365787504008234, + 0.07922535862041576, + 0.16580060707528066, + 0.16024104794897504, + 0.23339439051498695, + 0.2646113991223863, + 0.18190451793406165, + 0.21620559890867697, + 0.19785123757385248, + 0.21828096158683133, + 0.2119156987086889, + 0.14769408257623826, + 0.19418805847756984, + 0.1967409781570136, + 0.21992224135307073, + 0.2155586505539774, + 0.2087276458889665, + 0.23064068477567196, + 0.21689146363096648, + 0.24853945638772887, + 0.250911906978041, + 0.22466903235483063, + 0.20715639333869354, + 0.21850041760328073, + 0.21427757488699686, + 0.2566108854141741, + 0.21048323689611298, + 0.17506212982854527, + 0.16198451319997315, + 0.21895972468428695, + 0.12724854417538461, + 0.09841079051638105, + 0.14398556112404476, + 0.1255882306875922, + 0.12660382650054952, + 0.11392174019021915, + 0.1199284572585959, + 0.11739970794407928, + 0.10037103139469758, + 0.13827512237102188, + 0.1461592921022379, + 0.10178854955825252, + 0.13606002499913059, + 0.11432016274531513, + 0.1061095489942037, + 0.10349130778399858, + 0.13012381735652556, + 0.1221129066552401, + 0.1696556077864662, + 0.0843743126867049, + 0.15856966249025423, + 0.20182131209689705, + 0.10099551060747794, + 0.16347877499946073, + 0.2535982909975078, + 0.17296375174739853, + 0.3517694461825611, + 0.1266055807754516, + 0.15788170633408494, + 0.1562323781068171, + 0.24696918487291097, + 0.16309685121520195, + 0.14787608226277552, + 0.3937810734685932, + 0.3593216189444468, + 0.2068656472123894, + 0.14831916752390842, + 0.20365978470748267, + 0.31261473224087694, + 0.19459273016402998, + 0.16478628742426182, + 0.23845519057634113, + 0.2598581148268366, + 0.20482258548149798, + 0.2512168852810392, + 0.3505413987860778, + 0.276252129693236, + 0.1868868175994007, + 0.2248929435908566, + 0.26104102982859523, + 0.2534121865084927, + 0.12227708771032172, + 0.15258088632655176, + 0.16060507018208084, + 0.2726507386293701, + 0.2584081854369839, + 0.24374495888927902, + 0.1697007719675316, + 0.2270162535855155, + 0.14614446288532992, + 0.1704751424833256, + 0.32041151599605067, + 0.18293674999994786, + 0.13813496263445, + 0.11279722493632946, + 0.056800999159772846, + 0.27407694556428275, + 0.33332981740331663, + 0.2793681977811298, + 0.12912684127276766, + 0.3285778085492031, + 0.25052572652467614, + 0.3061124351354383, + 0.30024147617172603, + 0.14470827102981512, + 0.15681851555446005, + 0.21456075281684991, + 0.22812710575505704, + 0.2967835270306988, + 0.18282413334314365, + 0.25958160723023804, + 0.224719240328695, + 0.2050063966167298, + 0.07743177521590712, + null, + 0.3106305284981418, + 0.166344764409486, + 0.09884636571694654, + 0.21407832184116293, + 0.23169334524782115, + 0.1616927877489586, + 0.19737539793887104, + 0.29821479333599915, + 0.24343668386276968, + 0.08217465015347065, + null, + 0.1957787303047428, + 0.19354611757242296, + 0.18070665978892758, + 0.2065619736915255, + 0.17505231185877632, + 0.12896745926941733, + 0.13178050485562473, + 0.19615928583568984, + 0.16123496893476372, + 0.1008858426760283, + 0.17897941016906646, + 0.18836658575370654, + 0.20493884847035418, + 0.19880594637493346, + 0.2151810060082543, + 0.14200970096504956, + 0.23508420289803092, + 0.3409240515120919, + 0.26161524099056926, + 0.2573255687875423, + 0.2254910211141749, + 0.20167898099463835, + 0.3441146075949493, + 0.16517383495283428, + 0.21925513898589902, + 0.21739148266108393, + 0.261923339060027, + 0.17784012001506933, + 0.3469373792713492, + 0.2214152160379038, + 0.2149707980540519, + 0.07722397458256412, + 0.18973922737963608, + 0.109082671299508, + 0.15559454933891598, + 0.3389763999292997, + 0.11680357031349568, + 0.193838672866106, + 0.18420124321407716, + 0.17127390335293974, + 0.17043828145861506, + 0.18178108167300636, + 0.1869409191884171, + 0.18799300186262063, + 0.11197963184948435, + 0.06918753713904074, + 0.28860384040611065, + 0.1641255198181488, + 0.16836169437714832, + 0.29423647601331776, + 0.23228094433514151, + 0.22784969794238802, + 0.1882862693942581, + 0.21349456878182124, + 0.2853131245422792, + 0.16529940962031164, + 0.2201293602076694, + 0.27612566069336336, + 0.1950379499624833, + 0.1766361369707502, + 0.1591038450160599, + 0.22343444711920796, + 0.2720003013554971, + null, + 0.12399347440227304, + 0.18979644153395278, + 0.27283443556791914, + 0.3374391105962656, + 0.07557795456142018, + 0.12406905425971805, + 0.1174414692097831, + null, + null, + 0.21433847176020143, + 0.34858218130209184, + null, + 0.15170222440752268, + 0.24080105924469405, + 0.07554620940334987, + 0.1658805492914804, + 0.10746496390819814, + 0.1972235283350054, + 0.06206522390291015, + null, + 0.30152450997818625, + 0.24180773590909607, + null, + 0.11886562656576977, + null, + 0.29304327967396265, + 0.3423989610113775, + 0.21760098682647216, + 0.17017809728874955, + null, + 0.1843384449613224, + 0.19153250303832559, + 0.2515266136222784, + null, + 0.21349107251669644, + null, + 0.1967991949034412, + null, + 0.0587516169007136, + 0.186357601384145, + null, + 0.1953698368104638, + 0.24447802811561029, + null, + null, + 0.22911184605579182, + 0.2507905732748705, + 0.06896847768535579, + 0.057831484306083765, + null, + 0.3381447159906455, + null, + 0.24763080036349064, + 0.189990012336616, + null, + 0.2154081113955701, + 0.29626908484008746, + 0.319838314726192, + null, + 0.2718841307686186, + 0.2807973732627261, + null, + 0.30410435623296955, + 0.19524649772951574, + 0.30847355963741674, + null, + null, + null, + null, + 0.17680503503218825, + null, + null, + 0.16756191762332945, + 0.17165942978007026, + 0.2758353767581661, + null, + 0.13718580550911227, + 0.24097956985662058, + null, + null, + 0.36157504959736514, + 0.14592767490183253, + 0.25194414334638915, + 0.18848576795660182, + 0.172474782840237, + 0.16705849125554695, + 0.14137035986264382, + 0.21356211922339796, + 0.07812196386170175, + 0.24215494145990943, + 0.2862167485946107, + 0.27688570425735926, + null, + 0.15542715993998205, + 0.2598867200837133, + 0.09068942791094992, + 0.11381167950239891, + 0.07089388566372948, + 0.08323989644264912, + 0.1518490475596253, + 0.15747725589819242, + 0.2411058263896168, + 0.14518205715125052, + 0.27915747852215567, + 0.24586000062195804, + null, + 0.32169625965565446, + 0.18032295422116715, + 0.17898325787650954, + null, + 0.24721740616152238, + 0.22350708712483272, + null, + 0.25576550139708326, + null, + 0.16121944826565773, + 0.24572479285084445, + 0.23268781698281407, + 0.14478636766940897, + 0.1281213935803645, + 0.22522847354408204, + 0.21904663474829375, + null, + null, + 0.23553350478943177, + 0.12402303679025538, + null, + 0.248439671592945, + null, + 0.2116438257313733, + 0.17409649690513265, + 0.09709734575428748, + 0.17046054960647242, + null, + null, + null, + null, + 0.20905177728047247, + null, + 0.11913091019529909, + 0.266066988811243, + null, + null, + 0.3241795985137498, + 0.3297816908070702, + 0.32140268931101645, + null, + 0.36413474920743216, + 0.25325685120507, + 0.07978812665620355, + null, + 0.3209652728083315, + null, + null, + 0.3872218761239997, + null, + 0.36102826449650205, + null, + null, + 0.30713288553661183, + 0.2098313432621068, + 0.20408553223483777, + 0.19575614496218177, + 0.16159379260701812, + null, + 0.16119277484092767, + 0.0006311150109461728, + 0.2417049634784749, + -0.020177595239090414, + 0.016789565625980874, + 0.46220649315770607, + 0.024843224144978946, + 0.0707672878508136, + 0.08598997964965599, + 0.16087925766113384, + 0.0945470908398112, + 0.2249004461560465, + 0.12049002350000704, + 0.12385277239947595, + 0.21292684888231275, + 0.19984689092865504, + 0.1692873893253599, + 0.05775458606655954, + 0.11557537486184544, + 0.020660349076544487, + 0.04777567689682511, + 0.16256011901579673, + 0.03615312249760823, + -0.027746385072955945, + 0.004838113672359079, + 0.3561551871999712, + 0.5398654225890948, + 0.20922331808336173, + 0.17543441212063557, + 0.16074847818860122, + 0.16353725658758214, + 0.1551660582539232, + 0.0913969385860614, + 0.053639827038427346, + -0.03181545154625517, + -0.0023075640158305914, + -0.027299671857047226, + 0.012770976819822035, + 0.036397111858507726, + 0.120548717778977, + 0.01264404548578332, + -0.006318570151360844, + 0.0, + 0.013934979038116947, + 0.00039235756365851296, + -0.012049431320646354, + 0.0686605890693533, + -0.016441507559042577, + 0.14557973804467728, + -0.028188007671391013, + -0.008289772462002439, + 0.021026432520722976, + 0.0024711536650067066, + -0.009808089790848455, + 0.016641450445408865, + -0.010012092877732191, + 0.053184848952840556, + 0.07450722193143904, + 0.44045275844382403, + 0.044183162361760085, + 0.07628293220716921, + 0.10451983008813713, + 0.1560441232421276, + 0.14593617837183168, + 0.13097967195126622, + 0.1118179692710086, + 0.3051526936971663, + 0.09683907389766691, + 0.04383934361491844, + 0.04109903865658013, + 0.13491740588602544, + 0.29668917644337217, + 0.16650940840001593, + 0.24363987924407896, + 0.10767821098456896, + 0.12630596168077723, + 0.0749143130205853, + 0.14116908765031236, + 0.18497113792838857, + 0.36689820244126153, + 0.28672479721972455, + 0.22109709210329886, + 0.27705592002218626, + 0.24441974085830823, + 0.14717793651125746, + 0.07360861151035543, + 0.40432668095743735, + 0.15807426625617166, + 0.1867988147150302, + 0.14596902267783307, + 0.3254382726298606, + 0.2185194047262168, + 0.1805690278847041, + 0.42280095615383884, + 0.037340446566192356, + 0.02397485638738486, + 0.16651563442077608, + 0.07705446206678696, + 0.05310657027123281, + 0.2063760686519181, + 0.1038404474030794, + 0.11804072706402724, + 0.07071315924260979, + 0.059329926590649104, + 0.151391573532987, + 0.12440489227114163, + 0.25999559048306076, + 0.04086239889355686, + 0.5290555653332794, + 0.031226890297941352, + 0.11089468816269248, + 0.032608500593462174, + 0.21204342149198765, + 0.037938871771850305, + 0.09427702440173198, + 0.0512061146623163, + 0.043254768190666315, + 0.0514907507593577, + 0.03912563002772901, + 0.04202640053944412, + 0.053692699479162304, + 0.5443424943623263, + 0.11168758207896634, + 0.04968021688150274, + 0.3659098977904606, + -0.012609262580452899, + null, + 0.053820425657857184, + 0.11772839568523424, + -0.02677420014120426, + 0.04327573031864641, + 0.12967148180641414, + 0.10334756713162382, + 0.22285419804582204, + 0.2465995746136551, + 0.1540439671179643, + 0.007927682847752806, + null, + 0.09540998473570483, + 0.060557245922106734, + -0.0026043299916950896, + -0.019713167124243345, + 0.11451114824931313, + 0.09451248363790381, + 0.1622693383601333, + 0.06556094623735467, + 0.11249454904873206, + 0.22436190671877432, + 0.09963568914977616, + 0.050292103316042634, + 0.11217101864324384, + 0.0025212230385899194, + 0.09471555079627485, + 0.17155298786896417, + 0.0629147797522342, + -0.02236387578908103, + 0.1028800967336798, + 0.13627594974841328, + 0.0444611786500157, + 0.06313915420473729, + 0.01775780978471782, + 0.05409759098871833, + 0.09099606926496415, + 0.14172519499766642, + 0.13188850141635575, + 0.026415808577195117, + 0.07073778452137403, + 0.2315498507618392, + 0.10267474785886103, + 0.120586826182494, + 0.0912545121897535, + 0.10281127340174137, + 0.032237259099542774, + 0.08431439786889941, + 0.040244946917946445, + 0.03902994167014005, + 0.12003543046290553, + 0.011392080233630579, + 0.018864642484307158, + 0.06873927546845135, + 0.2511072906365681, + 0.0688040782320978, + 0.08201484572443858, + 0.001990924590398075, + 0.07038685559305895, + 0.16236676497295593, + -0.02906397411913029, + 0.10861222899045024, + 0.11988196597422866, + 0.09783595173329754, + 0.357383691933678, + 0.05400388752632817, + 0.06349542089753764, + 0.3245512145112595, + 0.1829768829904364, + 0.15661131090777422, + 0.00614786958782234, + 0.22174226640893555, + 0.08959912796099183, + 0.09181107378275326, + 0.1319979209959939, + null, + 0.02566018157011819, + 0.1536334445085626, + 0.11110520110931903, + 0.13438016466207617, + 0.20253431578550304, + 0.3769613744885117, + 0.0891114980211426, + null, + null, + -0.012183484162037623, + 0.14163760399978903, + null, + 0.040542887725011745, + 0.03625717006357631, + -0.01773691956504036, + 0.0953091606103341, + 0.0016645268171482912, + 0.23760569453033667, + -0.03378574904036731, + null, + 0.06455409205384342, + 0.042903489052978534, + null, + 0.09164307101991721, + null, + 0.09508630830134124, + 0.17806215555822122, + 0.41095852657156073, + 0.09771572511384291, + null, + 0.10756612077719867, + 0.0560798176447086, + 0.13062961344897572, + null, + 0.08139874826936111, + null, + 0.2008846673396132, + null, + 0.02271448794177913, + 0.23490939651068138, + null, + 0.12585013134560682, + 0.16254010685484208, + null, + null, + 0.13787742378337467, + 0.15769590369545866, + -0.009200790604029025, + 0.057191654892245064, + null, + 0.15215346635994603, + null, + 0.06872520237114906, + 0.23294128209682488, + null, + 0.16661682478321718, + 0.09030829081521673, + 0.10803016471717575, + null, + 0.20742444376012376, + 0.046914563557889914, + null, + 0.4364778295589719, + 0.27947082860648687, + 0.1675808233720268, + null, + null, + null, + null, + 0.15018260104560366, + null, + null, + 0.11993198229110265, + 0.13529599531358313, + 0.3852888330170668, + null, + 0.17974746309986836, + 0.15976223613668428, + null, + null, + 0.014269281878610176, + 0.003510802106800628, + 0.1570707364645834, + -0.009807104466578929, + 0.013875747070068558, + 0.013737551522167005, + 0.17486833504958227, + 0.05949105523090324, + 0.24116986775968266, + 0.010536713391313518, + 0.2815109702548031, + 0.1581130991108395, + null, + 0.04902901767564494, + 0.047154830243887655, + 0.06533060051675514, + 0.059834339030079266, + 0.02834675978802908, + -0.021328640585603027, + 0.020665718465732858, + 0.24437995294808681, + 0.15390127715603952, + 0.052948208649393445, + 0.02896033301907224, + -0.006817163429403707, + null, + 0.1399439147121941, + 0.05537770141162742, + 0.11299414176627628, + null, + 0.1428841126370557, + 0.1609612968920562, + null, + 0.14153166108465684, + null, + 0.08822791195410451, + 0.08215485937988691, + 0.0863170640096175, + 0.07953401327623721, + 0.26031392585000723, + 0.012999391778626591, + 0.02208815153317417, + null, + null, + 0.05241037172730288, + 0.01587683484633524, + null, + 0.21542055138650845, + null, + 0.2974628213377192, + 0.15614360185472134, + 0.08606792988800903, + 0.056592376615413775, + null, + null, + null, + null, + 0.05814289095567893, + null, + 0.15790321544548938, + 0.13529015930283383, + null, + null, + 0.243844231959324, + 0.022550427816182265, + 0.014626789160888134, + null, + 0.01897543697417134, + 0.11287123790213614, + 0.11652820562035239, + null, + 0.0009915042130697585, + null, + null, + 0.1673450487652101, + null, + 0.02456798979826964, + null, + null, + 0.11105533119278667, + 0.0474680592017677, + 0.10559276832731133, + 0.07343857299057632, + 0.09396581271871476, + null, + 0.10413481214488379 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9271523178807947, + 0.9407894736842105, + 0.9252669039145908, + 0.9113924050632911, + 0.956953642384106, + 0.9389830508474576, + 0.9295774647887324, + 0.9375, + 0.9326923076923077, + 0.925531914893617, + 0.9326599326599326, + 0.911864406779661, + 0.922077922077922, + 0.934375, + 0.9433333333333334, + 0.9375, + 0.9265734265734266, + 0.9326923076923077, + 0.9363057324840764, + 0.9411764705882353, + 0.9280575539568345, + 0.9354838709677419, + 0.9090909090909091, + 0.9442379182156134, + 0.9466192170818505, + 0.9473684210526315, + 0.9519230769230769, + 0.9559322033898305, + 0.9527027027027027, + 0.9344827586206896, + 0.9452054794520548, + 0.9385113268608414, + 0.9328358208955224, + 0.9331306990881459, + 0.9340659340659341, + 0.9274447949526814, + 0.9274809160305344, + 0.9373040752351097, + 0.8986013986013986, + 0.9295302013422819, + 0.931740614334471, + 0.9415807560137457, + 0.9312977099236641, + 0.9444444444444444, + 0.921311475409836, + 0.9324324324324325, + 0.9298892988929889, + 0.9175627240143369, + 0.9148936170212766, + 0.9387755102040817, + 0.9233576642335767, + 0.9338235294117647, + 0.9242424242424242, + 0.9319727891156463, + 0.9216417910447762, + 0.9319727891156463, + 0.9204152249134948, + 0.956953642384106, + 0.9383561643835616, + 0.9295774647887324, + 0.9379310344827586, + 0.940983606557377, + 0.9403973509933775, + 0.9419795221843004, + 0.9429530201342282, + 0.9556962025316456, + 0.9256756756756757, + 0.9563758389261745, + 0.9385665529010239, + 0.9509803921568627, + 0.9359430604982206, + 0.9354838709677419, + 0.936026936026936, + 0.9503311258278145, + 0.9498207885304659, + 0.9491525423728814, + 0.9517684887459807, + 0.9431438127090301, + 0.9551724137931035, + 0.9489795918367347, + 0.9403973509933775, + 0.9493670886075949, + 0.9383561643835616, + 0.9352750809061489, + 0.9333333333333333, + 0.934984520123839, + 0.9452054794520548, + 0.9575163398692811, + 0.9428571428571428, + 0.9367088607594937, + 0.9442508710801394, + 0.9450171821305842, + 0.9438596491228071, + 0.9319727891156463, + 0.9425675675675675, + 0.9618055555555556, + 0.9423728813559322, + 0.9307958477508651, + 0.9271523178807947, + 0.939873417721519, + 0.9366666666666666, + 0.9278688524590164, + 0.9430604982206405, + 0.9361022364217252, + 0.9556313993174061, + 0.9466666666666667, + 0.9101123595505618, + 0.9464882943143813, + 0.946843853820598, + 0.9391891891891891, + 0.9464285714285714, + 0.9362416107382551, + 0.948051948051948, + 0.9423728813559322, + 0.9455782312925171, + 0.9290540540540541, + 0.9344827586206896, + 0.9256756756756757, + 0.9422382671480144, + 0.9488054607508533, + 0.9392857142857143, + 0.9559322033898305, + 0.9375, + 0.9506578947368421, + 0.9271255060728745, + 1.0, + 0.9515570934256056, + 0.9198606271777003, + 0.9393939393939394, + 0.9440298507462687, + 0.9347826086956522, + 0.9413793103448276, + 0.9547038327526133, + 0.9285714285714286, + 0.9333333333333333, + 0.9331103678929766, + 1.0, + 0.9593220338983051, + 0.9440298507462687, + 0.9342105263157895, + 0.9306569343065694, + 0.9291044776119403, + 0.9375, + 0.9275862068965517, + 0.9298245614035088, + 0.9274809160305344, + 0.9330708661417323, + 0.9479166666666666, + 0.9191176470588235, + 0.9469964664310954, + 0.9185185185185185, + 0.9361702127659575, + 0.9194139194139194, + 0.935064935064935, + 0.9413793103448276, + 0.9347079037800687, + 0.9393939393939394, + 0.9422382671480144, + 0.9293286219081273, + 0.9343065693430657, + 0.9204545454545454, + 0.9340277777777778, + 0.9278996865203761, + 0.9303135888501742, + 0.9283018867924528, + 0.9280575539568345, + 0.9342560553633218, + 0.9602649006622517, + 0.939622641509434, + 0.9473684210526315, + 0.945054945054945, + 0.9249146757679181, + 0.9548872180451128, + 0.9486301369863014, + 0.9423076923076923, + 0.9416342412451362, + 0.9239130434782609, + 0.9303135888501742, + 0.9301470588235294, + 0.9300699300699301, + 0.9372822299651568, + 0.9402985074626866, + 0.933579335793358, + 0.931899641577061, + 0.9471830985915493, + 0.9630996309963099, + 0.9383561643835616, + 0.9090909090909091, + 0.921146953405018, + 0.9475409836065574, + 0.9305555555555556, + 0.9338235294117647, + 0.9214285714285714, + 0.9119718309859155, + 0.9198606271777003, + 0.9151943462897526, + 0.9044117647058824, + 0.9375, + 0.9379084967320261, + 0.9333333333333333, + 1.0, + 0.9342105263157895, + 0.9186440677966101, + 0.9302325581395349, + 0.9290540540540541, + 0.9389067524115756, + 0.9399293286219081, + 0.9144736842105263, + 1.0, + 0.9924242424242424, + 0.9372937293729373, + 0.935064935064935, + 1.0, + 0.9172413793103448, + 0.9194139194139194, + 0.9233576642335767, + 0.9215017064846417, + 0.9205298013245033, + 0.9137931034482759, + 0.9228187919463087, + 0.991869918699187, + 0.9285714285714286, + 0.9209621993127147, + 0.9924812030075187, + 0.91015625, + 0.9929577464788732, + 0.9454545454545454, + 0.9368421052631579, + 0.9295774647887324, + 0.9225589225589226, + 1.0, + 0.927536231884058, + 0.9330985915492958, + 0.9157894736842105, + 1.0, + 0.9225589225589226, + 1.0, + 0.9090909090909091, + 0.9915254237288136, + 0.93, + 0.9261744966442953, + 1.0, + 0.929368029739777, + 0.921311475409836, + 1.0, + 1.0, + 0.925531914893617, + 0.9313725490196079, + 0.9117647058823529, + 0.9158249158249159, + 1.0, + 0.9151943462897526, + 0.9922480620155039, + 0.9155844155844156, + 0.9312714776632303, + 0.9930555555555556, + 0.9346153846153846, + 0.9196141479099679, + 0.9272727272727272, + 0.9927536231884058, + 0.9142857142857143, + 0.9331103678929766, + 1.0, + 0.9344262295081968, + 0.9087719298245615, + 0.9220338983050848, + 0.9925373134328358, + 1.0, + 1.0, + 1.0, + 0.920265780730897, + 1.0, + 1.0, + 0.9285714285714286, + 0.9290780141843972, + 0.9230769230769231, + 1.0, + 0.9180887372013652, + 0.9212328767123288, + 1.0, + 1.0, + 0.9198606271777003, + 0.9311594202898551, + 0.9331103678929766, + 0.9155405405405406, + 0.9078498293515358, + 0.9251700680272109, + 0.9249146757679181, + 0.916955017301038, + 0.9152542372881356, + 0.9110320284697508, + 0.9267515923566879, + 0.915129151291513, + 0.9925373134328358, + 0.9298245614035088, + 0.9290540540540541, + 0.9142857142857143, + 0.936026936026936, + 0.9285714285714286, + 0.9270833333333334, + 0.9368421052631579, + 0.9137931034482759, + 0.921311475409836, + 0.9290540540540541, + 0.9347826086956522, + 0.9351535836177475, + 0.991869918699187, + 0.9356913183279743, + 0.9256756756756757, + 0.9358974358974359, + 1.0, + 0.9265175718849841, + 0.9390681003584229, + 1.0, + 0.9278350515463918, + 1.0, + 0.9294871794871795, + 0.9270833333333334, + 0.9240924092409241, + 0.9420289855072463, + 0.935374149659864, + 0.928082191780822, + 0.9230769230769231, + 0.9929078014184397, + 1.0, + 0.9183006535947712, + 0.9149659863945578, + 1.0, + 0.9097472924187726, + 1.0, + 0.920265780730897, + 0.9161073825503355, + 0.92, + 0.9169435215946844, + 0.9928571428571429, + 1.0, + 1.0, + 1.0, + 0.9178082191780822, + 0.9927007299270073, + 0.9377162629757786, + 0.946031746031746, + 1.0, + 1.0, + 0.9401408450704225, + 0.920265780730897, + 0.9333333333333333, + 1.0, + 0.9212328767123288, + 0.9259259259259259, + 0.9078014184397163, + 1.0, + 0.9490445859872612, + 1.0, + 1.0, + 0.9385665529010239, + 1.0, + 0.9223300970873787, + 1.0, + 0.9922480620155039, + 0.9343629343629344, + 0.9285714285714286, + 0.9243986254295533, + 0.9278350515463918, + 0.9154929577464789, + 1.0, + 0.9205298013245033, + 0.9887218045112782, + 0.9894736842105263, + 0.9963768115942029, + 0.9892857142857143, + 0.9968354430379747, + 0.9893238434163701, + 0.9965156794425087, + 0.9886792452830189, + 0.9863481228668942, + 0.9856115107913669, + 0.9965397923875432, + 0.9819494584837545, + 0.9863013698630136, + 0.9931506849315068, + 0.9925650557620818, + 0.9926739926739927, + 0.9753521126760564, + 0.9814814814814815, + 0.9928571428571429, + 0.9897610921501706, + 0.9755244755244755, + 0.9893992932862191, + 0.9866220735785953, + 0.9887218045112782, + 0.9928571428571429, + 0.9963636363636363, + 0.9900662251655629, + 0.9930313588850174, + 0.985663082437276, + 0.9881422924901185, + 0.9894736842105263, + 0.9926470588235294, + 0.9959677419354839, + 0.9822695035460993, + 0.989010989010989, + 0.9893617021276596, + 0.9920318725099602, + 0.9716312056737588, + 0.9895104895104895, + 0.9856115107913669, + 0.9823943661971831, + 1.0, + 0.9964664310954063, + 0.996078431372549, + 0.9862542955326461, + 0.9928057553956835, + 0.9849624060150376, + 0.9961832061068703, + 0.9924528301886792, + 0.9884615384615385, + 0.9959839357429718, + 0.9925650557620818, + 0.9924242424242424, + 0.99644128113879, + 0.9887218045112782, + 0.9851301115241635, + 0.9892086330935251, + 0.992831541218638, + 0.9857651245551602, + 0.9855072463768116, + 0.9825783972125436, + 0.9850187265917603, + 0.9829351535836177, + 0.9926739926739927, + 0.9927797833935018, + 0.9926470588235294, + 0.9857142857142858, + 0.9884169884169884, + 0.9809885931558935, + 0.989247311827957, + 0.9890510948905109, + 0.9818840579710145, + 0.9898305084745763, + 0.982078853046595, + 0.992831541218638, + 0.9891304347826086, + 0.9896551724137931, + 0.9893992932862191, + 0.9928057553956835, + 0.9886792452830189, + 0.9826989619377162, + 0.9897260273972602, + 0.9923954372623575, + 0.9851851851851852, + 0.9858156028368794, + 0.9926470588235294, + 0.9847328244274809, + 0.9861111111111112, + 0.9890510948905109, + 0.9853479853479854, + 0.9895470383275261, + 0.9893238434163701, + 0.993103448275862, + 0.9776951672862454, + 0.9824561403508771, + 0.996415770609319, + 0.986159169550173, + 0.9814126394052045, + 0.9897260273972602, + 0.989247311827957, + 0.9846743295019157, + 0.9890510948905109, + 0.9891304347826086, + 0.9928571428571429, + 0.9855072463768116, + 0.9930555555555556, + 0.9924528301886792, + 0.9962962962962963, + 0.9824561403508771, + 0.99644128113879, + 0.9859154929577465, + 0.9962546816479401, + 0.9834437086092715, + 0.985663082437276, + 0.982078853046595, + 0.9961538461538462, + 0.9823321554770318, + 0.9730769230769231, + 0.9802371541501976, + 0.988929889298893, + 0.9964912280701754, + 0.989247311827957, + 0.9853479853479854, + 0.9929328621908127, + 0.9925650557620818, + 1.0, + 0.9787234042553191, + 0.9928571428571429, + 0.9787234042553191, + 0.9841897233201581, + 0.9783393501805054, + 0.9895104895104895, + 0.9929328621908127, + 0.9887218045112782, + 0.9930313588850174, + 0.9847328244274809, + 1.0, + 0.9825783972125436, + 0.9852941176470589, + 0.9825174825174825, + 0.9930313588850174, + 0.9776951672862454, + 0.9893617021276596, + 0.9825174825174825, + 0.9858156028368794, + 0.9888475836431226, + 0.9925373134328358, + 0.9850187265917603, + 0.9891696750902527, + 0.9926739926739927, + 0.9846153846153847, + 0.985239852398524, + 0.989247311827957, + 0.988929889298893, + 0.9924528301886792, + 0.993103448275862, + 0.9926739926739927, + 0.9849624060150376, + 0.9864864864864865, + 0.9851851851851852, + 0.9891304347826086, + 0.989010989010989, + 0.9935275080906149, + 0.9825174825174825, + 0.9925093632958801, + 0.984251968503937, + 0.9897260273972602, + 0.992831541218638, + 0.9918367346938776, + 0.9814126394052045, + 0.9855072463768116, + 0.9777777777777777, + 0.9844961240310077, + 0.9893992932862191, + 0.992619926199262, + 0.9959349593495935, + 0.9858156028368794, + 0.9821428571428571, + 0.9895104895104895, + 0.9927007299270073, + 0.9825783972125436, + 0.9888475836431226, + 0.9816176470588235, + 0.9853479853479854, + 0.9753521126760564, + 0.9958847736625515, + 0.9851301115241635, + 0.9856115107913669, + 0.9788732394366197, + 0.9928571428571429, + 0.9896907216494846, + 0.988929889298893, + 0.9964664310954063, + 0.985239852398524, + 0.974910394265233, + 0.9820143884892086, + 0.9859154929577465, + 0.9807692307692307, + 0.9903225806451613, + 0.9897610921501706, + 0.9914529914529915, + 0.9929078014184397, + 0.98989898989899, + 0.9738805970149254, + 0.9926739926739927, + 0.9897260273972602, + 0.9931972789115646, + 0.9858657243816255, + 1.0, + 0.9833333333333333, + 0.9843260188087775, + 0.993485342019544, + 0.992, + 0.9859649122807017, + 0.9826989619377162, + 0.9807692307692307, + 0.9897610921501706, + 0.9825783972125436, + 0.9893238434163701, + 0.9893992932862191, + 1.0, + 0.989247311827957, + 0.9888888888888889, + 1.0, + 0.9850746268656716, + 0.992, + 0.9830508474576272, + 0.9857651245551602, + 0.9966216216216216, + 0.9894736842105263, + 1.0, + 0.9853479853479854, + 0.9859649122807017, + 0.9889705882352942, + 1.0, + 0.9822064056939501, + 1.0, + 0.9893992932862191, + 0.9841269841269841, + 0.9850187265917603, + 0.9818181818181818, + 0.9924812030075187, + 0.9890909090909091, + 0.989247311827957, + 0.9925925925925926, + 1.0, + 0.9862068965517241, + 0.9893992932862191, + 0.9835526315789473, + 0.9865319865319865, + 0.9915254237288136, + 0.9844357976653697, + 1.0, + 0.9893617021276596, + 0.9888059701492538, + 1.0, + 0.981549815498155, + 0.9825174825174825, + 0.9851301115241635, + 0.9922480620155039, + 0.9818840579710145, + 0.9894366197183099, + 0.9915966386554622, + 0.9931740614334471, + 0.9966555183946488, + 0.992831541218638, + 0.9852941176470589, + 0.9919354838709677, + 0.9841269841269841, + 1.0, + 0.9930795847750865, + 0.9908256880733946, + 1.0, + 0.9858156028368794, + 0.9931972789115646, + 0.9932432432432432, + 1.0, + 0.9895104895104895, + 0.9900662251655629, + 1.0, + 0.992, + 0.9930795847750865, + 0.9828178694158075, + 0.9854545454545455, + 0.976897689768977, + 0.9859649122807017, + 0.9854014598540146, + 0.9816176470588235, + 0.9897959183673469, + 0.9811320754716981, + 0.9849056603773585, + 0.9932203389830508, + 0.9787234042553191, + 0.9917355371900827, + 0.9861111111111112, + 0.978494623655914, + 0.9863945578231292, + 0.9894736842105263, + 0.986013986013986, + 0.993006993006993, + 0.9828767123287672, + 0.989010989010989, + 0.9832214765100671, + 0.9858657243816255, + 0.9851851851851852, + 0.9854545454545455, + 0.9921259842519685, + 0.9892857142857143, + 0.9928571428571429, + 0.9933333333333333, + 1.0, + 0.9826989619377162, + 0.9832775919732442, + 1.0, + 0.9868852459016394, + 1.0, + 0.9933554817275747, + 0.9893238434163701, + 0.9831081081081081, + 0.9803921568627451, + 0.9931506849315068, + 0.9813432835820896, + 0.9809885931558935, + 1.0, + 0.983739837398374, + 0.9859154929577465, + 0.9713261648745519, + 0.9922480620155039, + 0.9928825622775801, + 1.0, + 0.9863481228668942, + 0.9896193771626297, + 0.9825174825174825, + 0.9889705882352942, + 0.9781021897810219, + 1.0, + 0.9830508474576272, + 1.0, + 0.9894366197183099, + 0.991304347826087, + 0.9807692307692307, + 0.9932659932659933, + 1.0, + 1.0, + 0.996415770609319, + 0.9859649122807017, + 0.9857142857142858, + 1.0, + 0.989247311827957, + 0.9930313588850174, + 0.9863481228668942, + 0.9906542056074766, + 0.9902597402597403, + 0.991869918699187, + 0.9917355371900827, + 0.9854545454545455, + 0.9916666666666667, + 0.9849056603773585, + 0.9908256880733946, + 1.0, + 0.985239852398524, + 0.9792387543252595, + 0.993421052631579, + 0.9790209790209791, + 0.9796610169491525, + 1.0, + 0.9932203389830508 + ] + }, + "contact_maps": { + "mean_pearson": 0.019487273472909843, + "valid_tracks": 28, + "per_track_pearson": [ + 0.04081458015383671, + -0.036711409694901116, + 0.043084998546684676, + 0.01576171494300664, + -0.002445929120946461, + -0.0007610104612738161, + 0.006491668301607098, + 0.03212697937623195, + 0.08461809480986397, + 0.025972267093992755, + -0.0032193826660228187, + 0.053647996272669494, + 0.08124321378131096, + 0.025786206925746125, + 0.021896809034206245, + -0.011728535355311846, + 0.013000685370987948, + 0.05535391310488641, + -0.06686778615406057, + 0.041402546717417915, + 0.010961894078980893, + -0.025528232778043138, + -0.021206278231665276, + 0.05060674379984273, + 0.0594456626100659, + -0.0672671181282852, + 0.030098420559475286, + 0.08906494435117208 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.25558847529564294, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..525eec8738cb03afc861304e15552a7c6cac8cd4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S004/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.39635258936365514, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6965235643807073, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.1778776917280187, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7403571967826894, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.25762534461408215, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.26157671283777334, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2667099562067121, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.1633509969252289, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.2635279373317317, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.19721844991758614, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5022219249141129, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.16334218499410608, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.18509957659072374, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6cf49d92a2cc83f7e9876974abfdf57c122fa1d4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/README.md @@ -0,0 +1,82 @@ +# SeqFunc local01 at step 14,000 with transcript-direction junction decoding + +One model with **13,256,456 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and four dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The global parent trained for the first 2,000 updates. The local path was attached with zero residual outputs at step 2,000, retaining every parent parameter and optimizer moment, then training continued. This is the matched control for the uniform splice-context fork. + +This checkpoint records **14,000 updates**, 0 complete training permutations and +14,000 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 6 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup updates. The splice objective is `balanced_splice`; +other family losses follow the reference. Recorded accumulated training-loop wall time is +35.3 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +## Measured validation results + +Family-mean Pearson: **0.256811** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.449029 | +| dnase | 0.581962 | +| procap | 0.331749 | +| cage | 0.180663 | +| rna_seq | 0.200977 | +| chip_tf | 0.366987 | +| chip_histone | 0.352559 | +| splice_sites | 0.039608 | +| splice_usage | 0.130403 | +| junctions | 0.171500 | +| contact_maps | 0.019487 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.3964 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6965 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.1779 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7404 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2576 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2616 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2667 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.1634 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.2635 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.1972 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5022 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1633 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1851 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..887aa8a13bc6a3284c4731e5f06fc2a6ce24aadb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/local01-14000/package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..ab8daf031da0c33918dc6dca6e4641eb178f37fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/package_manifest.json @@ -0,0 +1,31 @@ +{ + "parameters": 13256456, + "source_checkpoint_sha256": "af8060229a79fbcb3e5ce34ee559af3724268ea7dc8e321bd12ce9d5f2d27e3a", + "files": { + "README.md": "4ac80b2a60d4eaf0ef4b1ada2c1595114da34426cc7f7008c03c147a5c4f4a1d", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "fdef062e6c8de5ac6f73c764d638d6cd994fd001370628e19bf4365db4d57d86", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "2871dc8893879c3e14a526dea659701d72c8c4d0e2f835b0bfb857034a2c2e6d", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "a6653431e850d448c414f48cc65cbaaec8c17a430c08376b281441a6f6bce987", + "training_config.yaml": "deab298b32f0aef33e66ddf538541ec470eee0c05a6ebc983684f2575287d791", + "training_provenance.json": "1a721e47432865cf50c2ff1a14ac3cc97b2926da28dff75023e7d008c55a9793", + "decoding_policy.json": "9e1f8466ebcb9edc902746e4eca4487b8e69df509fa512724a88219cada0ae95", + "validation_tracks.json": "25dc1eabed373620fc347c543a04b8b39080a99546ef45a99027b033f503395c", + "validation_variants.json": "41cd2d3e0548c520def53c102fbb8e99c5c7c473d807d62b0d5bc0376f7adbd4" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..dc0f0b20833fc99253b7dcf6023fd5e7a61b1128 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/predict.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..45bdf927f635baa77ab98cc8fcefbb60196aa0ff --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/train_single.py @@ -0,0 +1,155 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f9a3298d69b138a02fed00880182d4004b23327 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_config.yaml @@ -0,0 +1,28 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..6c451660c4de8331481dbbcfae4140a9747cbd2d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/training_provenance.json @@ -0,0 +1,66 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 0, + "offset": 14000, + "global_step": 14000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/local01", + "config": "/workspace/research/experiments/local01/config.yaml", + "resume": "/workspace/research/experiments/local01/step-012000.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 82040, + "workers": 6, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 2118.901816125028, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..932f05425559636ccf4632da6359bbe93284c1f8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.4490286571976123, + "valid_tracks": 167, + "per_track_pearson": [ + 0.5787180877972778, + 0.5991515013347578, + 0.35903155111506063, + 0.5193434419972531, + 0.35044756915873515, + 0.3437029574320618, + 0.34756048117451777, + 0.42540369245724163, + 0.47264460256463814, + 0.5209497485899214, + 0.4566865328180873, + 0.3897322480759855, + 0.2561485999623867, + 0.392830801227994, + 0.5730450863697303, + 0.3510127947274521, + 0.4247439334333387, + 0.3651042845681107, + 0.41584548725817033, + 0.40511820355764716, + 0.3720545185131241, + 0.3325212433700937, + 0.3794904461265274, + 0.4044846289545441, + 0.34088904670038417, + 0.3603870422288395, + 0.34918863670743455, + 0.3768702600974143, + 0.37958507518094714, + 0.3583781273707457, + 0.36833146617752877, + 0.3617025207634615, + 0.35761873715031967, + 0.2741533776590353, + 0.24876364328297623, + 0.42769623410994234, + 0.3562879171581113, + 0.47614307417969715, + 0.376639689062023, + 0.34804106233208476, + 0.4277149421346815, + 0.3856446880541281, + 0.36778459764982724, + 0.32680388208399097, + 0.4280376208326986, + 0.3693404429328073, + 0.2926603388401998, + 0.3390525379984045, + 0.3304521292534768, + 0.3358901517979463, + 0.3713851031989818, + 0.291369195709947, + 0.3348811477661533, + 0.4803003333422318, + 0.31847388698793383, + 0.5912899866101075, + 0.7050618993818616, + 0.41812667456178626, + 0.5093735640672955, + 0.4923613400628523, + 0.6138636281722541, + 0.3802104692385244, + 0.5855885756970621, + 0.5077466577288013, + 0.5412959212081566, + 0.37189575851373, + 0.45222198294493865, + 0.5367696541026322, + 0.6912142911857361, + 0.6188357722216369, + 0.40233711547805434, + 0.3287030980669896, + 0.28094874045698837, + 0.40550597303509167, + 0.39073632248756107, + 0.32863185404207995, + 0.3811561431503719, + 0.4066075827817357, + 0.4452432749912775, + 0.34815329539110534, + 0.40084296529664154, + 0.4339697572272894, + 0.3873856762228455, + 0.3626044497999108, + 0.35597436467198007, + 0.35691025801573173, + 0.31157023766608916, + 0.3363073722433857, + 0.32725185680472013, + 0.38117448203272436, + 0.3785619052543757, + 0.36997174134220223, + 0.36887085848978596, + 0.3917450291685058, + 0.450417009706941, + 0.27823992557334126, + 0.3155222023942281, + 0.2865205523520331, + 0.3557071844095252, + 0.38369098043498384, + 0.3167693743076178, + 0.34850457431170995, + 0.3738786695139217, + 0.3476567489255353, + 0.3787980146067932, + 0.39027981093851505, + 0.41761359232677997, + 0.38930952598736385, + 0.3896811573729035, + 0.3995248451092658, + 0.4218735532286697, + 0.3907429704980889, + 0.37784360755086716, + 0.3291786675279102, + 0.3368474362051795, + 0.3253297906209802, + 0.34052251784212867, + 0.36649769866383974, + 0.36129299757781, + 0.34810792744995317, + 0.5597513165982163, + 0.5460021078014532, + 0.6166491144926074, + 0.6154810007630878, + 0.5936062740268738, + 0.4845877936460461, + 0.7594485368541238, + 0.5844817709147023, + 0.6462148166221638, + 0.5812911817194366, + 0.632850306600059, + 0.46883697426637144, + 0.5015798861716791, + 0.47046089043070954, + 0.6080698047092125, + 0.6036079099576245, + 0.594101353709923, + 0.4372612090895903, + 0.3486512844607223, + 0.7029645986118448, + 0.6724949108336272, + 0.6841615662215763, + 0.6807315847140819, + 0.5054787172408788, + 0.7486562053843839, + 0.6547517935104803, + 0.5915206499866476, + 0.614528586838432, + 0.751781939765598, + 0.6909652024203696, + 0.5131597884359982, + 0.5198862050104495, + 0.5892171495087771, + 0.6319374980887114, + 0.6325310639208899, + 0.5042657847525656, + 0.635891209770898, + 0.4969548421046694, + 0.5086896230933758, + 0.5687633347115012, + 0.4548732898190214, + 0.6459349885441865, + 0.6807243888551492, + 0.5374119316844675, + 0.5081594689823072, + 0.4644304163979905, + 0.6303742633770874 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5819615351028306, + "valid_tracks": 305, + "per_track_pearson": [ + 0.4825282257095405, + 0.8974002730271095, + 0.749502436550507, + 0.4376828905737947, + 0.3830108847807131, + 0.4011917044597347, + 0.9208451261320248, + 0.6605305955440451, + 0.3416811029927286, + 0.28836991163471576, + 0.8485526763001894, + 0.6038023779910493, + 0.36602324670046693, + 0.4265459626010102, + 0.40900693942762534, + 0.7651063811057393, + 0.7519769154488417, + 0.8332204280345484, + 0.8263847144749069, + 0.8625045914197149, + 0.5794703075127715, + 0.5264584426256417, + 0.8796988997172452, + 0.430024316985493, + 0.45169426677341956, + 0.7492616418313209, + 0.39794117721266475, + 0.5897319446644322, + 0.42695689272464155, + 0.41687136352670573, + 0.5378633995590119, + 0.36494340162206135, + 0.8193912303093721, + 0.6037896913063845, + 0.8621877852173455, + 0.868006048865225, + 0.38031008727732524, + 0.804974467551571, + 0.8595415256782285, + 0.839636078560268, + 0.7990619046330576, + 0.6486011012064055, + 0.6831959616030377, + 0.5724135856366416, + 0.3980411929821742, + 0.6125831260513943, + 0.709101914891812, + 0.4611053985979649, + 0.528908445444118, + 0.3534388752025239, + 0.3042490440347367, + 0.49085115944150476, + 0.42180052531311213, + 0.8115408180272357, + 0.3608402898220272, + 0.3530011233629133, + 0.37393230536441274, + 0.39943654147910485, + 0.36187222643460853, + 0.4756765347985436, + 0.35078859081876285, + 0.557778983641464, + 0.4079158899136556, + 0.3957195007289911, + 0.4205409972673051, + 0.4151902354997426, + 0.4809152303109889, + 0.39405658851441255, + 0.32139719319031956, + 0.3497692637241536, + 0.4626645268430315, + 0.5608185232442054, + 0.4009653811055363, + 0.5352429552841572, + 0.2697163684900562, + 0.7853876031667651, + 0.8763889964084774, + 0.5902369183294703, + 0.3703770391580457, + 0.3791189988896404, + 0.41262372144998855, + 0.39215867160166024, + 0.43542258121652205, + 0.3969605438950681, + 0.4500823139519885, + 0.4315752366134483, + 0.5073280798299201, + 0.4391698230072379, + 0.5766697665139416, + 0.36362211736165195, + 0.5554945690275738, + 0.3638244602266766, + 0.6882167564369275, + 0.40154776318336677, + 0.6762109002226742, + 0.4358105492636843, + 0.6669945785646737, + 0.6237110081208828, + 0.2772826710923412, + 0.2854085998453261, + 0.3953352527091865, + 0.31780930334875246, + 0.2992469187811878, + 0.6401247760649295, + 0.6106326455090128, + 0.3703223424316646, + 0.3746139243969257, + 0.5323749109225467, + 0.8207545387896096, + 0.48938704396874655, + 0.4617391091371984, + 0.5115634303575717, + 0.42531684570827305, + 0.39197477736864733, + 0.7470606562233543, + 0.44617492430176564, + 0.4310197547683894, + 0.4879071631069926, + 0.5521642783226393, + 0.3847794750278131, + 0.4367224035791354, + 0.44074015116077486, + 0.4678391920179805, + 0.43959001649687257, + 0.543224747178906, + 0.4692161697771991, + 0.5136852623687543, + 0.6226287706822446, + 0.6616290819631337, + 0.5096670887301237, + 0.40680118671622156, + 0.5364894233236762, + 0.561318510830251, + 0.7521441734790806, + 0.38484285026662457, + 0.2609455591640401, + 0.520143220972116, + 0.4212361548541905, + 0.5544291280405642, + 0.5307509435489854, + 0.5362868969200196, + 0.4926008174758058, + 0.3964291406120515, + 0.7453802516590226, + 0.5418364081035901, + 0.7332123650951285, + 0.33263267709281763, + 0.3087638978465614, + 0.354238155705568, + 0.2905529374974485, + 0.8758223823514306, + 0.2702505238802112, + 0.8712975666663519, + 0.6447457970927334, + 0.42922127550273714, + 0.2859311974467197, + 0.301149201263264, + 0.32434863233154293, + 0.588028515617278, + 0.6306189406015921, + 0.6518511899872177, + 0.5771761935321705, + 0.4716700758270473, + 0.40115692995463303, + 0.6298680345324424, + 0.5157496581086833, + 0.43262276231009755, + 0.5433599481220323, + 0.4058489321164665, + 0.40872752538001755, + 0.32190405500458374, + 0.8107064703661099, + 0.36266813151500615, + 0.39262460568425667, + 0.5780010116922265, + 0.6544291899886129, + 0.40553652602947965, + 0.28510648619361983, + 0.8604114902647961, + 0.32309484184841447, + 0.7294956874886884, + 0.35185610713116644, + 0.5407675980391072, + 0.6561697949847166, + 0.4183078817725991, + 0.48810770312054585, + 0.4954281216013944, + 0.7956235347477538, + 0.3825445094412282, + 0.5502851252617066, + 0.46418876074977494, + 0.6426401908661429, + 0.8782246187000978, + 0.3349289045601117, + 0.5172354668515198, + 0.389005516439018, + 0.5570787232650555, + 0.6525185865503367, + 0.5171434063550875, + 0.3776519671427676, + 0.8503010075723495, + 0.8545136561051346, + 0.7282160815159273, + 0.8075321896239172, + 0.820137102973549, + 0.8042696857715371, + 0.8213672588735322, + 0.30301355242251055, + 0.3116441232576851, + 0.442641904914012, + 0.39955458568704627, + 0.3766432104080838, + 0.5306345878864145, + 0.8370611724324138, + 0.8297941854618072, + 0.6042414085122075, + 0.8025407944605704, + 0.797648994719154, + 0.8356447185591828, + 0.8841053121711021, + 0.4125645341654173, + 0.46532936810218234, + 0.5841951533867318, + 0.478759072466552, + 0.7911849149370651, + 0.4805132183989691, + 0.8856755203216273, + 0.8619167431281788, + 0.853539195244854, + 0.8815102852278999, + 0.8748030691599222, + 0.8464547148955549, + 0.8074472402483761, + 0.840626002614182, + 0.8834613897983155, + 0.3636450372010752, + 0.9031581825763413, + 0.8586240778783243, + 0.795770890104613, + 0.5261554811942016, + 0.8543772291026758, + 0.4472505693047547, + 0.7309171508575679, + 0.8532853235119024, + 0.4901464990907652, + 0.48215313147945527, + 0.5907646053354426, + 0.4616415284885569, + 0.4495362997208625, + 0.6834057229015857, + 0.860884200016329, + 0.636378716389505, + 0.7801336905815686, + 0.5425612695873796, + 0.8353631579388088, + 0.7815268747828015, + 0.8140703014268283, + 0.8827472520583913, + 0.8161716676025444, + 0.8766063853136333, + 0.8428897115327111, + 0.44988177710939536, + 0.8543887884464569, + 0.45526122562019705, + 0.45170442614839257, + 0.8481124905579103, + 0.884790807778116, + 0.8537046836638795, + 0.47666552175511717, + 0.5351547259734077, + 0.6063543064136611, + 0.7712425755636855, + 0.8652601151236203, + 0.622833354906427, + 0.8791653472412693, + 0.8944876067151276, + 0.5268880471773332, + 0.2638787245601644, + 0.45831750335841154, + 0.8813107845146603, + 0.3469979093849518, + 0.5361003543908566, + 0.857029629195837, + 0.8067780331110332, + 0.5241892271435185, + 0.8454454838175337, + 0.8760246454438088, + 0.8698598676537322, + 0.8526629412852625, + 0.8442254703182639, + 0.8700521812933121, + 0.8321992161964104, + 0.9058244817144404, + 0.7615346757542177, + 0.8125793735792926, + 0.8926746094592726, + 0.4688173121030755, + 0.47393266168426157, + 0.44579911722408755, + 0.4369128952742375, + 0.8901378837216493, + 0.7710907290579766, + 0.8272097396737335, + 0.7647090047806934, + 0.7376401636751615 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.3317488363965557, + "valid_tracks": 12, + "per_track_pearson": [ + 0.3261400809112019, + 0.3664624313873139, + 0.2863933619906205, + 0.36805408643131776, + 0.3608426609211243, + 0.36114425738366096, + 0.3011709810607287, + 0.3458949865051905, + 0.24822084547387102, + 0.3413506196714335, + 0.3301116730000494, + 0.3452000520221563 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.18066326957284914, + "valid_tracks": 546, + "per_track_pearson": [ + 0.1726681761944187, + 0.20444998335759548, + 0.17469811260136006, + 0.10567393240201972, + 0.05553939589000169, + 0.18464254978983816, + 0.08906861816787504, + 0.1187930161472794, + 0.18910829587781872, + 0.15817858944837998, + 0.20932229940045124, + 0.1108015382956947, + 0.12545456244359876, + 0.10245816956487758, + 0.07807499832925704, + 0.14246649974704498, + 0.17910295772034357, + 0.21745483045547478, + 0.2019907426311949, + 0.20112589595978936, + 0.1718483670231368, + 0.1638043704196551, + 0.2713768177157652, + 0.13911661244362694, + 0.14889946509213833, + 0.18176556213447814, + 0.20504643935946387, + 0.21091067842304698, + 0.22720398762974886, + 0.21532012535314043, + 0.13391510891746758, + 0.22241296833270585, + 0.14729891235799655, + 0.11244986290465893, + 0.22924430639147284, + 0.222388108317822, + 0.20465821098343373, + 0.228724647222783, + 0.19803238418417551, + 0.26356390664803986, + 0.267648463816309, + 0.2338322670423025, + 0.20063604341193128, + 0.19590235635205708, + 0.12081994582664772, + 0.16692893413487392, + 0.17334272460472994, + 0.1796396597483245, + 0.10720083684977567, + 0.14027107776606867, + 0.18957231182253786, + 0.2281242963729715, + 0.23776135359270728, + 0.16048390469452387, + 0.2501381969368553, + 0.2746704965410695, + 0.14031098966781305, + 0.254479736779293, + 0.1911134659539861, + 0.2150192947457678, + 0.25443727033121866, + 0.1510732877822669, + 0.2301584734131527, + 0.1652721061815186, + 0.23216159855690477, + 0.248767280510286, + 0.26351900722930643, + 0.15536653070987863, + 0.0981757400278823, + 0.09299616153957774, + 0.20112210421360754, + 0.23909783013540822, + 0.1948022801867839, + 0.20512518544039582, + 0.2268108124598396, + 0.18695213061348942, + 0.2211006543680953, + 0.2044632293061416, + 0.22944708237540298, + 0.23614595398899174, + 0.19062339331242262, + 0.11029991513081877, + 0.20832341144854052, + 0.24216025182511716, + 0.11600922227985422, + 0.21643616073428212, + 0.17267195595902757, + 0.24483004503559788, + 0.1495255688461516, + 0.23147289061257242, + 0.16863611940297848, + 0.19256557430441448, + 0.21494019030072098, + 0.09480512946935642, + 0.22189149522596996, + 0.1905208635938321, + 0.20249422086676658, + 0.22100818438266914, + 0.2018523157962885, + 0.11814627813006566, + 0.18768812434402699, + 0.2114962025872281, + 0.21041579015844433, + 0.29296404931725983, + 0.24210438186636368, + 0.26181575676695407, + 0.16964523998017725, + 0.21888367556315494, + 0.1657612641926429, + 0.17450322701733734, + 0.18980610215834084, + 0.18837985227418108, + 0.21258985905906125, + 0.20915026171950346, + 0.16376921320381532, + 0.20662066662648504, + 0.21251200323126385, + 0.2166758783460428, + 0.20404850663419993, + 0.23487772869264267, + 0.24054962855503567, + 0.2341081289941437, + 0.14145648597364865, + 0.15149263951657624, + 0.1854377349325125, + 0.2273134651343163, + 0.20422868107086678, + 0.22496203563102715, + 0.2977503918962978, + 0.21366613634644718, + 0.24053021590886567, + 0.16303805734427315, + 0.22830354622852866, + 0.22595065358263638, + 0.27607549755716726, + 0.19176269477788205, + 0.24650652866431433, + 0.1710395928588895, + 0.10297150937129589, + 0.21297696018833237, + 0.25235521093190494, + 0.2148448015305251, + 0.22824722061128347, + 0.18106439695606177, + 0.1698115140842062, + 0.19170138553198293, + 0.20145958753761836, + 0.24695477872719643, + 0.1742375481845308, + 0.17207633277833653, + 0.2084972623045976, + 0.1502848787476804, + 0.1923547095033159, + 0.21637004456949896, + 0.1892928711126083, + 0.1336066810922874, + 0.2244874791565176, + 0.2596680279126595, + 0.16145378288958803, + 0.12884095037550414, + 0.12305415125448384, + 0.1800682387190481, + 0.1799020365279142, + 0.2038416924578705, + 0.17703179419725185, + 0.14265785610612125, + 0.19984496047213454, + 0.1319037560156313, + 0.12076894069261535, + 0.22861251006387387, + 0.20688051322923748, + 0.1482125546245078, + 0.17953985123986904, + 0.2797165663757242, + 0.20579170457178841, + 0.11257708914116707, + 0.12380791068715653, + 0.1614434537494647, + 0.21453438281345855, + 0.2520460799346758, + 0.2388420840800608, + 0.23685789968958137, + 0.11371822242791063, + 0.23860534251747453, + 0.19283147346767035, + 0.15205409928502017, + 0.1698720308721409, + 0.1359331117711021, + 0.2550679289210337, + 0.21495991851383064, + 0.16651144439082724, + 0.1088197894701279, + 0.11139070620238169, + 0.18399266195756875, + 0.1830411816324703, + 0.1646064205994545, + 0.19516173708959275, + 0.19432190241734312, + 0.13819916055071238, + 0.18986074811920736, + 0.10251331756253702, + 0.08317966232997884, + 0.09474595832577153, + 0.08434448589617333, + 0.11314182843480543, + 0.1880300798051844, + 0.18641075384734396, + 0.15869868932482467, + 0.08743375239779505, + 0.17175742984010184, + 0.2358769158044101, + 0.20802181677180126, + 0.21183982359923145, + 0.20900236928133434, + 0.1949664902928648, + 0.19529771961007772, + 0.22708667162960314, + 0.25072953994130476, + 0.2249005559384223, + 0.21999461075856774, + 0.19948322291878562, + 0.14122339001408793, + 0.2013278431992787, + 0.2001494073418792, + 0.21154346844872096, + 0.2597763742553705, + 0.19565669493266866, + 0.16545134991305102, + 0.1918472354647725, + 0.2023471681735993, + 0.14463825553001297, + 0.14403360765874634, + 0.24781207775855307, + 0.1970955825386554, + 0.16277650016395043, + 0.19331984193626714, + 0.25733692165147337, + 0.13280934546918305, + 0.20106968012706583, + 0.19767522465928025, + 0.08048073555073869, + 0.13327375134234956, + 0.10883691754907344, + 0.19086821542330074, + 0.14247800522578166, + 0.17649202993575844, + 0.07082547438995754, + 0.24257872412014012, + 0.20280361318660253, + 0.21257643989846356, + 0.23115255834752035, + 0.13113235349912158, + 0.25749008627738484, + 0.10831443207021092, + 0.22676896407418734, + 0.15682032522436268, + 0.289928412754713, + 0.21536495398159894, + 0.2169553753529255, + 0.3148731386036633, + 0.17810100578001178, + 0.20342739002117718, + 0.09874402609377977, + 0.21172120834802, + 0.2031786436896366, + 0.13858068146065217, + 0.10309795750166087, + 0.10567153406602292, + 0.11861437764971648, + 0.2060168053369955, + 0.253808615028564, + 0.10518921847786952, + 0.09089768305691556, + 0.15750913493818378, + 0.2124690316687272, + 0.16853390710633856, + 0.11666712461078022, + 0.05766426988850531, + 0.17644941318847304, + 0.10752407009406169, + 0.11431911091084199, + 0.18625505094387754, + 0.15145961621982468, + 0.18936135778249769, + 0.11520013076093037, + 0.12309528183524364, + 0.10576427323055707, + 0.09798030896049242, + 0.13439113075981404, + 0.1675018275527015, + 0.19846812396858357, + 0.1879218509724187, + 0.18276942529591544, + 0.16243654743844127, + 0.15694591608202843, + 0.25230878580723903, + 0.13022418689057302, + 0.14011078878824265, + 0.18741894314806093, + 0.18765977686962004, + 0.19029835312358598, + 0.20581468541467712, + 0.2050018231567081, + 0.12038459022400438, + 0.19308526574827165, + 0.1408809094870893, + 0.1051586187682737, + 0.20815738883668586, + 0.21703992692330573, + 0.19940396446339892, + 0.2044106119086027, + 0.17637028157208004, + 0.24207677265298863, + 0.24226818294565894, + 0.21175774072525297, + 0.1954886015351272, + 0.18328358945245227, + 0.11370045141588091, + 0.16182644229565116, + 0.17301110038180004, + 0.1881488612585868, + 0.09985391392815608, + 0.13402049871526991, + 0.17001619083001798, + 0.2201358752936261, + 0.221888296849636, + 0.1479053535708754, + 0.22442196973055745, + 0.24560825524837168, + 0.13893991331668873, + 0.2255281870711701, + 0.1740070982918138, + 0.1876226323895866, + 0.23069259282905702, + 0.1445080822507166, + 0.21331761328068177, + 0.16280480082469706, + 0.22439630635288846, + 0.21506201212771314, + 0.23874621446393443, + 0.14143920439069615, + 0.09628627385100876, + 0.08804288801867405, + 0.1866328363793594, + 0.22151308484236615, + 0.18884997865964184, + 0.18899946977476123, + 0.21483472940005388, + 0.17598262368951428, + 0.19374678793172806, + 0.18725631479414112, + 0.20944877964129652, + 0.22550601742091925, + 0.18173968050461028, + 0.10797162780165785, + 0.19788410273175433, + 0.22034466157616212, + 0.11806439982615885, + 0.2019302223305541, + 0.19424194244137302, + 0.2305622471467936, + 0.1380465222434513, + 0.2270650706345365, + 0.16366362995387615, + 0.17403322278630304, + 0.20121279940363176, + 0.09581243422846143, + 0.2017457081362786, + 0.1745732470683879, + 0.18001862184859194, + 0.22027077443831855, + 0.18524203207545697, + 0.10775184892432976, + 0.1771181894506629, + 0.20934149854443773, + 0.1912602798972745, + 0.2689275343008004, + 0.2189673189778695, + 0.24830982054293213, + 0.1552323864537385, + 0.2014904145637359, + 0.16111424705589653, + 0.15676114774632727, + 0.17419040542573772, + 0.1747984564019772, + 0.19604269444650702, + 0.2000805206018612, + 0.15228266903225512, + 0.18961590522938812, + 0.19208466219678175, + 0.20365951243105512, + 0.18449929608645577, + 0.22346966460968334, + 0.2170039173026218, + 0.21236337895620253, + 0.13884972947256377, + 0.15359850404105427, + 0.1796028184643768, + 0.20479482263019322, + 0.184823551938803, + 0.21630136042555928, + 0.2766047397636958, + 0.1953245320621593, + 0.2229237397269222, + 0.15617976803205377, + 0.19902111933386676, + 0.18959509132488153, + 0.24884871537941772, + 0.18527935395538406, + 0.2308850666820824, + 0.15782683516980525, + 0.100107765473016, + 0.19682921489348512, + 0.22823962793588112, + 0.1918636609066415, + 0.22297177602680554, + 0.1632791388251168, + 0.16027864105780268, + 0.1706573258402939, + 0.17708946240568935, + 0.21017315467506187, + 0.1731453394831159, + 0.1512747523126322, + 0.17609147269353476, + 0.1279472258191604, + 0.16309532875745072, + 0.16476320795289692, + 0.16690553314071052, + 0.11224482526219258, + 0.19376898598377834, + 0.2232652735967115, + 0.14386801892542755, + 0.12286155005994216, + 0.10980724689643084, + 0.15671758492575905, + 0.158779530670779, + 0.2015739058675396, + 0.16905003440192978, + 0.12562086501288408, + 0.18595731524560358, + 0.14158263071873367, + 0.11064938323537822, + 0.2136336133405868, + 0.18935615871068795, + 0.16157892705899618, + 0.14851857468113497, + 0.2632791282399886, + 0.19109862603655162, + 0.10262103460831143, + 0.12232813157543616, + 0.16997554959430625, + 0.16900868204577751, + 0.252114334519964, + 0.21642040209252442, + 0.2276083714146563, + 0.09724542995918524, + 0.22281024126564852, + 0.18775944201511716, + 0.16567257626064089, + 0.1421602090836331, + 0.13046175178959546, + 0.21381895546203403, + 0.19134241279051822, + 0.16482929105349658, + 0.12440918541188073, + 0.0875319100022925, + 0.13164246249343295, + 0.18867456040625893, + 0.15608467623546024, + 0.19165890975487, + 0.18313351333769315, + 0.14218967715230216, + 0.15243337190109288, + 0.08103569323165048, + 0.06049685858308272, + 0.07652148919090973, + 0.07602518573459809, + 0.07681566732608199, + 0.18054072963445178, + 0.14529705767118148, + 0.1270397605273858, + 0.07966990189708109, + 0.17112120232875858, + 0.23631363821495005, + 0.2130679659584696, + 0.20782827266474244, + 0.2096949744023833, + 0.195806686641128, + 0.19428168567992884, + 0.2201823059445729, + 0.25134138922498284, + 0.22236651703673596, + 0.21243259058304503, + 0.2005846034331201, + 0.1359860318820037, + 0.20244329997386426, + 0.19232228038118268, + 0.21179980906689733, + 0.25391361932810186, + 0.17547800328402038, + 0.16126264088930017, + 0.15980583573418583, + 0.1837911722606348, + 0.10841985235022841, + 0.09571375836029448, + 0.23432571408604796, + 0.18660823756504746, + 0.16225636672256885, + 0.18515132070326978, + 0.23614130542168466, + 0.1361436540039818, + 0.1851872720840193, + 0.1776215263892009, + 0.07218544517090508, + 0.10253564055250439, + 0.10024280525057849, + 0.18970674409649899, + 0.12039549296252203, + 0.16621003217280853, + 0.07051372035005142, + 0.2367497584386492, + 0.20453387154416466, + 0.1991309411811481, + 0.22517509948269473, + 0.14777539920215754, + 0.22458541921470285, + 0.10311445517755129, + 0.1998282468597243, + 0.1550222435308118, + 0.2764768838082069, + 0.20868571857130666, + 0.21663348534066318, + 0.3032002464954743, + 0.17614226927872598, + 0.17801480093795252, + 0.09391615427579954, + 0.19293072564403826, + 0.18796644696013684, + 0.14812103079095326, + 0.09524191332724692, + 0.09674778007262756, + 0.11308968693620508, + 0.20872037873151622, + 0.2472456344877532, + 0.09828835650843112, + 0.07919245931528236 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.2009765256073382, + "valid_tracks": 613, + "per_track_pearson": [ + 0.23554563044125829, + 0.20602623740225573, + 0.1883331301503943, + 0.25136592596511803, + 0.22914660518105964, + 0.18920218052659182, + 0.21900527982267845, + 0.22861692731647065, + 0.23319050369613065, + 0.25698948682981515, + 0.17560412564620556, + 0.22134465946152646, + 0.15077690018693182, + 0.27154688091465407, + 0.2644948435447308, + 0.27450692619580663, + 0.2511793847914716, + 0.2339828631875139, + 0.2102832477666289, + 0.23205557054154707, + 0.1992294578223936, + 0.21391621846043554, + 0.23229731052051916, + 0.21880217509751418, + 0.1606532242491078, + 0.20576264581687667, + 0.1648091940216792, + 0.2213844549681071, + 0.22662451716845003, + 0.23729295668934347, + 0.2746224643238935, + 0.20933660208460883, + 0.2888372005112373, + 0.1894702501312026, + 0.2970942910723345, + 0.16293027812410554, + 0.23148293494058791, + 0.22869475427911026, + 0.206944157883126, + 0.22706237919849467, + 0.2586085722098654, + 0.20637398591967085, + 0.22345800203460484, + 0.21994117298548724, + 0.24609755653956217, + 0.19451390755499517, + 0.23057268312509857, + 0.2711130237442279, + 0.24786098932866021, + 0.2034697484478525, + 0.21871686367497092, + 0.2290880461540137, + 0.22626803778309026, + 0.2479169276475476, + 0.17079665208914102, + 0.24749026653087286, + 0.23140770886889298, + 0.22783103799052273, + 0.23067073106560645, + 0.20580263210245694, + 0.17967828110980022, + 0.20475961087407987, + 0.2076485308059104, + 0.18776059087215877, + 0.22702761051148754, + 0.2009357603854917, + 0.2289343815284083, + 0.21912425596621607, + 0.2312723444705509, + 0.21574040704532968, + 0.20997983666339892, + 0.22834850119910752, + 0.24374536474526542, + 0.21333262562210248, + 0.2274501768204644, + 0.24168704396494384, + 0.2272217065299782, + 0.22673864410285113, + 0.27560658411365446, + 0.2206211323733586, + 0.2070960389973582, + 0.20279271041906752, + 0.2646159527828368, + 0.24294813959986863, + 0.1726239287337746, + 0.19571753483010917, + 0.2320551039650409, + 0.25334343040954554, + 0.2299364837803433, + 0.2466035361009463, + 0.1831284655146108, + 0.1885157798601589, + 0.21217585278936898, + 0.21830165636531093, + 0.2288703166395457, + 0.2224307295265582, + 0.20693904553917164, + 0.22672619768381658, + 0.2365016864539691, + 0.2002693742942307, + 0.2077201681062996, + 0.24301939699826702, + 0.21988743023369015, + 0.21423838252320831, + 0.23567712173056304, + 0.2618699261030163, + 0.2124905243053831, + 0.27487091106127515, + 0.2508810373604454, + 0.23367820725996127, + 0.20406565747837965, + 0.31649988844810034, + 0.288175126324399, + 0.17294459504711507, + 0.2218854851910514, + 0.22221268662034316, + 0.2814530768662763, + 0.249183417459261, + 0.2456253164998071, + 0.22507660887246628, + 0.17288548022691902, + 0.23216253643728324, + 0.2170494486403018, + 0.24783529505340957, + 0.24753028256224005, + 0.2340076441057989, + 0.24992720811907557, + 0.16615114293440023, + 0.21514862787742792, + 0.20261939034247184, + 0.29630280443484996, + 0.24364849492877305, + 0.2098810132102792, + 0.16428792196316513, + 0.17845764008088685, + 0.22026437730549198, + 0.2627170954964987, + 0.2714755651383875, + 0.2704503683951755, + 0.2561114519774726, + 0.24381503279818617, + 0.19524044946675947, + 0.2705536460878809, + 0.2924338869215419, + 0.20277247875895602, + 0.20397619452300902, + 0.1782049166485818, + 0.22017737039931182, + 0.17943334291477864, + 0.24354059369968795, + 0.17960926203297453, + 0.2529038668980252, + 0.19822476226953978, + 0.20353988467648543, + 0.25887166357344527, + 0.1889595665058616, + 0.22819419497656812, + 0.263480466460366, + 0.2006702241564689, + 0.259003746712166, + 0.2913445980570118, + 0.2212169027036543, + 0.2475585773054586, + 0.2400574503983318, + 0.2360783431598858, + 0.23454767565484136, + 0.239222520835071, + 0.21456737620057367, + 0.22389394809807325, + 0.25925065795301355, + 0.18581402235967026, + 0.19212927046820824, + 0.2383910954524104, + 0.24879634894254918, + 0.21408764836549118, + 0.19801870736252403, + 0.2142681349720007, + 0.2635302523343798, + 0.2745406646545422, + 0.17787971898143173, + 0.23901434630271143, + 0.1888244988956557, + 0.19509502069753337, + 0.17938613013344476, + 0.19910267476920632, + 0.24295896027688027, + 0.20632968793364526, + 0.12331613198823657, + 0.17163993180660322, + 0.3037030219367529, + 0.32543884478723395, + 0.18392409074746896, + 0.16623592813426682, + 0.22649431334949927, + 0.17324545117660212, + 0.18991528639292937, + 0.19842823266899492, + 0.2109933622249976, + 0.16574585159045732, + 0.2180928191459438, + 0.29993583065343987, + 0.16639440477895084, + 0.33919401307224656, + 0.24589177216092717, + 0.190598321409746, + 0.2047131303877718, + 0.20183227062318004, + 0.1771475588395764, + 0.18919151656630856, + 0.16646853708287224, + 0.16542370979334486, + 0.16454047378128422, + 0.16211520976196686, + 0.17813196548204138, + 0.18104570273632342, + 0.21284097006171807, + 0.16548361031998254, + 0.20501599154726452, + 0.1744984741779595, + 0.18382367900103894, + 0.31850012829863333, + 0.18753284945288895, + 0.2831682271175704, + 0.24683127057130286, + 0.24938227059215454, + 0.28185123729766737, + 0.25138949288681034, + 0.271979184150333, + 0.20980988487906954, + 0.17631954950362097, + 0.22044026835889344, + 0.31715966972560217, + 0.19889426753031603, + 0.25036593942025354, + 0.2095467891118419, + 0.2303845026395889, + 0.14998774512406202, + 0.23152118340107225, + 0.23251908343364586, + 0.2321256358690178, + 0.23319247562978015, + 0.17244835625004284, + 0.22776095463295437, + 0.1838255166275759, + 0.18451119132614044, + 0.2469226864788969, + 0.30675667545963026, + 0.21973360567883943, + 0.2211702740562431, + 0.19346770820141837, + 0.20719099140674277, + 0.22049006298740867, + 0.16974168479839236, + 0.1927992779982286, + 0.22119139182147923, + 0.25199941442630164, + 0.21051631997357192, + 0.19918537631089656, + 0.20681664674148362, + 0.18168494691213663, + 0.2526201968325216, + 0.2253091340210493, + 0.22372544076558257, + 0.253382603291522, + 0.2241637712880344, + 0.2087687729877054, + 0.1918052567047429, + 0.1814010480736744, + 0.19995634819307662, + 0.22190992927579428, + 0.22647094479190774, + 0.13305579694929046, + 0.16393101089239168, + 0.14756275773108615, + 0.196239325367566, + 0.1725914935662401, + 0.1246863597952131, + 0.13933874868743154, + 0.17721070046180043, + 0.17775596566860188, + 0.19996988297844842, + 0.15999760969963245, + 0.1360747064570286, + 0.12742445822851614, + 0.17197598120156776, + 0.17163589988956496, + 0.1680150636856594, + 0.180238490753293, + 0.17564121421103773, + 0.17732143252067875, + 0.17049565185668145, + 0.14281779355438445, + 0.17971115841805307, + 0.15326195148509192, + 0.13759409615721283, + 0.10243244474968186, + 0.15296354485571703, + 0.14125111451437264, + 0.17625050668394104, + 0.16167504914601166, + 0.16396742744107645, + 0.19936266923590149, + 0.17972546105524598, + 0.15665791260935027, + 0.14674048217171545, + 0.24844393135620896, + 0.1267579066816808, + 0.17728140354764377, + 0.1734149855370335, + 0.16240143937380178, + 0.16954359609895955, + 0.1867071134022365, + 0.16909203156042096, + 0.17134411724904985, + 0.16534725612949974, + 0.17144088933259385, + 0.1379664639241948, + 0.1527519499030133, + 0.19003331008720267, + 0.16499546343478994, + 0.16863143445852216, + 0.17444570622430575, + 0.17719729326020112, + 0.15212944044613277, + 0.17463552550135808, + 0.1036688129714865, + 0.1638722810408192, + 0.17024304775118873, + 0.1628143819395391, + 0.17670708345256447, + 0.15523786613044746, + 0.14933614455793032, + 0.1671897883929436, + 0.17551866315128004, + 0.12910197183740685, + 0.1802950626800305, + 0.1630626126412974, + 0.19149487616542704, + 0.17746724511993325, + 0.1912593109155336, + 0.17333116941577864, + 0.16475972477118256, + 0.19116038738640798, + 0.1896966292697144, + 0.18007526501981153, + 0.18332792888326682, + 0.19927339126279517, + 0.1843076568522989, + 0.1858387332638462, + 0.20318690207183024, + 0.1652034479415218, + 0.15542296282567158, + 0.14174745243281392, + 0.19950576880505572, + 0.20052151365145512, + 0.14582845366722288, + 0.16286183320499253, + 0.16792686306082538, + 0.16851529435561577, + 0.16092783903016286, + 0.15054776432706898, + 0.11808297308984765, + 0.14519204367689778, + 0.16043937856796592, + 0.15727837336170078, + 0.16552584589864835, + 0.16864800643191322, + 0.16926913608857735, + 0.16819895390892076, + 0.16962616043129178, + 0.13919912389023506, + 0.1413248883675485, + 0.16034968873776556, + 0.15315073320521272, + 0.17629495781701165, + 0.16466122627284985, + 0.18401548566538173, + 0.17184005221148949, + 0.19170419950008424, + 0.19861917978415145, + 0.1678396070815798, + 0.14277321048472968, + 0.23980742054259915, + 0.20290162206911583, + 0.13889880138309854, + 0.18470981188189478, + 0.14857537219785105, + 0.202323794209049, + 0.1729557056565432, + 0.17907578569969404, + 0.15617681815178505, + 0.13099249951143443, + 0.1759103904500507, + 0.148768073123103, + 0.19579598800573506, + 0.17856597664001583, + 0.1781403487164693, + 0.164247299236794, + 0.126457422489266, + 0.1391904870287869, + 0.13891863497671195, + 0.2224790118539089, + 0.18067883658565254, + 0.1366383852305969, + 0.1181012175182539, + 0.13594383794281129, + 0.19988097760208956, + 0.19140543683068623, + 0.2107278643198629, + 0.18866834111692327, + 0.18407319580355883, + 0.1555804583057045, + 0.13204604364371736, + 0.20194443866478554, + 0.22310110583536044, + 0.14343530746365857, + 0.13935238482722917, + 0.14040300177634557, + 0.1656695849619853, + 0.14754397788514353, + 0.1728263652862161, + 0.1419897091707307, + 0.19271663895893593, + 0.14487214458629308, + 0.13492226598445695, + 0.1794140044863306, + 0.13980250407903913, + 0.18121185397923947, + 0.16222133339692835, + 0.1812288869942675, + 0.17413214270369673, + 0.21765640889578414, + 0.14065012719328304, + 0.1656850735589843, + 0.13060689640983122, + 0.1300604694697789, + 0.12680600436260112, + 0.12637373529324009, + 0.12759713122474453, + 0.1488418746680989, + 0.18593832513823208, + 0.17549502412093065, + 0.146306786994217, + 0.21032361471044814, + 0.16666061082053288, + 0.1384598913018664, + 0.1428951917406091, + 0.16701705783409235, + 0.22104515137196706, + 0.22264899789314785, + 0.132259903458532, + 0.20043851826367337, + 0.16761997539991638, + 0.11629123688418772, + 0.12571916484700857, + 0.13553819215824234, + 0.18227783480953383, + 0.16798932812717285, + 0.1031059396794889, + 0.13926259643944064, + 0.25985630573232993, + 0.26269590152953093, + 0.15229100056622005, + 0.1322341230828651, + 0.17942293647726423, + 0.13880403486828882, + 0.15957782405092416, + 0.16291855555934046, + 0.18946256302495323, + 0.145497633000943, + 0.1795178902785303, + 0.25602260656610176, + 0.15225374285471452, + 0.3023935131958923, + 0.18820421884246105, + 0.16219708669977584, + 0.16503304669123411, + 0.1481130535667473, + 0.11302159018264711, + 0.17592990347663842, + 0.12130513110709604, + 0.161137886929316, + 0.1536574821166151, + 0.1731249680761389, + 0.16889380044227448, + 0.15331563248895738, + 0.18133439936323137, + 0.17173865760906748, + 0.14619747457104357, + 0.17647974469593108, + 0.16645460315723365, + 0.24444344298503087, + 0.16643192272872476, + 0.21392172161490502, + 0.1542131472637783, + 0.15851303182601859, + 0.21377379760817358, + 0.14282093804098694, + 0.2053097781755193, + 0.10659910337920485, + 0.1599908344454838, + 0.12983946532173754, + 0.2757402576803516, + 0.1759180565099438, + 0.20296741128088985, + 0.17863815901093233, + 0.18838561332950668, + 0.10194044423377785, + 0.19198150370584136, + 0.18235870516460814, + 0.16133914491005033, + 0.16926521813659762, + 0.14692408153849013, + 0.2038821211556549, + 0.17683039665922215, + 0.18896648511090774, + 0.18822568110177074, + 0.26438740727495397, + 0.1747926804015978, + 0.17008669719557823, + 0.1414016065853467, + 0.13163361424641587, + 0.17126671999169785, + 0.13674784022473208, + 0.15151468794930006, + 0.17894064759091496, + 0.18439242809994458, + 0.11669631775313835, + 0.1666679056977758, + 0.13457117764247958, + 0.16345235264194677, + 0.18516502492504455, + 0.11270713635904142, + 0.1612214039369453, + 0.18869895494068387, + 0.17861570923994102, + 0.1885401561994078, + 0.17171722785898164, + 0.094133990277567, + 0.16954493310314647, + 0.18335104362207716, + 0.17539560723916423, + 0.19714972514048257, + 0.09879724269311678, + 0.26649338310416987, + 0.3041512266400287, + 0.2787026311202462, + 0.2564078803636943, + 0.2613092196052215, + 0.2171637621003818, + 0.2182132721090965, + 0.24965213245900372, + 0.19372498472732852, + 0.20946000709591236, + 0.21540639761738542, + 0.20134585491740928, + 0.20279622286989804, + 0.2085449251204644, + 0.2197893379255217, + 0.20206552677208878, + 0.21104901624346098, + 0.18817541547452094, + 0.2689607015697996, + 0.2676331625412983, + 0.2674801792369937, + 0.26619171253482227, + 0.30090973534611476, + null, + 0.28375492533037916, + 0.3370664695479783, + null, + 0.06323932219192019, + 0.4004253839180688, + 0.06944196669536425, + 0.2627819166823951, + 0.24289229955822064, + 0.30968395358415796, + 0.37011016676625563, + 0.36179691553501897, + 0.384191413460685, + 0.37732973012871573, + 0.3419001241572681, + 0.2946711320573942, + 0.2246294095690373, + 0.3827715338856736, + 0.18074980712931787, + 0.29700984684246884, + 0.263030593856712, + null, + 0.2565775203001909, + null, + null, + 0.26161210209410696, + null, + 0.24201560686165188, + 0.21151712095741593, + null, + 0.30272723382147604, + null, + null, + null, + null, + null, + null, + null, + 0.25706931138646205, + null, + null, + null, + null, + 0.2092049766131927, + null, + 0.2636095244878663, + null, + 0.21775112353422602, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.27678922056956456, + 0.1528666849431263, + 0.24992463367497467, + 0.25835807949740996, + 0.1706957944392102, + 0.23523990099544087, + null, + 0.29886180208842295, + null, + 0.24113795847737807, + null, + null, + 0.23887940425428458, + 0.2742931202766058, + 0.23170485704004978, + 0.237628979205647, + null, + null, + 0.20252211793131386, + null, + null, + 0.24471031041219002, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.23771348656357738, + 0.23682050268311297, + 0.25007183977269376, + 0.2232892200439769, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.36698738437050543, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.18605359449331416, + 0.25289971569315023, + 0.27528881363331564, + 0.39607796874979206, + 0.32310346969566484, + 0.2028468283413249, + 0.29450672040407516, + 0.34113895712910475, + 0.2818715157714315, + 0.27800257787958804, + 0.38293113176652477, + 0.3383346168403477, + 0.2568633752079149, + 0.33498679624750705, + 0.29160096038069105, + 0.2965596948938549, + 0.3172760936735992, + 0.2636134187775228, + 0.39261310088978724, + 0.2826942605203927, + 0.28766243645707235, + 0.2984055896807618, + 0.2737384251572345, + 0.3964741228010041, + 0.28340048057595635, + 0.2867119401062806, + 0.21008821249492896, + 0.3424714762632227, + 0.32117498722338916, + 0.36692848367903547, + 0.30307393959027074, + 0.34727887503096305, + 0.46163829027898795, + 0.25336545317487796, + 0.6170323487787197, + 0.267509047436743, + 0.18063229536533285, + 0.1510263847688611, + 0.29372613078305526, + 0.20057641614071678, + 0.24366026213310105, + 0.43918033119191824, + 0.2761619748662937, + 0.35796144304115934, + 0.2642792247886365, + 0.3601536662661621, + 0.26780710405146707, + 0.18908754661296057, + 0.4253668409782453, + 0.39452466279638104, + 0.35705882805241235, + 0.3469826410575407, + 0.36087548784345547, + 0.31329377371448186, + 0.30327533032225834, + 0.3132410034979766, + 0.23813747226455245, + 0.3328559030971655, + 0.3150044854205327, + 0.3370494331901398, + 0.2236554917343509, + 0.18264503019018463, + 0.37574598083267147, + 0.37241939334199675, + 0.2529441133976179, + 0.3029198358685377, + 0.31226509555744053, + 0.3857406712938147, + 0.3938055896042359, + 0.321514720699943, + 0.2965602140982005, + 0.6624054587566737, + 0.2913593490506823, + 0.2796184355649715, + 0.2706501000741514, + 0.36502862270352265, + 0.20558751056657265, + 0.5783069171824576, + 0.3235237330332286, + 0.32644372218209616, + 0.33149915165928745, + 0.355269194535597, + 0.4318767514872335, + 0.3400559874505129, + 0.25768543104231256, + 0.22102077866603675, + 0.40313967778352666, + 0.3098741798869462, + 0.7850087890216524, + 0.4159697300960074, + 0.2894078956990322, + 0.29350695698063317, + 0.3227396388517881, + 0.26050609285351983, + 0.30444188658420906, + 0.43538698869284026, + 0.28404555623104544, + 0.2637318842716572, + 0.2839226961581419, + 0.2345711945130131, + 0.22188384637084047, + 0.3500931646861991, + 0.31507632969463634, + 0.3114653888599944, + 0.35843843077123905, + 0.13020785978170227, + 0.14954200132466758, + 0.16574761154206943, + 0.3674521817198699, + 0.23600849920135047, + 0.14119385602334592, + 0.20150169497651382, + 0.17503485131790297, + 0.2578680204412213, + 0.22144154141417047, + 0.41357205254300766, + 0.30722526408259054, + 0.2570225214039928, + 0.20667415541622638, + 0.30316410421116363, + 0.19402012910355965, + 0.30822227074907965, + 0.2574745528932893, + 0.13354205799476218, + 0.24984750273656087, + 0.18872037061792316, + 0.22203559100704906, + 0.18510153648150457, + 0.2101577237941112, + 0.50403146045214, + 0.24345332338155123, + 0.15871182181488058, + 0.29006717352868755, + 0.5014125133577018, + 0.16005296892324591, + 0.2118907088291903, + 0.20418685240877887, + 0.24379408112785672, + 0.15714790731177383, + 0.16691152740782594, + 0.15733368883464502, + 0.39501816336538215, + 0.4238640612511172, + 0.1883956082897433, + 0.2872066102917414, + 0.3519493233294043, + 0.30885826696121976, + 0.13847045341842296, + 0.3120755813618935, + 0.23018399943055082, + 0.25072577553796005, + 0.19859596666756252, + 0.21735839815053284, + 0.36492605766271274, + 0.2948353796444221, + 0.28989160451487234, + 0.14627055118015567, + 0.18945711270424925, + 0.17782678686425216, + 0.27625615194108294, + 0.23991634598882844, + 0.22679543282947337, + 0.2485455263724565, + 0.2745577808678634, + 0.22470528688683714, + 0.24717644282868872, + 0.22545251649952497, + 0.3526481094299012, + 0.14894664004096606, + 0.2524683852060254, + 0.18688198512663437, + 0.17542911726329044, + 0.42248310881267925, + 0.1760334661025144, + 0.13138913317646553, + 0.3349106606617415, + 0.15496920175228268, + 0.2330260618916088, + 0.2252609855854983, + 0.19847514760861823, + 0.23732394651951372, + 0.21629303345496595, + 0.19378560859888683, + 0.30130839511275825, + 0.20244048823737604, + 0.2348680728001292, + 0.32602906019000993, + 0.17692699638566128, + 0.19409884307873784, + 0.3001900825279414, + 0.15223888904167626, + 0.2057899971250178, + 0.3577645076408821, + 0.275295598180099, + 0.18416591384880085, + 0.18171882488712168, + 0.16894055650566414, + 0.1503868308055996, + 0.1844688616553421, + 0.3503823793465296, + 0.19681828235071513, + 0.22293524645493068, + 0.19562543145471584, + 0.17529010735737327, + 0.3286345035875313, + 0.1948390912089053, + 0.2074330698458699, + 0.28872799656349224, + 0.15567061404404245, + 0.37619975287064794, + 0.1673084312434885, + 0.3070909436939991, + 0.23386048118941166, + 0.28758691896945027, + 0.832648611714524, + 0.29649676442360123, + 0.3391644506902616, + 0.34871329297455084, + 0.3972185524852638, + 0.5126235533047324, + 0.5837682465576233, + 0.592767705234659, + 0.31429896665357027, + 0.3292747035278923, + 0.41899705808202586, + 0.7086633371659438, + 0.2419101193716395, + 0.27613941976993633, + 0.509091225082812, + 0.48186766691030714, + 0.4432853926925452, + 0.358825563635336, + 0.2309653640418933, + 0.301842333881495, + 0.567983872561656, + 0.34592420972467014, + 0.21957770770115242, + 0.5910889980186711, + 0.5150465898994514, + 0.3999979021091964, + 0.43767584532356285, + 0.6845016352476812, + 0.4103188271979532, + 0.4314570762943625, + 0.3924218435059037, + 0.7470077102351553, + 0.2516716427195521, + 0.6217666666664271, + 0.32285177824481553, + 0.5156090813903257, + 0.37920910342237946, + 0.3700076886297759, + 0.6856223154004268, + 0.4629856071417176, + 0.23204097601830972, + 0.3273346251599295, + 0.6235206082687014, + 0.28493982981786004, + 0.4064667568461697, + 0.3550309017242522, + 0.2665953518242508, + 0.3955017979615849, + 0.3802816635871203, + 0.3455086372461645, + 0.3569818784087158, + 0.44222359529574173, + 0.2779348911657768, + 0.2550038566436117, + 0.6054909239973406, + 0.3133017598024955, + 0.4148983059591454, + 0.5128154635588197, + 0.2577274568757254, + 0.6402238076963658, + 0.24668454640413293, + 0.25838997867458274, + 0.5318036407814287, + 0.4300752673875781, + 0.5591826938299979, + 0.7143260444799578, + 0.6375762135294426, + 0.32702704148978234, + 0.4323630053341845, + 0.5021742144651438, + 0.22495767179175621, + 0.24547509934907855, + 0.5402249689412176, + 0.4407196866998587, + 0.6875639260089609, + 0.4054324872622091, + 0.29741266897102403, + 0.39227734297260014, + 0.31807457537730444, + 0.43473963806939253, + 0.6835559875266891, + 0.5990820596700522, + 0.38776207596523143, + 0.4826640605503489, + 0.44744967969748417, + 0.4880181880293719, + 0.7192265613447215, + 0.26078210408349095, + 0.4039921607692692, + 0.18750609393255394, + 0.5682967081251813, + 0.2564691569404414, + 0.7172599914847065, + 0.3374739732847154, + 0.3791486996494276, + 0.6707074029043528, + 0.4729186507572058, + 0.2569204548542268, + 0.5386313057412859, + 0.33689746452229646, + 0.48198243533911467, + 0.4847878325493918, + 0.2796042877073579, + 0.46989980401821885, + 0.3257287628059703, + 0.4196345648466626, + 0.28670997595296643, + 0.20094074714214497, + 0.3256679538045994, + 0.46198874494264647, + 0.5281138890864272, + 0.6354104838984955, + 0.31612367149019927, + 0.23623315077483645, + 0.3874478637033172, + 0.3987993080512689, + 0.3960760823583088, + 0.4407264016647176, + 0.37496154634626266, + 0.2982861915246932, + 0.27302855323158687, + 0.45813268588094236, + 0.7514720562330949, + 0.49986398087145095, + 0.569442312345614, + 0.5690465618581382, + 0.7544852546631636, + 0.505647990301987, + 0.2499671204133375, + 0.5544976813624711, + 0.3162460553686853, + 0.47016977361121687, + 0.3698908056675137, + 0.3259693570697146, + 0.4538288944502046, + 0.7604024927403812, + 0.20736913366385648, + 0.42307565994165763, + 0.4213363706857411, + 0.5971548593378089, + 0.35531753416314854, + 0.20234902245525524, + 0.35572691575174487, + 0.7068824375153333, + 0.32979043362537436, + 0.3858384983698464, + 0.4786085498632732, + 0.4101266549772203, + 0.3576450977265701, + 0.4357092476739072, + 0.5842836850884253, + 0.42588912620521374, + 0.3465696384788862, + 0.48931752731449274, + 0.41379948009140444, + 0.5376701656288129, + 0.45955181260475564, + 0.47075842106524174, + 0.5002150681596358, + 0.23624325527721657, + 0.2075963261006962, + 0.3904759294085451, + 0.2578412912628354, + 0.6782525599244604, + 0.237401199482277, + 0.5373255912321055, + 0.46313588463048194, + 0.6660246485595159, + 0.24396696712453833, + 0.2586862084849346, + 0.25861035537711424, + 0.2421993354414269, + 0.23078544588860606, + 0.21430109513269896, + 0.5878364299901442, + 0.2749730651119398, + 0.6642244521576793, + 0.5522128685273249, + 0.3657315779966556, + 0.5083359074986078, + 0.34152485144914246, + 0.46084988378651004, + 0.4646409629954072, + 0.4315700132413203, + 0.3295150190948383, + 0.42174751858562975, + 0.3893920713771147, + 0.4850011290109179, + 0.34283820869088044, + 0.24606856924552503, + 0.5736250405264319, + 0.41966575479218526, + 0.6400547958521636, + 0.5790557094246808, + 0.45890136501181594, + 0.2508966833100447, + 0.25762472107371626, + 0.2820533300549754, + 0.3495820277295755, + 0.733574445725315, + 0.42690954472826936, + 0.31048828754262653, + 0.7196561997957187, + 0.3588036371896697, + 0.5980612151929305, + 0.5004241466152364, + 0.33942494981785065, + 0.4421792435054773, + 0.38495168606147734, + 0.3870403192221798, + 0.5595403189841985, + 0.4852004885351695, + 0.4757112893516709, + 0.34371484312854, + 0.39034032982011996, + 0.33666383971422187, + 0.3950571487079336, + 0.7514633999489825, + 0.4179521296325508, + 0.4197896605311326, + 0.37411037522494195, + 0.21532851636269892, + 0.6506010574261583, + 0.3579911506020503, + 0.2441179261080642, + 0.4805947396880102, + 0.33430197472831547, + 0.24885968425853477, + 0.4620079386628156, + 0.3676440488867963, + 0.7607797897381351, + 0.2843393299485738, + 0.31062535316884377, + 0.24626398067172453, + 0.38923365443695307, + 0.31541579370060996, + 0.23304364678618794, + 0.30941882177021146, + 0.21854086695140165, + 0.34389011046025975, + 0.7199685400620145, + 0.23170990928957885, + 0.3608937794733225, + 0.46020316582353715, + 0.3670400570833297, + 0.3295514514203539, + 0.43550088302279194, + 0.33218022074805487, + 0.47715178077541764, + 0.4595915580883572, + 0.1547355529826961, + 0.5990702961162322, + 0.6523326991451418, + 0.45215882434752197, + 0.4217616401396936, + 0.4889580779335189, + 0.32419135015954914, + 0.3017504317150771, + 0.3981294472301202, + 0.7610997029572355, + 0.27166553500114243, + 0.5783186664899526, + 0.25344077822220246, + 0.41793007702569224, + 0.40710435108229986, + 0.41402683655538347, + 0.3189497031329361, + 0.20368803593368798, + 0.25025134649871494, + 0.2560702484636093, + 0.4684246593418133, + 0.3741499550525201, + 0.19142978592050489, + 0.35616682411873024, + 0.40778697807354614, + 0.23133198442359507, + 0.5094917675193085, + 0.36806040980985516, + 0.2991553283614306, + 0.39571687752650714, + 0.2192801638070743, + 0.37523639179734386, + 0.3820225848956433, + 0.5484075742323696, + 0.41218865438192337, + 0.765216282499321, + 0.2311157875820788, + 0.5286883856055992, + 0.7222838732046093, + 0.6112119195847519, + 0.1741007340193953, + 0.4247009948181881, + 0.38388024299749385, + 0.25968621124849417, + 0.28416694666887077, + 0.2759020683047752, + 0.40226499966868867, + 0.46347635630331474, + 0.285293349156511, + 0.5835431309594394, + 0.7433662003574371, + 0.3724414857018654, + 0.32138549092516483, + 0.3723686810272882, + 0.4130213240250631, + 0.443876100891244, + 0.46234197379894604, + 0.44764182366243005, + 0.23449025619232303, + 0.6140656830428848, + 0.43178240647279736, + 0.5121381836625698, + 0.6857141959281503, + 0.2516950496311891, + 0.5666743918572801, + 0.34375686704693575, + 0.38636610829295653, + 0.31196817571200425, + 0.46887833762671166, + 0.3996808087864588, + 0.2601712833188833, + 0.3199731879941164, + 0.38161970956248603, + 0.5223763599464406, + 0.5496776997722321, + 0.36438188578419367, + 0.3900348311317513, + 0.35778372946801096, + 0.7150912072340079, + 0.33500792399409574, + 0.4281104302829602, + 0.4102391118572346, + 0.42864921495093344, + 0.5645593845434204, + 0.20806184438896239, + 0.5376324839453901, + 0.3082230755437918, + 0.48858497789814687, + 0.3608441608547429, + 0.251434862929209, + 0.25489017758351296, + 0.4063934010314237, + 0.24142957259307096, + 0.4660519511223867, + 0.7052017773574277, + 0.7105676436361289, + 0.3104967641185964, + 0.24407823863376207, + 0.6793420023081937, + 0.4029954263595172, + 0.3225208809947289, + 0.45037014448743345, + 0.31657007036517, + 0.5649372019506219, + 0.40409822413047897, + 0.5298557091609908, + 0.5452610340998364, + 0.3249933482278543, + 0.46284065163220417, + 0.6312661709640282, + 0.45569019501353086, + 0.27942362793556286, + 0.33653945982295647, + 0.23130910611083072, + 0.4988735969297002, + 0.2883530453284474, + 0.4660774060414859, + 0.4536481803083875, + 0.2727691518039384, + 0.7012528844750182, + 0.25854974920555474, + 0.7052337324223124, + 0.3570195094747355, + 0.4942889237492147, + 0.7265869632363433, + 0.41355826054887296, + 0.5154656175536263, + 0.2627687417538505, + 0.6823772504120978, + 0.3267268096183758, + 0.2300569369654212, + 0.5052357111150394, + 0.47524265223803486, + 0.6109954476183926, + 0.5203560640538167, + 0.31072049480631664, + 0.434482705719762, + 0.20706948265053754, + 0.35971949456119945, + 0.5411317240805825, + 0.46114975628704746, + 0.27103967189152256, + 0.487188104407859, + 0.5478978606189799, + 0.5443870819752106, + 0.3882845944629728, + 0.2440768121000574, + 0.47493620394135067, + 0.4170168830085711, + 0.5268662454337044, + 0.4040878766929246, + 0.2633000983279371, + 0.3403347837448227, + 0.4499710851642015, + 0.3365077157443439, + 0.7098526315794417, + 0.46842740798108584, + 0.40539490054584604, + 0.4465153745220625, + 0.3250658591470602, + 0.3835162516535359, + 0.5101378389267005, + 0.5193059922466093, + 0.6942847159152924, + 0.49574629762626377, + 0.5513947487763665, + 0.4813662632016046, + 0.4164555697211996, + 0.5456554717334713, + 0.31590911993452153, + 0.5925378915363857, + 0.5691036650586155, + 0.7122853432418919, + 0.4655937986981408, + 0.5144813787874076, + 0.510090122889373, + 0.3173798175557842, + 0.5101815842224237, + 0.2525756556002707, + 0.6507719030112672, + 0.4241318353341681, + 0.7184387470622334, + 0.7393650820848614, + 0.3631309199757274, + 0.3978445593334518, + 0.45374253677209825, + 0.26474842124131254, + 0.4311501415207122, + 0.7405638702177568, + 0.45754951284959117, + 0.4546467719494046, + 0.555733871859977, + 0.44444640740921465, + 0.27661219966253325, + 0.5798613690285646, + 0.37687176579626475, + 0.3219335612233297, + 0.7188768506095619, + 0.3419050599758866, + 0.3548419055728725, + 0.3244054042200332, + 0.44982055353728473, + 0.5314856111343385, + 0.58216585245083, + 0.6563227911844007, + 0.5994255358654501, + 0.5628518133860972, + 0.3927969151512086, + 0.5826011682446383, + 0.5128143879371558, + 0.7658491005487835, + 0.7196901070141587, + 0.24751137163891193, + 0.28684278278687586, + 0.234858961359655, + 0.46598424565809127, + 0.3649800433489822, + 0.7724054668147958, + 0.313800440001297, + 0.365138548984094, + 0.5186135281167668, + 0.43571671451410404, + 0.5019909032088141, + 0.5955146147163812, + 0.4034181484021365, + 0.6015923909879987, + 0.3182417521639567, + 0.6522672381817181, + 0.7641134801083025, + 0.749030652377317, + 0.5561304506925238, + 0.3758158621323106, + 0.49661134571807547, + 0.7195626596822822, + 0.6658638518643839, + 0.5975099266441531, + 0.4227204675809448, + 0.3867180600734786, + 0.4571855167977552, + 0.7199587374606786, + 0.3622340343463004, + 0.36530510001560146, + 0.2540983052695302, + 0.5667301179957732, + 0.22673957092205468, + 0.32129254906912014, + 0.6282985831910676, + 0.5151962342403341, + 0.6652007228319367, + 0.4867211189949589, + 0.47175374196882863, + 0.6453777550605095, + 0.6097486118369962, + 0.5380493251939654, + 0.33553798142529784, + 0.6203184496363455, + 0.4799771789848933, + 0.35094179295847033, + 0.28272134579003055, + 0.4472465439554263, + 0.6140904003499805, + 0.4511161398953027, + 0.5822641266387751, + 0.533416675863123, + 0.3768249882123123, + 0.3493168843098487, + 0.6655174515378048, + 0.44406203897433866, + 0.5534175329931768, + 0.7078214522146168, + 0.37069065919799155, + 0.2896713215497781, + 0.28302531133743525, + 0.27803148224952295, + 0.6951575830764959, + 0.7197993588189363, + 0.5348266522724829, + 0.375518861971335, + 0.3952452219598125, + 0.3227291553629372, + 0.767484400949806, + 0.28738648035533143, + 0.4386137201859171, + 0.27177518441549037, + 0.4572901302320857, + 0.29583519233928185, + 0.5304399853574819, + 0.4121431717515912, + 0.22774730001766133, + 0.6708275094254347, + 0.4674918660128971, + 0.35577011716613816, + 0.3001935381693491, + 0.3092947863919248, + 0.3532393027928146, + 0.3660389074129678, + 0.32243957885145624, + 0.32002243558209986, + 0.3590491905037309, + 0.3251567561732045, + 0.35773083490209684, + 0.2825116509329533, + 0.38560605854737123, + 0.1819843368032579, + 0.19247291496177082, + 0.22514710879168823, + 0.3457856073964063, + 0.34887190293438225, + 0.3429814123030942, + 0.22416012856371578, + 0.2198537702401259, + 0.21667985755942676, + 0.6048950153712426, + 0.23962162594767308, + 0.20507131321825847, + 0.7595602574535322, + 0.30460278658596734, + 0.33388933844753466, + 0.26414996189320256, + 0.40844676731713103, + 0.22944866122440402, + 0.3257367486336916, + 0.299981738876954, + 0.39061317595646394, + 0.18327978661098138, + 0.36596446686017914, + 0.35822326970791074, + 0.4508113751166983, + 0.292988702253214, + 0.43872530041689406, + 0.2552844848178471, + 0.3448646036249608, + 0.3167312868897407, + 0.2836584221478152, + 0.43236309376270204, + 0.27374665046826246, + 0.4384517095897985, + 0.3005085627220206, + 0.2866858508716812, + 0.7972503361583261, + 0.20337577156504516, + 0.3283510052462013, + 0.31351171303798164, + 0.33068049356121204, + 0.23775839029821344, + 0.2745036603095118, + 0.4234708790601137, + 0.5124345885252063, + 0.3207162523254542, + 0.3385859287707956, + 0.19182982435494778, + 0.3684095056166137, + 0.3234359496713354, + 0.3464007856708311, + 0.3759028591546345, + 0.31060667486376803, + 0.2710254151601476, + 0.4271220215736897, + 0.3080246141413608, + 0.4035477875631576, + 0.18514087840664076, + 0.2765172043882844, + 0.4074626333926181, + 0.42428737774187947, + 0.337451315526277, + 0.2848926378696535, + 0.2701015661031775, + 0.8087137356035442, + 0.265107112854556, + 0.2460256962054871, + 0.29404881313098924, + 0.19589793273606187, + 0.23525435336614198, + 0.2440680095263587, + 0.17375972344388702, + 0.307660962800022, + 0.3262730027431138, + 0.28963758525529065, + 0.2430583992083702, + 0.3696547757610292, + 0.45357491194422067, + 0.20325081066570816, + 0.31225220011746124, + 0.5157111325974654, + 0.226888978267082, + 0.26865047417978094, + 0.4554359056297132, + 0.4004894045462279, + 0.22965616756055715, + 0.1977910010971038, + 0.23783400056598436, + 0.34799828118766724, + 0.40866639191537085, + 0.1896090237910511, + 0.2842244380767044, + 0.2620934956439851, + 0.34967200268236853, + 0.25860590284519197, + 0.17866786056274597, + 0.4160561992553546, + 0.4184428434548725, + 0.16862914108956592, + 0.3987933620840432, + 0.17853018151692865, + 0.24150251688662014, + 0.2189576551337448, + 0.7669798699722239, + 0.38722280247309876, + 0.3767228715134438, + 0.16977186968144029, + 0.3011833638382645, + 0.2732774268451316, + 0.6723352630672181, + 0.32531120326067375, + 0.257658810949493, + 0.40621983549865825, + 0.3542424182809436, + 0.33235856698746813, + 0.3806658031095602, + 0.43282774518841083, + 0.20341162402594823, + 0.22524795593695648, + 0.39547211341943417, + 0.31176013978505057, + 0.33391427029239673, + 0.30917561065415894, + 0.24093725917059577, + 0.1938436958132043, + 0.19479394645092168, + 0.3301528081205292, + 0.2895486720741964, + 0.5716409002992088, + 0.228250803390257, + 0.1955663280638364, + 0.21231143600848842, + 0.45345401770678506, + 0.2615791992119083, + 0.1607334166077325, + 0.21120837625377717, + 0.26441583143600833, + 0.28993684736838754, + 0.18532111190582054, + 0.49082025175409694, + 0.34440464072127835, + 0.20997310335890523, + 0.28919997891228716, + 0.3581979657450814, + 0.2681333159617006, + 0.2755729026269338, + 0.4401553986964813, + 0.2429784634322976, + 0.6240762041860991, + 0.3233477455508727, + 0.4287138809914076, + 0.3194423583149055, + 0.3129909698344175, + 0.3194887395919418, + 0.22780311031343525, + 0.1881529055345112, + 0.33838335925140745, + 0.5591330430852578, + 0.2563371605858244, + 0.16465920660162117, + 0.4504550799934871, + 0.3645223076012088, + 0.3636970628583224, + 0.5583565923067165, + 0.2548840876212227, + 0.3353785550172912, + 0.2016464734307399, + 0.3039252494885436, + 0.352729302433324, + 0.2101015497100214, + 0.4006999138728002, + 0.3373298139617666, + 0.3224488196053553, + 0.14577204539613545, + 0.31774322145253714, + 0.19462880792355702, + 0.21130797895187947, + 0.21908930060984452, + 0.13256148963491599, + 0.36802763200639454, + 0.4719804928583583, + 0.38239746292300053, + 0.41795462156901947, + 0.35469524395694335, + 0.1702817505989871, + 0.25598608683923507, + 0.3920131412034653, + 0.29813096999081284, + 0.32330109692789366, + 0.38936559955458094, + 0.30110276463792224, + 0.27613279785559747, + 0.46951067800512414, + 0.32137051082062673, + 0.23724077106493244, + 0.4623548454277142, + 0.24906477272518457, + 0.36729918485939966, + 0.2188527972720173, + 0.4110019928658122, + 0.5708103330812774, + 0.7749303591886911, + 0.35806906604028604, + 0.30135333555263666, + 0.3667611656153252, + 0.29750891041370864, + 0.29269843550126745, + 0.2796995642991855, + 0.3245081554351843, + 0.24565390798041067, + 0.18793768002729633, + 0.39290689671883894, + 0.23820520106459503, + 0.2594651723510979, + 0.19310625108124277, + 0.15593235076261514, + 0.33070115180132276, + 0.3408043420356443, + 0.3072979290278456, + 0.3039946662508196, + 0.3347865650086959, + 0.20227136674307913, + 0.13267408721789223, + 0.24607925737078093, + 0.3858502967764921, + 0.2086315505642643, + 0.33220180598876625, + 0.2408955335216578, + 0.2065777237055741, + 0.13637905248416715, + 0.451201428951066, + 0.18827646784094793, + 0.2239027030359999, + 0.3192057219979729, + 0.2640927403373323, + 0.2362931608672019, + 0.5589804048801202, + 0.3534753301673819, + 0.15787502807067474, + 0.3779363997356756, + 0.324259823619128, + 0.30285134832107047, + 0.2308766872978665, + 0.2039012747738416, + 0.24358191369031626, + 0.4046403061059753, + 0.1489764242733816, + 0.33793139463789845, + 0.14195457588322596, + 0.23993981105123235, + 0.26484518556276737, + 0.17080381342687936, + 0.3522596912752966, + 0.34859988692808713, + 0.23750013658318325, + 0.30673899939056987, + 0.3079727970313154, + 0.18633724231345544, + 0.30922955475638114, + 0.3366536384952241, + 0.4602814127187056, + 0.43453220901517586, + 0.31463903419373757, + 0.3418660373511292, + 0.31167463218118097, + 0.365399093886907, + 0.4086371121471241, + 0.4950168390081574, + 0.3434036564362416, + 0.2189251341089122, + 0.4680101078204892, + 0.2595988689588098, + 0.38413738502572226, + 0.37517014419927003, + 0.24615496140945933, + 0.4420307184991694, + 0.1519012629930153, + 0.15470521565742706, + 0.18763341306320838, + 0.14293840583153222, + 0.3069198548729355, + 0.28422509709841715, + 0.22312484140224695, + 0.3320202827566742, + 0.5064034351649201, + 0.29920884753776916, + 0.46124415193798274, + 0.761419416398278, + 0.4722514639905688, + 0.141262038271857, + 0.3793645257457848, + 0.3785548769430308, + 0.3286131075943226, + 0.17073038900528228, + 0.5388350159721846, + 0.3810742433817045, + 0.23035796763117547, + 0.3947417085775712, + 0.2829984342156634, + 0.39218327792762536, + 0.46176457640099994, + 0.31666652650673727, + 0.48350153140149643, + 0.3576753289902164, + 0.27177725036856726, + 0.42990494765275433, + 0.430935067577077, + 0.17016176855499807, + 0.18186357847637377, + 0.15908614909430252, + 0.5464333342913333, + 0.44344120334534987, + 0.8027271453107567, + 0.4670864288203019, + 0.3114404793642851, + 0.25336394819264574, + 0.43353329062540097, + 0.4142493696575435, + 0.5756681473592533, + 0.2663790161908105, + 0.4665294108010287, + 0.23549830126738236, + 0.2653529164486089, + 0.13480645113920278, + 0.1469598432731199, + 0.3210218569194839, + 0.12446330900601381, + 0.30590634015176676, + 0.4208390931873941, + 0.30042240480707505, + 0.14915388093917936, + 0.1745659385615523, + 0.3471539624190287, + 0.16492699026269572, + 0.40622041569786604, + 0.366404340589939, + 0.35411543646190635, + 0.33995427368024883, + 0.24822448400531766, + 0.5179409823478082, + 0.3981004441860417, + 0.2664734760421621, + 0.5572714240353875, + 0.2453921203371935, + 0.3312952743516471, + 0.25297434351966375, + 0.14473772550129904, + 0.2082338046546444, + 0.2642724312108993, + 0.36661752896227323, + 0.22696727047338258, + 0.6662441565420716, + 0.38284084341799046, + 0.19983126301613025, + 0.25248114658110915, + 0.1692107972480479, + 0.48812095589572807, + 0.19959946054589095, + 0.2584433367791381, + 0.2824346988210421, + 0.2952497056027492, + 0.15995464749805482, + 0.3371147222101234, + 0.22135424543509943, + 0.3679280251505258, + 0.7507399567491453, + 0.32967640590026126, + 0.27203709603824244, + 0.2767021082967714, + 0.28388235773910186, + 0.10792913108040819, + 0.25835574144101603, + 0.3289229152692079, + 0.7307111874364293, + 0.5045571860319207, + 0.3411919484349804, + 0.273159477338689, + 0.2784174090270741, + 0.2937775494185288, + 0.1760995119734915, + 0.35004789954986737, + 0.42641480425798445, + 0.21952939466539448, + 0.32505358411349006, + 0.30172703944485646, + 0.39135881434180836, + 0.31442347321857145, + 0.48367005926131335, + 0.32122541364759216, + 0.23508281060251793, + 0.3778672074129413, + 0.36068445279602906, + 0.49290350448428305, + 0.47153781756154717, + 0.20946202834615413, + 0.6988543702647709, + 0.6924891981204229, + 0.1790833439035082, + 0.22281966168945763, + 0.2962464079072148, + 0.38303908020118455, + 0.1958445286854277, + 0.2944821417474599, + 0.4298884994890377, + 0.4442123203226628, + 0.3949701366988139, + 0.30347537874539476, + 0.27584795849850025, + 0.2873767321443108, + 0.2768837808498725, + 0.3877941399858318, + 0.3370691409375647, + 0.3257280482302164, + 0.4327869315118264, + 0.34278070445337955, + 0.493465026814079, + 0.40151663637439095, + 0.5806562897373565, + 0.4057690675269979, + 0.34756914786608284, + 0.4334200550764225, + 0.3955078168154765, + 0.29019546111127403, + 0.5934450250813368, + 0.5618602938086247, + 0.3860901914294042, + 0.3244629686324915, + 0.3180132571920619, + 0.46947232024071683, + 0.23522366119742313, + 0.32963115825903283, + 0.31908738286896476, + 0.434735287108827, + 0.37460159274539134, + 0.33325766179397937, + 0.28376922890952927, + 0.463086571804689, + 0.43401639615748466, + 0.41114366071131003, + 0.423826860466582, + 0.3155283167755431, + 0.20559632716663778, + 0.12381791493612909, + 0.2232421827087729, + 0.4667092396524798, + 0.4042512751333207, + 0.13003142974138943, + 0.37845981581236543, + 0.2960932747070228, + 0.30651758164615456, + 0.27282213779574976, + 0.32140095535019014, + 0.2651897480716006, + 0.39627729975890014, + 0.5934571818791856, + 0.23726487903441437, + 0.366368426781588, + 0.5050402992133249, + 0.3074875330912043, + 0.3944542812722244, + 0.5450092435420117, + 0.4063551351040068, + 0.4069583978053078, + 0.3366267556873539, + 0.3235649073415287, + 0.18623131634346793, + 0.32995051084923543, + 0.4349381422248452, + 0.45530107503957457, + 0.42183966730000716, + 0.24211420702434555, + 0.4164612479456033, + 0.38212093759030513, + 0.44703534573330533, + 0.378708355756685, + 0.43930626870999057, + 0.3600635702143233, + 0.4111183329724241, + 0.4011607479564389, + 0.3746428678394253, + 0.20008877925477545, + 0.40257433732062287, + 0.4263602135272406, + 0.21433441530928818, + 0.31068198154804955, + 0.4181163597765165, + 0.33233984838623204, + 0.28598152579797564, + 0.45437799358191344, + 0.46676533607826126, + 0.37309841133227206, + 0.39280285368316553, + 0.26578198905053685, + 0.3826817890537515, + 0.30121043401435094, + 0.3121702632522353, + 0.3295149343857568, + 0.3918289848440541, + 0.32354440262963374, + 0.38726269088008713, + 0.37065122696936414, + 0.7028832989400071, + 0.5244045729089474, + 0.47474156613706203, + 0.511592871244369, + 0.4429050195346732, + 0.395406934952951, + 0.517257152941879, + 0.3151838927632771, + 0.3484461243151327, + 0.4425733767802326, + 0.29653401595325585, + 0.41275287092161705, + 0.4690251010491613, + 0.41637777945289106, + 0.4253723363184862, + 0.4174049793827201, + 0.2879816940275261, + 0.45908251496174357, + 0.46879566295464176, + 0.3340549991844644, + 0.38097589580453184, + 0.34522502104359004, + 0.31304414999539626, + 0.3871320159740237, + 0.3941898010500427, + 0.37755463095923525, + 0.4159272649390872, + 0.34713789309835735, + 0.4301624457197151, + 0.34001944279701496, + 0.3645313144219601, + 0.40798202223902785, + 0.4090206423776473, + 0.3841199678154216, + 0.4442106990174219, + 0.36613167614155223, + 0.3182680111070414, + 0.32381510426884813, + 0.3902916156453149, + 0.38018186139026433, + 0.4157643188897808, + 0.4254579749835348, + 0.3920457410299545, + 0.3806719491988647, + 0.32372602118303284, + 0.37350175640063066, + 0.4659285797706179, + 0.3475881154531543, + 0.18581815703218205, + 0.2930913771510924, + 0.3885171364312127, + 0.2710166939606777, + 0.2771082926527546, + 0.4186725303824527, + 0.4197063623870332, + 0.2527244353222533, + 0.2829827212853183, + 0.3085645453928915, + 0.2866166975862248, + 0.4269086196993659, + 0.22329909038687526, + 0.22311656112160028, + 0.2363614240323497, + 0.24233676740689564, + 0.17500392277593985, + 0.23271393475025445, + 0.21057056237202515, + 0.23221410714530255, + 0.20879938853412291, + 0.33336085613910776, + 0.23376642827205102, + 0.28047849714431383, + 0.2885963400057917, + 0.36249915799037735, + 0.24878642599830078, + 0.3471959634416379, + 0.29257548607804357, + 0.2927115627793225, + 0.2941800049687774, + 0.36336522596309007, + 0.32506288602725686, + 0.5240382875911606, + 0.2534889268792121, + 0.33462159525704654, + 0.43815584809750524, + 0.4236544816004041, + 0.3416309135725086, + 0.39062099185150423, + 0.35341585059299685, + 0.3880754681740758, + 0.4291447781157108, + 0.33777220687386067, + 0.541046671254246, + 0.38109588395686095, + 0.3642861124744676, + 0.3110940681899487, + 0.38634136899178184, + 0.3207279391855444, + 0.39142885838512254, + 0.3779520548648185, + 0.3916867492915712, + 0.42870793189766226, + 0.5604863585541684, + 0.37874516297456506, + 0.3415840550158766, + 0.44294195254778956, + 0.5122660697198569, + 0.47124229609004326, + 0.4110675938226276, + 0.456250872568822, + 0.4739292665455846, + 0.3438505082860616, + 0.3397570612647006, + 0.3773552281501464, + 0.3096933347099818, + 0.5207458397543216, + 0.4657699800656456, + 0.3545729853620521, + 0.3236024507330876, + 0.4343268801158952, + 0.370424776459385, + 0.4537669163187805, + 0.4051968342275225, + 0.3138870401557942, + 0.46726550377048004, + 0.23850321658847656, + 0.3557106609095587, + 0.44517447961352846, + 0.2811713696555483, + 0.353949176585692, + 0.26517877863671313, + 0.29727946800498267, + 0.31636951487380155, + 0.330239246384545, + 0.38372275164111125, + 0.2838196168029728, + 0.2700480237059723, + 0.3642500590122027, + 0.33071829384060974, + 0.3284079542495342, + 0.39412020480137167, + 0.30376212549209675, + 0.37307968658182317, + 0.4625807741157777, + 0.5337788899577366, + 0.3919455694504519, + 0.2974455896966319, + 0.43267566199587487, + 0.39344187103276784, + 0.33973267455663286, + 0.3555978511838084, + 0.26210515202861284, + 0.2666193208592378, + 0.23721925238537472, + 0.32193312463027535, + 0.24699480388445183, + 0.6238661423746091, + 0.3955227697239285, + 0.351750798029465, + 0.27545037309424447, + 0.2819640473184799, + 0.2954173398872383, + 0.30425861513745867, + 0.27515206309345885, + 0.2794030085278408, + 0.2692587985996827, + 0.3411710819137018, + 0.30568308227661817, + 0.40593805793361254, + 0.34858034544926386, + 0.30454456386187784, + 0.46006721190542305, + 0.45613269042925986, + 0.31260795197505425, + 0.3591657709648099, + 0.3525906205392218, + 0.3362945183217051, + 0.2949312139055417, + 0.48000978559692664, + 0.4078323342195921, + 0.5515377669894799, + 0.39757458246712063, + 0.3897739387507764, + 0.40863080735913093, + 0.41463088201900755, + 0.29078685843245244, + 0.48404854908112066, + 0.37596555497164397, + 0.21301359778450255, + 0.38351944084702094, + 0.47874116647105547, + 0.313376569671995, + 0.3695725085173783, + 0.7152305483311189, + 0.38159242249141917, + 0.4311616155421005, + 0.2428930084524839, + 0.40183214507394516, + 0.20350354481692493, + 0.19975357677368333, + 0.24393869871390464, + 0.1967654623300893, + 0.40102922223296694, + 0.1396367252028199, + 0.24921976535043672, + 0.3768703040285714, + 0.36108022377365484, + 0.3544900931203441, + 0.3683560742236908, + 0.26599751088550455, + 0.3058199370577074, + 0.4664600879756473, + 0.4801218875542452, + 0.28146017573527, + 0.3889357562311251, + 0.3248580275314019, + 0.7004447055686296, + 0.2997438715347221, + 0.5249727137669752, + 0.7197435513257489, + 0.6957726864152782, + 0.376757776563167, + 0.29542362332221667, + 0.2385173791725022, + 0.3859510148677174, + 0.4265159239022005, + 0.4503718556668207, + 0.5041324240993744, + 0.3648855156833504, + 0.5119272658996364, + 0.3175889321321663, + 0.30778638910730266, + 0.6736034414183364, + 0.6602439018807097, + 0.3671017830972141, + 0.35160706786643736, + 0.3897697289765821, + 0.34984488702547106, + 0.3900148985128119, + 0.6190767034742352, + 0.6270146641790615, + 0.3622185640467356, + 0.37694145996854467, + 0.3787999972870505, + 0.2783043747679959, + 0.4386595589003717, + 0.2485701029827418, + 0.39609364843451267, + 0.37427907058160936, + 0.2947475237470404, + 0.37839682832802224, + 0.29571148197559577, + 0.23358196928351777, + 0.4608397742791646, + 0.33212867177774, + 0.33487603873973615, + 0.30724816396971333, + 0.4166701971509831, + 0.3423092042567635, + 0.2672406766872706, + 0.2927171164279, + 0.20796182645609543, + 0.19096434295751327, + 0.21078745990993739, + 0.33097528515530716, + 0.349727746227323, + 0.2276205614461423, + 0.2957369093032005, + 0.21843677919296092, + 0.35960550720359735, + 0.33241454392950365, + 0.4230666735448857, + 0.25403656423862536, + 0.3307514347429417, + 0.42886365664916637, + 0.26097151655235384, + 0.42215500395881606, + 0.3605578652536481, + 0.3385361033475377, + 0.35410275682584275, + 0.27612813888714693, + 0.17094043608595885, + 0.21553953907739853, + 0.2183311940001192, + 0.4314538714126609, + 0.34158336906507736, + 0.37908323745001893, + 0.33426783400318183, + 0.35970137013667475, + 0.2755974109082173, + 0.2973599590733853, + 0.28862965396488616, + 0.629195225702457, + 0.3668175785022197, + 0.4059336363685931, + 0.2933537210790512, + 0.3848450483247846, + 0.4110093621295449, + 0.34743168469882085, + 0.32501026693859114, + 0.39408712363246085, + 0.36425789267680697, + 0.4574836261962677, + 0.3340803466222192, + 0.39196755841622233, + 0.3563562790627538, + 0.3455723092117635, + 0.3143244876443409, + 0.39129570642781186, + 0.33028848161340185, + 0.39883508097744086, + 0.2507185449138907, + 0.23885580733577846, + 0.24109333961324306, + 0.31314533225693, + 0.3680291798405156, + 0.30718371465758376, + 0.3194497529085334, + 0.3775580214339566, + 0.34705182939234397, + 0.4001006644034775, + 0.3676830538869277, + 0.23983534797640593, + 0.3381588412729586, + 0.20463345128642207, + 0.34317662773206375, + 0.17104548432732222, + 0.6262539620841252, + 0.26592805931159186, + 0.3286842413532772, + 0.46505274532321944, + 0.3850716554039118, + 0.21503971497805893, + 0.3220040434124037, + 0.26324181697694665, + 0.7531972339458506, + 0.22799436369064174, + 0.417539885677871, + 0.44968224699706044, + 0.433698905100008, + 0.6128020147006036, + 0.3004938030237645, + 0.4328474340604189, + 0.38348429598190564, + 0.28472410575138024, + 0.22841156688859243, + 0.38582324985156663, + 0.22648578564062574, + 0.6903184860477919 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.352559395518743, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.36629062864173684, + 0.7316013908668074, + 0.7179632002613122, + 0.7970644700487808, + 0.6927585554625931, + 0.5944043897206508, + 0.561421567844021, + 0.6204717787770606, + 0.3769664508310355, + 0.647338926147443, + 0.250298819301485, + 0.40384974358243597, + 0.2671500054650273, + 0.2632789550966217, + 0.23743755476391043, + 0.23851787513496311, + 0.2783376644113127, + 0.4420937740936593, + 0.3117803181316562, + 0.3239360247116592, + 0.2754492859504329, + 0.6448374708631298, + 0.2543844031844192, + 0.5681177940168167, + 0.31441523335963695, + 0.11448843123639596, + 0.2927674381867137, + 0.3586802245677119, + 0.35321312658837833, + 0.4225322375239587, + 0.24089848731647143, + 0.34324686990591396, + 0.2856943580994963, + 0.30341769302555965, + 0.35908053731825323, + 0.2930579112624999, + 0.247538337111022, + 0.4276942730936061, + 0.4318697153536675, + 0.6046589104546188, + 0.7646929968347111, + 0.7584400232960389, + 0.7469198737839043, + 0.618950257390443, + 0.32805488774265423, + 0.2804893768306813, + 0.7561030214470476, + 0.2731499380517031, + 0.25028584366250767, + 0.3831451380862664, + 0.28222253826591287, + 0.7541964618608558, + 0.7734961034363035, + 0.3288165123507215, + 0.25080151585481114, + 0.2984674031818475, + 0.257355209690155, + 0.32029781223811105, + 0.30402342337740307, + 0.3952048805548804, + 0.33716327892917647, + 0.34809494118253287, + 0.2974479205622664, + 0.39888462293202426, + 0.3232133340418648, + 0.28496425580356316, + 0.30153406302042846, + 0.4203898811302008, + 0.28900082379186876, + 0.3473945168785042, + 0.38330208739007066, + 0.32525083131359567, + 0.3049258014620504, + 0.29455376662687127, + 0.3218281446558546, + 0.2937961260825502, + 0.21969209620865163, + 0.22569344382067014, + 0.3388177316451495, + 0.27938848730954213, + 0.3660016173558387, + 0.26251273484782534, + 0.30474166647689244, + 0.23668708758334658, + 0.3388765329051456, + 0.4115062951865901, + 0.3285576857899993, + 0.21208718841775445, + 0.30755027263752577, + 0.2578044043621502, + 0.26521894344325064, + 0.2704521041580359, + 0.3785221460271985, + 0.4475041350316156, + 0.2083737169004958, + 0.46203456192390074, + 0.2694970623958501, + 0.41238347459457414, + 0.22404612989885603, + 0.34812380751335503, + 0.3548605563264817, + 0.42262951950118216, + 0.19023609532455746, + 0.34370178006231783, + 0.25255953481598437, + 0.2755307978459698, + 0.35307272607678, + 0.26723433474883634, + 0.2814382781019253, + 0.7856965955310375, + 0.779341007972849, + 0.7848712711211581, + 0.8010301221017408, + 0.5625241430262851, + 0.667749586725831, + 0.6841838304661298, + 0.7347160660125367, + 0.6703330021227172, + 0.5666403041454635, + 0.6349397354176173, + 0.557035676794621, + 0.7487522253997871, + 0.6406738832181199, + 0.37738118397105774, + 0.2941728644664008, + 0.2371472267740975, + 0.27729128701137795, + 0.44274704202333226, + 0.2506890225061245, + 0.23982816294449427, + 0.2900999699566197, + 0.3418320092158728, + 0.23772198001187206, + 0.3262035467419456, + 0.34019009262883304, + 0.3537521602558632, + 0.301277204623485, + 0.26737004516974416, + 0.29757246664855375, + 0.3128887708521389, + 0.25591600350466936, + 0.2731753833264643, + 0.2140823719703151, + 0.5720364391580425, + 0.25455833193822214, + 0.4990340713084567, + 0.3015452647103513, + 0.39723330924271094, + 0.20029695769251296, + 0.5346611076979312, + 0.23976675125574687, + 0.4394045233664376, + 0.2639218241598074, + 0.20562590960168298, + 0.2778087848205708, + 0.25938352911624135, + 0.2155813289132319, + 0.1303535184071216, + 0.2659784142071175, + 0.2785854961201844, + 0.40052126255398324, + 0.3477030851027322, + 0.4046974795241169, + 0.3840693533820968, + 0.29035773526304, + 0.36664072016789023, + 0.31494228564777554, + 0.3450435056288105, + 0.3281498661813067, + 0.2962801941363404, + 0.21143052701634085, + 0.13891525942094876, + 0.33906875295508443, + 0.2349114962809757, + 0.20006217202106466, + 0.2087377999105369, + 0.3479669131098725, + 0.39853422842043484, + 0.529823960777625, + 0.3349566636201588, + 0.19483330779776056, + 0.37169216910249653, + 0.374332238518509, + 0.4008470031962531, + 0.23433012464900846, + 0.41490309829996397, + 0.3331697021542826, + 0.4001613832763909, + 0.19666420428287387, + 0.29397500371906005, + 0.27346858672090446, + 0.20596153211178547, + 0.40084150053363776, + 0.3355217929758427, + 0.429378744615427, + 0.24957504356862859, + 0.5739816104389409, + 0.25185471419284877, + 0.17081281099898357, + 0.3637452885230618, + 0.2383853851062424, + 0.22197563621701105, + 0.16861281050819976, + 0.28848451954882415, + 0.579889328061551, + 0.19775541072363706, + 0.20653246510560075, + 0.3696793280197371, + 0.2755182182093906, + 0.3350759026060387, + 0.38168371628160763, + 0.28081013864361054, + 0.201740433125383, + 0.283954344897424, + 0.2793210606558612, + 0.41799667559657505, + 0.41250771501836736, + 0.4694883164911268, + 0.42562512733220287, + 0.23272006156066388, + 0.2891103663897236, + 0.38720389739985084, + 0.22143254070167975, + 0.29385035823541966, + 0.25099347414919654, + 0.13207367547500418, + 0.4030177980790768, + 0.2855125006103151, + 0.21268962406340636, + 0.472572697595076, + 0.202456400772721, + 0.14101740417669248, + 0.34488859224318585, + 0.34383356251632025, + 0.24101648963900796, + 0.49875839948877393, + 0.2783765752214343, + 0.262988926336143, + 0.20960303853474824, + 0.3845589863492821, + 0.2480724083726478, + 0.40802375322441964, + 0.24166499914463252, + 0.32334659046184344, + 0.3260640063468774, + 0.3307614912766334, + 0.2859584187907189, + 0.2771394713809101, + 0.2904422205074027, + 0.2871761553460043, + 0.40247991777964454, + 0.23855768077759312, + 0.3351114504836701, + 0.2545374106241418, + 0.2798860334553774, + 0.31772155456187057, + 0.33403144985235, + 0.29841079128793424, + 0.3617486168812424, + 0.2249082113416926, + 0.25770544067988, + 0.4323226843437381, + 0.226669519306973, + 0.2867153591860957, + 0.2636003368925685, + 0.20698815328013126, + 0.2549277226876691, + 0.23491230395232612, + 0.22687710372335204, + 0.23510447445859878, + 0.21327281878128823, + 0.26385800014418176, + 0.2630664207342478, + 0.312701810342033, + 0.32120043257563113, + 0.3975780039213577, + 0.1976219000345652, + 0.3451667562696652, + 0.42936000666595364, + 0.22293465803302615, + 0.21925211033201128, + 0.4644957007693398, + 0.19871046336479403, + 0.20490191661109056, + 0.32976999599614837, + 0.31210426046713596, + 0.34732126087461634, + 0.26379476298288856, + 0.39757622936758824, + 0.3205911762580085, + 0.3072221610127578, + 0.26009735712687326, + 0.33206774110247195, + 0.3398828659549598, + 0.3119807048289953, + 0.3481821760143691, + 0.41766169317604285, + 0.38051755543837573, + 0.6165100639220954, + 0.2526477714493638, + 0.5132394651007749, + 0.6059286616155873, + 0.6061767514989326, + 0.6045107043127795, + 0.4261194200885856, + 0.41552564685551757, + 0.3219870117919307, + 0.6454533903085214, + 0.3812543250747423, + 0.42403525664055325, + 0.24108654273481578, + 0.4052685457755035, + 0.3363789478915128, + 0.2915744798730479, + 0.33213154607249307, + 0.26291448482501706, + 0.37896260986440256, + 0.2376676952850595, + 0.27626446884985634, + 0.29614755069758253, + 0.3072045449245335, + 0.34833855565594307, + 0.2286328193234524, + 0.27628742843571114, + 0.3017950057191177, + 0.31961117358604885, + 0.26651515712526275, + 0.310495736907426, + 0.47224365374918414, + 0.34716286941584446, + 0.3001601541850633, + 0.32263594935977596, + 0.35213046437944034, + 0.24534273090590691, + 0.2735577755327546, + 0.24872502704217667, + 0.33333090016693784, + 0.2652826282695996, + 0.3525701391480226, + 0.43817449335689335, + 0.4280374010610559, + 0.31319543678637457, + 0.2578391020018306, + 0.3777986040481426, + 0.22249684050618793, + 0.275818389145443, + 0.3758666038400189, + 0.2653551825728106, + 0.7981153338886705, + 0.46242432769065456, + 0.36329028039525174, + 0.33944442907624517, + 0.39784803454195034, + 0.4076739187354931, + 0.3334899107739403, + 0.44500850515624374, + 0.2982364498438842, + 0.43406484236630544, + 0.3389118598639568, + 0.4135259463924025, + 0.4056167986094416, + 0.3918488594613867, + 0.3701063187527096, + 0.28916075711363975, + 0.29412635509629487, + 0.4452672447237841, + 0.47484644438891654, + 0.35740265173300373, + 0.5311219171115301, + 0.3631142885909491, + 0.4937159754267374, + 0.33821008245679457, + 0.36643749406369813, + 0.38692328937293985, + 0.4576983209334986, + 0.445006346198674, + 0.389412722190785, + 0.43223275653532217, + 0.40989685955563965, + 0.41906928269655314, + 0.3052317440844491, + 0.2999991110125969, + 0.28495333697628944, + 0.19396195467243219, + 0.29611915479621803, + 0.281219836024802, + 0.2471612000117066, + 0.24772452004368184, + 0.344489558210405, + 0.25789083946745733, + 0.24912001670084769, + 0.2729124908415793, + 0.4330496661617938, + 0.4825117401187068, + 0.42464542750408024, + 0.4568949745619066, + 0.36732985708601534, + 0.45544750464833395, + 0.29235127666144384, + 0.5785574522756761, + 0.37200648101503186, + 0.4365901843072475, + 0.34919221711678233, + 0.44981817815536546, + 0.43215644225390104, + 0.39682585450922864, + 0.3560930530535173, + 0.38335294964818795, + 0.38993989806707696, + 0.4096873151042715, + 0.22200782281201326, + 0.2846617280528562, + 0.2572902821637186, + 0.27325481765420295, + 0.41107827082058707, + 0.4565615579607734, + 0.23695626133642136, + 0.4642154756226303, + 0.37469661600925175, + 0.3581579540043524, + 0.2969076998216083, + 0.35949760444645446, + 0.3589091000941917, + 0.2523140266732751, + 0.26907398184861414, + 0.8239337957443074, + 0.3382743705769631, + 0.33646537701753726, + 0.37954547683277806, + 0.36529069731415564, + 0.3677834500872214, + 0.3359545108409, + 0.48923524551177394, + 0.3776679031427387, + 0.2841263795769821, + 0.33389559250860745, + 0.358224256588027, + 0.36801648928939446, + 0.34905828521199705, + 0.3500889235453269, + 0.4099299670768068, + 0.4016480707946923, + 0.23798019904785167, + 0.3949788239898603, + 0.24415781304787207, + 0.3232543289249381, + 0.18668145607380499, + 0.3853328625830226, + 0.2964994803756505, + 0.23601820690098652, + 0.36594492680628216, + 0.16642799807541997, + 0.21880257712865742, + 0.47720812045181127, + 0.43720086927142204, + 0.37465468486999515, + 0.3704611471934616, + 0.3352832246893677, + 0.349935600130519, + 0.4892790235560481, + 0.2985299184187114, + 0.5148592302650287, + 0.27019338690476075, + 0.22190734339479395, + 0.17208906595000978, + 0.3249060998479363, + 0.4012609819975288, + 0.4095484202009503, + 0.44226771113538726, + 0.3467792712335919, + 0.3846198096301066, + 0.2803109056443852, + 0.37129550001544354, + 0.3085777062620411, + 0.21682077803779062, + 0.2676200032438107, + 0.28070640959911003, + 0.3443549680140643, + 0.26153166568272296, + 0.28018360122458263, + 0.2957987940084932, + 0.30853040822053723, + 0.40630800345922546, + 0.32889865743184427, + 0.2574828456459174, + 0.4081162169685708, + 0.3493222127738676, + 0.4458800906201543, + 0.37863925763883693, + 0.47403998136241277, + 0.4874583911916353, + 0.6374070619663859, + 0.6202813676772101, + 0.43716681257354634, + 0.48316099547838354, + 0.36035044050488957, + 0.484597870231657, + 0.3552802709751935, + 0.3620719336089774, + 0.3196156907882365, + 0.5351055009528223, + 0.3303715759900732, + 0.47202745140568797, + 0.6155145066683075, + 0.5951508317410314, + 0.5323175715940559, + 0.6153342327451891, + 0.5021548404125155, + 0.42507372259450993, + 0.3669871598346392, + 0.32582077318057534, + 0.22764575229801584, + 0.28164067239932206, + 0.3875608496148256, + 0.4408333641244913, + 0.2947385800317466, + 0.2322487884129826, + 0.1591526865872648, + 0.3455795555070984, + 0.32754096446973713, + 0.3504746134173553, + 0.25263536593113284, + 0.4666856502086036, + 0.3366419310405578, + 0.38792182394210206, + 0.27529950674650705, + 0.3379610459290783, + 0.4221718920442834, + 0.260967861725663, + 0.28322209306522805, + 0.3371815315390847, + 0.28244900555858043, + 0.2783040666432296, + 0.26447849127794637, + 0.2773743961280985, + 0.27567056011461843, + 0.29665085282223547, + 0.30135825582921744, + 0.3254642004394761, + 0.20497451939481845, + 0.47407518944515803, + 0.3355996273471617, + 0.3993747674847591, + 0.41446084885930085, + 0.29235950240356134, + 0.37666193256812086, + 0.37042012195154916, + 0.3925067397334782, + 0.49062864477930473, + 0.31207420346728787, + 0.3492385129758697, + 0.38912556837001333, + 0.2592575486463234, + 0.3631653825806738, + 0.3329369357544677, + 0.2521268189539668, + 0.3671235950682948, + 0.3750096634706674, + 0.3391395262535734, + 0.22652124711908994, + 0.3069908286817063, + 0.17836529890197828, + 0.2830460117842632, + 0.5052292631444385, + 0.4298621236683947, + 0.45570401758535084, + 0.4392818615968832, + 0.5263489147985608, + 0.4377689979332688, + 0.4750408084491828, + 0.4295889360731599, + 0.4782160792976924, + 0.31189485291608693, + 0.28974235035839196, + 0.3284893981770256, + 0.25882958395475797, + 0.22581677019024549, + 0.3009689972287804, + 0.29604111952354634, + 0.26419859739105367, + 0.2191322097891563, + 0.29486101466653564, + 0.22835332761026492, + 0.3603919432897884, + 0.2566759780660859, + 0.4560295161182336, + 0.4656282092490403, + 0.481962894077874, + 0.23748189043202952, + 0.31671299296380306, + 0.5423030195425095, + 0.251388056775152, + 0.19614334658218877, + 0.21347375156621134, + 0.807191667136356, + 0.4447562985253446, + 0.3513918321789796, + 0.4621737356461, + 0.409119338062679, + 0.4148433007757541, + 0.34710233596882756, + 0.3819130072161786, + 0.41536391738028045, + 0.37273536278839775, + 0.4589453436292879, + 0.42609125897838757, + 0.19242556212037787, + 0.28224553452178813, + 0.2069938261620358, + 0.2766762680610375, + 0.22014230430023424, + 0.41223890985979045, + 0.3455471230484393, + 0.26990893805281635, + 0.37376537487838274, + 0.40729375489650094, + 0.2761412304050554, + 0.3663652901187348, + 0.22047141520860655, + 0.36998402027329924, + 0.41752090394425484, + 0.26794310183637393, + 0.27332089469096965, + 0.21665861580931145, + 0.4336736155628054, + 0.21228921685784077, + 0.27085723773718057, + 0.44140227286090156, + 0.35244914696857965, + 0.3703233711904024, + 0.2632459743567765, + 0.2893690248691715, + 0.3341231785623575, + 0.30469912145355055, + 0.3859699174299929, + 0.25956707107569577, + 0.40666308609991786, + 0.20666594385913026, + 0.4092649851848102, + 0.35964020981048067, + 0.32053949296768686, + 0.4181073428470443, + 0.3294924477460568, + 0.43737243308989554, + 0.27535706638720686, + 0.4236225808864425, + 0.3132549283911886, + 0.23965287402950303, + 0.37287397911262254, + 0.24931114225081383, + 0.42131776071560245, + 0.17094882722601681, + 0.6489087244460938, + 0.6562170280758005, + 0.5437666742989197, + 0.5443843647816765, + 0.5838953593698442, + 0.5818793030457606, + 0.22586810050652142, + 0.3081301719215782, + 0.2801448471928764, + 0.33160153457923475, + 0.2779545681794383, + 0.2952978843199669, + 0.2991062226515471, + 0.3171350966805998, + 0.3599155126027414, + 0.26520650872590396, + 0.296362530890156, + 0.25114889569471316, + 0.43580368210461845, + 0.23567957974043002, + 0.3527958514413518, + 0.35369371416261586, + 0.28722158288287214, + 0.4175933374306093, + 0.3130037299018572, + 0.42631708855242634, + 0.20754825560177562, + 0.39495251479431176, + 0.2998597595543761, + 0.29287691998513304, + 0.2532213017447966, + 0.2948515347589505, + 0.2685479047153871, + 0.2923582392338897, + 0.4690902758613812, + 0.41900156062460225, + 0.499897999790351, + 0.29291596158315686, + 0.3684189101746629, + 0.31021515748092976, + 0.2989128448331175, + 0.39811499993272975, + 0.26418375164161007, + 0.1784123108406575, + 0.3531380370559176, + 0.2732970156229332, + 0.29034673191074645, + 0.35940459821305865, + 0.2733142420343855, + 0.21401618147550053, + 0.2181140673425142, + 0.28200087235320564, + 0.2849518073621364, + 0.22941687462509075, + 0.16930127033342143, + 0.2509458049467125, + 0.4018854359155703, + 0.5441154641000091, + 0.4780719316753888, + 0.43851025059633786, + 0.48315955211908523, + 0.3913475475096094, + 0.45181469122505125, + 0.4503787865747798, + 0.38508374766466974, + 0.30807670220318006, + 0.3361015223140859, + 0.4052026151814398, + 0.5508254691907915, + 0.1266663848492591, + 0.2633359523952902, + 0.24758791058555962, + 0.20013867752765988, + 0.20761149937170237, + 0.3302953113537942, + 0.24487701509332233, + 0.32495036479710965, + 0.7054883259095126, + 0.5851665437828616, + 0.7366567262681761, + 0.7498544723470413, + 0.6210193860330003, + 0.697758768577576, + 0.7133454325394444, + 0.283783635192975, + 0.28049255395172523, + 0.3057637948007119, + 0.4055221156569354, + 0.26783304288359167, + 0.2879807760678996, + 0.32030522018273283, + 0.3080942048937178, + 0.32035776129361093, + 0.3029647494080191, + 0.23134987675255012, + 0.3989946271043395, + 0.3105925073750519, + 0.30947349974473815, + 0.3730393648401722, + 0.24757236297379678, + 0.27972151428479186, + 0.29674463626821146, + 0.2859600913770266, + 0.20595544213784558, + 0.3076726038607235, + 0.22063343015583256, + 0.4736763184533773, + 0.4364047217426452, + 0.30785290979050295, + 0.23434395468379693, + 0.14448454788335696, + 0.47418965803261687, + 0.5427930337319139, + 0.6229100546106322, + 0.2623439127688385, + 0.31963272699803524, + 0.1932541189427909, + 0.37333756926573813, + 0.2949495226238156, + 0.2560989690471945, + 0.42609975311633147, + 0.4036117384152573, + 0.4090772778221968, + 0.2740813133609591, + 0.38174588307140883, + 0.3631024845047149, + 0.413937908408567, + 0.3759468022504969, + 0.40742288160627643, + 0.3150148530414007, + 0.3141973806376536, + 0.3424684915939185, + 0.3187115105879245, + 0.22295999045489662, + 0.28229730631074, + 0.31289621657325223, + 0.3407588614002902, + 0.46105728125309725, + 0.14536778873932513, + 0.5195850953574693, + 0.6802236903365266, + 0.4951539911249801, + 0.33995751190838586, + 0.26844221878130436, + 0.28900997689661195, + 0.22471613145740513, + 0.3234670810990333, + 0.39362641842021795, + 0.37891862852816127, + 0.28127126253391455, + 0.23764211076068556, + 0.3723642106325479, + 0.4657978540076005, + 0.5083521304139414, + 0.5725832339229106, + 0.3371304253102388, + 0.4852995845341839, + 0.2667824515565486, + 0.3232850056444431, + 0.21382224719534934, + 0.19521564442604028, + 0.3161617252095432, + 0.2851898654002957, + 0.41589091353431457, + 0.3554062688778982, + 0.5793963594017759, + 0.6347860183057848, + 0.7298192992249799, + 0.5633178112589066, + 0.7391237802927201, + 0.3132745601987258, + 0.3092101165510909, + 0.3090420151193816, + 0.32804585093803185, + 0.26043237647717626, + 0.41611344567032854, + 0.36824967146648724, + 0.43676169237019963, + 0.3956017523268914, + 0.34494206930106197, + 0.34890936218852747, + 0.33316404941177385, + 0.26484139613604835, + 0.3183338988552016, + 0.2813197495808255, + 0.3206855276241134, + 0.37077942500314715, + 0.3485158026702594, + 0.40638134486521565, + 0.2735784464929686, + 0.3117281807658968, + 0.36955261322869287, + 0.2493068144194148, + 0.35898113468421444, + 0.3339028635633698, + 0.3399517372072442, + 0.3121704270804666, + 0.37714288353023995, + 0.26827598227822713, + 0.3311940090409064, + 0.33826804466617677, + 0.37862832287016257, + 0.284480756071729, + 0.4380399752037412, + 0.3272434929464557, + 0.3032048597842392, + 0.43736816198823664, + 0.35408062653439043, + 0.7177236047495661, + 0.18924659627964344, + 0.22455736601785714, + 0.2993449494174327, + 0.26967566254155523, + 0.1971923856712707, + 0.18378603689507803, + 0.2501714147868223, + 0.6526445676517776, + 0.5309827847183478, + 0.6151612279821856, + 0.27304447153064, + 0.2697133899873886, + 0.21569452134568945, + 0.27052249112450466, + 0.3103335079474996, + 0.15765040285631457, + 0.20150812674482627, + 0.30612500614185595, + 0.28402700103323625, + 0.33993225192808946, + 0.25623436498892876, + 0.3649337960836414, + 0.2949550986104996, + 0.25789686716916393, + 0.6680540246730998, + 0.43299234418027716, + 0.6651872079772576, + 0.5458570527405103, + 0.1828341013239223, + 0.4257845248829182, + 0.3927524327254155, + 0.42860413794373153, + 0.38223944115302744, + 0.32614501951149905, + 0.39557097964523646, + 0.38282276186843833, + 0.425905840869346, + 0.37634339495532715, + 0.28353406168542306, + 0.3108127422322657, + 0.2051331794131004, + 0.23250055738089345, + 0.6168949305137404, + 0.3419740850056944, + 0.5486941308100461, + 0.4454999098222382, + 0.38842561809728626, + 0.352529798282486, + 0.3461055978863772, + 0.3998664389953461, + 0.34350277427156684, + 0.18201626327348336, + 0.27263987428448727, + 0.3485034174361426, + 0.189193500026144, + 0.7367120819738502, + 0.3673904011301564, + 0.5466881832567049, + 0.7070670292944197, + 0.6576859656579236, + 0.31891193539390916, + 0.7845446795517137, + 0.45554421914948157, + 0.6350966569505302, + 0.33673614976402266, + 0.35991619527461677, + 0.27027071398317737, + 0.25209145132514593, + 0.37689641349506114, + 0.3658907408069636, + 0.36435169553145014, + 0.2706261127617317, + 0.2779561015371315, + 0.27903104895016684, + 0.30761914929499184, + 0.3683116436603128, + 0.18936278789501385, + 0.3974711805924283, + 0.3089514787267539, + 0.30835487900570313, + 0.3861866417227474, + 0.4263427059841198, + 0.24165082953416436, + 0.3284781983285937, + 0.3550762709314625, + 0.4611363129304021, + 0.402143947242335, + 0.3328607149256244, + 0.2245447887285828, + 0.44955154071425574, + 0.5048656758753655, + 0.49614878981430843, + 0.3705113705285348, + 0.5458544874088519, + 0.556583949986255, + 0.34952675137177874, + 0.399837932708222, + 0.4451173035351351, + 0.3985794969464103, + 0.3632249120683343, + 0.25499432170278236, + 0.20197595343802394, + 0.22949092022156792, + 0.28875090317908625, + 0.2158570016671487, + 0.22999661544503225, + 0.23645388513177823, + 0.2614153128609619, + 0.28105668252552446, + 0.21610760487679415, + 0.24863740612479432, + 0.23506976114586503, + 0.2506048861907752, + 0.3134684481080359, + 0.23494981611269863, + 0.24041233329919626, + 0.3664659789064584, + 0.45741860680269536, + 0.6124369936010535, + 0.2336764860837915, + 0.44309198717518383, + 0.43456726548122665, + 0.45710382568974367, + 0.440294608077155, + 0.3580440869187052, + 0.31880873874893856, + 0.2760918189915821, + 0.3236865158910961, + 0.3025750982958274, + 0.6020923853033392, + 0.3459300954283519, + 0.3831200003470225, + 0.2028590912836639, + 0.17680694274383968, + 0.33212679444816084, + 0.40443026497961426, + 0.3143914551080501, + 0.5333552414397893, + 0.2697173047744512, + 0.5875045758869077, + 0.4188880899376035, + 0.41629048978521155, + 0.43012171141609734, + 0.40128359685327847, + 0.35032819444357377, + 0.4404997284046873, + 0.42215295241715567, + 0.44110502779968447, + 0.42126677646947264, + 0.36928503551928854, + 0.28584430237552166, + 0.2101783333559105, + 0.3203680764309048, + 0.17906775668100322, + 0.3306661735769798, + 0.34908696321694554, + 0.36073358161181196, + 0.34384166758198353, + 0.36360938757138683, + 0.3479559932432104, + 0.4273146565908288, + 0.37834715797332064, + 0.29411830406499445, + 0.389950247654848, + 0.4449352124845602, + 0.2978997761141327, + 0.22204926213058, + 0.2825911005972334, + 0.3187736870046754, + 0.3440157563888024, + 0.19525198280926462, + 0.3516642751784331, + 0.2020702656176219, + 0.32383321872429927, + 0.3871062341335743, + 0.37501212768947373, + 0.16158159652663634, + 0.23315821412009877, + 0.2582328350451903, + 0.18703913769825248, + 0.18165665866810304, + 0.24560669465841928, + 0.35180744501520983, + 0.20547945276431995, + 0.27539315303848455, + 0.3866749824571506, + 0.2518893130987635, + 0.25872603040943754, + 0.3318957046974225, + 0.36324395368930157, + 0.12919755260610802, + 0.4039156121642507, + 0.28857012066344695, + 0.24492873887548136, + 0.2289049124419141, + 0.35860849272520906, + 0.29501030502983433, + 0.15430021119883974, + 0.3958542502958197, + 0.3315545424738874, + 0.318773976104654, + 0.39013089898586323, + 0.28718268265761765, + 0.2423733801434745, + 0.28654029184883767, + 0.18754701406571425, + 0.29152475116525745, + 0.2595077396705987, + 0.2959012227628453, + 0.30036113524062824, + 0.3243448399098133, + 0.3589828773461198, + 0.3169371609748325, + 0.24385642548298905, + 0.27719813520850894, + 0.3396712504367918, + 0.29312556662120387, + 0.3043128023615442, + 0.3797575237633639, + 0.3219737615576263, + 0.4286670598743424, + 0.47144141954466984, + 0.38787604333198833, + 0.3862442383945503, + 0.268333171940694, + 0.19431608697510278, + 0.28331703361264016, + 0.3591427512105243, + 0.2696383627433714, + 0.37903025231562226, + 0.35983710731503715, + 0.2631731239821132, + 0.365172756236146, + 0.26344707855113286, + 0.2244471773197006, + 0.31834314817707265, + 0.34823992455814, + 0.3595248163942301, + 0.3702621870343647, + 0.23670752536695444, + 0.48793532134512374, + 0.38749676526153326, + 0.4138240691700965, + 0.4214976250814173, + 0.4929637075430526, + 0.3419686654078747, + 0.2618359552399033, + 0.3339552338064686, + 0.33508082307842324, + 0.3328998049766994, + 0.2717477319549358, + 0.24970691779542292, + 0.2625906433175087, + 0.3320717852042375, + 0.21564668531752926 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.03960787550567231, + "valid_tracks": 4, + "per_track_pearson": [ + 0.04239889311807491, + 0.03786319848354249, + 0.04060818056968738, + 0.03756122985138447 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.13040267237774725, + "valid_tracks": 626, + "per_track_pearson": [ + 0.1306572653069055, + 0.12391256623662203, + 0.10047069411595314, + 0.11713822768626947, + 0.14189368388467746, + 0.13475586892667407, + 0.09249953170271356, + 0.09598874941602943, + 0.1396250270921677, + 0.12586638345822249, + 0.12691738008830802, + 0.1275479792989979, + 0.13022004528692166, + 0.1204568706596265, + 0.11692679599376118, + 0.12339367484888586, + 0.11450071270790364, + 0.10686755943422512, + 0.13865555736963492, + 0.1174245749433282, + 0.13865024852380323, + 0.11906008599065414, + 0.12956642104182586, + 0.08448950476430843, + 0.1175417342756212, + 0.12248983986985391, + 0.10583448599092242, + 0.13213293968772005, + 0.12790494725875284, + 0.11290098751119558, + 0.10895162351070331, + 0.11861969451123194, + 0.0871969434347343, + 0.12421715721299852, + 0.0962166597564298, + 0.13500473424608403, + 0.09061227948147034, + 0.1300091247926734, + 0.12952163175279027, + 0.13088767809677596, + 0.12664542103870513, + 0.11318076484020956, + 0.10997931112829473, + 0.10282808600561763, + 0.11030207861707121, + 0.11052514873062244, + 0.09195922975511528, + 0.11700023209534408, + 0.10204362452729622, + 0.08628205177766088, + 0.09939907390068904, + 0.09074176872461526, + 0.09818400868648026, + 0.1131656206877781, + 0.10959212977844329, + 0.10268101568538289, + 0.12645122261755296, + 0.12079953357318989, + 0.12300053551782425, + 0.12802350423988237, + 0.11014055177236125, + 0.13269237397190795, + 0.13520538765963255, + 0.12776720740526376, + 0.13259512396214, + 0.10863060574220637, + 0.1326593002420469, + 0.10451156327994245, + 0.11452609048061106, + 0.1118893144290243, + 0.13199009399090073, + 0.13320200302297464, + 0.11951985397377778, + 0.11379698540011071, + 0.1262917051276519, + 0.1265758430062536, + 0.11635550410769163, + 0.12958066879496444, + 0.12548389195097442, + 0.1305253775140466, + 0.11714507394910939, + 0.1223546963851826, + 0.1333692746772458, + 0.11072996291427406, + 0.11242702644512335, + 0.1253823352132722, + 0.12245677455391156, + 0.12622337029230826, + 0.11634951327467467, + 0.133613164642556, + 0.13002553345300236, + 0.12104010437629972, + 0.12355027964793908, + 0.12399238281996403, + 0.11052221093202809, + 0.12097743358110702, + 0.1217309750532783, + 0.12811847637383064, + 0.1288529862027313, + 0.1214812254545832, + 0.12445634239363154, + 0.12375583507985466, + 0.11919944980435211, + 0.12384051095325989, + 0.11936268210601153, + 0.11721650538049355, + 0.10142375332272083, + 0.12928393841657856, + 0.125515746549636, + 0.12120719549575729, + 0.11792224805721005, + 0.1218092096128794, + 0.13464197981621542, + 0.12441829380582743, + 0.1161090438197562, + 0.12342073159306781, + 0.11456389689879633, + 0.12246634931947892, + 0.11553626654710604, + 0.1089762570747978, + 0.11281679991406894, + 0.12288347098431038, + 0.11784812807798388, + 0.12601842246640219, + 0.10310839829141223, + null, + 0.12339466374869978, + 0.11138492135712333, + 0.11273929054507266, + 0.10905845938066623, + 0.10098615737071362, + 0.11310510307503505, + 0.12265716840339543, + 0.12121475931550937, + 0.13211286940573191, + 0.12138939118243601, + null, + 0.11040787830798585, + 0.11545932179703862, + 0.1042584924771297, + 0.09002058093189039, + 0.1195250676881928, + 0.09873113894400405, + 0.12156614170549235, + 0.11634464856099255, + 0.10433704972466626, + 0.08781850095490239, + 0.09859777880959578, + 0.08703432648901635, + 0.11192416307188839, + 0.10435677097064092, + 0.1142999428510802, + 0.12227293805604568, + 0.12700832027982895, + 0.10482360680125023, + 0.10904484335833974, + 0.10968658791568907, + 0.09747476011204723, + 0.12141584254556945, + 0.1068366873751187, + 0.11194209203349963, + 0.11118832993945217, + 0.1291932923071497, + 0.13226856994726738, + 0.10627700954846743, + 0.1070538244737696, + 0.11476614773747702, + 0.11146147727184495, + 0.11471386223721312, + 0.10951257418857561, + 0.10308125420228664, + 0.12120721730520027, + 0.1101346709484009, + 0.11069560871663346, + 0.10698728840869653, + 0.08611733360542742, + 0.1183258323308509, + 0.1139704999411507, + 0.11299911425818443, + 0.10938555851792994, + 0.1116947815315508, + 0.09837293948476399, + 0.09664112941160846, + 0.1262499888357524, + 0.11881646918082496, + 0.09002380706616245, + 0.1428404608943993, + 0.1260434138300435, + 0.12733909358010798, + 0.1206417954948336, + 0.1341260228349753, + 0.1127739867460164, + 0.1426148220216415, + 0.12647178472502735, + 0.1345922727926258, + 0.1336499125918867, + 0.13214235241040273, + 0.12350160501901981, + 0.13823475125754517, + 0.12654846640941936, + null, + 0.1397890464598405, + 0.13185695642047496, + 0.13432649958393525, + 0.13355693992051423, + 0.12637799098690344, + 0.12267257573067562, + 0.13279638764517984, + null, + null, + 0.14030053131717354, + 0.14367770403982374, + null, + 0.13813173949272214, + 0.1238582290830926, + 0.135592986510638, + 0.1262056373138844, + 0.13254879700406225, + 0.1303212023090997, + 0.1322921667620799, + null, + 0.14235054505223954, + 0.12280691840703022, + null, + 0.11562720682636088, + null, + 0.13069393996045758, + 0.14125892119125924, + 0.14427849767190368, + 0.14551637130271897, + null, + 0.1319151520865191, + 0.1312765763247291, + 0.12199778702125795, + null, + 0.11430184522586694, + null, + 0.13618746485060054, + null, + 0.11308979976320734, + 0.1355687655010785, + null, + 0.12992428411001908, + 0.12541739875448818, + null, + null, + 0.14586804236667217, + 0.14217155116597896, + 0.12644906599257097, + 0.1270009921630402, + null, + 0.13374989266690174, + null, + 0.13988011249036006, + 0.13918913535600463, + null, + 0.09682437092189741, + 0.11955164817753496, + 0.13530758825201725, + null, + 0.1254028439612634, + 0.1365144214329779, + null, + 0.13501290421817794, + 0.1317688862790525, + 0.13850037453149874, + null, + null, + null, + null, + 0.13873825765638148, + null, + null, + 0.12540076998555166, + 0.12674318625123998, + 0.13301260929451278, + null, + 0.1358442935456877, + 0.13575424362030142, + null, + null, + 0.13532794795193104, + 0.13903776064489684, + 0.12524376416343694, + 0.1352144050311425, + 0.14957834232940012, + 0.1363906290317537, + 0.11909172693089409, + 0.14224157216248315, + 0.11926015338757429, + 0.13639118366707326, + 0.13315251180386145, + 0.13115420346852066, + null, + 0.137482894049703, + 0.13427087169131274, + 0.1373668094463253, + 0.13918234532094528, + 0.11193800567946484, + 0.11717709309529914, + 0.1359554762476004, + 0.14214711118870874, + 0.13958883060947083, + 0.13202213032267024, + 0.12456223213738726, + 0.131721051287213, + null, + 0.1417991681857732, + 0.1424785360617939, + 0.13550093531774063, + null, + 0.13035820010906293, + 0.09701121478853343, + null, + 0.15008420130516895, + null, + 0.14977552041575928, + 0.14014688157185284, + 0.13988041652302652, + 0.12447964353446746, + 0.12220718452775751, + 0.12991331239084047, + 0.11986525684916573, + null, + null, + 0.13926218000656485, + 0.12302256838662115, + null, + 0.14346096514736859, + null, + 0.15218036302560015, + 0.13407248865969013, + 0.13970040848894996, + 0.1282184367566151, + null, + null, + null, + null, + 0.1413882417156448, + null, + 0.137907406589677, + 0.13041238370185024, + null, + null, + 0.13543711027672897, + 0.1489433434141096, + 0.13706670736676407, + null, + 0.13623101267491103, + 0.13854910180159874, + 0.13238917219852966, + null, + 0.14723458929281236, + null, + null, + 0.1442181296737932, + null, + 0.13233447991792766, + null, + null, + 0.13547386385389473, + 0.12449098955746415, + 0.12910045773187737, + 0.13054734812438004, + 0.12351158068039518, + null, + 0.14013190411230364, + 0.14577917467660978, + 0.1239436163432087, + 0.11040714531151402, + 0.1205010423142396, + 0.14992426919014892, + 0.12102157083170019, + 0.14752255521393198, + 0.1703975451104044, + 0.16958571750289955, + 0.17364352466303545, + 0.12636224019592315, + 0.1307601609874645, + 0.13725244518627283, + 0.13414495040981708, + 0.16246574074886544, + 0.1610646681864375, + 0.132294653859764, + 0.1698025553825608, + 0.1499313570702144, + 0.14476747804747622, + 0.1344195881282265, + 0.12657523284406025, + 0.12425208371871224, + 0.08246626788771964, + 0.18230105410021336, + 0.17102927508742446, + 0.12424823642976827, + 0.17730299894247886, + 0.15261673299336725, + 0.10287436984046654, + 0.14179937423077726, + 0.12386664177130507, + 0.11733788954568253, + 0.13848000246869877, + 0.10750930444207954, + 0.13389289459068085, + 0.1234646692666095, + 0.12377964426187654, + 0.1323000748996905, + 0.13080449914656478, + 0.11461217836638182, + 0.1490312727685537, + 0.1419146247641103, + 0.11954266510807127, + 0.1264910142996332, + 0.12145379831222759, + 0.09855257279115454, + 0.13048604447415196, + 0.12198975011248034, + 0.10023328529112846, + 0.10295281244522234, + 0.08798515315713962, + 0.13048718970549417, + 0.1098499227805957, + 0.09658241220013435, + 0.1606977243886449, + 0.13780592932052382, + 0.13163106333955846, + 0.17536721780160147, + 0.1237401576882913, + 0.10751597526449659, + 0.14127783599916052, + 0.15266984172891637, + 0.12414986141012839, + 0.1354213739834041, + 0.19150392689193235, + 0.14480652905726502, + 0.16438091155054438, + 0.17763035389344445, + 0.1570405576501629, + 0.16436379446370322, + 0.12704800900588023, + 0.1286237456940058, + 0.1300700839778146, + 0.12307160544512386, + 0.14929333559656002, + 0.13738201652889295, + 0.15085678222141208, + 0.13092072300897806, + 0.12624176075148524, + 0.14666498465751032, + 0.13325333394331096, + 0.16458815630167153, + 0.14236993872335083, + 0.16073286647098087, + 0.12464134148337729, + 0.127151492847172, + 0.12237586058321556, + 0.11655022151432011, + 0.1339716220236615, + 0.1338212116934364, + 0.1284331964749938, + 0.15710415144372522, + 0.1266437941123702, + 0.15868800380182235, + 0.12206468250299267, + 0.13691435717689085, + 0.13194289285642666, + 0.1314763211698808, + 0.1430423084012762, + 0.10926970596207537, + 0.12871084508242506, + 0.1349624386085073, + 0.1625391466346545, + 0.1638586015414857, + 0.1312634857420479, + 0.09803214111727636, + 0.14807631898858842, + 0.12779725676981074, + 0.1226450405901268, + 0.15545150145596306, + 0.11796314469360261, + 0.12225408729997082, + 0.11560323427043105, + 0.12054551819268572, + 0.17512471868722665, + 0.15454614867927438, + 0.1509424375327832, + 0.1542068033521314, + 0.11390336317356724, + 0.1369646504724708, + 0.16721895737726883, + 0.12356281539691212, + 0.18273680439324003, + 0.13375602105099485, + null, + 0.14306729676650468, + 0.16243991824013457, + 0.11580899121753603, + 0.11323332627149829, + 0.1323239250661116, + 0.1533511245582069, + 0.16120884630018967, + 0.15300326272371848, + 0.13589219250799156, + 0.11404402940874184, + null, + 0.1523941162091157, + 0.16384108592844107, + 0.15725947995657805, + 0.16648393400089062, + 0.11379680354758527, + 0.16478886874258933, + 0.12085032467083476, + 0.1331207361544298, + 0.10674221606995826, + 0.12225154135433655, + 0.16867566862587152, + 0.15948636857550735, + 0.1660451279196766, + 0.1297349421189938, + 0.12983812915897402, + 0.16778005164372717, + 0.15903552595311685, + 0.14236141272537844, + 0.1395522220938938, + 0.13696272742030352, + 0.12348931596061792, + 0.17989660930091983, + 0.13019305162480407, + 0.15237615255578635, + 0.1041583772840063, + 0.15282491496765263, + 0.11101579882417677, + 0.10766073394641516, + 0.10950671499089647, + 0.14894808451121824, + 0.1639379821179455, + 0.07363980272479505, + 0.13620440233825626, + 0.16646691610022257, + 0.1359657382624257, + 0.18006583799293316, + 0.17338479020415465, + 0.11422131617698018, + 0.134020400830209, + 0.13496917410406759, + 0.14285474637550943, + 0.1259661382257175, + 0.10665105850554711, + 0.15064269597357, + 0.15935399612012538, + 0.06183617536444805, + 0.13556431673293895, + 0.1270982797915753, + 0.14133847968316368, + 0.12905022941605276, + 0.11224621157904052, + 0.11951849512565944, + 0.15466743597231475, + 0.13796011936274108, + 0.12552304785584278, + 0.13666603461870455, + 0.17181381337938412, + 0.17194379833872078, + 0.16437138428866632, + 0.17709942938324477, + 0.17138081755839493, + 0.13148698605641268, + 0.13915219246953525, + null, + 0.1439273332835575, + 0.1324755023922641, + 0.11334766579617837, + 0.13425043453185326, + 0.14457880218456834, + 0.1479532100831199, + 0.12179144902530555, + null, + null, + 0.15910197596078574, + 0.1543097718425359, + null, + 0.1745883724730241, + 0.10482454592483047, + 0.1783299149266471, + 0.10489443129828532, + 0.16437356272655568, + 0.14219675774335572, + 0.13966353852059413, + null, + 0.1376207452072303, + 0.13373811301062627, + null, + 0.1676147837890567, + null, + 0.16839285860573697, + 0.1385509074341739, + 0.13157467024237746, + 0.15429158118072503, + null, + 0.16374808698393037, + 0.1561115640416497, + 0.12206288904093572, + null, + 0.1445887179439889, + null, + 0.1378929451988219, + null, + 0.11199040841908721, + 0.11768607214701542, + null, + 0.12527866868014514, + 0.1349246629334848, + null, + null, + 0.1418970141835635, + 0.1249691988348386, + 0.11667104966360141, + 0.13764312702293136, + null, + 0.10873593090289344, + null, + 0.13801238891205367, + 0.11705362134521444, + null, + 0.08437715567965681, + 0.1188788538479036, + 0.11790532910424058, + null, + 0.11624505413900581, + 0.130288757886464, + null, + 0.14513389141405278, + 0.14301480313761034, + 0.1529196573348648, + null, + null, + null, + null, + 0.1526881809874079, + null, + null, + 0.1107739546664221, + 0.1423240292158516, + 0.14891982452447394, + null, + 0.16238619761288609, + 0.15619448460871685, + null, + null, + 0.13183195626088912, + 0.12658584093179578, + 0.15100110485656762, + 0.13748061318473478, + 0.1284152322508227, + 0.17472513038968238, + 0.16381933064055365, + 0.1357452355182313, + 0.11136647117411391, + 0.1698503903715403, + 0.1464110519129561, + 0.12339280382533856, + null, + 0.11119468325901684, + 0.12393357287982676, + 0.1285546379988379, + 0.14391311772503962, + 0.1654141920563374, + 0.18782260843457424, + 0.11795564721802156, + 0.13879848489695593, + 0.14916859469902713, + 0.11322839717755087, + 0.16004896506070113, + 0.09887351475237946, + null, + 0.15966726693543587, + 0.15021474341335217, + 0.1499410228748348, + null, + 0.14933657081219004, + 0.15338618775694765, + null, + 0.1480927607772005, + null, + 0.13508262069600663, + 0.1206917614438717, + 0.19071905751410703, + 0.16511130665152793, + 0.15020308263166227, + 0.16341208651057054, + 0.13278681549425642, + null, + null, + 0.14491788237752018, + 0.12514862426072146, + null, + 0.11868215603520291, + null, + 0.1301513359026871, + 0.13655940306297387, + 0.12088401590184668, + 0.13522278313066527, + null, + null, + null, + null, + 0.12505397737789065, + null, + 0.13449023686136719, + 0.13573148588042896, + null, + null, + 0.1349250992361932, + 0.11439733075359701, + 0.10526022310876255, + null, + 0.11155080113624319, + 0.1256269935052763, + 0.12368942995047574, + null, + 0.14431900965062622, + null, + null, + 0.12602523912969776, + null, + 0.11091006370276686, + null, + null, + 0.15848263595466536, + 0.14303529234706538, + 0.14659695308915843, + 0.15313663068087677, + 0.1353088845647616, + null, + 0.12686860119081478 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508, + 0.7761419403548508 + ] + }, + "junctions": { + "mean_pearson": 0.17150042290546252, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2586037629812065, + 0.29529664516157605, + 0.18171330725530882, + 0.1148631840979154, + 0.28905341145663455, + 0.31309371663069296, + 0.018379684056428303, + 0.17971014899704077, + 0.20464678935754813, + 0.17116160083988663, + 0.19141296283047946, + 0.11287773394586481, + 0.07313801309149937, + 0.1604319028351027, + 0.14895427374376397, + 0.22388784054359717, + 0.260777780038976, + 0.17670781701118962, + 0.20393879139969007, + 0.17236217670274914, + 0.20460477580875622, + 0.1989142323424261, + 0.14080828671990786, + 0.18092050982952546, + 0.19506353271315652, + 0.20851940897479357, + 0.19476100615238323, + 0.19609578966279312, + 0.21762024658366444, + 0.21108286078643723, + 0.22628552739339022, + 0.23224068550829274, + 0.2198307847981137, + 0.20026917124535798, + 0.2129821919397798, + 0.20906676674044858, + 0.23638308693700122, + 0.20670207830089662, + 0.17296733243036025, + 0.15410413124928754, + 0.21699670677577573, + 0.12335528159045228, + 0.0958479556636043, + 0.13532208036247037, + 0.12175005856773413, + 0.1247641588459382, + 0.10966921113547677, + 0.1165038212339489, + 0.11147471722760906, + 0.09539363637819742, + 0.1297237796077672, + 0.13332778139330625, + 0.0938873242934333, + 0.1344285579223899, + 0.11140895186409892, + 0.09984097799394202, + 0.10039052260653152, + 0.12604438997616263, + 0.11426107777099397, + 0.14961949827308496, + 0.07090797700171203, + 0.15856966249025423, + 0.1924652881498565, + 0.0931110092363661, + 0.15553844813477324, + 0.24879528719672578, + 0.16516009249351515, + 0.3412088404049258, + 0.11106087940348203, + 0.1601601073450418, + 0.15033894835281464, + 0.2484929501824096, + 0.16017947042946368, + 0.14309190029737637, + 0.3743827286318056, + 0.3351251192650783, + 0.200953204348984, + 0.14497905463574703, + 0.19433105022730526, + 0.299234873781019, + 0.17850007346899294, + 0.15518571759330074, + 0.22835259161692167, + 0.25665977765102493, + 0.20482258548149798, + 0.24027980082041464, + 0.3448797425843491, + 0.2620649046712575, + 0.18129972006546474, + 0.21500259597792257, + 0.24433808486328898, + 0.2310688259488722, + 0.11945371287016991, + 0.14490752796602457, + 0.14386234386583438, + 0.2568118261585095, + 0.2552408639976393, + 0.22873705792228854, + 0.163322554084855, + 0.21606846505302085, + 0.1439063303162674, + 0.16188574978307355, + 0.3076167096534552, + 0.17122542511578667, + 0.1281881798196064, + 0.10428726221924017, + 0.05286553555843724, + 0.25222189617144636, + 0.31838980008980655, + 0.2618440667047028, + 0.12606064514088827, + 0.31761883469630076, + 0.23055573788998426, + 0.28423127095728645, + 0.29017851948630186, + 0.14470827102981512, + 0.14760584998945575, + 0.2078253011761094, + 0.21945225468531696, + 0.2795820597549262, + 0.166442935950352, + 0.2414694229813921, + 0.2121604964664002, + 0.20263922979858787, + 0.07743177521590712, + null, + 0.28856872084092533, + 0.166344764409486, + 0.0962737236107567, + 0.20751043273273356, + 0.2190233174280337, + 0.15497138316729736, + 0.18202505121554885, + 0.2887927448002014, + 0.23217504061116712, + 0.07935028366022016, + null, + 0.19134060648851225, + 0.17820328721464684, + 0.18293840184034704, + 0.19284563093526552, + 0.1607062635161467, + 0.10589572108710738, + 0.12186773053233636, + 0.1948353329717608, + 0.15426509299481625, + 0.09827428447471848, + 0.1732634889612173, + 0.1748516325507517, + 0.18792382031816002, + 0.19041312949768657, + 0.21396843024226572, + 0.1378163413299966, + 0.22752980456177693, + 0.3341713600667802, + 0.2746814637021293, + 0.2573255687875423, + 0.22217719670240135, + 0.19483066814028507, + 0.32765750245050995, + 0.17235536614938682, + 0.21569281095669177, + 0.20284547500199987, + 0.2452117600103696, + 0.17784012001506933, + 0.33475136870524147, + 0.2133118191238368, + 0.19144392191112916, + 0.07180024014873092, + 0.1828497499901468, + 0.08933087164204763, + 0.14701523821441992, + 0.3184238656449095, + 0.10873739363333586, + 0.184427168619828, + 0.1815509201739356, + 0.15872251036210974, + 0.15941475289746085, + 0.17297222631254594, + 0.18479220376137337, + 0.18419446649669421, + 0.10559676828169631, + 0.09040986976629123, + 0.2666695883362525, + 0.15522560621977968, + 0.16836169437714832, + 0.26364636338134806, + 0.22874537910461235, + 0.21469564939719982, + 0.18239954752347398, + 0.20105439690006519, + 0.297731063421077, + 0.1576032950443022, + 0.20699934551580904, + 0.26743603002905314, + 0.1902001561861125, + 0.16983126203172805, + 0.16173777607711728, + 0.2150239368961602, + 0.2611758105313655, + null, + 0.13401066266301442, + 0.17296466344147346, + 0.2546421359070479, + 0.33308196455679434, + 0.07219565177545478, + 0.11008271727556995, + 0.1121608663766058, + null, + null, + 0.20031031693933318, + 0.34595291338190903, + null, + 0.13512983178593097, + 0.2278587926687286, + 0.07423222987758203, + 0.16476870286113673, + 0.09852629287405333, + 0.1960119345938586, + 0.059186082959598235, + null, + 0.29064860099409096, + 0.23239203440398673, + null, + 0.13692960499006923, + null, + 0.2959323708355339, + 0.3250947104170493, + 0.21668925849999068, + 0.16605391699826738, + null, + 0.17681951254129882, + 0.18347094484789586, + 0.24310047534733914, + null, + 0.19577677555473097, + null, + 0.18928964242571084, + null, + 0.050308594045544386, + 0.20245323685617078, + null, + 0.18344379835864488, + 0.24105149405161677, + null, + null, + 0.21494647373094913, + 0.2425618775814362, + 0.06457772690936238, + 0.04557024846336202, + null, + 0.3296996049142048, + null, + 0.23314703990737434, + 0.16903202963087915, + null, + 0.21473013769850224, + 0.2899206005675968, + 0.32806394873791644, + null, + 0.25143545713862553, + 0.25504074990474374, + null, + 0.295924178460616, + 0.1965396056030082, + 0.3085957197838553, + null, + null, + null, + null, + 0.16766731933556742, + null, + null, + 0.15633756655218375, + 0.16298943040763667, + 0.28382211706410027, + null, + 0.12724787127469658, + 0.23404304955809851, + null, + null, + 0.3601671586234092, + 0.14067124598445568, + 0.23509208412129126, + 0.1761022341299516, + 0.17251697233965957, + 0.14859866743934788, + 0.1561214522266434, + 0.19576156968318847, + 0.07877209805344053, + 0.23719404533057858, + 0.2799258053310191, + 0.2761840811012366, + null, + 0.14124098167377236, + 0.2514068479614458, + 0.0787344649499614, + 0.10304765881227794, + 0.06718879174722828, + 0.08323989644264912, + 0.13760242389328445, + 0.15213353371989202, + 0.22658777041168304, + 0.13805229516085282, + 0.2744212864341173, + 0.24200983822908256, + null, + 0.30873448359659705, + 0.1727286226135009, + 0.17496430641674723, + null, + 0.23257951395593118, + 0.21210942201492858, + null, + 0.2681282091447614, + null, + 0.1471776190385603, + 0.23324315729913514, + 0.21381064257012036, + 0.13712773456977553, + 0.11607256737051071, + 0.21511567682465885, + 0.20630953420249995, + null, + null, + 0.21925267754899033, + 0.1184917522576864, + null, + 0.2439274760591703, + null, + 0.2339881303944177, + 0.15761851098370655, + 0.08065555375100684, + 0.1754327823748244, + null, + null, + null, + null, + 0.1978357493089635, + null, + 0.10819219713386428, + 0.2526524707849454, + null, + null, + 0.30985524991593427, + 0.3225983635485798, + 0.3095206743965141, + null, + 0.3558404049170877, + 0.25008423532562196, + 0.07151944732389161, + null, + 0.32883501907023205, + null, + null, + 0.37987939172758833, + null, + 0.3521868841505789, + null, + null, + 0.29623499963659816, + 0.20192182838588255, + 0.19677482432904997, + 0.18618502678599305, + 0.15978204972966364, + null, + 0.15073601798999042, + 0.023721839532814877, + 0.22624345492869594, + 0.0845325687437149, + 0.033944796498392917, + 0.48362803265498805, + 0.10622222623848654, + 0.035731126339211994, + 0.09700588187766102, + 0.19208828590877308, + 0.11333286449894035, + 0.23379787634531538, + 0.15224239861210476, + 0.13667909778472398, + 0.2541808225284439, + 0.2096979560125732, + 0.18529954827975714, + 0.07014241940992112, + 0.17519716175290373, + 0.1530285040523671, + 0.11435513496318453, + 0.20470878552347388, + 0.09333677939675263, + 0.015025082117102543, + 0.042548369650362315, + 0.37363875640416044, + 0.5401610646631549, + 0.21628431767901451, + 0.1827503837826422, + 0.2438510935611029, + 0.20811987369670307, + 0.15586566128452337, + 0.15140092481191453, + 0.01863720156025337, + 0.05492407767929697, + 0.0038010742216081543, + 0.03570049252937942, + 0.007329665015389243, + 0.0665083472700542, + 0.10546473428839474, + 0.024304425247842903, + 0.009390962325172857, + 0.025538743250352422, + 0.04673963807426896, + 0.030974461700682625, + 0.018480476168415014, + 0.17729305156335096, + 0.04366818179026154, + 0.20742247107563355, + -0.0028139663473230546, + 0.006183300035884772, + 0.08384314714047976, + 0.037633821940886165, + 0.0052541183624162865, + 0.006565455230208605, + -0.012816819504817473, + 0.1475189036306865, + 0.12092999410886995, + 0.40690279776166094, + 0.04552651679070232, + 0.08596995000267049, + 0.1630191951316596, + 0.21308661754734792, + 0.27481549261220567, + 0.24743726277263664, + 0.11195866714312439, + 0.27892448269107606, + 0.1285193098796911, + 0.051450373872866904, + 0.06850791865869277, + 0.17320247686870383, + 0.3378794313054251, + 0.17862260073914688, + 0.2903798152699362, + 0.1634711251777018, + 0.14490946325855583, + 0.11382257655213822, + 0.16653850200383594, + 0.17454917468033268, + 0.34118651515233683, + 0.32092050742285544, + 0.2510107894975448, + 0.2680150865219146, + 0.2681610268719389, + 0.16242436109420574, + 0.10744729664647805, + 0.4405841087768914, + 0.2730985689562599, + 0.27640032207617127, + 0.15596390701815818, + 0.3241509768252648, + 0.22754509884191776, + 0.2115282019809669, + 0.4330553741082247, + 0.04405105768731092, + 0.03739731729656318, + 0.14402654959767816, + 0.08383471606413104, + 0.07631191037098332, + 0.24766938624610543, + 0.13055909741422236, + 0.16983611399226012, + 0.09330891501774492, + 0.08583302557552895, + 0.1939345709023905, + 0.2105950709455903, + 0.40244485625480253, + 0.047338610493019816, + 0.5043584250926694, + 0.0639663107184224, + 0.129366971190861, + 0.05343580821356456, + 0.20537949173548592, + 0.06472440682467655, + 0.12007986442287348, + 0.14322644077515373, + 0.08833817788385571, + 0.06531316436183049, + 0.06060756931664553, + 0.05343477382494018, + 0.07384438236072494, + 0.3787615018616635, + 0.125647391222505, + 0.0435068447859086, + 0.42953282318736846, + 0.005016787687675359, + null, + 0.13072786895642569, + 0.11961382028828667, + -0.0004189475012502148, + 0.13043087515932494, + 0.22174096645568014, + 0.15664514570263827, + 0.2629323350669554, + 0.2809383500511027, + 0.23067645839586676, + 0.0526808998371867, + null, + 0.12365963089504736, + 0.07701634303957955, + 0.03335101764122241, + 0.0635351201175611, + 0.16931598338667478, + 0.10855810613374994, + 0.16768322498855995, + 0.1036716281612206, + 0.11881178982209227, + 0.26532183900788375, + 0.09841024569131643, + 0.08681107283009576, + 0.12479485955475776, + 0.027488487511981952, + 0.13203484285996425, + 0.19582202895010734, + 0.14103952840099424, + 0.13171740755104772, + 0.23369797725295616, + 0.2547702427027581, + 0.09214240783897587, + 0.08888772755998242, + 0.07217325789907257, + 0.09583255073250743, + 0.09624165400612054, + 0.14841803818214958, + 0.1514916237893046, + 0.07794015193625695, + 0.09894067465738744, + 0.24769596656919643, + 0.1494211917829896, + 0.13055347163141195, + 0.080798224292927, + 0.14128578484930632, + 0.03263326026809774, + 0.13555234994028068, + 0.08437151248261122, + 0.059215129407312064, + 0.1665292433688755, + 0.0593031040677087, + 0.04817237109865391, + 0.08716291988829565, + 0.2794436063998718, + 0.0849078903103481, + 0.0990524762668244, + 0.03819917367965668, + 0.08904837631864841, + 0.17578394723582483, + 0.053145480829057005, + 0.1432153403975694, + 0.12622486129509342, + 0.11926832459230832, + 0.3173863718264012, + 0.08771063502713396, + 0.1074975362491131, + 0.33861541235112874, + 0.1709021760408866, + 0.15501986722700223, + 0.04125109920749793, + 0.22416663332207437, + 0.13293963758897642, + 0.16706090010735694, + 0.2633412410246193, + null, + 0.13279483781082238, + 0.18839244551712866, + 0.16048494612723002, + 0.19701816039699668, + 0.2511863412938635, + 0.3882629837894673, + 0.13381744040558372, + null, + null, + 0.019045670095459852, + 0.17229048416336387, + null, + 0.15168035546199693, + 0.05244286845503128, + 0.010766703386595479, + 0.13533233099587644, + 0.0996162111076502, + 0.278372798532086, + -0.026245874694199853, + null, + 0.12423208822088339, + 0.11817885240343454, + null, + 0.04661165296003824, + null, + 0.09271348316804114, + 0.2642177663061739, + 0.2906261204413768, + 0.11184804070369375, + null, + 0.14335547194609016, + 0.08694988587991487, + 0.16729523987576567, + null, + 0.10260069879379614, + null, + 0.20157246045023838, + null, + 0.05793130419810435, + 0.23619419096822872, + null, + 0.155613285814553, + 0.21348569719009797, + null, + null, + 0.16952874423988748, + 0.19612659494149207, + 0.017119119233928923, + 0.11154200901876368, + null, + 0.143538153551836, + null, + 0.05752627801653035, + 0.29684334207483865, + null, + 0.1371270702967108, + 0.10551576837463435, + 0.1132692591463472, + null, + 0.2347348849012078, + 0.10494691767966499, + null, + 0.43164881935495414, + 0.3057411527012777, + 0.2933166056368512, + null, + null, + null, + null, + 0.19573274785446626, + null, + null, + 0.1410485667466124, + 0.1434540057099564, + 0.29026950517587335, + null, + 0.3126384430550862, + 0.1975515287183887, + null, + null, + 0.0582173532416801, + 0.026524296264065945, + 0.23288389729250725, + 0.036893163119795395, + 0.09621102820057582, + 0.058583420261252614, + 0.18807398182892918, + 0.13515292303732382, + 0.20522348338588234, + 0.06164143219942973, + 0.3189768348122162, + 0.18412753818333794, + null, + 0.11096710789239134, + 0.09877489083634373, + 0.15715428852679247, + 0.13644071659908674, + 0.07950895806048408, + 0.05846815024031309, + 0.06681050056568234, + 0.27967749452442736, + 0.19293132465113325, + 0.10741528994533103, + 0.048710573820134696, + 0.024124859388169176, + null, + 0.1608520573700804, + 0.25455228475388997, + 0.2567101347183399, + null, + 0.19729285365414015, + 0.16288778627350528, + null, + 0.17240475215407663, + null, + 0.24639818205434907, + 0.11502013851579192, + 0.19230157585348667, + 0.11552731269035153, + 0.28045938442695506, + 0.023857100311416718, + 0.07214020396317916, + null, + null, + 0.0439616665336178, + 0.043835378908051476, + null, + 0.22024422649570552, + null, + 0.29530980527948325, + 0.19268356409372933, + 0.130752658257148, + 0.10003777746787627, + null, + null, + null, + null, + 0.09199247307397457, + null, + 0.17168925680316569, + 0.1547324095415815, + null, + null, + 0.24052545319577706, + 0.11451722041210463, + 0.08497583245220451, + null, + 0.054983074077396644, + 0.19977506986001092, + 0.16548283649237092, + null, + 0.1636731360074751, + null, + null, + 0.14969985036222402, + null, + 0.09759783798737821, + null, + null, + 0.11617803098711633, + 0.0773563792126757, + 0.1674837405973848, + 0.15804596501530507, + 0.13677480759534216, + null, + 0.13685495959926838 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9238410596026491, + 0.944078947368421, + 0.9181494661921709, + 0.8955696202531646, + 0.9503311258278145, + 0.9355932203389831, + 0.9225352112676056, + 0.9340277777777778, + 0.9230769230769231, + 0.9219858156028369, + 0.9259259259259259, + 0.8983050847457628, + 0.9123376623376623, + 0.925, + 0.9366666666666666, + 0.930921052631579, + 0.9230769230769231, + 0.9294871794871795, + 0.9267515923566879, + 0.9215686274509803, + 0.9172661870503597, + 0.9258064516129032, + 0.9025974025974026, + 0.9368029739776952, + 0.9430604982206405, + 0.9407894736842105, + 0.9391025641025641, + 0.9491525423728814, + 0.9425675675675675, + 0.9275862068965517, + 0.9315068493150684, + 0.9255663430420712, + 0.9291044776119403, + 0.9240121580547113, + 0.9267399267399268, + 0.9211356466876972, + 0.916030534351145, + 0.9310344827586207, + 0.8951048951048951, + 0.9161073825503355, + 0.9215017064846417, + 0.9381443298969072, + 0.9236641221374046, + 0.937037037037037, + 0.9180327868852459, + 0.9290540540540541, + 0.9261992619926199, + 0.9139784946236559, + 0.9042553191489362, + 0.9319727891156463, + 0.9124087591240876, + 0.9264705882352942, + 0.9128787878787878, + 0.9285714285714286, + 0.917910447761194, + 0.9251700680272109, + 0.916955017301038, + 0.9503311258278145, + 0.9315068493150684, + 0.9119718309859155, + 0.9241379310344827, + 0.940983606557377, + 0.9337748344370861, + 0.9351535836177475, + 0.9362416107382551, + 0.9525316455696202, + 0.918918918918919, + 0.9530201342281879, + 0.9283276450511946, + 0.9509803921568627, + 0.9288256227758007, + 0.9354838709677419, + 0.9292929292929293, + 0.9470198675496688, + 0.942652329749104, + 0.9389830508474576, + 0.9453376205787781, + 0.939799331103679, + 0.9482758620689655, + 0.9421768707482994, + 0.9304635761589404, + 0.9430379746835443, + 0.9315068493150684, + 0.9320388349514563, + 0.9333333333333333, + 0.9287925696594427, + 0.9383561643835616, + 0.9444444444444444, + 0.9392857142857143, + 0.9272151898734177, + 0.9337979094076655, + 0.9312714776632303, + 0.9368421052631579, + 0.9217687074829932, + 0.9290540540540541, + 0.9548611111111112, + 0.9389830508474576, + 0.9204152249134948, + 0.9139072847682119, + 0.9335443037974683, + 0.9333333333333333, + 0.921311475409836, + 0.9359430604982206, + 0.9265175718849841, + 0.9488054607508533, + 0.94, + 0.9026217228464419, + 0.9364548494983278, + 0.9401993355481728, + 0.9256756756756757, + 0.9428571428571428, + 0.9261744966442953, + 0.9383116883116883, + 0.9288135593220339, + 0.9387755102040817, + 0.9290540540540541, + 0.9275862068965517, + 0.918918918918919, + 0.9350180505415162, + 0.9419795221843004, + 0.9285714285714286, + 0.9457627118644067, + 0.9301470588235294, + 0.944078947368421, + 0.9271255060728745, + 1.0, + 0.9411764705882353, + 0.9198606271777003, + 0.9356060606060606, + 0.9402985074626866, + 0.927536231884058, + 0.9344827586206896, + 0.9442508710801394, + 0.9210526315789473, + 0.9228070175438596, + 0.9297658862876255, + 1.0, + 0.9525423728813559, + 0.9328358208955224, + 0.9375, + 0.9197080291970803, + 0.914179104477612, + 0.9191176470588235, + 0.9172413793103448, + 0.9263157894736842, + 0.9198473282442748, + 0.9291338582677166, + 0.9409722222222222, + 0.9044117647058824, + 0.9363957597173145, + 0.9111111111111111, + 0.9326241134751773, + 0.9157509157509157, + 0.9285714285714286, + 0.9344827586206896, + 0.9381443298969072, + 0.9393939393939394, + 0.9386281588447654, + 0.9187279151943463, + 0.927007299270073, + 0.9204545454545454, + 0.9305555555555556, + 0.9153605015673981, + 0.9163763066202091, + 0.9283018867924528, + 0.920863309352518, + 0.9273356401384083, + 0.9437086092715232, + 0.9320754716981132, + 0.9403508771929825, + 0.9304029304029304, + 0.9180887372013652, + 0.9473684210526315, + 0.9417808219178082, + 0.9346153846153846, + 0.9377431906614786, + 0.9130434782608695, + 0.9198606271777003, + 0.9191176470588235, + 0.9265734265734266, + 0.9337979094076655, + 0.9365671641791045, + 0.9188191881918819, + 0.921146953405018, + 0.9401408450704225, + 0.9630996309963099, + 0.9212328767123288, + 0.9057239057239057, + 0.910394265232975, + 0.9442622950819672, + 0.9201388888888888, + 0.9375, + 0.9142857142857143, + 0.9014084507042254, + 0.9094076655052264, + 0.9081272084805654, + 0.8933823529411765, + 0.9375, + 0.9281045751633987, + 0.9233333333333333, + 1.0, + 0.9276315789473685, + 0.9050847457627119, + 0.920265780730897, + 0.9222972972972973, + 0.932475884244373, + 0.9293286219081273, + 0.9046052631578947, + 0.9915966386554622, + 1.0, + 0.9207920792079208, + 0.9253246753246753, + 1.0, + 0.903448275862069, + 0.9120879120879121, + 0.9124087591240876, + 0.9112627986348123, + 0.9072847682119205, + 0.903448275862069, + 0.9161073825503355, + 0.983739837398374, + 0.9192546583850931, + 0.9106529209621993, + 1.0, + 0.8984375, + 1.0, + 0.9272727272727272, + 0.9263157894736842, + 0.9225352112676056, + 0.9158249158249159, + 1.0, + 0.9202898550724637, + 0.926056338028169, + 0.9087719298245615, + 0.9921259842519685, + 0.9090909090909091, + 1.0, + 0.8993506493506493, + 0.9915254237288136, + 0.92, + 0.9228187919463087, + 1.0, + 0.9219330855018587, + 0.9180327868852459, + 1.0, + 0.9921875, + 0.9148936170212766, + 0.9215686274509803, + 0.9019607843137255, + 0.898989898989899, + 1.0, + 0.9045936395759717, + 0.9844961240310077, + 0.9025974025974026, + 0.9175257731958762, + 0.9930555555555556, + 0.926923076923077, + 0.9131832797427653, + 0.92, + 1.0, + 0.9035714285714286, + 0.919732441471572, + 1.0, + 0.921311475409836, + 0.8982456140350877, + 0.9152542372881356, + 0.9925373134328358, + 1.0, + 1.0, + 1.0, + 0.9069767441860465, + 1.0, + 1.0, + 0.9183673469387755, + 0.9219858156028369, + 0.9166666666666666, + 0.9916666666666667, + 0.9078498293515358, + 0.9143835616438356, + 1.0, + 1.0, + 0.9163763066202091, + 0.9239130434782609, + 0.9230769230769231, + 0.8986486486486487, + 0.9010238907849829, + 0.9115646258503401, + 0.9180887372013652, + 0.8996539792387543, + 0.9016949152542373, + 0.9039145907473309, + 0.9203821656050956, + 0.9114391143911439, + 1.0, + 0.9157894736842105, + 0.918918918918919, + 0.8964285714285715, + 0.9259259259259259, + 0.9251700680272109, + 0.9270833333333334, + 0.9228070175438596, + 0.906896551724138, + 0.9114754098360656, + 0.918918918918919, + 0.9311594202898551, + 0.931740614334471, + 0.991869918699187, + 0.9260450160771704, + 0.918918918918919, + 0.9262820512820513, + 1.0, + 0.9169329073482428, + 0.9283154121863799, + 0.9917355371900827, + 0.9278350515463918, + 1.0, + 0.9134615384615384, + 0.9166666666666666, + 0.9075907590759076, + 0.9347826086956522, + 0.9217687074829932, + 0.9212328767123288, + 0.916083916083916, + 0.9929078014184397, + 1.0, + 0.9052287581699346, + 0.9081632653061225, + 1.0, + 0.9061371841155235, + 1.0, + 0.9102990033222591, + 0.9026845637583892, + 0.9054545454545454, + 0.9169435215946844, + 0.9928571428571429, + 1.0, + 1.0, + 1.0, + 0.9075342465753424, + 0.9927007299270073, + 0.9273356401384083, + 0.9365079365079365, + 1.0, + 1.0, + 0.9295774647887324, + 0.9102990033222591, + 0.9259259259259259, + 1.0, + 0.9178082191780822, + 0.9225589225589226, + 0.8971631205673759, + 1.0, + 0.945859872611465, + 1.0, + 1.0, + 0.931740614334471, + 0.9918032786885246, + 0.9158576051779935, + 1.0, + 1.0, + 0.9227799227799228, + 0.922077922077922, + 0.9175257731958762, + 0.9209621993127147, + 0.9119718309859155, + 1.0, + 0.9105960264900662, + 0.9887218045112782, + 0.9824561403508771, + 0.9891304347826086, + 0.9821428571428571, + 0.9936708860759493, + 0.9893238434163701, + 0.9860627177700348, + 0.9811320754716981, + 0.9829351535836177, + 0.9784172661870504, + 0.9930795847750865, + 0.9783393501805054, + 0.9828767123287672, + 0.9897260273972602, + 0.9851301115241635, + 0.9853479853479854, + 0.971830985915493, + 0.9777777777777777, + 0.9928571428571429, + 0.9863481228668942, + 0.972027972027972, + 0.9858657243816255, + 0.9832775919732442, + 0.981203007518797, + 0.9892857142857143, + 0.9927272727272727, + 0.9834437086092715, + 0.9895470383275261, + 0.985663082437276, + 0.9841897233201581, + 0.9824561403508771, + 0.9889705882352942, + 0.9798387096774194, + 0.975177304964539, + 0.9853479853479854, + 0.9822695035460993, + 0.9840637450199203, + 0.9680851063829787, + 0.9825174825174825, + 0.9820143884892086, + 0.9788732394366197, + 0.9927007299270073, + 0.9929328621908127, + 0.9882352941176471, + 0.979381443298969, + 0.9928057553956835, + 0.9849624060150376, + 0.9847328244274809, + 0.9886792452830189, + 0.9769230769230769, + 0.9879518072289156, + 0.9851301115241635, + 0.9848484848484849, + 0.9822064056939501, + 0.981203007518797, + 0.9851301115241635, + 0.9856115107913669, + 0.985663082437276, + 0.9786476868327402, + 0.9818840579710145, + 0.9790940766550522, + 0.9850187265917603, + 0.9863481228668942, + 0.989010989010989, + 0.9855595667870036, + 0.9816176470588235, + 0.9821428571428571, + 0.9806949806949807, + 0.9809885931558935, + 0.985663082437276, + 0.9854014598540146, + 0.9746376811594203, + 0.9864406779661017, + 0.978494623655914, + 0.989247311827957, + 0.9855072463768116, + 0.9862068965517241, + 0.9823321554770318, + 0.9856115107913669, + 0.9849056603773585, + 0.9792387543252595, + 0.9794520547945206, + 0.9885931558935361, + 0.9814814814814815, + 0.9822695035460993, + 0.9889705882352942, + 0.9847328244274809, + 0.9861111111111112, + 0.9744525547445255, + 0.978021978021978, + 0.9825783972125436, + 0.9857651245551602, + 0.9896551724137931, + 0.9739776951672863, + 0.9754385964912281, + 0.989247311827957, + 0.9757785467128027, + 0.9776951672862454, + 0.9863013698630136, + 0.985663082437276, + 0.9770114942528736, + 0.9854014598540146, + 0.9855072463768116, + 0.9857142857142858, + 0.9818840579710145, + 0.9930555555555556, + 0.9811320754716981, + 0.9925925925925926, + 0.9754385964912281, + 0.9893238434163701, + 0.9788732394366197, + 0.9887640449438202, + 0.9801324503311258, + 0.978494623655914, + 0.982078853046595, + 0.9923076923076923, + 0.9752650176678446, + 0.9730769230769231, + 0.9762845849802372, + 0.981549815498155, + 0.9894736842105263, + 0.982078853046595, + 0.9743589743589743, + 0.9893992932862191, + 0.9888475836431226, + 1.0, + 0.975177304964539, + 0.9857142857142858, + 0.9716312056737588, + 0.9762845849802372, + 0.9783393501805054, + 0.9895104895104895, + 0.9893992932862191, + 0.981203007518797, + 0.9895470383275261, + 0.9847328244274809, + 1.0, + 0.975609756097561, + 0.9816176470588235, + 0.9790209790209791, + 0.9895470383275261, + 0.9814126394052045, + 0.9858156028368794, + 0.9790209790209791, + 0.9787234042553191, + 0.9851301115241635, + 0.9850746268656716, + 0.9812734082397003, + 0.9819494584837545, + 0.9853479853479854, + 0.9807692307692307, + 0.977859778597786, + 0.982078853046595, + 0.985239852398524, + 0.9924528301886792, + 0.9896551724137931, + 0.9853479853479854, + 0.9774436090225563, + 0.9797297297297297, + 0.9851851851851852, + 0.9855072463768116, + 0.9853479853479854, + 0.9902912621359223, + 0.9790209790209791, + 0.9850187265917603, + 0.9763779527559056, + 0.9863013698630136, + 0.989247311827957, + 0.9877551020408163, + 0.9739776951672863, + 0.9818840579710145, + 0.9703703703703703, + 0.9806201550387597, + 0.9893992932862191, + 0.988929889298893, + 0.991869918699187, + 0.9787234042553191, + 0.9785714285714285, + 0.986013986013986, + 0.9890510948905109, + 0.9790940766550522, + 0.9851301115241635, + 0.9852941176470589, + 0.978021978021978, + 0.9647887323943662, + 0.9958847736625515, + 0.9814126394052045, + 0.9820143884892086, + 0.9753521126760564, + 0.9821428571428571, + 0.9862542955326461, + 0.981549815498155, + 0.9929328621908127, + 0.977859778597786, + 0.967741935483871, + 0.9784172661870504, + 0.9788732394366197, + 0.9730769230769231, + 0.9903225806451613, + 0.9897610921501706, + 1.0, + 0.9858156028368794, + 0.9865319865319865, + 0.9701492537313433, + 0.989010989010989, + 0.9863013698630136, + 0.9897959183673469, + 0.9823321554770318, + 1.0, + 1.0, + 0.9780564263322884, + 0.990228013029316, + 0.992, + 0.9789473684210527, + 0.9792387543252595, + 0.9807692307692307, + 0.9863481228668942, + 0.9790940766550522, + 0.9857651245551602, + 0.9823321554770318, + 1.0, + 0.985663082437276, + 0.9888888888888889, + 1.0, + 0.9738805970149254, + 0.992, + 0.976271186440678, + 0.9857651245551602, + 0.9898648648648649, + 0.9824561403508771, + 1.0, + 0.978021978021978, + 0.9824561403508771, + 0.9852941176470589, + 1.0, + 0.9750889679715302, + 1.0, + 0.9858657243816255, + 0.9841269841269841, + 0.9812734082397003, + 0.9781818181818182, + 1.0, + 0.9854545454545455, + 0.985663082437276, + 0.9925925925925926, + 1.0, + 0.9827586206896551, + 0.9858657243816255, + 0.9802631578947368, + 0.9831649831649831, + 1.0, + 0.9766536964980544, + 1.0, + 0.9822695035460993, + 0.9850746268656716, + 1.0, + 0.974169741697417, + 0.972027972027972, + 0.9776951672862454, + 1.0, + 0.9782608695652174, + 0.9859154929577465, + 0.9915966386554622, + 0.9897610921501706, + 0.9933110367892977, + 0.992831541218638, + 0.9926470588235294, + 1.0, + 1.0, + 0.990990990990991, + 0.9896193771626297, + 1.0, + 1.0, + 0.9787234042553191, + 0.9897959183673469, + 0.9831081081081081, + 1.0, + 0.9895104895104895, + 0.9867549668874173, + 1.0, + 0.992, + 0.9896193771626297, + 0.979381443298969, + 0.9854545454545455, + 0.976897689768977, + 0.9859649122807017, + 0.9817518248175182, + 0.9742647058823529, + 0.9863945578231292, + 0.9735849056603774, + 0.9811320754716981, + 0.9898305084745763, + 0.975177304964539, + 0.9917355371900827, + 0.9826388888888888, + 0.982078853046595, + 0.9829931972789115, + 0.9859649122807017, + 0.9825174825174825, + 0.9895104895104895, + 0.9794520547945206, + 0.9853479853479854, + 0.9798657718120806, + 0.9823321554770318, + 0.9777777777777777, + 0.9818181818181818, + 0.9921259842519685, + 0.9821428571428571, + 0.9928571428571429, + 0.9933333333333333, + 1.0, + 0.9792387543252595, + 0.9765886287625418, + 1.0, + 0.9836065573770492, + 1.0, + 0.9900332225913622, + 0.9857651245551602, + 0.9831081081081081, + 0.9725490196078431, + 0.9897260273972602, + 0.9738805970149254, + 0.973384030418251, + 1.0, + 0.983739837398374, + 0.9788732394366197, + 0.967741935483871, + 1.0, + 0.9893238434163701, + 1.0, + 0.9829351535836177, + 0.986159169550173, + 0.9790209790209791, + 0.9852941176470589, + 0.9854014598540146, + 1.0, + 1.0, + 1.0, + 0.9859154929577465, + 1.0, + 0.9730769230769231, + 0.9865319865319865, + 1.0, + 0.9923076923076923, + 0.992831541218638, + 0.9824561403508771, + 0.9821428571428571, + 1.0, + 0.985663082437276, + 0.9895470383275261, + 0.9829351535836177, + 1.0, + 0.9935064935064936, + 1.0, + 1.0, + 0.9745454545454545, + 0.9916666666666667, + 0.9849056603773585, + 1.0, + 1.0, + 0.977859778597786, + 0.9757785467128027, + 0.9901315789473685, + 0.9825174825174825, + 0.976271186440678, + 1.0, + 0.9898305084745763 + ] + }, + "contact_maps": { + "mean_pearson": 0.019487273472909843, + "valid_tracks": 28, + "per_track_pearson": [ + 0.04081458015383671, + -0.036711409694901116, + 0.043084998546684676, + 0.01576171494300664, + -0.002445929120946461, + -0.0007610104612738161, + 0.006491668301607098, + 0.03212697937623195, + 0.08461809480986397, + 0.025972267093992755, + -0.0032193826660228187, + 0.053647996272669494, + 0.08124321378131096, + 0.025786206925746125, + 0.021896809034206245, + -0.011728535355311846, + 0.013000685370987948, + 0.05535391310488641, + -0.06686778615406057, + 0.041402546717417915, + 0.010961894078980893, + -0.025528232778043138, + -0.021206278231665276, + 0.05060674379984273, + 0.0594456626100659, + -0.0672671181282852, + 0.030098420559475286, + 0.08906494435117208 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.25681125891165696, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..525eec8738cb03afc861304e15552a7c6cac8cd4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S005/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.39635258936365514, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6965235643807073, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.1778776917280187, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7403571967826894, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.25762534461408215, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.26157671283777334, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2667099562067121, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.1633509969252289, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.2635279373317317, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.19721844991758614, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5022219249141129, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.16334218499410608, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.18509957659072374, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..52f3decd4fe2e9d275d08606b74575fafb701434 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/predict.py @@ -0,0 +1,130 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f9a3298d69b138a02fed00880182d4004b23327 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_config.yaml @@ -0,0 +1,28 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..2b40949199289aa3ddeddf4f8c91ffc14564ad13 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/training_provenance.json @@ -0,0 +1,66 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 0, + "offset": 22000, + "global_step": 22000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/local01", + "config": "/workspace/research/experiments/local01/config.yaml", + "resume": "/workspace/research/experiments/local01/step-018000.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 82040, + "workers": 6, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 3284.5674808800686, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..f53522ce1120d6c4dc5e9c9765b65fdb7dfedc87 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S006/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.4731655295963023, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6013286459601276, + 0.6111824037180612, + 0.37147828411906514, + 0.5510960940383585, + 0.3628910495098787, + 0.35651735457862743, + 0.3698694041456809, + 0.44516546955876274, + 0.49476822118825464, + 0.5486942994134172, + 0.47744487588850487, + 0.4171275953559535, + 0.27713151451750306, + 0.4122200020391918, + 0.5987517471586723, + 0.37675755430758534, + 0.4488173869299511, + 0.39123317111062045, + 0.44349256000001375, + 0.4299019006095824, + 0.39734091884287176, + 0.35548137688671233, + 0.4085280987680093, + 0.4255180785743437, + 0.3706661599136409, + 0.38713392309872113, + 0.37280718551353775, + 0.39863764060397694, + 0.404517342277504, + 0.3922603628550291, + 0.39144135654205847, + 0.3964485039221974, + 0.3837704941584931, + 0.2956368167374332, + 0.2707579846345881, + 0.4579926528553293, + 0.3863325643442393, + 0.506680594428396, + 0.40361673379936847, + 0.3766433187512069, + 0.4572472431164581, + 0.41797957001666863, + 0.3940699928828094, + 0.35075609667109076, + 0.45561139888994945, + 0.4018003700570312, + 0.3166162432963891, + 0.36738858794530455, + 0.3573874004100285, + 0.36712814640238445, + 0.396734897651804, + 0.3213784667137442, + 0.369206112868748, + 0.510818488433694, + 0.34428721330864037, + 0.6078935550759434, + 0.7185634911771055, + 0.4310656420242184, + 0.5222929114947963, + 0.5154030872329963, + 0.6272468211890894, + 0.3911571412292211, + 0.6001598636370123, + 0.5246759655524275, + 0.5515704562991312, + 0.3966391569027745, + 0.4657835095808229, + 0.5514588347635233, + 0.7009054069001606, + 0.6313073353116716, + 0.4246621634657868, + 0.3564446767644761, + 0.3106759083499761, + 0.4304644880858005, + 0.4102193112895741, + 0.3575501878114184, + 0.4032892318204387, + 0.4321223256126798, + 0.47424995004527415, + 0.37389831125368184, + 0.4333058467441242, + 0.46705779029997263, + 0.4118436662651715, + 0.3864681372286581, + 0.3839318019344053, + 0.3816427700317667, + 0.3415101022223562, + 0.36356345232533566, + 0.34998566012708854, + 0.407222975749994, + 0.4020230789359966, + 0.3969884665824216, + 0.3911083378533324, + 0.4136390653427931, + 0.477217407017503, + 0.3003236540806991, + 0.3383754225903856, + 0.31141737918056295, + 0.3778597380209459, + 0.40697340526553416, + 0.3417907538673221, + 0.36977980621885415, + 0.39781612660622523, + 0.3698551247841001, + 0.40829584777450284, + 0.42376999175356345, + 0.44126837584172907, + 0.41559243130239737, + 0.4173147697647144, + 0.4261325538489726, + 0.4541877392116222, + 0.4170613504874718, + 0.40356402432434124, + 0.34895808814000046, + 0.359560915175142, + 0.34655641894896244, + 0.3656142057744852, + 0.3897769084119796, + 0.3861160514786089, + 0.3740896099571337, + 0.5818377230968613, + 0.5636346325072724, + 0.6319969858831194, + 0.6470610796325881, + 0.6175647026411765, + 0.5105509861368986, + 0.7827635102107477, + 0.6167621747532082, + 0.6768347964728743, + 0.6000744358353137, + 0.6501473605330443, + 0.48700993314094143, + 0.5210474463817522, + 0.4921579021428993, + 0.6254225119688397, + 0.6251234415662205, + 0.6119176048883712, + 0.4546359936771379, + 0.36740761442230746, + 0.726151223917174, + 0.6920541304431157, + 0.7071748542080367, + 0.7017091273424998, + 0.5234260705528297, + 0.777042940826962, + 0.6897439502752724, + 0.6258569119594449, + 0.6345201784796692, + 0.7738672663024522, + 0.721643395154349, + 0.5457151797836002, + 0.5398115850430047, + 0.6286745232920614, + 0.6581945075104377, + 0.6630774689944396, + 0.5334532785510296, + 0.6544117570635666, + 0.5113267632557071, + 0.5323481087089901, + 0.6075431075386385, + 0.4882948558372131, + 0.6816119598214316, + 0.7039960282398324, + 0.5480347037942414, + 0.5327962293470551, + 0.4881489266086036, + 0.6583127191778553 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5972120538847244, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5051412390757164, + 0.9039791216211878, + 0.7599377040037648, + 0.4582112726624527, + 0.41094878600143886, + 0.42829203921786346, + 0.9273658984707428, + 0.6844153183530971, + 0.36179413418385664, + 0.29918772745479727, + 0.8596335653722132, + 0.6254443533490435, + 0.3834672779755635, + 0.44757138297183013, + 0.4277844076130149, + 0.779224825642009, + 0.7621141926702011, + 0.8376612783216764, + 0.834986446119553, + 0.8703883244037185, + 0.5956900555145016, + 0.5636597786905543, + 0.885186796826046, + 0.4436843152114603, + 0.4667632507736149, + 0.7560849480973113, + 0.40886101433876354, + 0.6043441968620606, + 0.4416351333949014, + 0.4508545453255306, + 0.5606128923372493, + 0.38992509680487375, + 0.825170169381775, + 0.6096367531891844, + 0.8676458458758008, + 0.874100989070171, + 0.3935263966534776, + 0.8105204930914245, + 0.8649452914950271, + 0.8453967347750222, + 0.8054081241387577, + 0.6552357974331766, + 0.690329024296813, + 0.5941260351187171, + 0.41754810011121934, + 0.6186859502760247, + 0.715623819163911, + 0.48028380418196753, + 0.5477689533582968, + 0.3647462074055072, + 0.32421200808679645, + 0.5135377361149885, + 0.4423097182696443, + 0.8176030007926326, + 0.37499636202336106, + 0.3680322237323357, + 0.3874208246824089, + 0.4179007313621277, + 0.38186753091970066, + 0.48904328510297645, + 0.36495262285049174, + 0.5790558451611819, + 0.4254068705074563, + 0.4117252168924671, + 0.43793204254636614, + 0.4343388638979915, + 0.49825625049206, + 0.4145707815950944, + 0.3341347126931504, + 0.3651450203238988, + 0.47765287711024035, + 0.5802841868652213, + 0.42225912392131143, + 0.568861547374955, + 0.28738480569997277, + 0.7918185755911138, + 0.8824428952010261, + 0.6108636763109566, + 0.3853403508521573, + 0.3978045654593289, + 0.4430023749844547, + 0.42069642568656684, + 0.4483563097772634, + 0.4208178554126363, + 0.46254806420978767, + 0.442776663459555, + 0.5292958262632764, + 0.4565903576515068, + 0.5943844751273505, + 0.38334919111226007, + 0.577490841240051, + 0.37546495987885975, + 0.6962934834341986, + 0.4190111328204881, + 0.6844195429114894, + 0.4544393681097317, + 0.6900083177559112, + 0.6428545355115505, + 0.2932375004568467, + 0.3008917476724384, + 0.41261076470544567, + 0.332510242012974, + 0.3107643385464482, + 0.6589872416278881, + 0.6309038183073117, + 0.3847121038984889, + 0.4045514831846905, + 0.5475215898169992, + 0.8277811980670334, + 0.5131103726357608, + 0.48020697592334743, + 0.5362245981202053, + 0.4486474203569618, + 0.411420906044964, + 0.7553254130217688, + 0.48373438678252745, + 0.45499151475348676, + 0.5156150184655025, + 0.5651508709188512, + 0.41020255569002045, + 0.4601273461996575, + 0.4633353571275484, + 0.492235580691646, + 0.4697370967709789, + 0.5614821028475304, + 0.4895725613683395, + 0.5387580981309158, + 0.6476642649037287, + 0.6705726297391721, + 0.5468391822289532, + 0.4265546130981197, + 0.5529301720449187, + 0.5803350870547929, + 0.7651136877045289, + 0.3988971113601048, + 0.27426388276296265, + 0.5294349377687377, + 0.4359587324520897, + 0.5671490961157516, + 0.5443052155393326, + 0.5499770590251257, + 0.506278909889507, + 0.4108096175553566, + 0.7554074354346134, + 0.5582910275068428, + 0.7414402199661823, + 0.3535973508441246, + 0.3248733359507268, + 0.3704222269092111, + 0.29420169389891593, + 0.8813016453616442, + 0.2843048379176497, + 0.8807537955841911, + 0.6582895088161341, + 0.44509501597784007, + 0.30107271995255097, + 0.3133443875452178, + 0.33305979824350995, + 0.6050993365002446, + 0.6478689848490811, + 0.6699778739183201, + 0.5936285015347443, + 0.4837558650512324, + 0.4158705829654101, + 0.6421280803655025, + 0.5275716372282644, + 0.4732146459922996, + 0.5577656132042892, + 0.41857152534076203, + 0.42762638377865686, + 0.33847479284800464, + 0.8162857933820427, + 0.3817186509184056, + 0.4131026590469168, + 0.6111399658229307, + 0.6669123272047408, + 0.41982770343204245, + 0.3062049989188756, + 0.866776368363699, + 0.3365138304591387, + 0.7377727459044833, + 0.3811573621504486, + 0.5591908558600549, + 0.6774079486557671, + 0.4320148675530522, + 0.521530320995965, + 0.5248119592765526, + 0.8009268883854046, + 0.3938839554456271, + 0.567962670836628, + 0.5032693572056718, + 0.6813592443092868, + 0.8841138510640218, + 0.35065326105621997, + 0.5325385926503946, + 0.404327334586611, + 0.578549211741897, + 0.6728678261876577, + 0.5341955283465095, + 0.3985313518685662, + 0.8564665625457476, + 0.8651800468443056, + 0.7397591403656755, + 0.8193328474863029, + 0.8297919918592871, + 0.8177006284577808, + 0.8282923501212779, + 0.3188904919607142, + 0.3287411355452429, + 0.46205314850599305, + 0.41500255941447667, + 0.3948526655545259, + 0.5563036608382869, + 0.846872454755895, + 0.8358028985055524, + 0.6294714172283105, + 0.8090833648623682, + 0.8059721915537695, + 0.8430174276599792, + 0.8913813624414137, + 0.434541639473131, + 0.4874306996782043, + 0.6012502956683935, + 0.5025988731827913, + 0.7997822657554415, + 0.514748969599269, + 0.8921184003180869, + 0.8668551805705957, + 0.8600036655714399, + 0.8876254341212961, + 0.8803522940367446, + 0.8540824822706901, + 0.8129707399344436, + 0.846319335975496, + 0.8905531352193478, + 0.3809258829340656, + 0.9103380560897857, + 0.8642636844749527, + 0.8036772321510267, + 0.5382003825955652, + 0.8594214098312858, + 0.46010679894899037, + 0.7375744137279391, + 0.8613390805959167, + 0.5154748174086311, + 0.502359437654293, + 0.6063061560219359, + 0.4822543778527146, + 0.4760833808746472, + 0.7006351908890982, + 0.8672137881391057, + 0.6524705329696113, + 0.7882152346330937, + 0.557944704983605, + 0.8473618823860584, + 0.789147227181865, + 0.8210665643046198, + 0.8892107024369943, + 0.820115914376293, + 0.8849731209376374, + 0.8540924603798358, + 0.4797414728427978, + 0.8646259606000696, + 0.4852414251489208, + 0.46852934688615977, + 0.8675912085475542, + 0.8935625342943153, + 0.8646353802150993, + 0.5065703364954367, + 0.5526621425648127, + 0.6219667563431475, + 0.7798532087272525, + 0.8717203133488263, + 0.6590581961785495, + 0.8883675541656951, + 0.9007846205682084, + 0.5363952742782602, + 0.2722477561601995, + 0.4684873079057694, + 0.8867299278528875, + 0.3663900987539562, + 0.549288913254928, + 0.8652438520658882, + 0.8153115995909519, + 0.5480926912700991, + 0.8526184111623369, + 0.8811800486705268, + 0.8753701130511087, + 0.8625882957691333, + 0.8505380653657254, + 0.8808779642936991, + 0.8377954691697456, + 0.9125376683089116, + 0.7683906965771488, + 0.8179589343370908, + 0.8972106430702682, + 0.49048606009902274, + 0.4974106901411681, + 0.47032027227964623, + 0.4602693919762573, + 0.8954821418675448, + 0.7800245708431833, + 0.8345809802840655, + 0.772389054690256, + 0.7440804471997101 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.3495633665373259, + "valid_tracks": 12, + "per_track_pearson": [ + 0.3425533541709063, + 0.3844043401138221, + 0.29199809937071514, + 0.3828719891697744, + 0.3794390714674845, + 0.381257182776449, + 0.3191611654168779, + 0.3690012243547561, + 0.25958703988924603, + 0.3622608502102254, + 0.35419721449430486, + 0.3680288670133492 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.19304508881650334, + "valid_tracks": 546, + "per_track_pearson": [ + 0.19963743222710378, + 0.2245151515192077, + 0.1905717877013149, + 0.11964248765098394, + 0.07152139261661361, + 0.20022819351560842, + 0.10106867153842654, + 0.1290257653113182, + 0.20798668361581507, + 0.1830363479663043, + 0.23181480374603183, + 0.12942730770467314, + 0.1329226942512388, + 0.1119747235267625, + 0.08085254604325375, + 0.15596375894549722, + 0.19040744523181877, + 0.22857310728786343, + 0.22359006305287013, + 0.21473128523548987, + 0.18212162110426455, + 0.17040126360568592, + 0.28483781868228264, + 0.1529631904095447, + 0.16180473401536394, + 0.1970944780904702, + 0.219256548728304, + 0.2227465584753554, + 0.2440023074515507, + 0.22365435150799345, + 0.1406612493425488, + 0.23494790911536337, + 0.16874516482976126, + 0.11733902513726399, + 0.24403061468319157, + 0.23482720756887082, + 0.21427616877248595, + 0.24242749635540378, + 0.20547741935152458, + 0.27880972865661247, + 0.28590929996093795, + 0.2472120040661485, + 0.21589706072807838, + 0.2118997737478342, + 0.12233607363539888, + 0.1763457301361941, + 0.18906741444835676, + 0.18812978007185852, + 0.1158625307602565, + 0.1480810316254365, + 0.20775290474403313, + 0.24425267506334164, + 0.2533875550355919, + 0.17333667660740007, + 0.2623149950364797, + 0.2889024685062696, + 0.1570654136871655, + 0.26755026636057483, + 0.20659965541319833, + 0.22905047697851189, + 0.2692611255493315, + 0.16406043901250753, + 0.240385072073791, + 0.1822525353955715, + 0.2541784342782522, + 0.26326053197534915, + 0.27889796720242344, + 0.16191239255661558, + 0.11640381493365784, + 0.10168532370698853, + 0.21650090327047475, + 0.2527586319652388, + 0.20843710031350093, + 0.22100748163448214, + 0.23995464007225392, + 0.20171968767240558, + 0.24159228369746347, + 0.21924575702950624, + 0.24350467771671797, + 0.24968535471685943, + 0.20581833051233842, + 0.1242262305456049, + 0.22059979057074525, + 0.2555231329849282, + 0.1238302562208637, + 0.24043279390417494, + 0.19690780724407506, + 0.25958309288505066, + 0.1646564556300625, + 0.2484155555634404, + 0.1782991062298271, + 0.20676256363205417, + 0.2287167301619044, + 0.1071025466199493, + 0.23802996967272044, + 0.20746141266119045, + 0.21865485090972858, + 0.231721982949357, + 0.21480639501723567, + 0.13164403197117963, + 0.2058683285985062, + 0.22408179098401737, + 0.22526439865839865, + 0.3069749996222225, + 0.2609961831153163, + 0.2722388659652408, + 0.17849372183258055, + 0.23193844631233212, + 0.1824615126169958, + 0.18382354117821031, + 0.20063956867027746, + 0.2018900224196736, + 0.22551772023714597, + 0.2234355308878649, + 0.17583778577584194, + 0.2212477468202168, + 0.22721034768789844, + 0.23306680172650548, + 0.2165855409016279, + 0.2514003314348282, + 0.25546836859226496, + 0.2517415918238057, + 0.15517145759938936, + 0.1708085390561727, + 0.1971338829640894, + 0.2443790067923984, + 0.21892046737991736, + 0.2412631119880329, + 0.31248343832985903, + 0.22812696518068007, + 0.2530451442308502, + 0.17863893096265313, + 0.24150781844679148, + 0.2442813253218422, + 0.28924263951710616, + 0.21131383227400916, + 0.2592095427657572, + 0.18149260278135318, + 0.10938880824478528, + 0.223665822691819, + 0.26981317268017013, + 0.22782262199895284, + 0.23966309024329413, + 0.19463606397808156, + 0.19661811920409347, + 0.20968159984376222, + 0.21746683505839598, + 0.2569697178614046, + 0.18931606082231153, + 0.18458512773621244, + 0.21879100168019727, + 0.16334760047593722, + 0.20438481638841896, + 0.23082828646117276, + 0.20570604909786092, + 0.14335485379481838, + 0.23435929688263094, + 0.2663390497324989, + 0.1722077859665428, + 0.13963256641031505, + 0.138458916139623, + 0.19661616127914874, + 0.18942267996996628, + 0.21676040674351088, + 0.19047963866110132, + 0.15314802042319106, + 0.21481403808741975, + 0.1467082159401736, + 0.1421755901994135, + 0.23710011645705883, + 0.21743434379111273, + 0.17035639225919696, + 0.19105641493499012, + 0.29457625839167884, + 0.2189078225747126, + 0.1351844551085417, + 0.1375893934874893, + 0.17304140673233123, + 0.22643265689507708, + 0.2674098233462751, + 0.25324687846295557, + 0.2558465690392431, + 0.11932852826270066, + 0.2540167813748298, + 0.20688217623705638, + 0.17012147095073318, + 0.17357670674017198, + 0.14705408152259006, + 0.27114203201565074, + 0.22758360990940454, + 0.18775562155677583, + 0.11823677356388777, + 0.12210036369962801, + 0.19811458868059706, + 0.19418184121973792, + 0.17901276337282448, + 0.20585180323943741, + 0.20782536935938553, + 0.15651539898292433, + 0.19758483733081295, + 0.1144250353652124, + 0.09179152362130627, + 0.10802719570948492, + 0.09841356381969758, + 0.12177050245512142, + 0.20240703348245884, + 0.19834031306445177, + 0.17088818581243326, + 0.10430234691178651, + 0.1955246908091444, + 0.25548559370241464, + 0.2177072780256793, + 0.2227198283447376, + 0.2215703308569964, + 0.20600354469743445, + 0.2022231102854161, + 0.23842422718887163, + 0.27160485888466, + 0.24169628339275126, + 0.2319127988735446, + 0.2106206718886003, + 0.15279371102740624, + 0.21531403288350967, + 0.21555879191656235, + 0.22518938055888008, + 0.2792215994367789, + 0.20645719870040782, + 0.17656976139214983, + 0.20117754038019917, + 0.21465794484943715, + 0.1597863172466666, + 0.16108603433815444, + 0.26115476295809953, + 0.20552129532496072, + 0.17366294235612487, + 0.20689065235013843, + 0.27196109661967227, + 0.14013787702394193, + 0.21495774503083007, + 0.2128060751200572, + 0.10063811690248212, + 0.14238965293177716, + 0.12626551187473947, + 0.2048367687813779, + 0.15354894624786145, + 0.18834203929715876, + 0.08476448539120633, + 0.25929774534832406, + 0.221649886859605, + 0.22761626620860356, + 0.24350363885923498, + 0.15392756162944213, + 0.27022236909555153, + 0.1221556993595787, + 0.2451123252359195, + 0.1905852256647855, + 0.3084006310503463, + 0.22512904647519594, + 0.22928673234909197, + 0.3339129817672829, + 0.19181784929021706, + 0.21617352312659852, + 0.11543380848941028, + 0.2312831970409721, + 0.2141613793539884, + 0.14928626612156823, + 0.11684626184586366, + 0.11965478541048111, + 0.1407618430273847, + 0.22368118068795853, + 0.273705640317898, + 0.12731849939667383, + 0.1096504116673761, + 0.17127488765657653, + 0.2332660388014542, + 0.1806490202656308, + 0.12482055977844757, + 0.06065954595572302, + 0.1886357685472996, + 0.11314098447807278, + 0.12601773352827988, + 0.20077253076326246, + 0.1722382249214084, + 0.2102782323304441, + 0.12189169557626903, + 0.1303490143151533, + 0.1129623041560028, + 0.10242927293865052, + 0.14510154485369764, + 0.17931439902735155, + 0.2056631825340719, + 0.20487091136591964, + 0.19472576416903198, + 0.17190007618288208, + 0.16558422748274895, + 0.26182371091973217, + 0.14206682716492103, + 0.1486677975387749, + 0.19911100922732988, + 0.19966343396539904, + 0.20082292862716586, + 0.2195123086683878, + 0.21022425651965784, + 0.12780912012855905, + 0.20346930938987307, + 0.1566719894506183, + 0.10883637925349288, + 0.21937556889305315, + 0.227006708810247, + 0.2090483415084892, + 0.21647457115165333, + 0.18448811274469015, + 0.25474259643421676, + 0.25550653208242724, + 0.2242723531408731, + 0.20620877894261877, + 0.19673229678257784, + 0.11349107276212253, + 0.16839689798699822, + 0.17939923502355357, + 0.19353139150322304, + 0.10504393044415064, + 0.14030885170480326, + 0.17949849215046546, + 0.23229386913670522, + 0.2334710089144904, + 0.16422776142227974, + 0.23517594026281116, + 0.25770593144308224, + 0.15624693681310023, + 0.2330981537958757, + 0.1867827254867687, + 0.20034020663338314, + 0.24078974309821108, + 0.15175681326749957, + 0.2221130214858683, + 0.1736869395601912, + 0.23357669819278465, + 0.22872907449477803, + 0.24946469928222464, + 0.1473919805049619, + 0.10486220716511097, + 0.09680074478986547, + 0.19783082719358475, + 0.23219956790139745, + 0.1973065228704249, + 0.20225550694697145, + 0.2251133644152629, + 0.18965839337798748, + 0.20841401383944214, + 0.19713667723748002, + 0.21798145252448714, + 0.2352329085867909, + 0.19409523838602816, + 0.11897882143522899, + 0.20808229317966195, + 0.2292675802151096, + 0.12335323745821589, + 0.21629408061951372, + 0.20588817121707334, + 0.24214016939425972, + 0.14920716726529076, + 0.2370584572536522, + 0.1715018397662041, + 0.18628026551605206, + 0.21155929929282705, + 0.10492698715208104, + 0.21469091398268283, + 0.18536037575005798, + 0.19180574367745387, + 0.22869341712738595, + 0.19425792846685408, + 0.1226792798259263, + 0.1894351211500589, + 0.21732537545380579, + 0.2013916033195542, + 0.2794266559113629, + 0.23359858276614623, + 0.25528929956009866, + 0.1631678421722943, + 0.21227744910054436, + 0.1674464849851793, + 0.16307865871357435, + 0.18099695254206977, + 0.18249881743524596, + 0.20849261354467177, + 0.21244790472071842, + 0.1621426740741858, + 0.20164236129570173, + 0.20063976848430817, + 0.21296040976588912, + 0.19469501562131947, + 0.23623962087925332, + 0.22627502754955922, + 0.22406378259989818, + 0.14662062620239075, + 0.16366987667272645, + 0.1924132243839972, + 0.2185846376729656, + 0.1970988636781091, + 0.22859920489211533, + 0.290094657585727, + 0.2063371710697399, + 0.2342897329444792, + 0.16845963791276997, + 0.2076734947665002, + 0.20190176706773372, + 0.2580907386414105, + 0.19964199550782205, + 0.24036333747704466, + 0.1646260975331347, + 0.10329302076652905, + 0.2061009248530624, + 0.23937860266005942, + 0.19969201400249806, + 0.232335601196532, + 0.17207540761338222, + 0.17473980503320258, + 0.18237811753783897, + 0.18973315611837038, + 0.21709511314737862, + 0.1770339613107639, + 0.16295883044623286, + 0.18267351779893082, + 0.13875857116726664, + 0.17318102655095383, + 0.19264303902603208, + 0.17983488931233993, + 0.12231473169609934, + 0.20268462619113842, + 0.2268685651320969, + 0.1482352387273149, + 0.12901322709524782, + 0.12468422423698362, + 0.1673794231908881, + 0.16640372255136585, + 0.2078654787638116, + 0.17837668762832684, + 0.13615400801931202, + 0.19761847189776857, + 0.14536399234463027, + 0.13144791433691583, + 0.21828069835860603, + 0.19968063610983952, + 0.16946367003772345, + 0.16322668672154766, + 0.2749366246609979, + 0.20464614320297678, + 0.12139261531884474, + 0.1294962882463985, + 0.17925695873926092, + 0.18104515993261452, + 0.26456508119256805, + 0.23212574935461736, + 0.2429481516440189, + 0.10391618262387824, + 0.23386926094664545, + 0.1957699080118236, + 0.17821758819733843, + 0.14945491822299437, + 0.13831902807093693, + 0.2273552433378328, + 0.2061969699739563, + 0.17945324531175053, + 0.13076374104392113, + 0.09635109884095344, + 0.1435185058326181, + 0.19502751345925576, + 0.16509258466728874, + 0.19852298658420012, + 0.19195642308035518, + 0.14966887656560565, + 0.1625584516312234, + 0.09095694440957718, + 0.06798385640305359, + 0.08992390245820558, + 0.08605332466559866, + 0.085287773521107, + 0.19155075431644406, + 0.15507310401040889, + 0.13574402967408977, + 0.09557043077824803, + 0.18068825638179228, + 0.24969104571591202, + 0.22023529479786655, + 0.21824048491512965, + 0.21796548878996266, + 0.20545622166534167, + 0.20042359759111245, + 0.2298997727907956, + 0.26431038615082003, + 0.23738638079338115, + 0.22520826018521095, + 0.2093389759962482, + 0.14432641000595353, + 0.214770067183866, + 0.2036588027556857, + 0.2208002069828014, + 0.2663409243201122, + 0.1873976307317009, + 0.1717601846991669, + 0.17095999818775745, + 0.1934824982280978, + 0.11556341883467, + 0.10642904998526519, + 0.24425765539885178, + 0.20014519634698336, + 0.16865156349152202, + 0.1924811005628897, + 0.24994436300228337, + 0.14196104948866003, + 0.19715374537562963, + 0.1905316759159516, + 0.09061681532235732, + 0.11438626012075903, + 0.112911263524962, + 0.2004833289934298, + 0.12433643046335566, + 0.18008686401677457, + 0.08196280706993488, + 0.24885444331132328, + 0.21768238591603314, + 0.21120599446154342, + 0.23461968647132883, + 0.1610295823536545, + 0.2345746999960491, + 0.11475533204767296, + 0.2123373901863533, + 0.189647359970566, + 0.2894185846091312, + 0.21785416715487851, + 0.22757503902937115, + 0.3189442609358214, + 0.18916446976478646, + 0.19118297591772349, + 0.10552968379998116, + 0.20916795871218244, + 0.1959857425910565, + 0.15961217709107609, + 0.10675084773989864, + 0.10817528247458782, + 0.12995097835724279, + 0.22077311676566344, + 0.2625901647710042, + 0.1190396303069823, + 0.09158069464460193 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.21072031803388058, + "valid_tracks": 613, + "per_track_pearson": [ + 0.23969040405078307, + 0.21739707510955347, + 0.19428939461393768, + 0.25461136258919387, + 0.23693353229587774, + 0.19519901969091985, + 0.22135168899737293, + 0.23676921972973458, + 0.2433535495835691, + 0.26781545477517615, + 0.18372550026135803, + 0.2296488106821544, + 0.15175337954575285, + 0.2795156920260795, + 0.2679784796185113, + 0.27680261566050673, + 0.26846507320529484, + 0.23569334454725652, + 0.21576703724415014, + 0.24534113573177826, + 0.20610615316974273, + 0.2260628316550285, + 0.23139972881341347, + 0.22629855963617512, + 0.16440122601271318, + 0.21394541906743528, + 0.16820532061429033, + 0.22092606897783773, + 0.2297252072935349, + 0.23812336394117775, + 0.2822462251505817, + 0.21817775588951976, + 0.2981797157416184, + 0.19372974832953027, + 0.3012381968494638, + 0.16103405988597116, + 0.23863743701816179, + 0.2322942567729669, + 0.20721940383964207, + 0.2334361900299618, + 0.26002979673628673, + 0.2089607444290647, + 0.23477894274260086, + 0.23135524272515995, + 0.2594444111198102, + 0.1934073001942424, + 0.22799821200385803, + 0.27659359258274796, + 0.25167049876188113, + 0.2130955539850386, + 0.23805951856715, + 0.23780781471310705, + 0.23263844256511965, + 0.25667140119768317, + 0.17244366652696622, + 0.24660127676135862, + 0.239878957943353, + 0.24251089800857753, + 0.23992350055949108, + 0.2190170101630955, + 0.19359660368225487, + 0.2201175328797935, + 0.219691449586224, + 0.1929782334730071, + 0.23995542688259097, + 0.21198749776954706, + 0.24334294067905493, + 0.2277186291394363, + 0.2360518832111463, + 0.2271772875563229, + 0.22682308665736342, + 0.2400705815575897, + 0.26162946336030946, + 0.22984510059273794, + 0.24432744094702258, + 0.2581248421888115, + 0.24244702218434447, + 0.2429056206907216, + 0.28550148600774594, + 0.23686583959865246, + 0.22398379894645215, + 0.20857573696509796, + 0.2754855445829973, + 0.2559812303214295, + 0.18343228165756983, + 0.20440526503442646, + 0.24901105546264754, + 0.25592417428418146, + 0.23955492632626246, + 0.25495575405644255, + 0.18078024685085506, + 0.196438942853841, + 0.22726265671618615, + 0.23137683298982, + 0.24182889924676945, + 0.24125184774296188, + 0.2254111923432986, + 0.2406953459078971, + 0.2372009725512687, + 0.20906561361252116, + 0.21463126008036104, + 0.24662658976387647, + 0.22802685897599012, + 0.224817514965492, + 0.24145746485407732, + 0.2711092957563707, + 0.2312640331799606, + 0.28329137147156125, + 0.26270376625162656, + 0.24012728964827892, + 0.20783903729323924, + 0.322083491315912, + 0.2954930908192649, + 0.1821856136566942, + 0.23286237714959576, + 0.23632273640436485, + 0.28914717605050116, + 0.2566000951411341, + 0.25601562059163296, + 0.23298051521373173, + 0.17969874595612959, + 0.24043357647184618, + 0.22949822356184615, + 0.26745266750010493, + 0.2620427656457899, + 0.23912325736862902, + 0.2519211050606024, + 0.17772953822612444, + 0.2201214916939145, + 0.2096795070860859, + 0.30496923497956135, + 0.2505020945687625, + 0.2115676790659294, + 0.16133632110209153, + 0.18193877551683796, + 0.23218257945271073, + 0.26915192601622695, + 0.2861873528818506, + 0.2765179491347518, + 0.27145546918949937, + 0.24005858760685989, + 0.19489612123638717, + 0.28857469312842243, + 0.2989236698552959, + 0.20970886265239422, + 0.213509056157845, + 0.1827024132118687, + 0.22682215254216592, + 0.17034689312886273, + 0.24834650611965386, + 0.18719733131556762, + 0.252105459165786, + 0.20468735559496515, + 0.20739387276303625, + 0.26636344561570263, + 0.19480839101546804, + 0.2320445759866727, + 0.2700091360887629, + 0.20891792107788537, + 0.269115609514404, + 0.30592813545724196, + 0.22854276163489576, + 0.25615228771520254, + 0.23874467065382526, + 0.23698948455913385, + 0.23284737239923872, + 0.23794769867673618, + 0.21535778477932943, + 0.22556137753946315, + 0.25574756709404023, + 0.1924126955037166, + 0.2009105208015619, + 0.2429173491007464, + 0.2489762732852538, + 0.21229061988862422, + 0.20399441325333678, + 0.21484820472841296, + 0.27063444488404687, + 0.2702829551334902, + 0.1883448525599003, + 0.23839998722636074, + 0.19664813553591387, + 0.1969886737109146, + 0.18465824680991827, + 0.19765076477018884, + 0.24366215963597476, + 0.21440151744123528, + 0.12172337451264971, + 0.1771987171347075, + 0.30485167794475043, + 0.3331975082736404, + 0.18830545011147037, + 0.1726198091846316, + 0.23354280385399032, + 0.1829700947271679, + 0.20367955553274264, + 0.2062917685802503, + 0.21209479187269223, + 0.17147593315847795, + 0.22880876385133805, + 0.29503778152611454, + 0.17576158704759715, + 0.3490271568294933, + 0.2534031817960256, + 0.20052541722618575, + 0.2120823267940438, + 0.21009569751499832, + 0.17191944688865515, + 0.2010355168607822, + 0.18282625878544367, + 0.17341899998120724, + 0.1726575249066728, + 0.17131067704961783, + 0.1867567240689794, + 0.19262292367326736, + 0.22560538431132268, + 0.17406082147022484, + 0.20130049358412955, + 0.1792546400575419, + 0.19471675552984213, + 0.32727954745718324, + 0.20057646697589931, + 0.2854168164996083, + 0.25014451550127115, + 0.2534112071072858, + 0.28894356230335655, + 0.2521487158964388, + 0.27623692754296514, + 0.20708254045713292, + 0.19075653265265505, + 0.21987530829603705, + 0.3219424956785737, + 0.2070659670552662, + 0.24512766561281327, + 0.21417024014562283, + 0.2248639746175545, + 0.1489628574365299, + 0.2418478870549343, + 0.2417175029603382, + 0.240070119412859, + 0.24322876430020102, + 0.18281295265298234, + 0.22918783620259636, + 0.19123870692655937, + 0.18851054532740216, + 0.2506925600857704, + 0.30292196972310376, + 0.22631860937667514, + 0.2261063880726497, + 0.20510632940063905, + 0.2071141615361029, + 0.2291145700661164, + 0.17672099001239447, + 0.19563693391141956, + 0.23046651588249903, + 0.25749290740882175, + 0.2071424872240855, + 0.212855567802449, + 0.20696307686890497, + 0.1932650819688181, + 0.25481282554055273, + 0.22854215561221652, + 0.23697886713734473, + 0.264736303736871, + 0.23318805063264947, + 0.2168170166780958, + 0.20393043660912696, + 0.1842290535067823, + 0.20848103938873894, + 0.2306908217128969, + 0.22326111128289433, + 0.138564406923603, + 0.18951357192214902, + 0.16960568421056402, + 0.21200211810129113, + 0.18981570661875016, + 0.13731480350455244, + 0.14781291576586367, + 0.20213392349245216, + 0.20911157355643958, + 0.2174048364927207, + 0.1808762758974657, + 0.15287962181104986, + 0.14293994265419274, + 0.183230629331559, + 0.17963597611928825, + 0.17951587866559746, + 0.1970201883734215, + 0.19075021682751522, + 0.20257343850745468, + 0.18304900201146485, + 0.15047631536270037, + 0.1986097985355104, + 0.15864662806338672, + 0.1502720316007557, + 0.11294367679578342, + 0.17856432161021962, + 0.16772027304915355, + 0.20044375815689708, + 0.17913492622346655, + 0.1829347927720383, + 0.2137057614338202, + 0.1895156614348994, + 0.16143755269813118, + 0.167183870979497, + 0.2673560993391546, + 0.1431106870221944, + 0.1941961661904578, + 0.18822895179385682, + 0.18059018785472425, + 0.18391033840232157, + 0.1998739407107766, + 0.1931911971959286, + 0.20492330513136928, + 0.19012749101647142, + 0.17750312862892303, + 0.14045943073943487, + 0.1630572014770884, + 0.20057608946477654, + 0.17293362273742108, + 0.1818360428253677, + 0.1970874478316941, + 0.19269128066515476, + 0.15642577482451167, + 0.1814660421712155, + 0.11480042171562096, + 0.1795826432657616, + 0.1796438157129462, + 0.19388828962194612, + 0.19107046313762696, + 0.17271301445451276, + 0.1687377534904519, + 0.19092797510660528, + 0.19596229378042065, + 0.13752284360775657, + 0.2073898343187699, + 0.19045067684497918, + 0.21637946392323681, + 0.19854166124296665, + 0.20887658804748982, + 0.19611557062442855, + 0.18724422084545744, + 0.21978826047863378, + 0.22084755093803682, + 0.21120490671408046, + 0.211864040653565, + 0.22815719356426456, + 0.21261512334469163, + 0.21084576906693286, + 0.21441001431494242, + 0.18525555181409364, + 0.17507946613585865, + 0.15703561591949086, + 0.2166351719592399, + 0.22149126207410616, + 0.15959991177900457, + 0.18685920403407588, + 0.18829312429259576, + 0.17013528711684725, + 0.1734265468549255, + 0.1586195609003189, + 0.13213763839796183, + 0.16474746669734827, + 0.18329527302197327, + 0.1782403132385336, + 0.17932988212746617, + 0.19514205527144027, + 0.19436992985542817, + 0.18584290733482406, + 0.18193266178680273, + 0.15747535567356646, + 0.1493639574056796, + 0.17105913740334697, + 0.16708308428004623, + 0.20453986362934662, + 0.17102927792745487, + 0.19452675633408542, + 0.19137004841200228, + 0.2037487596667846, + 0.2157913446182034, + 0.18040818168644446, + 0.1578869759815952, + 0.2484314742628691, + 0.21066800559584553, + 0.15722940189673648, + 0.2017122484307071, + 0.1619979715113008, + 0.21117190744073405, + 0.17798905986719007, + 0.18486675612843598, + 0.17020970076542122, + 0.16171550991393902, + 0.1865463637022718, + 0.1643968799954617, + 0.215569328239559, + 0.19026238843263865, + 0.18164202462482357, + 0.17330242025242587, + 0.13794476415397494, + 0.146150865624815, + 0.15503378000475188, + 0.2421129997764003, + 0.19102436656620406, + 0.14879556559704601, + 0.1392312963076462, + 0.1434707596810032, + 0.2137170270745632, + 0.1997498749282287, + 0.22839305434888263, + 0.19337674190483484, + 0.19605701592947058, + 0.162356166501176, + 0.1525410465103125, + 0.21647776708152447, + 0.22780434608623543, + 0.14651277446570238, + 0.14854705906317792, + 0.15146229358574378, + 0.18433881512768108, + 0.16847595039157373, + 0.18206150577736874, + 0.16443676175612998, + 0.19637503395446454, + 0.16190163320582296, + 0.1447187343776154, + 0.18317476048656547, + 0.15284801057182107, + 0.19710964124498892, + 0.1672745908178016, + 0.18907716007260567, + 0.18396404840164962, + 0.22870298515134682, + 0.15518576571578532, + 0.17553612732204937, + 0.13860412483880719, + 0.14168286920326553, + 0.1376050514570641, + 0.13438076078264274, + 0.13062669905127586, + 0.1713855509852802, + 0.1960590765709846, + 0.19739984320781384, + 0.1611667286796272, + 0.2332492356306061, + 0.1938924749389335, + 0.15105586047768357, + 0.1520747113526188, + 0.1767452227384499, + 0.2453397238305923, + 0.21762212685292262, + 0.14194836186345872, + 0.2036081607131202, + 0.19441132898960795, + 0.12283405467820543, + 0.14015195501143612, + 0.13998880867755006, + 0.1934742525240154, + 0.18400652182179222, + 0.11266136899670354, + 0.14571247689392353, + 0.2639266455404362, + 0.27131311207930314, + 0.18183391834174872, + 0.14865435796427115, + 0.19825698261042668, + 0.15731864804001316, + 0.17856808112850134, + 0.17556925462847756, + 0.19376986436309226, + 0.1589127643898668, + 0.19807422417531978, + 0.255992872788928, + 0.1703544616094957, + 0.3102139304387368, + 0.20757757542638897, + 0.18603528373365028, + 0.18331190710637627, + 0.16853835869197953, + 0.12229275693821538, + 0.20898846098377172, + 0.14100490101409718, + 0.16806173059411547, + 0.1663795999267496, + 0.18716133392049022, + 0.1796184434476091, + 0.18169185067692584, + 0.21646615126078367, + 0.17982020862492862, + 0.15517470142571127, + 0.19031611965359643, + 0.18475911080875615, + 0.25785220333332587, + 0.19163692857969372, + 0.22702888331213977, + 0.16865508857660788, + 0.17571266453419693, + 0.2260649988723124, + 0.15518387721398563, + 0.2203031845245295, + 0.11350513301339137, + 0.18234454497486, + 0.144801178777791, + 0.27777855657767325, + 0.20700810180772364, + 0.20612087787732175, + 0.20653393832728495, + 0.1875045301481661, + 0.10548845117612543, + 0.2144595211994412, + 0.20110504135930377, + 0.18043427505123896, + 0.18617101873922787, + 0.17402209027410737, + 0.2191527009487016, + 0.19260305937481537, + 0.1951939352326912, + 0.1996595514821971, + 0.267940824558023, + 0.20513942809204458, + 0.18109298870239257, + 0.1529188018374558, + 0.15617350788289824, + 0.18219127846527847, + 0.15316268564952287, + 0.16937463412208636, + 0.19393764530424412, + 0.20422050233386188, + 0.1282714652480351, + 0.17908328828338246, + 0.16154284095177252, + 0.1810977436886272, + 0.19134255287375795, + 0.1287538410909065, + 0.181098141422373, + 0.21087939733356373, + 0.19153763004736515, + 0.21001814459480553, + 0.19554719517461003, + 0.11075247020610095, + 0.19785602341243183, + 0.19690715904605677, + 0.181891401171546, + 0.19780002457473722, + 0.09643392888151571, + 0.2676250829887548, + 0.3044727703021241, + 0.2763276540885419, + 0.2530621917877391, + 0.2602387709708008, + 0.21557022077833957, + 0.21724866759728706, + 0.2469760799905791, + 0.18887342904281926, + 0.2041053192447853, + 0.21205722210897127, + 0.19782164334231345, + 0.19732240725292238, + 0.2048872678782233, + 0.2145582440131306, + 0.19934737624642168, + 0.20774598807750747, + 0.18635436332540767, + 0.26718174641773884, + 0.26427660435244693, + 0.268024500480644, + 0.26640495533471154, + 0.2950350233994811, + null, + 0.28205652721920305, + 0.33281063997971544, + null, + 0.07686580517439523, + 0.39840334856569637, + 0.0751012495165791, + 0.2558159018407948, + 0.24163109873856325, + 0.3056199806454504, + 0.36769990476171766, + 0.3590800949637853, + 0.3752924440910474, + 0.37348729774592465, + 0.33956389003506454, + 0.29368080855324036, + 0.2246815602892889, + 0.379314559824923, + 0.18580319401287984, + 0.29748114195188574, + 0.2649203895135359, + null, + 0.25608536386849884, + null, + null, + 0.2605633856958646, + null, + 0.23295457942934536, + 0.20902488268813568, + null, + 0.3022069588365508, + null, + null, + null, + null, + null, + null, + null, + 0.25836989366920915, + null, + null, + null, + null, + 0.22023782699116173, + null, + 0.26231662837454306, + null, + 0.2118232784031749, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.2757339373893144, + 0.1598165360353767, + 0.24625057438440298, + 0.2583100816154936, + 0.18847066195809922, + 0.23696408946571731, + null, + 0.2997178087238204, + null, + 0.23994355495593084, + null, + null, + 0.24083321862862594, + 0.27182723717541324, + 0.23051224981594112, + 0.23728985015100804, + null, + null, + 0.20613580117711028, + null, + null, + 0.24512065160344465, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.23900865839787808, + 0.23673103642238003, + 0.2508628558267105, + 0.2228029171550747, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4032880538674383, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.2059611649954854, + 0.31160698348064314, + 0.29397857017695656, + 0.4443777855236212, + 0.36803453943574854, + 0.23738807505132875, + 0.34822719531277646, + 0.40190718673893666, + 0.3373947563595317, + 0.3472971684373527, + 0.3938795210465233, + 0.36966765788207223, + 0.2693988462963414, + 0.4072418514988678, + 0.34888864819572113, + 0.353565040872697, + 0.38541403874118774, + 0.3205255495142339, + 0.453201711510987, + 0.3369161278222373, + 0.3433023626394589, + 0.32724498554057935, + 0.3035811952364979, + 0.45722028198604653, + 0.3276140471062388, + 0.35607007799790896, + 0.2559680928853111, + 0.3651575402294399, + 0.35936794126830124, + 0.379631521477237, + 0.3599021336938568, + 0.3902764381628737, + 0.4806574395039059, + 0.27369206198881507, + 0.6461341044389772, + 0.27410696652984406, + 0.20420525316660684, + 0.16164161641834504, + 0.3227275865988042, + 0.21672205970010625, + 0.2627387640342629, + 0.4532084419504838, + 0.29447377707308, + 0.39208618220294633, + 0.2855509521062304, + 0.3859902856530214, + 0.29290477904565576, + 0.20143262166400153, + 0.45693537821268115, + 0.42332486927911495, + 0.3794615412566519, + 0.3571840558284281, + 0.36828334782419947, + 0.34657228362485915, + 0.30896976491038386, + 0.3718945196424023, + 0.29520484181543194, + 0.4136901714897575, + 0.39836504130679873, + 0.4199459694512278, + 0.2618494052119527, + 0.19642278562384544, + 0.3912189010467751, + 0.4107066425126872, + 0.2701175498236084, + 0.3356318775131523, + 0.33405371984444804, + 0.4372045636581371, + 0.4598267360616762, + 0.38003485583509883, + 0.33513294489037215, + 0.6758260178125622, + 0.3387406234391633, + 0.3249983591236495, + 0.32391408861510385, + 0.3997951216096658, + 0.24853619737445712, + 0.6082063404232371, + 0.37651770049013933, + 0.3758660597270677, + 0.36357668438880675, + 0.3967226609369895, + 0.468147196663312, + 0.4136601781868159, + 0.30271946621019, + 0.2601316859965126, + 0.43506139260731425, + 0.3887643601700455, + 0.8002270716884109, + 0.44124658287798924, + 0.3376371182584523, + 0.32755760681957635, + 0.39018415594808137, + 0.3154908222407075, + 0.355939355664211, + 0.5028855033173675, + 0.31458836302464077, + 0.3125798075932069, + 0.33282757076033176, + 0.2734187555063799, + 0.24217427138970082, + 0.37154798277075246, + 0.32820374110837275, + 0.32189642207412517, + 0.36804392746362674, + 0.18235845772050416, + 0.2232930825810244, + 0.2437556758849438, + 0.3749816007091805, + 0.31979511398861404, + 0.18730375966459878, + 0.2646070734730338, + 0.23959208530900894, + 0.31736381294666866, + 0.3077056264225168, + 0.4301432031712071, + 0.3382943930278312, + 0.33100524620490424, + 0.2803930700475306, + 0.346021489239177, + 0.2765049119114364, + 0.3580470066876746, + 0.3317603103978675, + 0.2154730200990696, + 0.3115039741535035, + 0.25288320266694825, + 0.2693128376102961, + 0.25416588184567707, + 0.2749590277763031, + 0.5223402094603201, + 0.31655998841550503, + 0.21634385736754744, + 0.35230346971717824, + 0.5640699281685941, + 0.23761816700200677, + 0.2969903403854188, + 0.26950633116664496, + 0.27926937871966895, + 0.2623636986720806, + 0.2369297724634056, + 0.22630967496700616, + 0.42091298477216094, + 0.4885171396854973, + 0.2652077842874983, + 0.3518686276636406, + 0.4172593863034048, + 0.3824219848028069, + 0.2147805352727085, + 0.36547551443978943, + 0.28382493955911764, + 0.3278264682744966, + 0.26596692426830687, + 0.32720346193386, + 0.4132396228446377, + 0.3556453482272505, + 0.35125661795986723, + 0.22174083500920974, + 0.26111002400766825, + 0.24078476719334507, + 0.3470685999194596, + 0.3112118750992344, + 0.2897605701951654, + 0.3264742681286414, + 0.3264484872998315, + 0.30932846211701404, + 0.3097402529801342, + 0.2914728872024483, + 0.4144079832188278, + 0.21206237432839334, + 0.3056748867404308, + 0.27578174961128177, + 0.2279880502260606, + 0.45448563470294484, + 0.24980445386416916, + 0.20583081445291984, + 0.3682525218742193, + 0.21510929155283773, + 0.3151188555123096, + 0.3033416457686615, + 0.2591063343564694, + 0.29957981528717487, + 0.3045240014586629, + 0.26551750206985414, + 0.3485810843804997, + 0.2583518240422254, + 0.3084076977998182, + 0.3662720769285974, + 0.22291319618635574, + 0.24550323488221912, + 0.33743899949512607, + 0.21969368213960594, + 0.28456540687920695, + 0.3924673487149325, + 0.3250574112400604, + 0.2541171625790341, + 0.24747819442079871, + 0.2277240400535241, + 0.22606020394193316, + 0.22726927476476588, + 0.39891235917372825, + 0.2605709916983311, + 0.2904933539058671, + 0.25795466038374487, + 0.2528975933854791, + 0.39130785369506826, + 0.278819885680689, + 0.3038546182145071, + 0.3218672408006584, + 0.22647751826905396, + 0.43555807211914116, + 0.24260153330970888, + 0.3687364335181392, + 0.29097155976358263, + 0.340315209886279, + 0.8401884773142785, + 0.3692783099289347, + 0.36389549269729243, + 0.35592956932610564, + 0.430337782979694, + 0.5263168915610246, + 0.5950712921952303, + 0.6083927816583405, + 0.37775992905262573, + 0.36721000532444004, + 0.44578198247675144, + 0.7166989275288063, + 0.27658381770261115, + 0.3092680266823322, + 0.5237670782876087, + 0.4990698668260651, + 0.5107108197117082, + 0.4084535874245197, + 0.28446944624022275, + 0.3453903593576009, + 0.5822572996347807, + 0.39983136141519854, + 0.2936691134657863, + 0.6032398299016014, + 0.5313098144264219, + 0.4318878785387505, + 0.46436291716753525, + 0.6937769404488437, + 0.437904503813926, + 0.4731660509857655, + 0.42970211899689303, + 0.7546984383979597, + 0.2916060824415696, + 0.6280377606498362, + 0.3669359506717708, + 0.5330976063428925, + 0.4054101567408445, + 0.3793070335308112, + 0.6910465277122918, + 0.4872804516475326, + 0.2699752273147567, + 0.3527167620394178, + 0.6293250352852398, + 0.3246892922043074, + 0.43239091245465916, + 0.4224546037200168, + 0.3211697693671075, + 0.44176830056784006, + 0.442879386047536, + 0.38493396619086157, + 0.37915236935998703, + 0.4556075965355893, + 0.32005438786203955, + 0.2924909039122066, + 0.6167595786480669, + 0.3441049746030022, + 0.43502790894122423, + 0.5446591688314999, + 0.3040506348283847, + 0.6506604995429053, + 0.30244858314765044, + 0.3057054068622776, + 0.5457205969695629, + 0.45752648030568854, + 0.5730386733270063, + 0.7220321119153814, + 0.6477535396096774, + 0.3816839589566377, + 0.4487788592706646, + 0.524095771143839, + 0.2649122772659674, + 0.3034547261197109, + 0.5486203842387654, + 0.5003758193148967, + 0.7053449279678783, + 0.4289927756164248, + 0.3339665549645331, + 0.4411688997195936, + 0.352536080865304, + 0.45857236696459225, + 0.6875905163562449, + 0.6104789964874342, + 0.45783581758847086, + 0.5078752956407289, + 0.4758306277175329, + 0.5009604518197234, + 0.7276017762953402, + 0.28372553863762484, + 0.42255928310846386, + 0.25558313042748176, + 0.5802535751472118, + 0.2989227639857436, + 0.7271601565523619, + 0.38302148478216874, + 0.39723564075273493, + 0.6783564641569573, + 0.49802829064715853, + 0.29878445775898615, + 0.5479781760127376, + 0.37762066787988213, + 0.5056364330674082, + 0.49209528026612753, + 0.30934373365620466, + 0.4803587824210935, + 0.36317941735786424, + 0.44267729140485645, + 0.3299494619333398, + 0.23985145417845719, + 0.37383376655699385, + 0.4938801976201314, + 0.5543687175641959, + 0.6478184938923223, + 0.3655704966647084, + 0.27117152778519377, + 0.43391682135527215, + 0.4399672031967675, + 0.4090735900432153, + 0.47117336018122397, + 0.43967685474438023, + 0.34299432092292315, + 0.30396433771718767, + 0.4810001445206351, + 0.7585697517333583, + 0.5148250403928238, + 0.5853864871908206, + 0.5796785159668392, + 0.7606929876451368, + 0.5397935718948903, + 0.28898653813966463, + 0.5654816185268172, + 0.34993871269369703, + 0.49093395441816806, + 0.418953120306169, + 0.3705488992955516, + 0.4762926425867342, + 0.7698829409371125, + 0.21595076263268442, + 0.44225505556018035, + 0.4514296578215531, + 0.6073655081580461, + 0.36500622024315116, + 0.23539504672950762, + 0.3950400350083502, + 0.7129527335779541, + 0.36216224219170495, + 0.419384027041569, + 0.49242155123091763, + 0.4258355566441609, + 0.3771128337789886, + 0.4741412821061025, + 0.601298898621179, + 0.44580735757936535, + 0.3745233677289671, + 0.5235253791746532, + 0.44043794239770667, + 0.5510034895981862, + 0.5130298545572753, + 0.5081629091380747, + 0.5284931841742627, + 0.2762304828353791, + 0.2527625426267952, + 0.4714907113380268, + 0.28346794150245164, + 0.6834685298692126, + 0.2704581845397709, + 0.5456733117847071, + 0.5339899023848922, + 0.6753001711627769, + 0.2755104189840804, + 0.3056484190065947, + 0.30461440730061734, + 0.29635372040673347, + 0.26191023020674287, + 0.2495599875606036, + 0.5979994463110903, + 0.3165251399693646, + 0.6718910038341295, + 0.5635641931184039, + 0.41134754665342654, + 0.5420511304049, + 0.36865846001562014, + 0.5187287772223096, + 0.5116699819976039, + 0.44848450486949726, + 0.37036918776586497, + 0.4454475221955819, + 0.4364374045757675, + 0.5029299948201132, + 0.37211114349625424, + 0.29378590313634867, + 0.5964369058199815, + 0.4511904553484185, + 0.6459196092625901, + 0.6058238159334617, + 0.4794509286850127, + 0.28131136803243095, + 0.2970406990855399, + 0.32230510496387915, + 0.4100268106514921, + 0.7392196453250055, + 0.45678917558790705, + 0.3484194883223293, + 0.7287085577923191, + 0.40566691601937377, + 0.6123137640270422, + 0.5155190061267884, + 0.3773828711777176, + 0.4797475357463653, + 0.43227525991879046, + 0.4225908985016821, + 0.5691630719363024, + 0.5126847066346815, + 0.5005729336248197, + 0.4008252633473065, + 0.43352890127516885, + 0.4063965506722644, + 0.4556437352262116, + 0.760394086649823, + 0.4663163458374773, + 0.44126242127312, + 0.4178714734886738, + 0.2460670602821083, + 0.6555969285614033, + 0.3742577815785483, + 0.28657889144313675, + 0.4960807248440662, + 0.3875664046004106, + 0.29611335702081626, + 0.47336827716434304, + 0.4045715510691169, + 0.7688348712888964, + 0.30116423837281225, + 0.35762320920445456, + 0.28784145740775513, + 0.4124288936092655, + 0.3651026334300143, + 0.26604360818349965, + 0.361264169404189, + 0.25127512973786265, + 0.3881736984875864, + 0.7324598408536824, + 0.2706841287552226, + 0.38226710587393065, + 0.47419808470004937, + 0.39629027953707074, + 0.3592690905118103, + 0.4562013910894916, + 0.3427936382431436, + 0.5016334314182497, + 0.49407321657698383, + 0.17299612471911943, + 0.6108587049144462, + 0.6678969340458583, + 0.47726514376385726, + 0.4542844196360799, + 0.4972462595276168, + 0.351318485672492, + 0.3228012720878008, + 0.4022012293410447, + 0.7707767604734248, + 0.3179703975890745, + 0.5838924931932151, + 0.3038319821279704, + 0.4377969624351232, + 0.44713881424569696, + 0.4609965273913388, + 0.3542887590172274, + 0.277055717887924, + 0.2825401333666088, + 0.2604922917891587, + 0.476419320770288, + 0.40902881131923025, + 0.23417328215112274, + 0.3826002564529947, + 0.4305802454987508, + 0.26451546249784635, + 0.5345341794390519, + 0.4396961454741143, + 0.35265403237084403, + 0.43278270095139526, + 0.2598666041312768, + 0.39982542610536764, + 0.45307792591146395, + 0.5569339518294415, + 0.42149815438757887, + 0.7698846560501286, + 0.27563750013896954, + 0.5414436356374086, + 0.7272294037121492, + 0.616883883168812, + 0.19970422024211607, + 0.4640455527334266, + 0.40080982671190657, + 0.30063723723685626, + 0.3253318322694056, + 0.32608037737727497, + 0.43786865412864673, + 0.48205787955294427, + 0.31531870311829624, + 0.5913074732272403, + 0.7543997084199573, + 0.4140797405772714, + 0.3821772401876929, + 0.4117782881695713, + 0.45796903540570655, + 0.48408464476800067, + 0.4777837762132096, + 0.46353345077723185, + 0.27265892825451343, + 0.6290127205398683, + 0.4510383003426483, + 0.5346217037861536, + 0.6965020076163655, + 0.2822062658118065, + 0.5782625410284961, + 0.36655101936447004, + 0.42486189215591613, + 0.3340424118781699, + 0.49563402257739464, + 0.42463207323252544, + 0.29493229554195016, + 0.3838439539288857, + 0.39523275710679023, + 0.5296727980721624, + 0.5701568259520945, + 0.41258415731356296, + 0.43683552739636267, + 0.38840265068246105, + 0.7242029012512815, + 0.4005416166577361, + 0.47135594398100616, + 0.43759366227901136, + 0.4476980644213778, + 0.5871727933767996, + 0.24852767776206006, + 0.5486152350163652, + 0.3512755032635394, + 0.5116377696611485, + 0.3931625090747992, + 0.2965410102417954, + 0.2967413746279924, + 0.43933385337502995, + 0.27589916238055373, + 0.48856496167156077, + 0.7144196444986314, + 0.7169000730960627, + 0.3480264222003502, + 0.2910947451531687, + 0.6876687338233631, + 0.42631859765804186, + 0.35600448963307463, + 0.4855330577564648, + 0.33909833806749234, + 0.580130533825625, + 0.41884887180141067, + 0.5457678532353524, + 0.5696764907316908, + 0.3956504825294991, + 0.4903599584422227, + 0.6434457835106975, + 0.47409069575355783, + 0.32403254707872287, + 0.3749156298685635, + 0.27253198924690325, + 0.5159366770465775, + 0.318257574813407, + 0.49366061761712526, + 0.47032594563773483, + 0.3133677383874106, + 0.707409584106324, + 0.2948207232094278, + 0.7119241942446997, + 0.40420361001452926, + 0.5174209336175463, + 0.7319305458774209, + 0.4525825398686527, + 0.5366188622430428, + 0.33862681896816027, + 0.6932822212132811, + 0.3804294520261775, + 0.27602214825280424, + 0.5352208867197381, + 0.49504557998036364, + 0.6206165581438012, + 0.5313212993209517, + 0.3589341200992916, + 0.46300453621311677, + 0.24064121122896248, + 0.4147389455164781, + 0.5482967912989054, + 0.4821432627434857, + 0.31766096671172955, + 0.51452451277923, + 0.5638886916479815, + 0.5528148214471765, + 0.4086254267282192, + 0.2910432470159894, + 0.4893192716136527, + 0.4440904542200924, + 0.5399351413944782, + 0.4604212870503714, + 0.29647200173919386, + 0.3788993562405871, + 0.47761487797985186, + 0.3736681711972575, + 0.7166840828153536, + 0.49047902575259206, + 0.4241363034306429, + 0.47147254898094426, + 0.3602582614156472, + 0.40591106720667486, + 0.5241198567457247, + 0.5306281366108281, + 0.700558296967673, + 0.5140409116851617, + 0.5596709678966891, + 0.5051310995715274, + 0.4478446688989212, + 0.5565943326403516, + 0.3504523906150882, + 0.6079253826025588, + 0.5825713920693623, + 0.7176656646288089, + 0.49212813449325604, + 0.5279680769286212, + 0.5256205063795625, + 0.35876739185915496, + 0.5209147332266304, + 0.29823030686718677, + 0.6618047448492119, + 0.4556795335969799, + 0.7272410518638918, + 0.7535977810163242, + 0.4283668844153874, + 0.42271023165310173, + 0.47487838142209776, + 0.3095681590541258, + 0.47001328148641963, + 0.7498460682247204, + 0.47207941103124623, + 0.47817361590468777, + 0.5723600275874038, + 0.46639026451354587, + 0.31497480858853216, + 0.5917306959207254, + 0.4043817214253644, + 0.3637252972391023, + 0.725696205597284, + 0.3820878480992003, + 0.40660555416650773, + 0.36698057883522406, + 0.4672624272438, + 0.5451537094876615, + 0.5961355038739781, + 0.665136858742322, + 0.6087166963340187, + 0.5735065152145788, + 0.432930418941978, + 0.5915778773409794, + 0.5316459255989612, + 0.7734036312745928, + 0.7283900259847915, + 0.3050496576983453, + 0.33835545424025654, + 0.28150439355442936, + 0.49671832246066194, + 0.3959849805251218, + 0.7803414081436124, + 0.35311415858972584, + 0.39953693255531203, + 0.5402603875002187, + 0.4641561295177946, + 0.5393787998264797, + 0.6063580874851119, + 0.44583803949689366, + 0.6095088705999939, + 0.35580172895471845, + 0.6634169772192918, + 0.7720093529536461, + 0.7564636603300677, + 0.5632171165232639, + 0.40556036863128275, + 0.5273251049738973, + 0.7307190783241818, + 0.6742744131341432, + 0.6089337243648494, + 0.46038267463778754, + 0.4082151351566448, + 0.4772807257985687, + 0.7281740952327354, + 0.3927942862364927, + 0.3995101580359465, + 0.28421209994549945, + 0.5814274108584726, + 0.26074183125362166, + 0.37500158247848187, + 0.6543124758933659, + 0.5270646868011307, + 0.6699239747460503, + 0.521669642966866, + 0.49496927374745836, + 0.6536601423215885, + 0.6208652214757838, + 0.5512807092341205, + 0.36090223274131034, + 0.6328986258487607, + 0.5027334685929951, + 0.382391706700054, + 0.3217473634605418, + 0.46256370133949026, + 0.6285738838301074, + 0.4804105952316818, + 0.595854079771192, + 0.5584880105644754, + 0.4005210487150148, + 0.38205869665695136, + 0.6742401349235038, + 0.47220891101091916, + 0.5716235964896451, + 0.7150928263312692, + 0.39756571808284963, + 0.32918985497221076, + 0.3271506427615289, + 0.33595234341529695, + 0.7031595079105257, + 0.7283335258099692, + 0.5558232951979574, + 0.4026065025532031, + 0.4355295901118064, + 0.3641650733728891, + 0.7732999295050901, + 0.32189309023481294, + 0.4642318154086223, + 0.3116925339154101, + 0.4806538686529993, + 0.3421386394725173, + 0.5471652871385699, + 0.45580454483096144, + 0.26979492455772175, + 0.6829872235370427, + 0.49446271618125215, + 0.3964352401178954, + 0.3205790279160098, + 0.3400281718416035, + 0.38714174545625135, + 0.40091386883280233, + 0.3521070296685022, + 0.34781373539813515, + 0.3876086312891627, + 0.3591588282313378, + 0.39130180103196777, + 0.3128517029150828, + 0.4226470959790025, + 0.2014107022202609, + 0.23950324193345718, + 0.29214879456834286, + 0.36633379265637833, + 0.41318288404980713, + 0.3921610749715229, + 0.2961809114275043, + 0.2899812100478304, + 0.2868661326575352, + 0.649509890893103, + 0.285813717211278, + 0.26956994479314506, + 0.772635980861086, + 0.3371376805242071, + 0.3791333589648892, + 0.31459261084284956, + 0.42987006280453655, + 0.2554745410072378, + 0.36867795512020635, + 0.3303598849289945, + 0.4558923270024994, + 0.22970277251340235, + 0.39216235864838467, + 0.37761762088809114, + 0.48419035386189974, + 0.34513493093509506, + 0.4573291110303379, + 0.2799599047562622, + 0.42317027038063615, + 0.3833803180001879, + 0.3230383707672763, + 0.45028963256435, + 0.3280930101522669, + 0.45963926166046104, + 0.34813676366612395, + 0.329701588776157, + 0.8029211492203793, + 0.2673017164313259, + 0.39489141268742467, + 0.33925424704329854, + 0.37476838545034963, + 0.277883946526584, + 0.3430978293116923, + 0.447226075971702, + 0.5347485326307169, + 0.3367184106577987, + 0.3630596667149099, + 0.2412956223265985, + 0.39139181165732695, + 0.38566217595775754, + 0.3946265240090058, + 0.42462440207020274, + 0.34286349501066415, + 0.3067983539481225, + 0.45395195100561225, + 0.36870651398063287, + 0.4372430825705004, + 0.2533231394346832, + 0.30846043947510715, + 0.4607097572888264, + 0.44696153187909554, + 0.3771557423548262, + 0.3304871859753294, + 0.32825980782000186, + 0.8189939840524604, + 0.30357837450440256, + 0.29691510570066443, + 0.32433368491530046, + 0.2655913043712118, + 0.3058937792452022, + 0.30197046255403437, + 0.25228174103691914, + 0.3450531055837574, + 0.3436572543657687, + 0.34673375976892845, + 0.304742571424356, + 0.40409776966425276, + 0.4602808287941608, + 0.2935597168244561, + 0.3457707422899919, + 0.5354993873740342, + 0.28164749893696833, + 0.3329294437964481, + 0.4796352523962646, + 0.4262359452810308, + 0.2841137808926989, + 0.2340813973514013, + 0.28786838547628374, + 0.41184781929561803, + 0.423489004428752, + 0.24227876411080546, + 0.3214396965475563, + 0.308446243177123, + 0.40514577551588393, + 0.2886822526469946, + 0.24937726417592082, + 0.45730608046094606, + 0.4885668240959993, + 0.22565783984116655, + 0.40738625290076685, + 0.23669343054603492, + 0.28830275341517364, + 0.2812530521433982, + 0.7718507563334036, + 0.40644140110023236, + 0.41816829672272804, + 0.23161537276277533, + 0.33544472289484706, + 0.31303867377826333, + 0.6799939725898073, + 0.3683223409811377, + 0.30983815376967383, + 0.43006456105180496, + 0.38709602757012745, + 0.36414917851731526, + 0.4232380647244363, + 0.46095329208444813, + 0.228883876023643, + 0.27188885682928005, + 0.4134053276819923, + 0.35868393502557583, + 0.3743790942271373, + 0.3656449305132989, + 0.2940189202680186, + 0.24916776370273605, + 0.2437057470496293, + 0.3719967013435749, + 0.327832760330238, + 0.578550470077177, + 0.27531923533170294, + 0.24991524615017097, + 0.24577860384872974, + 0.5027807533043944, + 0.3137989621879811, + 0.21683711063333236, + 0.26457133546900047, + 0.33938290802186316, + 0.3509912868993759, + 0.25076352900323395, + 0.5207147615167655, + 0.4022264951645974, + 0.24353372819583605, + 0.35585566140776603, + 0.4034988968345329, + 0.3333033154385495, + 0.31433636726138603, + 0.45101737327347896, + 0.2820893040504712, + 0.6343384539814302, + 0.3518864428467697, + 0.4404604906402134, + 0.3843792589030337, + 0.3446864670995125, + 0.3750852084602496, + 0.3003888945731283, + 0.22630833226222843, + 0.3665846841494932, + 0.573467975160583, + 0.308342692079243, + 0.23164769490291315, + 0.4735789351924908, + 0.39262451246994773, + 0.39392679303968614, + 0.5862439587508295, + 0.3195969506056887, + 0.37841423234834176, + 0.27836630443101595, + 0.3407232058472182, + 0.41527060680093814, + 0.28697992091773455, + 0.43128294853580884, + 0.3762065367149923, + 0.3852666491215353, + 0.1990099545660858, + 0.3599621805074104, + 0.2719422444287568, + 0.25546501685608164, + 0.25569776350216483, + 0.23949132336802686, + 0.39581837796582836, + 0.4984858538053468, + 0.4138260114856988, + 0.4412937572665788, + 0.4097674266665669, + 0.24373449016502166, + 0.3048241155854859, + 0.44422674374427085, + 0.3288520331836574, + 0.3557466790450837, + 0.43625717206158443, + 0.34519172266762604, + 0.3339377935052306, + 0.48213901919639207, + 0.36779952047168313, + 0.2881254177515008, + 0.48199427191346333, + 0.2967646280062806, + 0.40497612950124284, + 0.2611206460762825, + 0.46026962075763, + 0.581447724864003, + 0.781384648403679, + 0.39033324639465367, + 0.3434876559663732, + 0.3946791572084027, + 0.339576937049721, + 0.3226291265289972, + 0.3216367573343825, + 0.37535826909242986, + 0.31241188715380647, + 0.26723364483109563, + 0.4187860132275095, + 0.3396100248574336, + 0.3189589215746901, + 0.24951652220478182, + 0.23267511867517557, + 0.3698324160053893, + 0.3790813149729156, + 0.34205607430036794, + 0.34676331896761425, + 0.3615774387321863, + 0.27213390761035466, + 0.19898703831236186, + 0.3274389243875298, + 0.4113937449271721, + 0.26500912933842036, + 0.3639734194547245, + 0.26928132880114874, + 0.2884344491894733, + 0.18727584225342433, + 0.49283114120654736, + 0.27318089193795125, + 0.2556437464467551, + 0.35897655673411094, + 0.3117613966131375, + 0.2740647057141949, + 0.5754648938133641, + 0.4292540077274976, + 0.19899287180625516, + 0.40811735630455415, + 0.3685179297037893, + 0.3571275318339913, + 0.2886962580956992, + 0.22699729148985948, + 0.30781211795376656, + 0.4690436606654073, + 0.21493173115836292, + 0.37384257281035776, + 0.19753301494439776, + 0.26011039011315606, + 0.2683776821226182, + 0.23469092760456658, + 0.35692131457824583, + 0.3737548290408656, + 0.30307175792660435, + 0.32529903166330776, + 0.3541135535821836, + 0.231741622199441, + 0.3473540722095462, + 0.3437712375606941, + 0.478819548297836, + 0.48213445279318423, + 0.3720725572766295, + 0.3766853391631555, + 0.3535185470648537, + 0.3935514132614756, + 0.4248995712323517, + 0.5113362408520385, + 0.3742083449798173, + 0.2534439655818218, + 0.4855126620252593, + 0.31364563806902984, + 0.42032874404821086, + 0.41570762418573054, + 0.303288271559116, + 0.46114162558691596, + 0.21417441020084582, + 0.253554746959299, + 0.23232641106907548, + 0.2079436725911554, + 0.32680598874563643, + 0.33002108233905214, + 0.2670930782955262, + 0.3724126208297835, + 0.5148775445322832, + 0.3541828066228248, + 0.4826949595901383, + 0.7685650227519043, + 0.5249720703746862, + 0.2212236745336048, + 0.39535045290974563, + 0.4150589918145518, + 0.37722794408951615, + 0.22459050750122914, + 0.558275791896197, + 0.4301001842491215, + 0.3238799186042615, + 0.42640716915429255, + 0.32175034253893187, + 0.42861198112854726, + 0.48652605771528623, + 0.35598459036728813, + 0.5045362207003836, + 0.42642522102402597, + 0.32044007661677487, + 0.4362746529946906, + 0.4748113216407911, + 0.2517093371948809, + 0.27604577413745995, + 0.21774109155359708, + 0.5606600383242252, + 0.4829254848534537, + 0.8097862677654883, + 0.5001339002138598, + 0.36846790660894335, + 0.26141967819679884, + 0.4465167309456696, + 0.4227426005964498, + 0.5872133850544886, + 0.3046101743210992, + 0.5068003006949657, + 0.2706710457835335, + 0.3088397909580367, + 0.22597203534446886, + 0.2213844183671285, + 0.37479932222541157, + 0.19042731803884613, + 0.3719695334713237, + 0.43022864685751644, + 0.3420478860975032, + 0.22904425145780397, + 0.23822551843171808, + 0.3777161172521852, + 0.22701966422720468, + 0.4808058471282069, + 0.3974340398574979, + 0.4240853480440935, + 0.38142839965011477, + 0.2900809169545098, + 0.5353097347710426, + 0.44194904942070307, + 0.3095630665950116, + 0.5723272376561165, + 0.26461463820741826, + 0.3905645687578044, + 0.2983137048363408, + 0.2403689228382088, + 0.24964532007089638, + 0.3087766024859434, + 0.4182587541699402, + 0.2776191917432516, + 0.6719279896200893, + 0.41933150664938923, + 0.2409165572323836, + 0.30454048253463106, + 0.2458059392441612, + 0.5264104669854084, + 0.23393525598073184, + 0.308429164991943, + 0.32583722578305957, + 0.3578281902143772, + 0.23972032773463423, + 0.36990703097029126, + 0.2531857869224777, + 0.38946454554132204, + 0.757234824292744, + 0.38955986004341253, + 0.31607329508407245, + 0.3327047450889518, + 0.3197175103538106, + 0.20846307420271543, + 0.3111223051557824, + 0.3948601966854757, + 0.7390981921735754, + 0.5224867626537413, + 0.4045267485813751, + 0.3260591529262068, + 0.3133831788937059, + 0.3553022589789066, + 0.22868777359080916, + 0.3927095384342941, + 0.4809793881395718, + 0.27729867754408005, + 0.3655245541151178, + 0.34451948999181875, + 0.4291567070761698, + 0.3209588360320599, + 0.4991837456606168, + 0.340294398254044, + 0.27035651269332633, + 0.39332828459413344, + 0.3995177335484157, + 0.5157186797993792, + 0.5101320485668449, + 0.22469618551559603, + 0.7070323947484172, + 0.703530689908521, + 0.194382496892989, + 0.26146674759996186, + 0.3375062123857787, + 0.42150577434680686, + 0.23111819001090686, + 0.33812891109483195, + 0.44138824813634225, + 0.45648257267009273, + 0.4285076820527602, + 0.3393797718395296, + 0.3239813213236981, + 0.3226607088100477, + 0.2989425676254542, + 0.4129339981431156, + 0.3540161124619342, + 0.35510030884768273, + 0.4762870361706437, + 0.3787813176274186, + 0.5079653112584495, + 0.42351233203896427, + 0.5917525929891084, + 0.4235676562534633, + 0.390674070566952, + 0.4710057620700464, + 0.4303929705355197, + 0.3122291026916861, + 0.6064850911220242, + 0.5837278610828751, + 0.4164996054954971, + 0.3491950620927999, + 0.34885421380515164, + 0.5071099772395478, + 0.31195721599374915, + 0.39341278961779175, + 0.37508875073102244, + 0.4540604592257188, + 0.4131833058787717, + 0.3646632749793512, + 0.33453176513245814, + 0.5075027577914069, + 0.4676723400272156, + 0.44250712946490045, + 0.45924122278973073, + 0.35581582073683327, + 0.22598189031693303, + 0.20487405550583815, + 0.28144937382749435, + 0.4979933415801336, + 0.43872144064541757, + 0.2208128304520413, + 0.41858550638836334, + 0.3270716744745912, + 0.31788445300454576, + 0.3194857578667684, + 0.3576957899099475, + 0.3141379468034288, + 0.42290097440640967, + 0.6091373539646074, + 0.2925404091623891, + 0.403454208381404, + 0.5115357866891004, + 0.3206621458414955, + 0.402923973988395, + 0.571826067877154, + 0.4366938146864898, + 0.4307245627754615, + 0.39315268647573415, + 0.3730252133874417, + 0.22922664134872864, + 0.3477040805187826, + 0.4684756114062495, + 0.47420013954443774, + 0.45453620204141765, + 0.289998727669516, + 0.4401424616294728, + 0.430501402933145, + 0.4857998755125784, + 0.3892116457331087, + 0.453433093958886, + 0.3740726228965481, + 0.4415959428080976, + 0.42964942391859634, + 0.39546260974834885, + 0.26516598014770243, + 0.43369377621108224, + 0.4488589270864968, + 0.23427170511417, + 0.3378734292205462, + 0.4256807784394697, + 0.3745155127867763, + 0.34229848087038023, + 0.4714176381529745, + 0.4774106590402368, + 0.41551295096413693, + 0.4362908332998213, + 0.3121566329961008, + 0.4335928151927656, + 0.35371558313594864, + 0.34910485454530216, + 0.3731154058048994, + 0.4286273497071974, + 0.3670500236969819, + 0.39659734017874687, + 0.37796842995055346, + 0.711286198722274, + 0.5320783731671695, + 0.4897093807927895, + 0.5238585458316394, + 0.456188623319985, + 0.4021389420035435, + 0.5272188730453037, + 0.33989361815240066, + 0.3716018963627411, + 0.45936482611230645, + 0.30871694239962627, + 0.4254047714838857, + 0.4861813567505643, + 0.4328697681588496, + 0.4384411725664732, + 0.442171372603534, + 0.3096268542394185, + 0.5081350189066534, + 0.4819336074125611, + 0.3546077332379798, + 0.40828211246295193, + 0.36057362111792957, + 0.3267214160618004, + 0.4261863714535014, + 0.4202766941451077, + 0.3996641171644747, + 0.4377646007115371, + 0.3682172311589045, + 0.4640785548633187, + 0.34852958530303957, + 0.374962279196752, + 0.42279630134399293, + 0.4173363833952529, + 0.3950583092169228, + 0.46032731361409, + 0.38616243345782025, + 0.33146882532612687, + 0.340128375203654, + 0.4127130349067839, + 0.39365910253246295, + 0.42956699971837864, + 0.4533779060687256, + 0.4098571678206011, + 0.4022610864402929, + 0.34018804315007056, + 0.3891014186415822, + 0.5102108982432704, + 0.37013070461226966, + 0.19594239624948107, + 0.33323429367265534, + 0.401029773669757, + 0.29407515745511287, + 0.2894200960707318, + 0.4354479713928045, + 0.4370198735735562, + 0.2546196093356415, + 0.30273680173047646, + 0.334121201134152, + 0.31927929015680234, + 0.49368775343624116, + 0.27022283795767876, + 0.27048612404017386, + 0.2902142850202491, + 0.29566228133570294, + 0.21311555678395533, + 0.2789148470854416, + 0.25212936552506915, + 0.28114769472278595, + 0.2421048024790265, + 0.3810484596176992, + 0.2681759335027538, + 0.3106111055983204, + 0.3194472863138492, + 0.3715573708375194, + 0.2750456391658697, + 0.37257779193558166, + 0.32193470449120837, + 0.32256416633126067, + 0.3252975388473171, + 0.37701310183338577, + 0.3482176197993219, + 0.5640195010082324, + 0.2914926074455348, + 0.3604158976917825, + 0.5107298552826947, + 0.4616616286004852, + 0.3588947737166068, + 0.4406356397186281, + 0.39732077902201013, + 0.43509084460340836, + 0.45728221669262364, + 0.3430738951867321, + 0.5653850301238084, + 0.4107859895489307, + 0.40554108873196554, + 0.34265383165884156, + 0.4106459716934519, + 0.3606052692435602, + 0.43689897051398086, + 0.4394511519139482, + 0.43112121437986517, + 0.45053890769782917, + 0.5648372128931473, + 0.4229061293188212, + 0.3827359542174169, + 0.4507334439514407, + 0.5162160416209625, + 0.5053597266792322, + 0.42984254621597723, + 0.5317400759635988, + 0.5271675843665544, + 0.37016091951447627, + 0.3651289692544647, + 0.415263785411151, + 0.33176080623462983, + 0.600498118346964, + 0.4749659870797041, + 0.401433010717546, + 0.3395604316580783, + 0.4507727818303472, + 0.3858430212248723, + 0.4974464292717541, + 0.42466351200659636, + 0.3804341965182474, + 0.49009903065301663, + 0.2734942995607281, + 0.3927145144165143, + 0.5049217159500288, + 0.3129888677110699, + 0.39369024765236316, + 0.2970009486751475, + 0.33319452253831666, + 0.35747452030403737, + 0.36599939789365843, + 0.43275317400297336, + 0.313333841437325, + 0.3053711205643479, + 0.3986912741928104, + 0.3673006213269948, + 0.3555070347751828, + 0.4236017212787, + 0.32213399450571656, + 0.4000002634670026, + 0.4853990420595242, + 0.546504023038552, + 0.43615692544220647, + 0.3346380428973556, + 0.44723331770604713, + 0.4210326212047745, + 0.36737292608038596, + 0.3960276433177055, + 0.2964833856384832, + 0.3040410953120274, + 0.2712618345514804, + 0.3500684991882343, + 0.25771896301599356, + 0.6378783727779207, + 0.41011965452125665, + 0.38673663646216017, + 0.28852525582927474, + 0.28797664426591263, + 0.31044438533209207, + 0.30943596015300034, + 0.27905439489731904, + 0.28670127248681915, + 0.2794386827367167, + 0.35172089067863427, + 0.35395916917488734, + 0.41513922910425705, + 0.3835218327127068, + 0.3442351143910963, + 0.4831248047420651, + 0.4776184710190782, + 0.3523092824044027, + 0.3943140054129665, + 0.40222961647546734, + 0.368173035097252, + 0.33400831738565906, + 0.4971729530261475, + 0.42903298849045884, + 0.5628306084049682, + 0.42768477576125974, + 0.4212778028838702, + 0.42876448046899834, + 0.4351662791351257, + 0.31343876306038704, + 0.5557050504329779, + 0.397965322254319, + 0.22142232976442713, + 0.4364811829291017, + 0.5417446514121806, + 0.3477145873061901, + 0.41394726763301176, + 0.723805147663172, + 0.41701601603443433, + 0.4817516218032316, + 0.2621740432999576, + 0.45127786905708217, + 0.24691525389432004, + 0.26497940944218373, + 0.2856919891898113, + 0.2390142952072589, + 0.4292683178422009, + 0.16979767889105638, + 0.2802530971517188, + 0.42487790364389716, + 0.3761539504390425, + 0.3823830779690974, + 0.37972679540738585, + 0.30108901263018806, + 0.3564634138168675, + 0.4983512196143385, + 0.5117243198762628, + 0.32825432138598737, + 0.4230557989937713, + 0.3777988496109023, + 0.7096230310465604, + 0.35531153593449083, + 0.5430783763523648, + 0.726356991502742, + 0.7022478585389347, + 0.414885435703171, + 0.3421924581591741, + 0.27363419725239435, + 0.42111393898970695, + 0.46384426500954795, + 0.47576352860539717, + 0.544176535474554, + 0.4023325617463771, + 0.5354076489839468, + 0.3899698894927203, + 0.3571183910112994, + 0.681902956751277, + 0.6679878860143731, + 0.4294873629983448, + 0.3899972855450229, + 0.4384685355453423, + 0.3893236230458348, + 0.42370194113033677, + 0.6295845951715647, + 0.6458725854889624, + 0.39505760914763616, + 0.41031953121172626, + 0.4118458229339016, + 0.3342196420549494, + 0.47122080492827606, + 0.3022212242970777, + 0.4516098867960625, + 0.45086532192242734, + 0.3284775377091107, + 0.4097505074079694, + 0.35486704239788586, + 0.2989288168189717, + 0.4801930521106024, + 0.4051261212247119, + 0.3676551902806973, + 0.34454035443011416, + 0.4505696322255207, + 0.3860669878781093, + 0.3063291972174266, + 0.3234082421178379, + 0.22272400488751945, + 0.21233214352886756, + 0.24147605461652197, + 0.3937370805448189, + 0.4233866532523845, + 0.2889431301406871, + 0.30769835437606724, + 0.2878052074449826, + 0.3876669309846373, + 0.38631471798602535, + 0.46418245520599977, + 0.30551321174983803, + 0.3725304845573117, + 0.4739987237546256, + 0.3196304897294091, + 0.48611059387492334, + 0.43114508449793776, + 0.3788779470356583, + 0.3778727623581999, + 0.3208152057893859, + 0.20367754231682933, + 0.2589396745138346, + 0.25375653787809666, + 0.4842548012683973, + 0.3811567479122509, + 0.3956519776806246, + 0.425670648137242, + 0.45034080480054917, + 0.2923637635028856, + 0.3667673097622826, + 0.3319451673061349, + 0.6347645099396754, + 0.40958903508323286, + 0.438653121640497, + 0.3383042894134842, + 0.42444305934876003, + 0.45591879603719826, + 0.40128233433767935, + 0.37331691293569363, + 0.449701356799192, + 0.4160222429485354, + 0.5163402094246753, + 0.38323492874669135, + 0.44777852992959616, + 0.4105798135729472, + 0.39151063373796263, + 0.36416755267175865, + 0.4381790409009921, + 0.3835552801710088, + 0.45169984295061366, + 0.30999567563608804, + 0.2899952881937466, + 0.29055841816107536, + 0.36630226638874186, + 0.4138884386426719, + 0.32537891880114583, + 0.375012538111093, + 0.4240407249658556, + 0.389908448333997, + 0.41780111282990856, + 0.39911274918598716, + 0.25411160904971625, + 0.37871392279979227, + 0.215091192650451, + 0.3850593170913676, + 0.19760762082326017, + 0.6332891502406175, + 0.29924748926631667, + 0.40481935150762255, + 0.5126304580743981, + 0.43150820333757195, + 0.24136594016061885, + 0.37416910974125894, + 0.3022827992526133, + 0.758530288001689, + 0.2575861440418835, + 0.46933103149042416, + 0.47423439369094056, + 0.4731643887706389, + 0.6181723563846656, + 0.37628200603154627, + 0.508414084712605, + 0.42356963016603427, + 0.33571930428255714, + 0.277243789715861, + 0.42195820780096366, + 0.2768931017586974, + 0.6971191859156717 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.38721842855496186, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.397597310765855, + 0.7411656485570381, + 0.7272770851150084, + 0.8087828893319039, + 0.7072571212754197, + 0.6280334642360244, + 0.5814471412265654, + 0.6411858642009013, + 0.3878410666572805, + 0.6764722916349213, + 0.28694821745326643, + 0.44342656173913403, + 0.2911368539619454, + 0.2916326312058275, + 0.27103226376588124, + 0.2572585397518227, + 0.30442623611142483, + 0.49923659696231554, + 0.34666626976759485, + 0.3563448216005758, + 0.303716367245645, + 0.653686875572357, + 0.2670981368165173, + 0.573122854610182, + 0.34227578798328195, + 0.1541731169384448, + 0.3580390049226718, + 0.40119290986038686, + 0.4236139120983669, + 0.4558810878531269, + 0.25451122776682455, + 0.3654456993570827, + 0.30402578846020123, + 0.31685921532109423, + 0.3928427102680385, + 0.32200411465305107, + 0.2759138686570593, + 0.47716503440692215, + 0.4916859610830955, + 0.6182355049233396, + 0.7767281382033696, + 0.7710851760305864, + 0.7589744861380278, + 0.6337350971191423, + 0.3593764741560558, + 0.31558686967868543, + 0.7661768378077596, + 0.28369355508687616, + 0.29916393813775877, + 0.4220119897575826, + 0.3158169633624967, + 0.7672052813053382, + 0.7857626499244871, + 0.36277839478971435, + 0.2886189426074232, + 0.31661694903348797, + 0.3060311070822784, + 0.35317920400051894, + 0.34282608511304086, + 0.46522349680707714, + 0.40346806552298226, + 0.40074646011510606, + 0.3568026286128722, + 0.4618443573873398, + 0.39313742519719347, + 0.3311056646665632, + 0.3168599867438221, + 0.47582458465017063, + 0.32607912995498073, + 0.40220551674340294, + 0.4406167774543942, + 0.3852740500353711, + 0.36140141503564455, + 0.3488939240437687, + 0.37379944534278287, + 0.35572672379900544, + 0.25237408258096705, + 0.2461137170735381, + 0.3478721562533705, + 0.2982150749109447, + 0.4037031515392064, + 0.2713082847351936, + 0.3081341796004297, + 0.25521357870187467, + 0.3700982691364768, + 0.4545644651150782, + 0.3715545543043927, + 0.25117042269854134, + 0.31919829887456386, + 0.30329874812510627, + 0.28682782550369085, + 0.30580786776595176, + 0.44761036773711543, + 0.46972067337061557, + 0.2276568510497995, + 0.5128236263967175, + 0.30753909531428253, + 0.4785409279161955, + 0.25667187586242507, + 0.40974771593773157, + 0.3913327401579082, + 0.46901811563818163, + 0.1992521520083957, + 0.355385902215181, + 0.2616601331005409, + 0.28796437120620355, + 0.41367103717139153, + 0.28398095349571184, + 0.2935158003220632, + 0.8004844080117955, + 0.7953233865526446, + 0.799785381032302, + 0.8155263573263618, + 0.5704601639029159, + 0.6745451725684666, + 0.6961852985693293, + 0.7467461013661749, + 0.6864870630691913, + 0.575965325172013, + 0.6457327987089806, + 0.5654683021599248, + 0.7635591274160242, + 0.6467152531211549, + 0.40022244647085076, + 0.3037717120060782, + 0.2456182524757113, + 0.2936274759198247, + 0.499694277023641, + 0.2706869185662609, + 0.2539706687028234, + 0.32613297452355117, + 0.36207574469488846, + 0.25294477760071876, + 0.3350809373425051, + 0.3661278332850296, + 0.38884158063894353, + 0.3430350739656413, + 0.2833968940733727, + 0.32049442509389087, + 0.3523643620940478, + 0.279293977517095, + 0.28748809771907874, + 0.2373552837545276, + 0.5777996888321186, + 0.2803792206937706, + 0.5172220269757709, + 0.33894553660676136, + 0.4057678755995305, + 0.21971069786057673, + 0.5397862984107402, + 0.2685200021470265, + 0.46137165319509843, + 0.2905668108303144, + 0.21801972352679894, + 0.3132450015772807, + 0.28431908699239294, + 0.23789837263899294, + 0.15185847320617354, + 0.27941382260924064, + 0.2932189711618829, + 0.44157757052072316, + 0.40642950568847086, + 0.4610487341287633, + 0.4417915349076895, + 0.30388999438261466, + 0.41300315792530995, + 0.35474591396214494, + 0.3869535246146233, + 0.37889413387837034, + 0.3743257212366025, + 0.2364247121724719, + 0.14441648271706567, + 0.3545165417672294, + 0.2703953864102751, + 0.2197972442745117, + 0.2237378752217547, + 0.383021973240482, + 0.4507854340852328, + 0.5446945051830616, + 0.41049098991025257, + 0.2120632934256051, + 0.44400446839574165, + 0.43526516222498063, + 0.46514770568098124, + 0.26166355697817567, + 0.48067423278995347, + 0.3849214460406465, + 0.4664527060310047, + 0.2181969243479358, + 0.34563621881745066, + 0.3110415259947869, + 0.23041247785757427, + 0.4623033483416923, + 0.38927766789881574, + 0.49910162442435396, + 0.2768070378158088, + 0.5837679579189846, + 0.27148096877184763, + 0.18701077595225957, + 0.36730852160566824, + 0.25429667624691005, + 0.24702841794035116, + 0.18873225173981692, + 0.30932500442851657, + 0.5906124839574132, + 0.2154388694464989, + 0.21951122255116226, + 0.3959355767860876, + 0.31510266753965477, + 0.35606327737390436, + 0.4190749859868041, + 0.3033224616488591, + 0.2339011644119619, + 0.3179613887289928, + 0.2967533811627624, + 0.48561724408119994, + 0.4851479112936456, + 0.5300116977244599, + 0.5074054976179762, + 0.24743211949920388, + 0.30531968959417605, + 0.4010164320941577, + 0.2364318425959813, + 0.3016078977788803, + 0.26426128577417785, + 0.13953283668615887, + 0.4194483254066594, + 0.29201955696371595, + 0.22553986759634986, + 0.49451672727525253, + 0.21851828252472189, + 0.17959034830071421, + 0.36460745999508853, + 0.356170290805964, + 0.25722975308325186, + 0.5668234754743944, + 0.2924086932259068, + 0.27460858725254217, + 0.2202539097816417, + 0.39539438765432433, + 0.264353146264558, + 0.4281031626496788, + 0.2480714459950818, + 0.3679162116857143, + 0.3403190893239283, + 0.3408679882735844, + 0.30050248989333006, + 0.29633353711288324, + 0.3059957981025147, + 0.30427709992950003, + 0.42013992432412683, + 0.2547477388578071, + 0.3494822974905962, + 0.2682794019379183, + 0.30405936610169054, + 0.3498774974342965, + 0.35305869695224723, + 0.3209019713018081, + 0.3958091274675297, + 0.2584921477863119, + 0.2789481933042435, + 0.4719065878391003, + 0.23651997266558591, + 0.301095917732027, + 0.27996465725593606, + 0.21590979789271447, + 0.2795652182672088, + 0.24971084057965914, + 0.24265390092445469, + 0.26121082734414547, + 0.23573710302303752, + 0.28837196499475354, + 0.28817104884181305, + 0.3293387834158483, + 0.3403637520580696, + 0.4276000440585255, + 0.20775025847935075, + 0.37137457386798706, + 0.4404085880344947, + 0.23407004880674576, + 0.23604135925861214, + 0.5376200941709677, + 0.21302875190282422, + 0.21759252729031017, + 0.38357752785344706, + 0.3709543431873846, + 0.403920496318995, + 0.3293808415642086, + 0.45901084290974653, + 0.3935872667053365, + 0.3886581341415895, + 0.3204364166362046, + 0.39421894676748653, + 0.3727153681238745, + 0.3505309629571492, + 0.3623480226543506, + 0.44152055654963873, + 0.3897224713026689, + 0.6439292475948437, + 0.2613037311385725, + 0.5372442783696983, + 0.6219492506168115, + 0.6276418035701836, + 0.6245367504908305, + 0.44186825338435626, + 0.43094933678033387, + 0.3650674146238307, + 0.6532119381084979, + 0.39483037692909867, + 0.44806899471580164, + 0.2851157845064014, + 0.42369054566587355, + 0.34811567344812583, + 0.3358861822272172, + 0.37588297046488456, + 0.2792923824925967, + 0.39972724129155823, + 0.2502081229809549, + 0.28955425019222286, + 0.3124730720524211, + 0.31760496167078084, + 0.36237667579050775, + 0.23462042006951528, + 0.30158228860068065, + 0.3214776356539076, + 0.3295205600680799, + 0.2810253001458278, + 0.3276544926249017, + 0.5093792036426942, + 0.3982341757404875, + 0.3412149144247038, + 0.4105064395355569, + 0.40408696763284485, + 0.25730742998762174, + 0.2856940628847243, + 0.33513358703671065, + 0.3639999529847336, + 0.2840692343760599, + 0.390842623809828, + 0.49202317917972577, + 0.48521852437951996, + 0.3425319879992828, + 0.2710606263785625, + 0.42616417329768225, + 0.2390612674638213, + 0.28883521388695843, + 0.41342313826340593, + 0.2875334775500538, + 0.8138999889018531, + 0.4659044716235079, + 0.3672104985441826, + 0.3424593175876498, + 0.4066605437056832, + 0.4130676080099133, + 0.3477405780758416, + 0.4518566134712132, + 0.3033183732193047, + 0.4416082195474628, + 0.34248100580036217, + 0.4235143949007692, + 0.4121880627100938, + 0.3960068502925382, + 0.3739330616409338, + 0.2963856871814381, + 0.30561021562219903, + 0.45991476693779737, + 0.4823810353518302, + 0.36182524601835825, + 0.5414792869839261, + 0.3832351504964773, + 0.5069307188546157, + 0.3422709206020859, + 0.3759347728492412, + 0.39113472987301573, + 0.5218139242504406, + 0.517408864088551, + 0.46035800270362504, + 0.4862145743381143, + 0.4598288584791804, + 0.46944516688494803, + 0.3524924878291253, + 0.34116130781395887, + 0.3073878465634842, + 0.23733667333984784, + 0.3360318938694459, + 0.3023673035859026, + 0.2815920421480529, + 0.30522973066158704, + 0.40943193647297, + 0.30420322270536226, + 0.2665182589002472, + 0.295594249097126, + 0.4985294389059964, + 0.5409594521191627, + 0.4724895702401285, + 0.5102279062759939, + 0.3753940463124902, + 0.5217288686811916, + 0.31492267153667935, + 0.5850662589330827, + 0.3962064255390704, + 0.5056256861242853, + 0.398612502613787, + 0.49762879959201184, + 0.4971102655987371, + 0.44783367001932634, + 0.4101816449138541, + 0.4503129774315738, + 0.4592444905526741, + 0.48082060283541916, + 0.2647847035324125, + 0.3232656147546029, + 0.2730720304208412, + 0.31239805040150115, + 0.4604585832900736, + 0.5149236731373347, + 0.2665007576282804, + 0.5328520446237243, + 0.4399740987913996, + 0.4155032711787456, + 0.33471026985109625, + 0.4112390784014402, + 0.417816693129025, + 0.3173566991581242, + 0.3252359454932762, + 0.840836354528705, + 0.38657823887299797, + 0.40083942815157453, + 0.4377246182283666, + 0.4410068136233474, + 0.4116094951593394, + 0.363486348925292, + 0.5488102486840278, + 0.41034901275241886, + 0.33115641366922655, + 0.3861073272616099, + 0.38617512193321274, + 0.39453144660315237, + 0.37342884728789333, + 0.38288173637054157, + 0.4214143173151579, + 0.4128844566767009, + 0.25314647112848504, + 0.4061795450391082, + 0.25686149914702117, + 0.3516536250409042, + 0.19430074596350005, + 0.40395635011929415, + 0.3241439410039789, + 0.2444119487053781, + 0.4312506343512217, + 0.17679794003351942, + 0.2342344579746214, + 0.5206349465667272, + 0.45104160644316627, + 0.41776207481749017, + 0.41905123686854084, + 0.3693528751136291, + 0.3794280067503435, + 0.5570075784491185, + 0.3102031333725869, + 0.5287292612551722, + 0.31535282534234343, + 0.2346776706663807, + 0.18639813720282827, + 0.35363359559981783, + 0.43885308982986354, + 0.4865572823504055, + 0.517285252841385, + 0.40291735089414643, + 0.45636729812696275, + 0.3212082537553653, + 0.4356262713131833, + 0.3474346685289307, + 0.26570687544166016, + 0.3171809315783325, + 0.3290949608717205, + 0.39529607460227933, + 0.3108759924712797, + 0.344431634189767, + 0.36682543504570164, + 0.3715413066192281, + 0.4297239939498389, + 0.3567594110813034, + 0.31927417527959845, + 0.449525483422159, + 0.40944918787672796, + 0.493262509947121, + 0.4491613958015244, + 0.5226668840407535, + 0.5357943528262485, + 0.6476553029454929, + 0.6330469197034854, + 0.4591149676106005, + 0.5048810119285897, + 0.3904496261063583, + 0.5014424519079962, + 0.3708835419862814, + 0.41997418155470206, + 0.353590776539421, + 0.5569767454089913, + 0.3608066499886764, + 0.5133099754887788, + 0.6304189875699039, + 0.6053325928149059, + 0.540247633469741, + 0.622427172182909, + 0.5159900355645899, + 0.4364746806456422, + 0.3775929251586727, + 0.3562738332221743, + 0.24587183968788287, + 0.3219877008192342, + 0.4109748889640148, + 0.49136112394755815, + 0.36005597776061854, + 0.28305590235929345, + 0.19798114042314732, + 0.4073337344143119, + 0.3999917914508869, + 0.37664234063789165, + 0.26490570465257623, + 0.4809846068146234, + 0.3587490845267391, + 0.4035952788993298, + 0.310181402042206, + 0.41088560536585356, + 0.4368957753818759, + 0.29117121574861743, + 0.3374010485334153, + 0.39281309576687823, + 0.32695524541112525, + 0.32457286630978155, + 0.32515685790973536, + 0.3268249416596575, + 0.2896739236508333, + 0.34804809312996904, + 0.3570044069540035, + 0.36159332414685985, + 0.22885818662783328, + 0.5263709490883117, + 0.3575668048824457, + 0.45569571943549353, + 0.4718447798087251, + 0.3225264411119937, + 0.42494848017699166, + 0.43254015443945415, + 0.4344752459436814, + 0.5366892344193653, + 0.3623094119964938, + 0.3930368803238393, + 0.43177380619495437, + 0.31466996240422224, + 0.42092852262178876, + 0.3803713346192237, + 0.28579146182655385, + 0.40127954735183874, + 0.4196544758195065, + 0.3762746306124278, + 0.2507377621673322, + 0.3625910502291904, + 0.2225712869720462, + 0.33562593267953617, + 0.579756934978982, + 0.4950790822601452, + 0.5098812367120831, + 0.5113272385315338, + 0.5899350125344524, + 0.4987538316619577, + 0.5456780634955244, + 0.5074305671589656, + 0.5519215788116664, + 0.3709174035276166, + 0.3527375230579681, + 0.3678269977929797, + 0.2858556518453979, + 0.24482103908924382, + 0.32175492925453375, + 0.31228427465598385, + 0.27730433014655964, + 0.23649142154339942, + 0.3488814215557226, + 0.2517591331736366, + 0.4087937256871359, + 0.2861973668932646, + 0.4930582639228278, + 0.5297301669662421, + 0.4925267912498651, + 0.2527711128997722, + 0.33264937163460123, + 0.5499203627258346, + 0.2650429738786589, + 0.21654384602529145, + 0.229669552658904, + 0.81807499796823, + 0.4855766784843697, + 0.38544256702286817, + 0.5206231972755647, + 0.4774613734961803, + 0.4555743382233916, + 0.3790133935634213, + 0.43225380907250543, + 0.4756795753450096, + 0.4062136991605387, + 0.5244056403044158, + 0.4806419382151316, + 0.21148738069360107, + 0.3340309518316049, + 0.2283287921142058, + 0.31355413667634785, + 0.24987066594074384, + 0.45572126731193496, + 0.36792051379233864, + 0.29030907316605375, + 0.4147355799254271, + 0.47281332860355124, + 0.29356930441084167, + 0.4149049406449704, + 0.24066433559961067, + 0.4033499660475157, + 0.4566559361341027, + 0.2948319595496265, + 0.2942437882297128, + 0.22903671795070799, + 0.5034580629573173, + 0.2288795131834501, + 0.29611656995921004, + 0.4722092494837467, + 0.36309473824963445, + 0.38455652062728096, + 0.2807234947259456, + 0.2956008342891745, + 0.3672677846272809, + 0.3400598199391828, + 0.4115074027617374, + 0.29262458454316675, + 0.44968850261592147, + 0.2386361758043122, + 0.4582468464671177, + 0.3872573766600373, + 0.3504081468930473, + 0.4345246688872954, + 0.34898848383236597, + 0.44886550692720867, + 0.2979530936782391, + 0.46058754714544, + 0.3455492925143263, + 0.27386268117486884, + 0.3976908522049151, + 0.2738319105308737, + 0.4652356195222987, + 0.2052855010311476, + 0.6579386551741782, + 0.6634074713675692, + 0.5545975320234312, + 0.5559123097674057, + 0.5952777437499289, + 0.5940601404124793, + 0.24480473830096486, + 0.32843621390855565, + 0.29554284786410806, + 0.34524761782190433, + 0.29643742190338696, + 0.3130677725387827, + 0.33155502905603246, + 0.36550626122311886, + 0.3862344608069894, + 0.29541499249047104, + 0.3325181446139241, + 0.27803346245234273, + 0.4501253413857104, + 0.2407747692117018, + 0.38350270878866366, + 0.3936957678082391, + 0.3344050258612119, + 0.4486977204634364, + 0.35623093318169935, + 0.4740513731746844, + 0.2351142741839148, + 0.4397014419642119, + 0.34173589204491905, + 0.3254180343622615, + 0.28753157631948884, + 0.3392138923965416, + 0.31159566366662816, + 0.3383394456121815, + 0.47946232714705717, + 0.43129520365304075, + 0.5151992370663189, + 0.3175787559949438, + 0.3765624788104333, + 0.35502829836371663, + 0.34943932927443366, + 0.43616999370825804, + 0.3308888044131038, + 0.2062738277384488, + 0.38715527384832643, + 0.33544756904283624, + 0.33048102065944257, + 0.40358195463744834, + 0.3137266687011792, + 0.25162854851219013, + 0.23340390592810858, + 0.3082489883757558, + 0.32855730759421564, + 0.27275475382892983, + 0.19918698536171056, + 0.2945851822255015, + 0.4748764223102548, + 0.5806805557456994, + 0.5057336529015699, + 0.47001259135681844, + 0.5058851433158839, + 0.44576760195357784, + 0.5158684928823232, + 0.46287880758678246, + 0.45340967478382566, + 0.3190291311599916, + 0.36012108605543874, + 0.4314504391906732, + 0.5717522506119554, + 0.15447385056340296, + 0.28214097675934285, + 0.2672123139888133, + 0.25037857633287364, + 0.266024670860071, + 0.3659717200691271, + 0.2719415838032753, + 0.343179923593162, + 0.7232644244228266, + 0.6015271050951733, + 0.7501450594641826, + 0.7658773159203731, + 0.6290543727468113, + 0.7084607662198431, + 0.7287012637970023, + 0.33132947068064256, + 0.32864334708067183, + 0.3561144184373371, + 0.4697310605361428, + 0.3020200534055507, + 0.34372571865493357, + 0.33256113392597864, + 0.3335419639982097, + 0.3489887314844633, + 0.32661168864018597, + 0.24350558564130526, + 0.4103818533856532, + 0.3538494080139996, + 0.3302602838875952, + 0.38808880064091456, + 0.28776930431488196, + 0.29464701204133137, + 0.30372152208747805, + 0.28866979501022555, + 0.23156322725389136, + 0.31508698768595894, + 0.23841163717823546, + 0.49034762271342663, + 0.44842734185594363, + 0.32136406821442914, + 0.25584639211544913, + 0.18935789311801207, + 0.48115808790138864, + 0.5501363795916925, + 0.6272886390445926, + 0.2841151788242587, + 0.33789418981440045, + 0.22915348756075368, + 0.4254555820597103, + 0.30622517320475756, + 0.2676900854829668, + 0.4496229361042637, + 0.46492577275677927, + 0.4339494353100011, + 0.2982207072365898, + 0.4068865203953432, + 0.39924778747387685, + 0.4502811994205553, + 0.41518277332797754, + 0.45519008825937096, + 0.34874203313520324, + 0.3939966509159828, + 0.4022332521642188, + 0.4076949046996722, + 0.23784362237252768, + 0.32615450827505293, + 0.37017837086271554, + 0.37844815739525534, + 0.52715609540966, + 0.18479279635529997, + 0.5342169386275772, + 0.6939765204915892, + 0.5248461874717151, + 0.3568024927372156, + 0.2779137806651649, + 0.31124708605146456, + 0.23609762672534143, + 0.3664692926493296, + 0.44098051099237334, + 0.41283469805639933, + 0.3129371471728522, + 0.2550663961815856, + 0.43718252968273763, + 0.4768313250293298, + 0.5136280570479861, + 0.5786156542979011, + 0.4222725202935166, + 0.49851180616173774, + 0.3226939801072063, + 0.42858484110753403, + 0.24257722253122171, + 0.244737686938647, + 0.3584100496437081, + 0.3336234752812936, + 0.47825464683561686, + 0.44571198880541457, + 0.587951016666427, + 0.642339776107798, + 0.7400500943321563, + 0.5790975218329781, + 0.7522073579786068, + 0.37688269860497386, + 0.35562965734662044, + 0.3587493960185815, + 0.4194663447190098, + 0.28122893603424104, + 0.45885704702563257, + 0.4128072274691964, + 0.45914407341334723, + 0.4401153862763562, + 0.3722942917766795, + 0.38521671285316555, + 0.39906738125338564, + 0.33456266640054333, + 0.39451638366356795, + 0.36569229596409075, + 0.39448374807618647, + 0.44978240136634845, + 0.4599386068070845, + 0.4483630010319428, + 0.3417429822979003, + 0.3826904093269517, + 0.4306780842313979, + 0.30181494756369187, + 0.3989754804521913, + 0.36503321879223843, + 0.3892579654424112, + 0.34981760654037763, + 0.39746678770543925, + 0.32028962047895865, + 0.3898418049450255, + 0.393656351565251, + 0.4318102396568835, + 0.2973128707836594, + 0.44983109129799137, + 0.41203254679264306, + 0.35656620275256956, + 0.4885855497613958, + 0.45682960346403084, + 0.7314085382414609, + 0.20501015235226772, + 0.27218647544411445, + 0.3269033641876457, + 0.3158456957019923, + 0.23939537157778698, + 0.2013156958197486, + 0.2660403932420232, + 0.6614910747327488, + 0.5360733864953001, + 0.6221862238212846, + 0.30579079389301494, + 0.303010351439668, + 0.24481628381601753, + 0.3364514697531652, + 0.37281908767007316, + 0.20071897645536368, + 0.23783404086062493, + 0.3706507624314158, + 0.33985522730124373, + 0.4015828644373182, + 0.340406035385944, + 0.3795934714735601, + 0.32929487811284247, + 0.27380924125219996, + 0.6825185221574372, + 0.4411342439344193, + 0.682436320592845, + 0.5589130086266894, + 0.21224335101045538, + 0.4672846408950582, + 0.44167811042974214, + 0.4632420882376424, + 0.4318872577119608, + 0.3478385607682421, + 0.4272491292788729, + 0.43824582250617294, + 0.46453963666732395, + 0.4266782701507487, + 0.3029857057186648, + 0.37940047275700767, + 0.2307905071576504, + 0.2511468628745962, + 0.6217533440120921, + 0.43093950718529284, + 0.5571456160414962, + 0.4586457641906173, + 0.417255466423205, + 0.3972665483740383, + 0.3680142471187338, + 0.4557747954418613, + 0.36725215693228064, + 0.2139087242463665, + 0.33915895672443747, + 0.42226653341002446, + 0.22731029148427498, + 0.7490978467389875, + 0.39279243261306734, + 0.5518851486645314, + 0.7131181917813236, + 0.67345572526269, + 0.3414947234233901, + 0.79358057060097, + 0.4641704384198309, + 0.6462917381906736, + 0.3583156449476101, + 0.4120731556211386, + 0.3295749639896622, + 0.31237918737231984, + 0.42349954226819814, + 0.45726717123523986, + 0.3893838145984808, + 0.31031956228609553, + 0.3384005370722445, + 0.339418868986373, + 0.3645480282539028, + 0.4643687817283485, + 0.22359255392396568, + 0.4461409386752097, + 0.3870467311301256, + 0.36541930961064756, + 0.44667139938950956, + 0.516191323819047, + 0.2755034394773082, + 0.35996416492123695, + 0.3875731199198659, + 0.4909719516927855, + 0.4596044080699294, + 0.3716805232209838, + 0.25061742792711295, + 0.4557299125216722, + 0.5153678394821527, + 0.5018538276286869, + 0.38108256027514975, + 0.5588802852559904, + 0.567462206263981, + 0.3897871406093858, + 0.4522971204544265, + 0.49366820871748246, + 0.45473440406332577, + 0.40806974302791893, + 0.3039393243953033, + 0.2531266545314413, + 0.2766084967580695, + 0.32483254014906193, + 0.26417381291539543, + 0.24342853832206818, + 0.2859902833318389, + 0.3114069356040293, + 0.3293069412976357, + 0.26814160000485654, + 0.28859471501456757, + 0.2799559713263158, + 0.2972853519659702, + 0.36103457077134204, + 0.28176066244399883, + 0.25648144141649276, + 0.3808906673180997, + 0.4769118192528092, + 0.6215533436462015, + 0.25555580108942755, + 0.4797568741636748, + 0.4932326675967699, + 0.49789751377152763, + 0.5071174640682211, + 0.3801096909971367, + 0.4141773061834552, + 0.34041929290117323, + 0.4360118267213592, + 0.36876340743925334, + 0.6098625056836896, + 0.4136691107209265, + 0.4487698371890088, + 0.24227596695351913, + 0.22548610509436987, + 0.36562915722358663, + 0.44375585505596266, + 0.32385100941909223, + 0.5392611970590518, + 0.3009335209455349, + 0.5955490107949002, + 0.45546517054855257, + 0.46925029768850224, + 0.4697760004134698, + 0.45769204023957943, + 0.37640732911637764, + 0.48540961775927505, + 0.4783100780767292, + 0.47990612315218, + 0.4786408070530979, + 0.3926233039669983, + 0.30360690373836013, + 0.23867831938640616, + 0.33740781269506265, + 0.21336737222727464, + 0.3555257152541472, + 0.3888067421582697, + 0.39572148515492844, + 0.3792386786493435, + 0.4075326895925398, + 0.37750967164930055, + 0.46543271848378726, + 0.4232874338096224, + 0.30431008806087856, + 0.4322672684386149, + 0.49928491438974903, + 0.31023551212933825, + 0.26517442424356097, + 0.31199949983941994, + 0.3485294680820037, + 0.42054269883578327, + 0.2296348786985299, + 0.39636019174418846, + 0.23247109536548244, + 0.37422250517048306, + 0.43583932424037386, + 0.4496744203200757, + 0.20863814427954025, + 0.2770945391617699, + 0.2956168196470272, + 0.20187304504918935, + 0.19200755585436233, + 0.29875924872656145, + 0.40150740811986124, + 0.24377313880548612, + 0.3116913386766913, + 0.404690053071502, + 0.28477796507426734, + 0.29428408945809365, + 0.39808578320607196, + 0.4507586649395339, + 0.18577943428793867, + 0.4227725274298279, + 0.3456522136976684, + 0.2910861689473099, + 0.2646715205269199, + 0.41761875438845253, + 0.3655006972022166, + 0.1846059397559281, + 0.4380923261323077, + 0.37631421036045015, + 0.346928718444624, + 0.45003058387812733, + 0.312943773109309, + 0.2591903500895202, + 0.31017674510414495, + 0.2222450854559638, + 0.332032900927983, + 0.30958229313853436, + 0.3296719152055621, + 0.35904313809521565, + 0.36201609379751476, + 0.44142896134569237, + 0.3760457706893173, + 0.2989087289474206, + 0.3424721083960292, + 0.3851843492561018, + 0.3651738947571142, + 0.33314795545679254, + 0.43309498405897023, + 0.37220811889534927, + 0.4804543768148956, + 0.5156691150368506, + 0.4757344441549841, + 0.41489530868442687, + 0.30646014187710313, + 0.22134651646033257, + 0.3501608312458654, + 0.4170641891719483, + 0.3569600265292123, + 0.40939775994395633, + 0.4059621009459798, + 0.27737956762026683, + 0.4191749330375119, + 0.28775227461450126, + 0.2364015536659561, + 0.36487324748380073, + 0.38369137154882177, + 0.4154006008625409, + 0.42238459331373185, + 0.2604313627065534, + 0.5377395650701542, + 0.44175946036892727, + 0.44182444233607304, + 0.4803007160162302, + 0.5494047764604797, + 0.3889132569682163, + 0.2795606583398785, + 0.3552387487104395, + 0.3601643525942641, + 0.3536721092060918, + 0.3345304036295984, + 0.2940166297474131, + 0.30951295001781143, + 0.36268403936985694, + 0.2565037768060527 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.04044274918123718, + "valid_tracks": 4, + "per_track_pearson": [ + 0.044021783319805524, + 0.03822383603995233, + 0.041003728207205055, + 0.03852164915798582 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.16648503868770018, + "valid_tracks": 626, + "per_track_pearson": [ + 0.1983298019515548, + 0.1922119912839516, + 0.17811877964197564, + 0.19170990711970926, + 0.19667860691552325, + 0.19451989256549304, + 0.14296983519024878, + 0.1667341588535513, + 0.2091738725888962, + 0.19044142317398619, + 0.19276788020348673, + 0.1937095502109928, + 0.1983240736116098, + 0.193994615917778, + 0.1842973428603371, + 0.1904347425259197, + 0.19765186161583323, + 0.186776192747932, + 0.19731047648002753, + 0.20601981585291712, + 0.20766802328322348, + 0.18963800896531882, + 0.1922435041151215, + 0.1524306003398744, + 0.18978768297300405, + 0.18199417476910776, + 0.17331449279530906, + 0.19691907271108408, + 0.18445389670606652, + 0.17528032030057944, + 0.17699564606197998, + 0.17565297927998505, + 0.15025535178872873, + 0.20294751366214533, + 0.14929415392170578, + 0.19785176613873512, + 0.14232592752465192, + 0.19400369028327652, + 0.18860597592774597, + 0.21145352388875951, + 0.2052301609664116, + 0.1745276966077065, + 0.17653909054844663, + 0.1702390964349996, + 0.1723512087998118, + 0.15973812040818042, + 0.17795472434840584, + 0.18623304561478088, + 0.17575820343952317, + 0.1677398036652207, + 0.18467130171094853, + 0.15739792340350492, + 0.16537339624989966, + 0.182935823438077, + 0.1783217104532488, + 0.16483436617068534, + 0.16989262090985272, + 0.17551353938217149, + 0.19424357103986947, + 0.19251980412008296, + 0.1755577184048381, + 0.19259229137968034, + 0.1954231115545591, + 0.18393766217389554, + 0.19315165250314825, + 0.18347575960188686, + 0.19510814380338004, + 0.1843757769946205, + 0.18874729832119808, + 0.17190254759403545, + 0.1803462173870484, + 0.19234770895230852, + 0.18375448788465096, + 0.17776416201385592, + 0.1933605923217053, + 0.1901558256800328, + 0.18307809862858018, + 0.17084810810310158, + 0.17882277017878762, + 0.1733726755723364, + 0.1825222541300837, + 0.1794969080134945, + 0.17818247057105926, + 0.1764142873837774, + 0.17942992290930573, + 0.198885516767753, + 0.18054338146768623, + 0.18285537224708168, + 0.1857801428169083, + 0.17816738762244627, + 0.17981957031526422, + 0.17634234099073068, + 0.17943920458764956, + 0.19413037892504603, + 0.1781206750572119, + 0.19750855364111392, + 0.19336551558629578, + 0.19306579644632416, + 0.19177838742593678, + 0.18583323792591222, + 0.1919144394268746, + 0.18175808797873766, + 0.18743505559219287, + 0.2030392295454213, + 0.19377550745264166, + 0.17049771577130476, + 0.15447279955672974, + 0.1766788792796389, + 0.19920258379165684, + 0.19572392759641466, + 0.19163266714637953, + 0.1971897533478506, + 0.19599570167170052, + 0.1980697193770598, + 0.19838719565114282, + 0.1907105410654986, + 0.18467978625203998, + 0.18335769944647878, + 0.1945589528776203, + 0.19059629168248857, + 0.1837905726120799, + 0.19588981520474838, + 0.19476814674687176, + 0.1946642390498577, + 0.1642807550142336, + null, + 0.1975686020831963, + 0.18100138583498532, + 0.1957993361289156, + 0.19186234990937415, + 0.16369233412875162, + 0.20029549326149382, + 0.1949837745709017, + 0.16951600434417124, + 0.18040483952175748, + 0.19613059925233506, + null, + 0.19345119556836698, + 0.19983590022136505, + 0.15721636939693207, + 0.16718648036955275, + 0.18030136136477387, + 0.1816195567451379, + 0.1889977438162276, + 0.1853956742203525, + 0.18615402257100197, + 0.17611362747882234, + 0.19048932488792616, + 0.17276179760457108, + 0.17370144327592402, + 0.1574262014927781, + 0.18267571464545537, + 0.1772510746220806, + 0.18323038954569035, + 0.16820473327468677, + 0.16515540023933306, + 0.17500651940363196, + 0.162182155705468, + 0.1776884827371618, + 0.1738341199443916, + 0.18048166641974775, + 0.1802800533700728, + 0.19661008935482865, + 0.19519102700355326, + 0.1868264631034538, + 0.18756236307745347, + 0.17798082062393755, + 0.18415632768305398, + 0.15279117591955918, + 0.18154817131472625, + 0.17362343411124745, + 0.18248329348252004, + 0.18193705036959876, + 0.18538438719645103, + 0.18281325972608842, + 0.14550384502486047, + 0.19111505723479422, + 0.18078247710899648, + 0.18211008013548646, + 0.18688050325025882, + 0.17143337451887564, + 0.1701809394511362, + 0.1662914281088844, + 0.19715875354488896, + 0.188287866607062, + 0.12280057971956718, + 0.19769830826359927, + 0.200286862090029, + 0.19061956692245277, + 0.1937417590200049, + 0.18409099944988436, + 0.1745389357094909, + 0.1868076649536253, + 0.18937354884396776, + 0.19280779693756075, + 0.19409870222700853, + 0.19175309294049223, + 0.18442496539345093, + 0.20658580096324286, + 0.19165456807074566, + null, + 0.1845227604510843, + 0.18686200692843064, + 0.18561269597506602, + 0.18900230142874871, + 0.1812971864119507, + 0.17891677415401866, + 0.18420656352999007, + null, + null, + 0.17760446547300213, + 0.18202287197807626, + null, + 0.204518684588419, + 0.17247539225288377, + 0.16658783501735014, + 0.18625853365292563, + 0.19366287057799186, + 0.1893631054975812, + 0.18339008989732092, + null, + 0.18623155149833973, + 0.17342253128164828, + null, + 0.1668043398014824, + null, + 0.17592388714991658, + 0.18762856933919053, + 0.2004896869484134, + 0.17146465489435567, + null, + 0.166697975884125, + 0.16797192434010708, + 0.18087281795096674, + null, + 0.19183463766464093, + null, + 0.1893716818730588, + null, + 0.1846986507723338, + 0.1898347720185931, + null, + 0.1805243188887873, + 0.18892404628634601, + null, + null, + 0.2017597936633015, + 0.19197710953462996, + 0.2034827991271821, + 0.19214652215722533, + null, + 0.189155917163022, + null, + 0.19536996642007973, + 0.18262694020267692, + null, + 0.15422229258595582, + 0.19332996877082656, + 0.1867486561076851, + null, + 0.19099523478500366, + 0.1956744573836637, + null, + 0.181185301669733, + 0.18049353891490152, + 0.18320625584307323, + null, + null, + null, + null, + 0.19033287245135175, + null, + null, + 0.19944104944168806, + 0.1835524626017263, + 0.17852158359188647, + null, + 0.19622874105830476, + 0.18621981241284377, + null, + null, + 0.17856580994402868, + 0.18749222400836216, + 0.17420567846999555, + 0.19254798648702642, + 0.1945194394408605, + 0.19269792961030055, + 0.1857046032821387, + 0.19489043264595124, + 0.18596606208330577, + 0.19003333374688003, + 0.1786780278735871, + 0.19528742083811615, + null, + 0.19519172476107946, + 0.19078792165142158, + 0.1809114865690379, + 0.1864348263679641, + 0.1844910056499285, + 0.17999848724021356, + 0.1870126508657214, + 0.17609442554598326, + 0.18103854144361578, + 0.18281586290736676, + 0.18061489990781124, + 0.17640166764416018, + null, + 0.19334269092233092, + 0.18776414185460888, + 0.1887655185817466, + null, + 0.19333211402159936, + 0.1616478027132452, + null, + 0.20045698626524913, + null, + 0.18475995435982873, + 0.17887823198518046, + 0.19073766500219186, + 0.17504249010807815, + 0.17756539475517416, + 0.1925149599097821, + 0.18957069497435677, + null, + null, + 0.1952075950268508, + 0.18683829880277356, + null, + 0.189230167503396, + null, + 0.1916873815269924, + 0.1815371163626324, + 0.18294279799422633, + 0.1878581124193724, + null, + null, + null, + null, + 0.19999823600357852, + null, + 0.18490494221565126, + 0.1923432752818846, + null, + null, + 0.18701372702282468, + 0.19758475912617737, + 0.19312813227421383, + null, + 0.1851032802415165, + 0.18172673549061813, + 0.18910857217310698, + null, + 0.19258327205295495, + null, + null, + 0.19105869534332962, + null, + 0.18690797254081407, + null, + null, + 0.18672430093361322, + 0.19272012314767445, + 0.18401106642544618, + 0.1872866330550267, + 0.18988604930792827, + null, + 0.1909152098185491, + 0.16574278526516326, + 0.15837516645967326, + 0.1219196294325723, + 0.12840515478673378, + 0.16377829611225445, + 0.1451079969623015, + 0.12135529850744015, + 0.1430513323393603, + 0.15838615048552265, + 0.1590788177500101, + 0.16746219918126545, + 0.1603858667594284, + 0.13598817317323658, + 0.16082378164345135, + 0.13256683895131008, + 0.13558373473892585, + 0.15911791532955097, + 0.1627951491367808, + 0.17180669772527093, + 0.16652837744361879, + 0.16185411105364056, + 0.1452350255043974, + 0.13603672278069825, + 0.10200626259866513, + 0.1686080777069642, + 0.14853194027305927, + 0.16099147112418988, + 0.16300346970786395, + 0.1462023431785146, + 0.1477153416702511, + 0.1432039571046541, + 0.1296725222551417, + 0.10823623621956398, + 0.13606899511887457, + 0.09784262790118672, + 0.1355857285852787, + 0.09319971685680735, + 0.1607169984466354, + 0.16706785220221865, + 0.14933134594879902, + 0.14261504408846354, + 0.11163412830955197, + 0.11668887736919822, + 0.12647132045528406, + 0.14162965645266354, + 0.10626872334518218, + 0.1228013793090776, + 0.11478818189254636, + 0.12450100313658788, + 0.11344253410991352, + 0.1132394118417776, + 0.11709468483207547, + 0.12921580415440953, + 0.129840557527682, + 0.10866468479826907, + 0.13058072149847813, + 0.14377061064165148, + 0.1563269409650639, + 0.15131688565214887, + 0.1618980818824328, + 0.14860674336328508, + 0.17070823983058114, + 0.1538680759692162, + 0.1546216394224192, + 0.16534416541309332, + 0.17578650869094253, + 0.15088272087916121, + 0.15772781258319285, + 0.1577540543386536, + 0.13820169723830628, + 0.1746324291514484, + 0.16134266633193506, + 0.15468078387854012, + 0.15504281738545916, + 0.15560519220441876, + 0.15121446719465376, + 0.14992371507122648, + 0.15707929201175558, + 0.14285217748141266, + 0.15776327275570148, + 0.1593188982901679, + 0.15727875498856816, + 0.16323724617624996, + 0.15133443071134323, + 0.16053256690556306, + 0.1555085981721976, + 0.15369494134874576, + 0.1589770159939104, + 0.14344551274596334, + 0.1586767032826519, + 0.16006655217616259, + 0.15963244467298568, + 0.16718440686148694, + 0.15845840373879114, + 0.15392864858669836, + 0.16302535618243433, + 0.15574176372128135, + 0.1673305802395763, + 0.17323503132163084, + 0.17463646167275726, + 0.15324603661621208, + 0.15582611674711677, + 0.15956587174811526, + 0.16228094906264928, + 0.15910794228900937, + 0.16759682438857607, + 0.10379511233991019, + 0.15645849276561505, + 0.15269326849408205, + 0.14868584374868743, + 0.14283228316310018, + 0.1507873708043591, + 0.15709191893066496, + 0.14040182610759644, + 0.15422182489026587, + 0.159583906340814, + 0.1538981183772886, + 0.15744812510365122, + 0.13001119831895874, + 0.13853490879904373, + 0.1602196137304365, + 0.15744854994656002, + 0.16330947235989912, + 0.16913444820808696, + 0.11713472750146998, + null, + 0.1587615920431429, + 0.13442403632005367, + 0.1469066208275922, + 0.1327440149456633, + 0.14176556928732809, + 0.15977306894977092, + 0.1595587483456473, + 0.15545709959016107, + 0.14971984037048774, + 0.14583050707461853, + null, + 0.1570996985250592, + 0.14785982196385236, + 0.1373877128635349, + 0.14515417460496322, + 0.14447588255790125, + 0.14146056368871093, + 0.15673800108552255, + 0.15292847676345198, + 0.14804528453598467, + 0.1424871625870095, + 0.15353823528646088, + 0.1554673678260602, + 0.15641391148393546, + 0.13564682346502283, + 0.15192861685109885, + 0.1691979663770763, + 0.13025712738606784, + 0.11964148947341795, + 0.13697261788575668, + 0.12458179522168147, + 0.162584791112711, + 0.15430734399688334, + 0.15248537733163167, + 0.13898027444616853, + 0.14245730342367258, + 0.16481750735566042, + 0.13764952155962798, + 0.15154938924054218, + 0.14491601798653952, + 0.17125552602592325, + 0.14069966265226314, + 0.15337055684280695, + 0.1407211609530491, + 0.15035427506792448, + 0.1699983522929503, + 0.14855047153987128, + 0.15667261645889402, + 0.15303834161502827, + 0.13710224699577772, + 0.1530883061032439, + 0.15198383967447657, + 0.1492262500821584, + 0.1395564055164067, + 0.15820506120816086, + 0.1547243119255314, + 0.14083516707411878, + 0.1813228213307383, + 0.14791537566169174, + 0.14613934069560267, + 0.158645551645438, + 0.1423213665783451, + 0.16160582129512058, + 0.15515385531945045, + 0.16267354232881018, + 0.1507587160286599, + 0.14538811330972382, + 0.14378799936378936, + 0.14104565207463696, + 0.13289467720574177, + 0.13508807816863813, + 0.1562408007629002, + 0.16280676273970532, + 0.158444886731231, + null, + 0.16575599081595024, + 0.15288074319679115, + 0.13303497635288689, + 0.14924989019927448, + 0.16266765747679357, + 0.15184086580859135, + 0.1337133399975363, + null, + null, + 0.1444460713494144, + 0.14466138018666472, + null, + 0.15212765226072103, + 0.1242756887703595, + 0.15059708560188825, + 0.13411732808952775, + 0.13795287016136284, + 0.14173741740530263, + 0.1676517152626033, + null, + 0.15501563881245767, + 0.15913644849052763, + null, + 0.12774539610562274, + null, + 0.14217354851325856, + 0.1537201211529368, + 0.14050774239311728, + 0.1457220246771606, + null, + 0.14097190319113898, + 0.1469406040885218, + 0.15666576301929452, + null, + 0.12566448513019263, + null, + 0.14853775552375928, + null, + 0.15316200085340506, + 0.1473385265048855, + null, + 0.14318326145618146, + 0.16971118666014004, + null, + null, + 0.16447336905622248, + 0.15095823667061592, + 0.15078843131724318, + 0.1363332693877073, + null, + 0.1319612091100123, + null, + 0.15051566832860697, + 0.15085276945362056, + null, + 0.12206632184003328, + 0.13313375085710707, + 0.14337570872605243, + null, + 0.16158981040532913, + 0.14062401948047146, + null, + 0.1743603704977596, + 0.1667300186357043, + 0.1692350568733498, + null, + null, + null, + null, + 0.1764427159138771, + null, + null, + 0.15906529321342278, + 0.1478744788863684, + 0.1699366781895173, + null, + 0.17624499480801456, + 0.17406522793253493, + null, + null, + 0.15222150799088122, + 0.1436804348876132, + 0.14977071431156297, + 0.1497840971306891, + 0.1503739725568607, + 0.16000235052848577, + 0.11243257965096781, + 0.14059006892625064, + 0.12635044786624364, + 0.14577756458628274, + 0.15492927344085444, + 0.1333111208844536, + null, + 0.13498805837079778, + 0.1395312182224206, + 0.16200900923704675, + 0.1628433835355164, + 0.14700153850559206, + 0.16626581933594767, + 0.14620771647793682, + 0.15890246076004202, + 0.16055259897608573, + 0.1360796008327379, + 0.13497764503034615, + 0.13361371934071817, + null, + 0.13833550167523787, + 0.1688267821433579, + 0.17398158389474286, + null, + 0.14116791735231202, + 0.14282272813118618, + null, + 0.15158661237992085, + null, + 0.15393828466194337, + 0.1478447933251011, + 0.15803886831760866, + 0.14602865649707586, + 0.15961433744100692, + 0.14464019027250927, + 0.12780276399547422, + null, + null, + 0.15059769313817548, + 0.16042031271056437, + null, + 0.14704303392800072, + null, + 0.1538544655999303, + 0.14324001181498092, + 0.14059723284876735, + 0.16487575055322493, + null, + null, + null, + null, + 0.1443414537990795, + null, + 0.15089241965549735, + 0.15907683958476557, + null, + null, + 0.14930188137332712, + 0.13086802954971893, + 0.12927669432869834, + null, + 0.1328803656343007, + 0.13997107490974095, + 0.14452374048905353, + null, + 0.14568358281830704, + null, + null, + 0.14111763099544092, + null, + 0.13814711218245582, + null, + null, + 0.1380611084595924, + 0.15547305144572532, + 0.1661473086849242, + 0.176280687181453, + 0.16401024025021982, + null, + 0.14789756191915676 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923, + 0.747678369195923 + ] + }, + "junctions": { + "mean_pearson": 0.13949657164361998, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2939079936773143, + 0.22621934635283225, + 0.28885039943989566, + 0.20890931316824315, + 0.27697983085917893, + 0.2706931238476557, + 0.03433358198374349, + 0.12072216110390095, + 0.22505999575336746, + 0.15934459023934097, + 0.18558062758294186, + 0.1225880516432769, + 0.06314606692435962, + 0.08924186592887359, + 0.18962719030694825, + 0.18411988605518326, + 0.12154339070118786, + 0.07908133259276594, + 0.22513123701053495, + 0.19906045490672836, + 0.27395589608631005, + 0.2466915197229336, + 0.2259385368852549, + 0.1483861339061643, + 0.18109454164117186, + 0.22343353637463484, + 0.15676843341771854, + 0.2380915474364873, + 0.14077127505509024, + 0.1417254865261197, + 0.21722271694103346, + 0.2896387792724388, + 0.22915559917075426, + 0.3116197529512436, + 0.237270044102795, + 0.2808739930311326, + 0.285425353000523, + 0.20392605345384768, + 0.20693544832824062, + 0.12030859728909798, + 0.14083249796351172, + 0.19480399113273644, + 0.2007654446658249, + 0.1904088589350666, + 0.18478764203308637, + 0.13881624001869244, + 0.14507304903535972, + 0.19301510193101182, + 0.16109960626052935, + 0.16198221910059166, + 0.2099269807561582, + 0.23751005746299064, + 0.12992951506245048, + 0.26053308967067357, + 0.16352697845065267, + 0.22472632400935455, + 0.16033165042702097, + 0.08221729493349324, + 0.18397687161937826, + 0.14187239983639138, + 0.04465187660878714, + 0.14834775052733487, + 0.16607652310414237, + 0.09958973058245091, + 0.06175931495253915, + 0.23425526243153116, + 0.19408490610248189, + 0.20992317008994638, + 0.13039998417586357, + 0.13531726949393305, + 0.18612069398232203, + 0.2157610811920659, + 0.1499812211010772, + 0.06369031973875755, + 0.40588663251408486, + 0.3477490553842121, + 0.17578518325204437, + 0.11176444252614919, + 0.13181533120787775, + 0.24893778490633214, + 0.17673292508944552, + 0.18508848237383813, + 0.1812292027718249, + 0.16510406569177005, + 0.163897163657825, + 0.2555323750147468, + 0.2585924483085289, + 0.2497441071593875, + 0.24895922137782153, + 0.20139817060141832, + 0.24169901725916992, + 0.19015920272403536, + 0.10329465912658188, + 0.18228680077982415, + 0.20139325142178768, + 0.33441573366285937, + 0.23851185051311852, + 0.1726063804022771, + 0.19747322597175124, + 0.20261345065461298, + 0.12671426459101326, + 0.18202100228311577, + 0.38009818778791926, + 0.2037091350564104, + 0.18277130142215328, + 0.1311430467417337, + 0.07546878380172752, + 0.23875132406267863, + 0.3375665436955635, + 0.3037754046097626, + 0.10740248324341196, + 0.39136951268036996, + 0.3073015991526228, + 0.3795251747086496, + 0.3004667957460752, + 0.11956730521353731, + 0.07969754488219254, + 0.16406330670460295, + 0.10404095339490584, + 0.2360373416663839, + 0.19987263905522984, + 0.14668518446463044, + 0.12374309538050386, + 0.20742947753633775, + 0.06925717268214271, + null, + 0.29930770167944876, + 0.17074002528730103, + 0.05986491409512803, + 0.17491351589338883, + 0.23548125952251361, + 0.1809570455362694, + 0.12629140800746916, + 0.21737114183286607, + 0.2870016883088111, + 0.12921543891252885, + null, + 0.2995538607563337, + 0.29913312971333683, + 0.22780460483098247, + 0.22448396454829841, + 0.2117510549422206, + 0.19519421580899105, + 0.20343026703629435, + 0.17872096005555838, + 0.21755102071756252, + 0.23417583075106516, + 0.19347815635839244, + 0.13832604872672544, + 0.18079352226431827, + 0.16757873167997375, + 0.2442588108427645, + 0.16425709478056627, + 0.2427483298276737, + 0.37661398220673364, + 0.28380822132704686, + 0.35214531595905296, + 0.19232454229378632, + 0.17256306667398644, + 0.20730414428844965, + 0.1619628735685277, + 0.2539096596797384, + 0.23232204716647092, + 0.26063951349238795, + 0.09479641358534797, + 0.3066319218208779, + 0.25694185987301066, + 0.15398344551751297, + 0.062091557707668235, + 0.1748996839334713, + 0.20060888230442486, + 0.2157386844879578, + 0.2709676792237972, + 0.07695334258739321, + 0.14593816161380932, + 0.18977646656284639, + 0.16200875352270727, + 0.16722057517045175, + 0.14693621267420495, + 0.2030476930647631, + 0.07818784404533488, + 0.11367065333571136, + 0.04463164475529913, + 0.21836717674566047, + 0.222003953588347, + 0.2277796035994216, + 0.28963700388552904, + 0.21389539642788882, + 0.28477825937014023, + 0.21619508516219352, + 0.2943010558325339, + 0.326898152494585, + 0.18095547694582947, + 0.11538980394398418, + 0.13329981435897142, + 0.11076569059138941, + 0.08885126787183444, + 0.15718215656845266, + 0.30501646784387454, + 0.38856655966904635, + null, + 0.17712851863228993, + 0.189224456505632, + 0.2291415120668534, + 0.26497748077871863, + 0.10081744799476844, + 0.13740333747667657, + 0.07628871054463021, + null, + null, + 0.21411476631885842, + 0.2743034367399137, + null, + 0.133653531369868, + 0.11874545049926437, + 0.0874056155550852, + 0.15986864158067154, + 0.10036556708808896, + 0.16167357464654872, + 0.068811624563754, + null, + 0.30662805474459853, + 0.15265193498386095, + null, + 0.11799223143606159, + null, + 0.2016384479977732, + 0.24182788813044168, + 0.1607610966156757, + 0.1572152365596869, + null, + 0.07518181584829936, + 0.04247260754044395, + 0.20137134995455278, + null, + 0.15415012823475324, + null, + 0.10112173393213562, + null, + 0.11072875362253691, + 0.14145790289089913, + null, + 0.18140078941955964, + 0.19352913413255307, + null, + null, + 0.22381865793660102, + 0.25308757218294076, + 0.04921108787709853, + 0.07775895927924331, + null, + 0.21283115027622718, + null, + 0.1919812312463311, + 0.19409305673134944, + null, + 0.16956900661573845, + 0.26857388950578936, + 0.22868174788398637, + null, + 0.24133455724384503, + 0.25120750152970456, + null, + 0.27266490480053784, + 0.18064718150855372, + 0.26190388189111047, + null, + null, + null, + null, + 0.1368856358623288, + null, + null, + 0.16684037919892836, + 0.24914639053444232, + 0.23424375765120165, + null, + 0.11268582343629653, + 0.28379371431539197, + null, + null, + 0.1846391873423697, + 0.09460751945680622, + 0.21781637570704826, + 0.10651433106249, + 0.1263813290257419, + 0.09221309097476374, + 0.13175204796515488, + 0.1458829948072346, + 0.030146336882898323, + 0.09383172116676348, + 0.3384292618468933, + 0.1728571275429953, + null, + 0.18521771241229174, + 0.20497582888626265, + 0.03974189897858646, + 0.14812770633124828, + 0.08933358201034824, + 0.061495507677304724, + 0.1040529325209558, + 0.14881062577015663, + 0.22150896312649987, + 0.15631296083055873, + 0.2085660590351718, + 0.17314435514542395, + null, + 0.15425205045465562, + 0.1877836026461758, + 0.19290239396892286, + null, + 0.24603297007575026, + 0.1790046341659175, + null, + 0.17368089892306926, + null, + 0.17101250071500462, + 0.14192169338340616, + 0.21246992633775663, + 0.11115980891912039, + 0.16860191309845718, + 0.18535207553901734, + 0.22927972078075676, + null, + null, + 0.12076529847149033, + 0.11071964978377714, + null, + 0.19165161982300902, + null, + 0.19693309760690458, + 0.08783213771589737, + 0.07859589712141712, + 0.1780657005348477, + null, + null, + null, + null, + 0.035738505460832216, + null, + 0.07470493850694282, + 0.2260903939225216, + null, + null, + 0.14895325901336204, + 0.2384529506889697, + 0.16586777184127302, + null, + 0.2323097007356938, + 0.21025261024102915, + 0.05862900741844739, + null, + 0.20319878955435278, + null, + null, + 0.2538921618248756, + null, + 0.1672190922462114, + null, + null, + 0.18182726807125563, + 0.19743990688049268, + 0.17550959101846003, + 0.19807286779844274, + 0.15579816941658667, + null, + 0.16101415874728964, + 0.015409755210263148, + 0.11521703044819853, + 0.24243122162027034, + 0.0748547064161385, + 0.23417904007404833, + 0.09706196635028308, + -0.019612817057850115, + 0.025267088565569938, + 0.06369134775463725, + 0.1196936899258509, + 0.1066596196463312, + 0.10804739856036998, + 0.06306977407072077, + 0.05601978049270827, + 0.16243146341227246, + 0.13059648832977472, + 0.02245539495589094, + 0.050778434977647016, + 0.1313572424755252, + 0.08340731380698506, + 0.08293983735265713, + 0.22437981619897657, + 0.1318055649853598, + 0.12609811541691093, + 0.08737533115442407, + 0.11095586205603386, + 0.10110307956231473, + 0.07062535275439803, + 0.058828253468310436, + 0.057430806870372955, + 0.06618931799939627, + 0.1053957138034984, + 0.01920663386816842, + 0.19140749358930875, + 0.026399694504921786, + 0.16969600018933112, + 0.04523631929626014, + 0.05786848560796611, + 0.057977125916587366, + 0.09437238365350516, + 0.03847479325956878, + 0.046952420032051086, + 0.04488059325214309, + 0.006738748220552326, + 0.04967360397686112, + 0.0034705827452424223, + 0.039657179634355194, + 0.20402148922928504, + 0.05502218863706585, + 0.0764330761161955, + 0.05112586121061464, + 0.0922440094591787, + 0.048189338942152535, + 0.060776934840573116, + 0.01453656180149994, + 0.10639720270588006, + 0.13983235503914498, + 0.16110537128089533, + 0.10185937885804175, + 0.047804107492268606, + 0.09526092674880371, + 0.03935524482803524, + 0.09480331778695233, + 0.1719587989634844, + 0.11913204480378833, + 0.15034185775611014, + 0.10760469079543958, + 0.06310381610138537, + 0.06246243275567287, + 0.07498538677518622, + 0.14671791731899636, + 0.06881934763476333, + 0.12081018077816598, + 0.11945098277892982, + 0.049802973549381464, + 0.07744488084474295, + 0.03609439672317635, + 0.09876967225788831, + 0.05821831037236524, + 0.07746794524235784, + 0.07684630505870652, + 0.1303799941926847, + 0.10877501985856418, + 0.046941163912521885, + 0.1138941259593769, + 0.1530345870463696, + 0.17473160420396402, + 0.08192399240950109, + 0.10530262950021321, + 0.051874323790119706, + 0.09815905729534123, + 0.05363975205046008, + 0.05470711674760522, + 0.03300965956637957, + 0.07118503714814023, + 0.07262213273488421, + 0.06989314761560132, + 0.012001435727621705, + 0.15986470728128774, + 0.02817391942422359, + 0.09395087012230909, + 0.08793219976268306, + 0.09513258836026628, + 0.15770943209878435, + 0.16740642802027522, + 0.10972732630780209, + 0.041593065815932896, + 0.1694081083835398, + 0.08350908362674714, + 0.128737140140591, + 0.06562394344099888, + 0.12208186931882027, + 0.06989925073663118, + 0.12703647397217044, + 0.12649125491419985, + 0.05643903809788537, + 0.07800940365437728, + 0.032608691596191584, + 0.05897466543394895, + 0.0953019256339778, + 0.12071155740593835, + 0.08824700859013362, + 0.05855987702224789, + 0.18785472886457272, + 0.0842997709712214, + null, + 0.19192717813358862, + 0.09435605768128144, + 0.025426792609454713, + 0.22246239763193204, + 0.09413273896110312, + 0.0233561953561616, + 0.032152620011706746, + 0.1470880653674793, + 0.10902473561229722, + 0.08795320870022955, + null, + 0.12069764911906447, + 0.06643898697223248, + 0.03385174417341387, + 0.07230294212608968, + 0.13117894678744832, + 0.012263962370098469, + 0.10622828534237257, + 0.1203487816858339, + 0.01790050464382027, + 0.07176547060109371, + 0.014871249823746466, + 0.03126336083977924, + 0.10614088842809943, + 0.13401378179980875, + 0.1304546533783496, + 0.06180334117294663, + 0.10472162106009489, + 0.12357372356306481, + 0.1652830117556219, + 0.17390443593447313, + 0.054923233323895856, + 0.14299343008030552, + 0.13611401207216353, + 0.06564979928902878, + 0.06202744394939602, + 0.09480560134904525, + 0.1182582996500235, + 0.18600756077431402, + 0.14010131029844478, + 0.1729547516230375, + 0.07149247612196813, + 0.07269285753587713, + 0.15688959295524973, + 0.10647854159195855, + 0.016823518090390184, + 0.10715937893572622, + 0.10505607427757929, + 0.05739216834502475, + 0.1055619432414244, + 0.15169009224680285, + 0.07932055973770832, + 0.07205877284432897, + 0.1838316344126951, + 0.037576656949752256, + 0.04278378796450761, + 0.23441153063225548, + 0.14396410262586734, + 0.061702476944002174, + 0.13564902492908038, + 0.08349450832366433, + -0.003395662597879498, + 0.11359021487716216, + 0.17518054677008782, + 0.2323234565140931, + 0.33563258200088697, + 0.13237138403796977, + 0.18764279296210928, + 0.02848283403240754, + 0.018048418112273718, + 0.028863777674191512, + 0.09328392274956826, + 0.10270883288447095, + 0.10483393717056415, + null, + 0.17691417169069318, + 0.11195980595432212, + 0.08777289234613951, + 0.08522301933845551, + 0.2006146503649276, + 0.07196654751559084, + 0.10203029242163927, + null, + null, + 0.10441270338928199, + 0.14037494665912315, + null, + 0.027295241811937463, + 0.005867180266364679, + 0.031169625869343548, + 0.08696322804120978, + 0.0400274880263661, + 0.04247045460471229, + 0.04028657034657976, + null, + 0.05351091941483158, + 0.016554762621762576, + null, + -0.033279613968624486, + null, + 0.07096061631042672, + 0.0988759732320907, + 0.10215515736044486, + 0.04084449161886963, + null, + 0.04480526296311857, + 0.07845222414472236, + 0.058641990807965484, + null, + 0.040076524892406276, + null, + 0.1550847536948786, + null, + 0.007312250460246574, + 0.14737332154843052, + null, + 0.09520546173160906, + 0.10083721307002998, + null, + null, + 0.1234393985120741, + 0.11444430490877974, + 0.16077660696567742, + 0.12950668831113635, + null, + 0.028295180907239068, + null, + 0.0065895764414565395, + 0.08986991972355672, + null, + 0.14642538289292273, + 0.06353327164605566, + 0.1143020016090323, + null, + 0.08534865002859714, + 0.08465473360706911, + null, + 0.12667370343068038, + 0.16016767169112506, + 0.17072489897245774, + null, + null, + null, + null, + 0.11598808582024211, + null, + null, + 0.10185045400144642, + 0.10010412136769421, + 0.10972991384927769, + null, + 0.09685026799177625, + 0.12524061033907963, + null, + null, + 0.053335082516162466, + 0.0792574823465752, + 0.1205345295291422, + 0.045414517700688196, + 0.05811171535765219, + 0.024033820283881006, + 0.22379087154109406, + 0.07546952805440511, + 0.05187984105219922, + 0.03911463165538955, + 0.10474158871432547, + 0.05824903638167321, + null, + 0.08987551874237865, + 0.05745510398989993, + 0.1510525855328759, + 0.10257662757410153, + 0.0452919654530072, + 0.015760549548137857, + 0.09434083208120547, + 0.035916908993848376, + 0.10725305324173035, + 0.10988681840695586, + 0.05732898322533825, + 0.08071712860986041, + null, + 0.07649188420362614, + 0.07247029311475024, + 0.12352273689784801, + null, + 0.09542467005729599, + 0.016298652236568314, + null, + 0.08691317829409972, + null, + 0.19044701749707096, + 0.09317395611929746, + 0.09701851834281112, + 0.09912433961198075, + 0.1266462411411258, + 0.11294361602238599, + 0.16267672368231095, + null, + null, + 0.0885024634057245, + 0.03665177930455491, + null, + 0.036665421478112185, + null, + 0.14199689908212812, + 0.12007082014864069, + 0.1571361477998821, + 0.08716597777141724, + null, + null, + null, + null, + 0.07224436654426566, + null, + 0.08637916952773673, + 0.18922755576977857, + null, + null, + 0.03375970195703859, + 0.22265077088940852, + 0.09398014125316502, + null, + 0.08253415812744568, + 0.08597767766812253, + 0.20021720510108806, + null, + 0.14936905376524615, + null, + null, + 0.0707935262222616, + null, + 0.09260431848488705, + null, + null, + 0.01347515406414022, + 0.043155240930048744, + 0.09509822047051661, + 0.07688137931014379, + 0.08333540658840206, + null, + 0.12066228741834112 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9536423841059603, + 0.9375, + 0.9501779359430605, + 0.9462025316455697, + 0.9701986754966887, + 0.9457627118644067, + 0.9647887323943662, + 0.9479166666666666, + 0.9519230769230769, + 0.9468085106382979, + 0.9427609427609428, + 0.9355932203389831, + 0.9448051948051948, + 0.940625, + 0.9566666666666667, + 0.9407894736842105, + 0.9440559440559441, + 0.9455128205128205, + 0.9522292993630573, + 0.9477124183006536, + 0.9496402877697842, + 0.9419354838709677, + 0.9415584415584416, + 0.9405204460966543, + 0.9537366548042705, + 0.9539473684210527, + 0.9615384615384616, + 0.9525423728813559, + 0.9493243243243243, + 0.9448275862068966, + 0.9452054794520548, + 0.9644012944983819, + 0.9365671641791045, + 0.9574468085106383, + 0.9413919413919414, + 0.9495268138801262, + 0.9427480916030534, + 0.9404388714733543, + 0.9230769230769231, + 0.9395973154362416, + 0.9453924914675768, + 0.9553264604810997, + 0.9427480916030534, + 0.9518518518518518, + 0.9475409836065574, + 0.9425675675675675, + 0.9446494464944649, + 0.9354838709677419, + 0.9290780141843972, + 0.9523809523809523, + 0.9416058394160584, + 0.9595588235294118, + 0.9318181818181818, + 0.9489795918367347, + 0.9328358208955224, + 0.9421768707482994, + 0.9446366782006921, + 0.9503311258278145, + 0.9486301369863014, + 0.9401408450704225, + 0.9482758620689655, + 0.9573770491803278, + 0.9470198675496688, + 0.9522184300341296, + 0.9362416107382551, + 0.9525316455696202, + 0.9324324324324325, + 0.9530201342281879, + 0.9351535836177475, + 0.954248366013072, + 0.9572953736654805, + 0.9451612903225807, + 0.9461279461279462, + 0.9602649006622517, + 0.942652329749104, + 0.9525423728813559, + 0.9517684887459807, + 0.9431438127090301, + 0.9551724137931035, + 0.95578231292517, + 0.9403973509933775, + 0.9588607594936709, + 0.9452054794520548, + 0.9352750809061489, + 0.9543859649122807, + 0.9442724458204335, + 0.9452054794520548, + 0.9640522875816994, + 0.95, + 0.9493670886075949, + 0.9512195121951219, + 0.9518900343642611, + 0.9543859649122807, + 0.9455782312925171, + 0.9391891891891891, + 0.9618055555555556, + 0.9457627118644067, + 0.9377162629757786, + 0.9437086092715232, + 0.939873417721519, + 0.9433333333333334, + 0.9442622950819672, + 0.9359430604982206, + 0.9488817891373802, + 0.9590443686006825, + 0.9566666666666667, + 0.9325842696629213, + 0.9498327759197325, + 0.9435215946843853, + 0.9459459459459459, + 0.9464285714285714, + 0.9395973154362416, + 0.9577922077922078, + 0.9525423728813559, + 0.9455782312925171, + 0.9358108108108109, + 0.9448275862068966, + 0.9425675675675675, + 0.9386281588447654, + 0.9556313993174061, + 0.9428571428571428, + 0.9559322033898305, + 0.9375, + 0.9473684210526315, + 0.9352226720647774, + 1.0, + 0.9550173010380623, + 0.9337979094076655, + 0.9393939393939394, + 0.9477611940298507, + 0.9492753623188406, + 0.9517241379310345, + 0.9442508710801394, + 0.9360902255639098, + 0.9508771929824561, + 0.939799331103679, + 0.9919354838709677, + 0.9559322033898305, + 0.9514925373134329, + 0.9506578947368421, + 0.948905109489051, + 0.9291044776119403, + 0.9485294117647058, + 0.9448275862068966, + 0.9368421052631579, + 0.9389312977099237, + 0.9448818897637795, + 0.9479166666666666, + 0.9227941176470589, + 0.9434628975265018, + 0.9333333333333333, + 0.950354609929078, + 0.9413919413919414, + 0.961038961038961, + 0.9482758620689655, + 0.9347079037800687, + 0.946969696969697, + 0.9530685920577617, + 0.9434628975265018, + 0.9416058394160584, + 0.9356060606060606, + 0.9513888888888888, + 0.9467084639498433, + 0.9407665505226481, + 0.9320754716981132, + 0.9460431654676259, + 0.9411764705882353, + 0.9668874172185431, + 0.939622641509434, + 0.9508771929824561, + 0.9560439560439561, + 0.9385665529010239, + 0.9548872180451128, + 0.952054794520548, + 0.9384615384615385, + 0.953307392996109, + 0.927536231884058, + 0.9407665505226481, + 0.9522058823529411, + 0.9440559440559441, + 0.9477351916376306, + 0.9440298507462687, + 0.9520295202952029, + 0.942652329749104, + 0.954225352112676, + 0.959409594095941, + 0.9417808219178082, + 0.9326599326599326, + 0.931899641577061, + 0.9508196721311475, + 0.9409722222222222, + 0.9485294117647058, + 0.9357142857142857, + 0.9330985915492958, + 0.926829268292683, + 0.9222614840989399, + 0.9227941176470589, + 0.9444444444444444, + 0.9477124183006536, + 0.9566666666666667, + 0.9922480620155039, + 0.9572368421052632, + 0.9491525423728814, + 0.946843853820598, + 0.9358108108108109, + 0.9614147909967846, + 0.9575971731448764, + 0.930921052631579, + 1.0, + 0.9924242424242424, + 0.9603960396039604, + 0.9512987012987013, + 0.9923076923076923, + 0.9517241379310345, + 0.9267399267399268, + 0.9343065693430657, + 0.9385665529010239, + 0.9503311258278145, + 0.9448275862068966, + 0.9429530201342282, + 1.0, + 0.953416149068323, + 0.9312714776632303, + 0.9849624060150376, + 0.9296875, + 1.0, + 0.9490909090909091, + 0.9438596491228071, + 0.9436619718309859, + 0.9494949494949495, + 1.0, + 0.9456521739130435, + 0.9330985915492958, + 0.9368421052631579, + 1.0, + 0.9393939393939394, + 1.0, + 0.9285714285714286, + 1.0, + 0.9533333333333334, + 0.9496644295302014, + 0.9927007299270073, + 0.9256505576208178, + 0.9377049180327869, + 1.0, + 1.0, + 0.9468085106382979, + 0.954248366013072, + 0.9444444444444444, + 0.9393939393939394, + 1.0, + 0.9363957597173145, + 0.9922480620155039, + 0.9415584415584416, + 0.9484536082474226, + 0.9930555555555556, + 0.95, + 0.9389067524115756, + 0.9381818181818182, + 0.9927536231884058, + 0.9285714285714286, + 0.9565217391304348, + 1.0, + 0.9540983606557377, + 0.9333333333333333, + 0.9525423728813559, + 1.0, + 1.0, + 0.9922480620155039, + 1.0, + 0.9401993355481728, + 1.0, + 0.984251968503937, + 0.9421768707482994, + 0.9361702127659575, + 0.9519230769230769, + 1.0, + 0.9419795221843004, + 0.9452054794520548, + 1.0, + 0.9923076923076923, + 0.9372822299651568, + 0.9420289855072463, + 0.9464882943143813, + 0.9425675675675675, + 0.9283276450511946, + 0.935374149659864, + 0.9556313993174061, + 0.9342560553633218, + 0.9322033898305084, + 0.9181494661921709, + 0.945859872611465, + 0.9298892988929889, + 1.0, + 0.9473684210526315, + 0.9527027027027027, + 0.9392857142857143, + 0.9528619528619529, + 0.95578231292517, + 0.9548611111111112, + 0.9438596491228071, + 0.9379310344827586, + 0.9442622950819672, + 0.9425675675675675, + 0.9420289855072463, + 0.9385665529010239, + 1.0, + 0.9421221864951769, + 0.9459459459459459, + 0.9551282051282052, + 1.0, + 0.9456869009584664, + 0.946236559139785, + 1.0, + 0.9312714776632303, + 1.0, + 0.9455128205128205, + 0.9444444444444444, + 0.9471947194719472, + 0.9311594202898551, + 0.9455782312925171, + 0.9315068493150684, + 0.9370629370629371, + 0.9929078014184397, + 1.0, + 0.9379084967320261, + 0.9285714285714286, + 1.0, + 0.9350180505415162, + 1.0, + 0.946843853820598, + 0.9328859060402684, + 0.9381818181818182, + 0.9368770764119602, + 1.0, + 1.0, + 0.992, + 1.0, + 0.9417808219178082, + 1.0, + 0.9342560553633218, + 0.9619047619047619, + 1.0, + 1.0, + 0.9507042253521126, + 0.9435215946843853, + 0.9407407407407408, + 1.0, + 0.9383561643835616, + 0.9393939393939394, + 0.9184397163120568, + 0.9919354838709677, + 0.9681528662420382, + 1.0, + 0.9922480620155039, + 0.9556313993174061, + 1.0, + 0.948220064724919, + 0.9921259842519685, + 0.9922480620155039, + 0.9420849420849421, + 0.948051948051948, + 0.9381443298969072, + 0.9450171821305842, + 0.9366197183098591, + 1.0, + 0.9370860927152318, + 0.9736842105263158, + 0.9824561403508771, + 0.9891304347826086, + 0.9821428571428571, + 0.9841772151898734, + 0.9786476868327402, + 0.9825783972125436, + 0.9735849056603774, + 0.9795221843003413, + 0.9820143884892086, + 0.9896193771626297, + 0.9819494584837545, + 0.9828767123287672, + 0.9828767123287672, + 0.9888475836431226, + 0.9853479853479854, + 0.971830985915493, + 0.9740740740740741, + 0.9821428571428571, + 0.9795221843003413, + 0.972027972027972, + 0.9823321554770318, + 0.9665551839464883, + 0.981203007518797, + 0.9821428571428571, + 0.9709090909090909, + 0.9834437086092715, + 0.9790940766550522, + 0.978494623655914, + 0.9762845849802372, + 0.9789473684210527, + 0.9816176470588235, + 0.9838709677419355, + 0.975177304964539, + 0.978021978021978, + 0.975177304964539, + 0.9880478087649402, + 0.9645390070921985, + 0.9790209790209791, + 0.9712230215827338, + 0.9683098591549296, + 0.9781021897810219, + 0.9893992932862191, + 0.984313725490196, + 0.979381443298969, + 0.9784172661870504, + 0.9887218045112782, + 0.9809160305343512, + 0.9886792452830189, + 0.9846153846153847, + 0.9839357429718876, + 0.9851301115241635, + 0.9886363636363636, + 0.9786476868327402, + 0.981203007518797, + 0.9776951672862454, + 0.9820143884892086, + 0.985663082437276, + 0.9857651245551602, + 0.9746376811594203, + 0.9860627177700348, + 0.9662921348314607, + 0.9658703071672355, + 0.9853479853479854, + 0.9855595667870036, + 0.9816176470588235, + 0.9821428571428571, + 0.9884169884169884, + 0.9809885931558935, + 0.982078853046595, + 0.9781021897810219, + 0.967391304347826, + 0.9796610169491525, + 0.982078853046595, + 0.974910394265233, + 0.9746376811594203, + 0.9689655172413794, + 0.9823321554770318, + 0.9784172661870504, + 0.969811320754717, + 0.972318339100346, + 0.9794520547945206, + 0.9771863117870723, + 0.9777777777777777, + 0.9822695035460993, + 0.9816176470588235, + 0.9770992366412213, + 0.9756944444444444, + 0.9781021897810219, + 0.9743589743589743, + 0.9790940766550522, + 0.9750889679715302, + 0.9724137931034482, + 0.9851301115241635, + 0.9859649122807017, + 0.985663082437276, + 0.9757785467128027, + 0.9628252788104089, + 0.9828767123287672, + 0.974910394265233, + 0.9693486590038314, + 0.9744525547445255, + 0.9818840579710145, + 0.9714285714285714, + 0.9782608695652174, + 0.9826388888888888, + 0.9849056603773585, + 0.9851851851851852, + 0.9754385964912281, + 0.9857651245551602, + 0.9823943661971831, + 0.9850187265917603, + 0.9735099337748344, + 0.974910394265233, + 0.982078853046595, + 0.9730769230769231, + 0.9823321554770318, + 0.9769230769230769, + 0.9802371541501976, + 0.985239852398524, + 0.9824561403508771, + 0.985663082437276, + 0.9816849816849816, + 0.9823321554770318, + 0.9776951672862454, + 1.0, + 0.9787234042553191, + 0.9785714285714285, + 0.9822695035460993, + 0.9841897233201581, + 0.9638989169675091, + 0.9825174825174825, + 0.9823321554770318, + 0.981203007518797, + 0.9825783972125436, + 0.9809160305343512, + 0.9906542056074766, + 0.9790940766550522, + 0.9742647058823529, + 0.9755244755244755, + 0.9790940766550522, + 0.9814126394052045, + 0.9787234042553191, + 0.9825174825174825, + 0.9822695035460993, + 0.9814126394052045, + 0.9888059701492538, + 0.9812734082397003, + 0.9783393501805054, + 0.989010989010989, + 0.9769230769230769, + 0.977859778597786, + 0.982078853046595, + 0.977859778597786, + 0.9849056603773585, + 0.9793103448275862, + 0.978021978021978, + 0.9774436090225563, + 0.9864864864864865, + 0.9851851851851852, + 0.9746376811594203, + 0.9816849816849816, + 0.9805825242718447, + 0.9825174825174825, + 0.9887640449438202, + 0.984251968503937, + 0.9863013698630136, + 0.974910394265233, + 0.9836734693877551, + 0.9888475836431226, + 0.9855072463768116, + 0.9777777777777777, + 0.9844961240310077, + 0.9823321554770318, + 0.981549815498155, + 0.983739837398374, + 0.9822695035460993, + 0.9821428571428571, + 0.986013986013986, + 0.9890510948905109, + 0.9825783972125436, + 0.9851301115241635, + 0.9926470588235294, + 0.9743589743589743, + 0.9753521126760564, + 0.9794238683127572, + 0.9739776951672863, + 0.9784172661870504, + 0.9753521126760564, + 0.9821428571428571, + 0.9862542955326461, + 0.988929889298893, + 0.9752650176678446, + 0.981549815498155, + 0.982078853046595, + 0.9784172661870504, + 0.9823943661971831, + 0.9769230769230769, + 0.9838709677419355, + 0.9761092150170648, + 0.9914529914529915, + 0.9787234042553191, + 0.9797979797979798, + 0.9776119402985075, + 0.967032967032967, + 0.9794520547945206, + 0.9761904761904762, + 0.9681978798586572, + 0.9920634920634921, + 0.9916666666666667, + 0.9843260188087775, + 0.9804560260586319, + 1.0, + 0.9789473684210527, + 0.972318339100346, + 0.9807692307692307, + 0.9761092150170648, + 0.975609756097561, + 0.9750889679715302, + 0.9823321554770318, + 1.0, + 0.974910394265233, + 0.9814814814814815, + 0.9923076923076923, + 0.9850746268656716, + 0.984, + 0.976271186440678, + 0.9750889679715302, + 0.9797297297297297, + 0.9789473684210527, + 0.9925925925925926, + 0.9706959706959707, + 0.9859649122807017, + 0.9742647058823529, + 1.0, + 0.9750889679715302, + 1.0, + 0.9787985865724381, + 0.9920634920634921, + 0.9662921348314607, + 0.9818181818181818, + 0.9924812030075187, + 0.9709090909090909, + 0.978494623655914, + 1.0, + 1.0, + 0.9758620689655172, + 0.9787985865724381, + 0.9835526315789473, + 0.9797979797979798, + 1.0, + 0.9766536964980544, + 1.0, + 0.975177304964539, + 0.9776119402985075, + 1.0, + 0.977859778597786, + 0.9755244755244755, + 0.9814126394052045, + 0.9922480620155039, + 0.9818840579710145, + 0.9753521126760564, + 1.0, + 0.9829351535836177, + 0.9832775919732442, + 0.985663082437276, + 0.9852941176470589, + 0.9838709677419355, + 0.9920634920634921, + 0.9819819819819819, + 0.9792387543252595, + 0.9908256880733946, + 0.9908256880733946, + 0.9822695035460993, + 0.9829931972789115, + 0.9763513513513513, + 0.9922480620155039, + 0.9825174825174825, + 0.9801324503311258, + 0.9915966386554622, + 1.0, + 0.9757785467128027, + 0.979381443298969, + 0.9745454545454545, + 0.9702970297029703, + 0.9789473684210527, + 0.9708029197080292, + 0.9816176470588235, + 0.9863945578231292, + 0.9886792452830189, + 0.9735849056603774, + 0.9728813559322034, + 0.9822695035460993, + 0.9917355371900827, + 0.9652777777777778, + 0.978494623655914, + 0.9795918367346939, + 0.9719298245614035, + 0.9755244755244755, + 0.9825174825174825, + 0.9794520547945206, + 0.9706959706959707, + 0.9798657718120806, + 0.9787985865724381, + 0.9851851851851852, + 0.9854545454545455, + 1.0, + 0.9785714285714285, + 0.9785714285714285, + 0.9833333333333333, + 0.991869918699187, + 0.9792387543252595, + 0.9832775919732442, + 1.0, + 0.9639344262295082, + 1.0, + 0.9800664451827242, + 0.9822064056939501, + 0.972972972972973, + 0.9764705882352941, + 0.9794520547945206, + 0.9701492537313433, + 0.9809885931558935, + 1.0, + 0.991869918699187, + 0.9823943661971831, + 0.974910394265233, + 0.9922480620155039, + 0.9857651245551602, + 1.0, + 0.9795221843003413, + 0.9757785467128027, + 0.9755244755244755, + 0.9779411764705882, + 0.9854014598540146, + 0.9921259842519685, + 1.0, + 0.9813084112149533, + 0.9859154929577465, + 0.991304347826087, + 0.9730769230769231, + 0.9865319865319865, + 1.0, + 0.9923076923076923, + 0.978494623655914, + 0.9789473684210527, + 0.9714285714285714, + 1.0, + 0.978494623655914, + 0.9790940766550522, + 0.9829351535836177, + 1.0, + 0.9772727272727273, + 0.991869918699187, + 0.9917355371900827, + 0.9781818181818182, + 1.0, + 0.9735849056603774, + 0.981651376146789, + 0.992, + 0.974169741697417, + 0.9653979238754326, + 0.9769736842105263, + 0.9755244755244755, + 0.9796610169491525, + 0.9915966386554622, + 0.9830508474576272 + ] + }, + "contact_maps": { + "mean_pearson": -0.002049182393212716, + "valid_tracks": 28, + "per_track_pearson": [ + -0.0016930648718103878, + -0.00011279486006055398, + -0.0605475627784429, + 0.0006469772013722384, + -0.0017719303932812317, + -0.0015245680876786012, + -0.00011636734984409063, + -0.0006068662832769321, + -9.643571435590143e-05, + 0.0008268027944473424, + 0.0003844822027854281, + 0.001455881771835578, + -0.000832752260780232, + 0.0, + -0.002115659215550962, + -0.004427497005003978, + 0.010380976927792347, + 0.001102744203667497, + -0.0037967871334827676, + -0.00054875619538323, + 0.0, + 0.001458185129694075, + 0.0010049864660418087, + -0.00483557223401072, + 0.008168586552663826, + 0.0005669139317062131, + 0.0004647445150052297, + -0.0008117743240051396 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.2689625469464074, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/README.md new file mode 100644 index 0000000000000000000000000000000000000000..bc8e0ea71d152a5eb3818e2d39d53cb70eca306b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/README.md @@ -0,0 +1,77 @@ +# SeqFunc bridge03 at step 22,000 + +One model with **13,281,416 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and four dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. This connection was attached with zero output while preserving every parent parameter and optimizer moment. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The initial compact model trained for 2,000 updates with balanced splice loss, then a zero-output local path was attached. At local01 step 18,000 a zero-output projection from local pooled features into the global trunk was attached, preserving all parent weights, optimizer moments and RNG states. This candidate trained the next 4,000 intervals with the same learning rates as the local01 control. The only scientific change is the added connection. + +This checkpoint records **22,000 updates**, 0 complete training permutations and +22,000 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup updates. The splice objective is `balanced_splice`; +other family losses follow the reference. Recorded accumulated training-loop wall time is +56.3 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +## Measured validation results + +Family-mean Pearson: **0.272973** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.473191 | +| dnase | 0.597237 | +| procap | 0.349566 | +| cage | 0.193042 | +| rna_seq | 0.210687 | +| chip_tf | 0.403283 | +| chip_histone | 0.387228 | +| splice_sites | 0.040422 | +| splice_usage | 0.166443 | +| junctions | 0.118197 | +| contact_maps | 0.063410 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5115 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6235 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0124 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7862 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2750 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2243 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2682 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.1845 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.2032 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2161 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5703 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1528 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2112 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..e5ac10d82f9f60a3e2398cfc99c7aa54006897f8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/package_manifest.json @@ -0,0 +1,32 @@ +{ + "parameters": 13281416, + "source_checkpoint_sha256": "1f1bfe3f3ee43c0febfc85184bccda7d302344014af7331a9ba94a89f5972e3f", + "files": { + "README.md": "49e79998f21f21c7af0d5c1f25061d4e1a845dfda2d7b09a8a91742692fd7a73", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "58cb6a3db4d24356332b46e5c78bfa51e5b99dde77d59e197b2addc2ed4bb44b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "2725592456c55c188b22c376f0cb249e9f46b4ae1370d2226df7143da358a8b2", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "f6e13a3c7aaf110d6a7bfbbae719d08e1be4abb8d1c576a35ecbeb91465d73d7", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "ec77e3de38f4a9990992a3c27b2935a4692e24c07ef68790b1034ed21a0b8fd9", + "training_provenance.json": "4381e253305f50628dd9c60c3f1ea02c85b242379dded59d8acb38233d33ace8", + "validation_tracks.json": "0452a25dae24c187eacd8ed6b6ed55e726da14d11729ca48a600f0abf7071707", + "validation_variants.json": "54cc6d9e4a08f64fdc1899b95cc5a8aa1187187913508e2b0b1f407a7bce8a71" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..52f3decd4fe2e9d275d08606b74575fafb701434 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/predict.py @@ -0,0 +1,130 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..c432425931d84585d994ee839351ff8bd59c8426 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/model.py @@ -0,0 +1,450 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + + def __post_init__(self) -> None: + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..30ffd5e8aad11fba7e5362c55cb47a05839ad65b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_config.yaml @@ -0,0 +1,30 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..7998d384cab5bbfefbe3c3a4a6770e379abb660e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/training_provenance.json @@ -0,0 +1,84 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "splice_context_mode": "offset", + "local_to_trunk": true + }, + "epoch": 0, + "offset": 22000, + "global_step": 22000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/bridge03", + "config": "/workspace/research/experiments/bridge03/config.yaml", + "resume": "/workspace/research/experiments/bridge03/start.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 22000, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 3378.2578727870714, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "local_to_global_trunk", + "parent_sha256": "cd64f7ccf50d6335cac129447dd451d5b829fc7d3857f780985b47aab7101842", + "step": 18000, + "initial_outputs_unchanged": true, + "computation": "average aligned local features into 1-kb tokens, zero linear projection, add before transformer" + }, + "interventions": [ + { + "kind": "local_to_global_trunk", + "parent_sha256": "cd64f7ccf50d6335cac129447dd451d5b829fc7d3857f780985b47aab7101842", + "step": 18000, + "initial_outputs_unchanged": true, + "computation": "average aligned local features into 1-kb tokens, zero linear projection, add before transformer" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..9dd110d759cd34e0216d4e8e6087887276d59cbf --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.47319080337453506, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6013569800083834, + 0.611230826606813, + 0.3715117945868288, + 0.5511424457436707, + 0.3629125711945762, + 0.35654874664890407, + 0.36989550633852447, + 0.4451403188891816, + 0.4947862642058731, + 0.5487417968153977, + 0.4774852488126404, + 0.41718372668167625, + 0.2771671525279424, + 0.4122614102308198, + 0.5987590328673628, + 0.3767810748988027, + 0.4488039383182791, + 0.39130416252459943, + 0.4435090578658996, + 0.4298864166820064, + 0.3973432022754257, + 0.3554749768779291, + 0.40861399186658753, + 0.42548698014389397, + 0.3707584577037519, + 0.3872132348892331, + 0.37283706711057535, + 0.3986515152361764, + 0.40455104366511874, + 0.3923066605898128, + 0.39141428506747583, + 0.39637208561282855, + 0.38375193312457767, + 0.2956162453517259, + 0.2707746425137148, + 0.45799869189698456, + 0.38641390142512944, + 0.5066500236423994, + 0.40366001627091774, + 0.37660099848628786, + 0.4572512841355684, + 0.4181203415320314, + 0.39404308164317814, + 0.3507138608420332, + 0.45565305972080206, + 0.40184392431118315, + 0.3167246937828745, + 0.3673306786328815, + 0.3574736644435944, + 0.3673296167848582, + 0.396687430845057, + 0.32150547295418175, + 0.36939499418081473, + 0.5108054593493221, + 0.34433082849927765, + 0.6079266511064885, + 0.7186509219901122, + 0.4310927297712832, + 0.5223282010372912, + 0.5154275846996749, + 0.6272784430785567, + 0.39121889290288936, + 0.6001879926507501, + 0.5247495992133304, + 0.5515549111245756, + 0.39667024568334003, + 0.4658759544967921, + 0.5514952894121927, + 0.7009484843709354, + 0.6313365483181583, + 0.4246587501443308, + 0.3565040133893579, + 0.3107802241412696, + 0.43051099540251525, + 0.4102547716876519, + 0.35768142507088035, + 0.4032784479579138, + 0.4321671457459026, + 0.4742687939135322, + 0.37396404968046915, + 0.4332545753673459, + 0.46714822386209137, + 0.4118492414916431, + 0.3865321575640476, + 0.383946248458017, + 0.3816749414612892, + 0.3416790004727847, + 0.36361445838940115, + 0.3500184534620256, + 0.40721863366147504, + 0.4020352585517828, + 0.39700860942740923, + 0.39113497032494526, + 0.41366239134494975, + 0.4772287489897358, + 0.30033088130719665, + 0.3383490827593295, + 0.31143092806558553, + 0.3778968124153396, + 0.4069560958668667, + 0.341861941040783, + 0.3697940377472689, + 0.3978033847241105, + 0.3698657879295328, + 0.40839660875640604, + 0.4238325148536602, + 0.4412739771825903, + 0.4156108454041656, + 0.4174085462806695, + 0.42613168444285493, + 0.45423136217540483, + 0.4170674866066782, + 0.40363817539378155, + 0.34893054584432914, + 0.35957748239549736, + 0.34656049725483623, + 0.36566289239636535, + 0.38980052534925397, + 0.38611041766618065, + 0.37413957744168813, + 0.5818546380753078, + 0.563705243521356, + 0.6320165947326114, + 0.6470620662935201, + 0.6175543342383838, + 0.5105765661712526, + 0.7827605746421785, + 0.616767870800456, + 0.6768357825582676, + 0.6000625205649728, + 0.6501728878353985, + 0.4869939789230486, + 0.5210728599984886, + 0.4921864960123308, + 0.6254844461382348, + 0.625155330767958, + 0.6118997088125304, + 0.45467764461857374, + 0.36739901904774885, + 0.7261627967362708, + 0.6920300637085525, + 0.7072323069791313, + 0.701714804551732, + 0.5234255364996532, + 0.7770716894879003, + 0.6897072248445472, + 0.6258463812198516, + 0.634514381019221, + 0.7738630794166514, + 0.7216249736957288, + 0.5457252855143914, + 0.5397438566590256, + 0.6286178652069961, + 0.6582516097547308, + 0.6630960538524545, + 0.5334665889293448, + 0.6544289752094904, + 0.5113792613428266, + 0.5323458470644185, + 0.6075548880651679, + 0.48834282899917464, + 0.6816124634671773, + 0.7040198126527397, + 0.5480678727553623, + 0.5327971510655569, + 0.48812922922221164, + 0.6582668828985359 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5972369929990613, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5051464287081975, + 0.9040064855485875, + 0.7599811893591262, + 0.4582454837933557, + 0.4109466902566876, + 0.4282630957048445, + 0.9273840097753159, + 0.6844059007426798, + 0.36179562333821996, + 0.2991670393230131, + 0.8596535558019618, + 0.6255074544944994, + 0.3835047011415663, + 0.4475894437042361, + 0.42776923204165274, + 0.7792464508875344, + 0.7621453968695244, + 0.8376713047993213, + 0.8350132800042932, + 0.8704146439884979, + 0.5957969832543013, + 0.5636376786924457, + 0.8852082197808491, + 0.4437107050837591, + 0.4667850479693932, + 0.7561056007461083, + 0.4088639237357805, + 0.6043946958625348, + 0.441650329571576, + 0.4509137079181166, + 0.5606220938593567, + 0.38999492057278295, + 0.8251877934266737, + 0.6096487582481648, + 0.8676531667546116, + 0.8741266199797021, + 0.3935250488964507, + 0.8105418627116354, + 0.8649676597872599, + 0.8454100659201653, + 0.8054168946425659, + 0.6552472662005258, + 0.6903586013830724, + 0.5941764682151254, + 0.4176149449829323, + 0.6186939017212982, + 0.7156464377785566, + 0.48030916764969556, + 0.5477641871527885, + 0.3647482031464813, + 0.3242449193254172, + 0.5135392352608715, + 0.4423417009864615, + 0.8176287365262694, + 0.3750288324687304, + 0.36804940895333277, + 0.38746978962430406, + 0.4178972037414227, + 0.3819109734539283, + 0.4890929571106253, + 0.36497796032909285, + 0.5790922178718865, + 0.4254310778727347, + 0.41174308083837663, + 0.43794310806046866, + 0.43438309371094, + 0.49828787718542183, + 0.4145770307030255, + 0.3341421657411775, + 0.36514191508802035, + 0.47767111774696464, + 0.5802995322875978, + 0.4222640292641831, + 0.5688609945051096, + 0.28741977706449945, + 0.7918288677141055, + 0.882469864869311, + 0.6108961582875497, + 0.38534732227108115, + 0.39782713145030535, + 0.44307138960852704, + 0.420605836856261, + 0.44837179589245446, + 0.4208249971577822, + 0.4625959455470755, + 0.4427998426757078, + 0.529318240942104, + 0.45660677533363137, + 0.5944018177367659, + 0.38336093131765275, + 0.5775486063549733, + 0.37546356357072763, + 0.6963015784915666, + 0.41902209540161983, + 0.6844451491141329, + 0.4545061710880882, + 0.6900918630189271, + 0.6428855550251171, + 0.293290834356576, + 0.3008796951404523, + 0.4126632650125963, + 0.3324968932579727, + 0.31077711447169704, + 0.6589489448396658, + 0.6308945006525427, + 0.38473305837399613, + 0.4045527490178858, + 0.5474814678906212, + 0.8278231861446294, + 0.5131966266032941, + 0.4801410748049326, + 0.5362578354459829, + 0.448644057040946, + 0.4114133228883751, + 0.755352503683002, + 0.4837671099803786, + 0.45505089204739885, + 0.5156146869702245, + 0.5651765907728481, + 0.4103925714153132, + 0.46014608673376345, + 0.46339305106014644, + 0.49231513037860936, + 0.4698185612252677, + 0.5615131813856594, + 0.48954577806080457, + 0.5386860834954825, + 0.6477177673394762, + 0.6705915045152536, + 0.5467838696385618, + 0.426599420947635, + 0.5529368612616599, + 0.5803306875182386, + 0.7651446309087153, + 0.39893844677282575, + 0.2743336068438853, + 0.5294721007034671, + 0.4359647461447, + 0.5672857701576846, + 0.5444082964186903, + 0.5499912851457619, + 0.5063472046332705, + 0.41080357615963864, + 0.7554115634397447, + 0.5583453732596939, + 0.7414336257705796, + 0.3536230260772644, + 0.3249113742826109, + 0.37039597149822867, + 0.29423287934749665, + 0.88132164269304, + 0.28431343057004727, + 0.8807689000329385, + 0.6583149424846829, + 0.4451226130093942, + 0.30109285591893054, + 0.31333738774845815, + 0.3330717852363559, + 0.6050946760578954, + 0.648044156042965, + 0.6701066659949098, + 0.5936444055192259, + 0.48388760797934244, + 0.41592817721766284, + 0.6422585664210787, + 0.5276855471139255, + 0.4732631534557596, + 0.5577892459322403, + 0.4185388347652587, + 0.42768424367024294, + 0.33851642107277624, + 0.8163129560013446, + 0.3817916183884341, + 0.41314913537824083, + 0.611175842116569, + 0.666912040667777, + 0.4197770844139869, + 0.3062577325540888, + 0.8667889010851785, + 0.3365176477705042, + 0.7377808207526263, + 0.38109029315600396, + 0.5592363040089487, + 0.6774969816395758, + 0.4320291351140382, + 0.5215552976508007, + 0.5249196656935252, + 0.8009589898539154, + 0.39389106354042475, + 0.5680090424161645, + 0.5033003335480184, + 0.6813565194267117, + 0.8841505204189964, + 0.3506400083925581, + 0.5325796479312582, + 0.4043093767670347, + 0.5785724563427923, + 0.6728914435525277, + 0.5341613937447999, + 0.39854396560277844, + 0.8564889204424364, + 0.8652151754258923, + 0.7397854661872288, + 0.8193662991615421, + 0.8298077991630294, + 0.8177391213313542, + 0.8283264169989877, + 0.3188909479179335, + 0.32877560999580585, + 0.4620541334955505, + 0.4149822659272492, + 0.3948729336334369, + 0.5563044485609598, + 0.8469084551090093, + 0.8358264980735723, + 0.6295026799099597, + 0.8091105357000382, + 0.8059829508556513, + 0.8430430196823874, + 0.8913996453225206, + 0.43459661165513114, + 0.48746767119188505, + 0.6012632013997647, + 0.5026447324691392, + 0.799817654423142, + 0.5147789093087264, + 0.8921409336842042, + 0.8668716960065588, + 0.8600227186024813, + 0.887652780096815, + 0.8803558092006152, + 0.8541092476886623, + 0.8129857038387429, + 0.846345321319629, + 0.8905593228324823, + 0.38094692496207555, + 0.9103627241671178, + 0.8642856701590852, + 0.8037080681896945, + 0.5381983311450295, + 0.859435687149562, + 0.46015357302901805, + 0.7375771800279055, + 0.8613547450367912, + 0.5155028674006044, + 0.502354220313448, + 0.6063524220351131, + 0.48228687914777546, + 0.4760801733454178, + 0.7006585705784053, + 0.8672315883091891, + 0.6525923471644431, + 0.7882485508835309, + 0.5580066933923811, + 0.8473988596265536, + 0.7891803935967545, + 0.8210927720996523, + 0.8892529884044479, + 0.8201358099387646, + 0.8850031507160516, + 0.8541049896479801, + 0.47977083890399985, + 0.8646450033021486, + 0.4852748216758936, + 0.46854127571905346, + 0.8676179803741704, + 0.893574462976209, + 0.8646620071272132, + 0.5065920085440799, + 0.5526710469698969, + 0.6220205823454538, + 0.7798607193634521, + 0.8717324754528205, + 0.6591096976105665, + 0.8883877911929138, + 0.9008138101555477, + 0.5365514546963454, + 0.2722332113247191, + 0.4684903468825709, + 0.8867388934593328, + 0.3664076369113273, + 0.5493782130940472, + 0.8652720466341627, + 0.8153334238124229, + 0.5481068119649307, + 0.8526252929441862, + 0.8811922053126112, + 0.8753638559991634, + 0.8626250994782892, + 0.850578090406638, + 0.8808970538598544, + 0.8378385714436191, + 0.9125621801517521, + 0.768394542664312, + 0.8179519849912746, + 0.8972138453831112, + 0.49051167546900615, + 0.49740247811967714, + 0.4703383810229919, + 0.46031192818372796, + 0.895503977384706, + 0.7800539995521896, + 0.8345923373631745, + 0.772414543119302, + 0.7441019241144452 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.34956550154905325, + "valid_tracks": 12, + "per_track_pearson": [ + 0.342583975020021, + 0.3844080157361664, + 0.29193647947308327, + 0.3828237890532391, + 0.37945738757018704, + 0.381257042275856, + 0.31919838821547963, + 0.3690742289657949, + 0.25954457337855685, + 0.3622746990207944, + 0.35421949895828836, + 0.36800794092117173 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.1930421274386517, + "valid_tracks": 546, + "per_track_pearson": [ + 0.19955618236466635, + 0.2245851065309552, + 0.1906140466539765, + 0.11951416133946903, + 0.0716866516781868, + 0.20025783154930754, + 0.10037635762241008, + 0.1290160018314467, + 0.20829654159378191, + 0.18292753761212366, + 0.23189861145548754, + 0.1293719330711211, + 0.13300535977251393, + 0.1119465372521382, + 0.08104786075745787, + 0.15598731738622787, + 0.19037613022233976, + 0.22864570215507782, + 0.22357622681218245, + 0.21472609149540284, + 0.18210673807964625, + 0.17031013750146534, + 0.2848138823306632, + 0.15294705019829297, + 0.16193296521188438, + 0.19718722927556412, + 0.219204721982228, + 0.2227202508038051, + 0.2441571893806723, + 0.22360630264566606, + 0.14062909172566893, + 0.23492936246369364, + 0.16877043942280848, + 0.11739686111775059, + 0.24415030938101737, + 0.2348478616097727, + 0.21415928375288093, + 0.24245726404614198, + 0.2053248672442707, + 0.2787880612253706, + 0.285961912752194, + 0.24728958610731439, + 0.21593821565400537, + 0.21190599742150745, + 0.1223972532814035, + 0.17640524857307918, + 0.18910935502665063, + 0.1880088699300437, + 0.11594512941425876, + 0.14800237401984495, + 0.20780401929029935, + 0.24429556581660122, + 0.25341979159673284, + 0.17342939582674777, + 0.2622484163779034, + 0.28891412424029267, + 0.1570630463925638, + 0.2675609976918749, + 0.20682275179151005, + 0.2290441286062082, + 0.2693270075763393, + 0.1639657568677222, + 0.2403702534166539, + 0.18225634324556567, + 0.25423228258622793, + 0.26328641570140526, + 0.27889333346678663, + 0.16187118135793882, + 0.11642732700216267, + 0.10168264688455979, + 0.21642774683829125, + 0.2526590736579851, + 0.2083910526237493, + 0.22113146778662718, + 0.2399838725370858, + 0.20165659037002168, + 0.2416771199727865, + 0.2193273515044356, + 0.24353355232011278, + 0.24962053390215755, + 0.20582012536330974, + 0.12428454812010258, + 0.22064660984707377, + 0.2554397684879941, + 0.12383663241130934, + 0.24047680967864232, + 0.19696293133800763, + 0.25954330111492685, + 0.1646362641604071, + 0.248473427888448, + 0.17820410538443254, + 0.20680377428438632, + 0.2286706953421968, + 0.10708699098803119, + 0.23800897740504456, + 0.20750676045035568, + 0.21857938108542418, + 0.23186613913832116, + 0.2149646462225257, + 0.13172744340793444, + 0.20581948056690919, + 0.22398507696229347, + 0.2252351568745013, + 0.3069988078546611, + 0.2609979962253995, + 0.27215494314249716, + 0.1784716044138763, + 0.23194510661697834, + 0.1825735102567508, + 0.1838100830224675, + 0.20076291017660391, + 0.20187855083963427, + 0.22560725842707285, + 0.22343612827199444, + 0.1758400003375151, + 0.22128539346918824, + 0.22728320496149237, + 0.2330519488477772, + 0.21657221763956533, + 0.25148018125348315, + 0.2555094041264715, + 0.25183577099560106, + 0.15513301241221117, + 0.17094543723450298, + 0.19715695401570896, + 0.2444161114138086, + 0.21894036327244218, + 0.24127217729770853, + 0.31250343670032726, + 0.2281041432419601, + 0.2529502823535138, + 0.1787240442624391, + 0.24145395585756999, + 0.24430226006019642, + 0.28917584617849645, + 0.21126221124051564, + 0.25914341311574585, + 0.18142444526737173, + 0.10938153394964355, + 0.22352646997898176, + 0.269758533272295, + 0.2277954384825295, + 0.2396377646024378, + 0.19467360387255214, + 0.1967196566602217, + 0.20977289051661838, + 0.21746376908686005, + 0.2569152089717521, + 0.18938931460634598, + 0.18461558623629612, + 0.21868509891315718, + 0.16327357924567087, + 0.20432392176753209, + 0.23074344965755433, + 0.20571438463735295, + 0.1434080826334781, + 0.23437116626980645, + 0.26626433271877353, + 0.17205608527557542, + 0.13949471972950242, + 0.13854184120732296, + 0.19648330106152118, + 0.18934609621553128, + 0.21681719351473444, + 0.19043645532908088, + 0.15322090448533512, + 0.21480955090218962, + 0.14669399688260568, + 0.14213634067898812, + 0.23704016754245244, + 0.2173980025534816, + 0.17030858732263826, + 0.19108415002232412, + 0.29457669931368624, + 0.2189601328703026, + 0.13516430564307283, + 0.13757918027108978, + 0.1731134405076223, + 0.22639035373641483, + 0.26735095489658195, + 0.2532353416806349, + 0.2559319853674132, + 0.11925594043881045, + 0.2539828242386957, + 0.2067763505350997, + 0.17015366270549406, + 0.17369873830098648, + 0.14705560774908408, + 0.27114328905692453, + 0.22755827292619055, + 0.18772844615217213, + 0.11819317566683311, + 0.12207927922301387, + 0.1979710265534275, + 0.1942355428537205, + 0.179058434359934, + 0.20576985380615978, + 0.20780967772876976, + 0.15661604292754736, + 0.19761761717375464, + 0.11439102371594093, + 0.09172182285580623, + 0.10794562835667958, + 0.09844199617482663, + 0.12207397518894834, + 0.2024037787964843, + 0.19824996677514606, + 0.17097143824082228, + 0.1043553256741376, + 0.1953569559853853, + 0.25541336997753067, + 0.2175752506044569, + 0.22263040664792957, + 0.22152635108910243, + 0.2060019450341039, + 0.20224320979718538, + 0.238336676383899, + 0.2715927879817231, + 0.24163357620210454, + 0.23189378387302312, + 0.21050946448414087, + 0.15280564332654661, + 0.21527593226421704, + 0.21552022405464866, + 0.2251058712650847, + 0.27924014056764596, + 0.20643188461278014, + 0.17662020144158266, + 0.20119456142269115, + 0.21466291991824338, + 0.15966947006622842, + 0.16110431181711435, + 0.2612575727660771, + 0.2055300221795156, + 0.17364458162206528, + 0.20686104655279364, + 0.27197083191510446, + 0.14013007621162973, + 0.2149533122424987, + 0.21270512815860365, + 0.10055703109546674, + 0.14241734333777592, + 0.12612836512019504, + 0.20478549930262918, + 0.15352583706024817, + 0.18837850340996026, + 0.08467824683098811, + 0.2592552942006553, + 0.22157731568349606, + 0.22762131864846716, + 0.24346883035670572, + 0.15404978292659008, + 0.2701643171315074, + 0.12214413795266454, + 0.24504491594863834, + 0.19062132828906744, + 0.3083329188371513, + 0.22502380211145376, + 0.22922875465815074, + 0.33394881420820727, + 0.19179240133829875, + 0.21624488215851526, + 0.11541580325088342, + 0.23132427439483821, + 0.214086391475855, + 0.14931679256637584, + 0.11676441484666773, + 0.11960837828823204, + 0.14063277001538624, + 0.2235938525227398, + 0.2738007568155193, + 0.12730627350677323, + 0.10967376758419962, + 0.17123943984457055, + 0.23331538203003133, + 0.18066746689465368, + 0.12493377053823768, + 0.060626055660009975, + 0.1886127195963779, + 0.1127968158577476, + 0.12603841745859826, + 0.2009681734401198, + 0.1721756478827082, + 0.21031606659027002, + 0.12192612905823681, + 0.13033808670638075, + 0.112967935979915, + 0.10261511815382761, + 0.14509570045485368, + 0.1793511212054257, + 0.20566691894935918, + 0.20484299227447061, + 0.19476868635409497, + 0.17194358089500858, + 0.16561291429595273, + 0.2618059167331183, + 0.14209794756609123, + 0.148682272989394, + 0.19901002219126912, + 0.199686964635623, + 0.20075433484920283, + 0.21966909892346448, + 0.21024669204088284, + 0.12777133207356992, + 0.20342777102397716, + 0.1566188230061999, + 0.10881955060798904, + 0.21935099690299603, + 0.22705839112668472, + 0.20902241216748263, + 0.21649676559265021, + 0.1844738796568263, + 0.25475853000757986, + 0.2555679568318227, + 0.22426987199096812, + 0.20625214705744518, + 0.19678496161853362, + 0.11378162974721312, + 0.16851887397888043, + 0.1794292927324039, + 0.19362804966356748, + 0.10494977497839451, + 0.14029018763073925, + 0.17960076454089793, + 0.23225806571767704, + 0.233494924228106, + 0.16420083093165472, + 0.23523792303128405, + 0.25768327190118495, + 0.15620734123847094, + 0.23307464515353332, + 0.18666951177963714, + 0.20034454754019096, + 0.2407410481794789, + 0.15172753963916943, + 0.22210533457220052, + 0.17375982097472945, + 0.23362005123736518, + 0.22862028080970975, + 0.24935531176831602, + 0.14743629081189527, + 0.10491735502222808, + 0.09677217813160675, + 0.19780637955008618, + 0.23212643668801977, + 0.19730979773470322, + 0.20232138660561097, + 0.22508662932845544, + 0.1897000047762809, + 0.20838729793563093, + 0.1970385205456463, + 0.21811927733941863, + 0.2352480210902977, + 0.19410717470310596, + 0.1189734532416716, + 0.2080173704055007, + 0.2292118331502473, + 0.12334352298874772, + 0.2163273038389398, + 0.20605070740501907, + 0.24215672399611823, + 0.14927802076562657, + 0.23706021207574515, + 0.1715612713049819, + 0.18627184005201453, + 0.21161046596990407, + 0.10496554039893093, + 0.2147092832884699, + 0.1853337576362698, + 0.1918873892702912, + 0.22871068690716143, + 0.19421375670596072, + 0.12272406126738189, + 0.18953609578029199, + 0.2172783819633657, + 0.20134881363396576, + 0.27941653708768055, + 0.23358685592148376, + 0.25531462531335, + 0.1632142051259845, + 0.2123336988092235, + 0.1674364037599529, + 0.1630164128931391, + 0.1810279959887571, + 0.18258332030623586, + 0.20850012888284364, + 0.21250427876083683, + 0.16218625012532525, + 0.2016781976780125, + 0.20065655608866947, + 0.21302797030413764, + 0.19473153979992913, + 0.23629712053266183, + 0.22621991228750843, + 0.2240139154463825, + 0.14655341801508978, + 0.16374173493175817, + 0.19239369761353511, + 0.21859393925453247, + 0.1971370880008696, + 0.2285387628149081, + 0.2900852075180928, + 0.2063792957287801, + 0.2343402512927579, + 0.16850836967713792, + 0.2076232348549899, + 0.201951880705826, + 0.2580870323222759, + 0.19958033749117485, + 0.24031215377050708, + 0.16456130533946295, + 0.10340666535876968, + 0.20608440024060196, + 0.23940797919729506, + 0.19968963674619064, + 0.23238891271536782, + 0.17207355637722102, + 0.17473275398823412, + 0.18258570065767812, + 0.18968622119411174, + 0.2170389832343234, + 0.17712191931338844, + 0.16301564533708093, + 0.18251457967063867, + 0.1387876320742056, + 0.17313595060050904, + 0.19260091985389693, + 0.1798956734315381, + 0.12227081444669484, + 0.20249063391365618, + 0.22682698961280418, + 0.14832904003720193, + 0.12892554061519954, + 0.12474500581917501, + 0.16753084589845157, + 0.16642508194830832, + 0.2077916621658573, + 0.17827902021485317, + 0.13599086017698916, + 0.19775649820145771, + 0.1454189439426982, + 0.13140169716904124, + 0.21835067018717585, + 0.1997999715434767, + 0.16947308794143306, + 0.16320093138381422, + 0.2749078621764768, + 0.2046673901516236, + 0.12147220735963989, + 0.12952226949870124, + 0.17929510759939477, + 0.18094673445655726, + 0.2644995811803432, + 0.2320883303586097, + 0.243015780576851, + 0.1038970953813028, + 0.2338640887573672, + 0.1959086771130411, + 0.17807689669360646, + 0.1494378244957304, + 0.1383410467176855, + 0.2273335616884033, + 0.20629804962355866, + 0.1795540467029413, + 0.13077631014904922, + 0.09642531231287918, + 0.14361762037053938, + 0.1950530871851025, + 0.1651629441744796, + 0.19854373282630458, + 0.19198474946311822, + 0.14960335690477974, + 0.1623561953309846, + 0.09133150575983846, + 0.06796169360456275, + 0.08966105895981427, + 0.08578318492434432, + 0.08503408355196798, + 0.19150913542242312, + 0.1550005799577552, + 0.13576421075062123, + 0.09549847164886092, + 0.18042650506656002, + 0.24968001632266265, + 0.22042201769957373, + 0.21829483221999305, + 0.21798487166886774, + 0.2054288735921829, + 0.2004250641775354, + 0.22990929179678193, + 0.26431086969676126, + 0.23740435753391906, + 0.2251821052454753, + 0.2093256739309808, + 0.14436875922429873, + 0.21472511225645538, + 0.20364162842713399, + 0.22080772908208804, + 0.26646300500864495, + 0.18738833547214462, + 0.1717388180437895, + 0.170876901083507, + 0.19356766323301192, + 0.11533845323253158, + 0.10630562174127979, + 0.244295021788761, + 0.20010240134702367, + 0.1685937116429196, + 0.19241547633858896, + 0.24983106654931073, + 0.14192165761756045, + 0.197109878275414, + 0.1905447702272679, + 0.09063581202804785, + 0.11438744072178621, + 0.11279196745579735, + 0.2004464301879699, + 0.12436890271130757, + 0.18001225104356305, + 0.08164948799503802, + 0.24882058706442794, + 0.2176918178298283, + 0.21123938301555756, + 0.23459326274031234, + 0.1612442217341263, + 0.23457697986721, + 0.11465443090398364, + 0.21236737999764738, + 0.18950088224694445, + 0.28954152336082994, + 0.21785846933931405, + 0.22755544912908476, + 0.3189009165846459, + 0.18917542760114403, + 0.19112308000662492, + 0.10554739598671549, + 0.20922997340654872, + 0.19586396319860885, + 0.15963747010154228, + 0.10666468583134212, + 0.10817336276834513, + 0.12989475271851386, + 0.22073215896115628, + 0.26270337259704996, + 0.1189918650410344, + 0.09150785338667813 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.21068735878499603, + "valid_tracks": 613, + "per_track_pearson": [ + 0.2396121765083079, + 0.21742617378297072, + 0.1944567561852619, + 0.254575053872246, + 0.23684763819818888, + 0.19535534380279307, + 0.22150627243633472, + 0.2367186297400461, + 0.2433750676995738, + 0.26769853996768667, + 0.18358557052842428, + 0.2297276275925011, + 0.15200581387533388, + 0.2795054582754713, + 0.2679380292372057, + 0.2767658563931556, + 0.2684881806686551, + 0.23572991964052437, + 0.21572088145498813, + 0.2452829304031348, + 0.20615403998670326, + 0.22586807046619314, + 0.2312258900009004, + 0.22639425233225888, + 0.16398057072939187, + 0.21394235009659757, + 0.16834604491942048, + 0.22094684036348278, + 0.2297318506671091, + 0.2382664267822145, + 0.282093980760062, + 0.21824816842758085, + 0.2980225758085653, + 0.19369530211423946, + 0.30112750744964956, + 0.16105423554419265, + 0.23865509265207965, + 0.23226581513267025, + 0.20721583934470458, + 0.23332193820083144, + 0.25998874580978554, + 0.20883027935329718, + 0.23483883128481073, + 0.23147560089067842, + 0.2594532928340572, + 0.1931830854508425, + 0.22776281937048637, + 0.27661515694708855, + 0.2518365315488076, + 0.21299805647334466, + 0.23810849523606784, + 0.23775063710614092, + 0.2327442468915728, + 0.25648832480562644, + 0.17232725711663466, + 0.24676513170241396, + 0.2399492667391458, + 0.24233208166004017, + 0.2398456194331544, + 0.21888488699445308, + 0.19363327561891383, + 0.2197264727613775, + 0.21967774427839767, + 0.19296804062558154, + 0.2398228662689071, + 0.21202176408134799, + 0.24326636193248322, + 0.2276154601960646, + 0.23587188027974534, + 0.22696137601723868, + 0.2268604291025291, + 0.23999693908879957, + 0.26166643864182254, + 0.22987937658134236, + 0.24423419798657708, + 0.2580626719438884, + 0.24237900162529127, + 0.24268873311819386, + 0.28560593958505337, + 0.23688032232891254, + 0.22388578383581353, + 0.20838984444022549, + 0.275448047083708, + 0.2556824027837078, + 0.18333253950484757, + 0.2041759956016737, + 0.24907237965010373, + 0.2561179220420642, + 0.23944771002761514, + 0.2550131960249841, + 0.18070976203383124, + 0.19654134576498722, + 0.22718084012261158, + 0.2312000354156662, + 0.24182485155566963, + 0.24120432898888633, + 0.22533058965167574, + 0.24064158534797417, + 0.2368558375102774, + 0.20890667634578078, + 0.21491418749918528, + 0.24650354800844246, + 0.22818480934926258, + 0.2246922352444411, + 0.2413418449743371, + 0.2712404747622371, + 0.23100784464485552, + 0.2833208367561689, + 0.26264140019435095, + 0.24015098968322632, + 0.20768413896127005, + 0.32194412084059254, + 0.29546219913163785, + 0.1821737199553992, + 0.23285778089682252, + 0.23655721451746697, + 0.28936148797365857, + 0.25665826813456777, + 0.2559447151471447, + 0.2333183769595399, + 0.1796537171549368, + 0.24046227451132815, + 0.22964150280847279, + 0.2673902348007867, + 0.26197450560995683, + 0.23915466286622994, + 0.252008749320876, + 0.17772643214523284, + 0.2201441335091361, + 0.20977221698840856, + 0.3049221692291915, + 0.250447980301304, + 0.21166412680704774, + 0.16138561473451332, + 0.18184371307022465, + 0.23233064129971778, + 0.26919789199826155, + 0.2860277684856349, + 0.276692213051496, + 0.271507581406338, + 0.24009842220841893, + 0.194841338167427, + 0.2885492434870452, + 0.29887982978716804, + 0.20960534417637003, + 0.21362776043940976, + 0.18262286043520948, + 0.22662755279441635, + 0.1707560726356683, + 0.24829822745635857, + 0.1870056160156746, + 0.25203087741725183, + 0.2048022856632488, + 0.20739205030262217, + 0.2664608907177063, + 0.1946522145008973, + 0.2322629117760048, + 0.2703644990948379, + 0.20863642437508462, + 0.2693442843927822, + 0.30580198596520913, + 0.2285395245166197, + 0.2563225553022507, + 0.23868846437540917, + 0.23677481235409886, + 0.23271524447290873, + 0.23781601107491457, + 0.21528687922868847, + 0.2257624071125744, + 0.2556584685932397, + 0.19223862346705495, + 0.20083319645106373, + 0.24283631735561187, + 0.2491218384600011, + 0.2123444531165577, + 0.20402801721174701, + 0.2148556638787894, + 0.2706074093104324, + 0.2703990316004476, + 0.1884731220812994, + 0.23851781435715202, + 0.1967824688574875, + 0.196867582190135, + 0.18461444272206487, + 0.1974398985239025, + 0.24374787551908708, + 0.214119682910095, + 0.12151660611755885, + 0.1772349342447161, + 0.3045530922195927, + 0.3330761009951277, + 0.18843976472568602, + 0.17238046416207317, + 0.23342211327961576, + 0.18283849993141846, + 0.20338472080518116, + 0.20622876172739796, + 0.21200420056523495, + 0.17141839207349277, + 0.22870845921249466, + 0.29506648687047166, + 0.17561278279353268, + 0.3490942778420528, + 0.2534191923053743, + 0.20037642547523593, + 0.21195315810070503, + 0.21007168994448486, + 0.17190136928954503, + 0.20099208639618668, + 0.18281030286826808, + 0.17313816955276187, + 0.17244965779786803, + 0.17117933719298442, + 0.18653720606159763, + 0.192521441277054, + 0.22529466700934933, + 0.17375644512176383, + 0.20108448367591877, + 0.1790061038321579, + 0.1946794607887882, + 0.3273645418040744, + 0.2002949474044493, + 0.2852787904202679, + 0.2501894560621998, + 0.2533703368637356, + 0.2887573660780674, + 0.2522001556724362, + 0.27617541911277627, + 0.20709241900468198, + 0.19056574466802445, + 0.21985571710244656, + 0.32198004916898393, + 0.20697530255325863, + 0.24524134663911898, + 0.21406913272487407, + 0.2248152865180252, + 0.14908726782446471, + 0.2418601886408755, + 0.2417359844129213, + 0.24006048900071816, + 0.2434246674189741, + 0.18262849096157655, + 0.22897952506688754, + 0.19100740212689465, + 0.18841555369796706, + 0.25061308533155396, + 0.30286748503933314, + 0.2261625206690557, + 0.22606323181235918, + 0.2050862316355344, + 0.2071110392001243, + 0.22900196669281955, + 0.17672143522770808, + 0.19582809129715484, + 0.23046793093520151, + 0.2574757622680812, + 0.20693637081268618, + 0.21288306218997868, + 0.2068925127060573, + 0.19310843885473342, + 0.2548624210724129, + 0.22860102642608884, + 0.23694563574252922, + 0.26461461708732903, + 0.23320637548470363, + 0.21659852264832424, + 0.20367958397660357, + 0.1840719495241105, + 0.20836599342321216, + 0.23067636592492444, + 0.22311442979349927, + 0.13863875346552393, + 0.18963945510777586, + 0.16947804383676757, + 0.21205036875133312, + 0.189808732650034, + 0.13725770133359128, + 0.14759642363761702, + 0.20221053086186525, + 0.20912703815141837, + 0.2173380615581059, + 0.18087536086269024, + 0.1527787599226931, + 0.1430811234217776, + 0.1833549995454915, + 0.1797256781248388, + 0.17954418037326025, + 0.19699757318726205, + 0.19075582476129876, + 0.20254800475122725, + 0.18296503795236554, + 0.1506383722070428, + 0.19866509609990443, + 0.15869156606611734, + 0.15024901895806045, + 0.1129645445908297, + 0.1785972050478994, + 0.16774912953219062, + 0.20045420891626753, + 0.17908277539656628, + 0.1829265735305624, + 0.2136096771906913, + 0.18948986057194298, + 0.16130454429989052, + 0.16727003896080456, + 0.2673356126431994, + 0.14305846966230235, + 0.19427774271755477, + 0.1881398475567588, + 0.1805135763782955, + 0.18377701781970773, + 0.19996268755228083, + 0.19323164197465414, + 0.20491250803614985, + 0.19016172612127363, + 0.1773994258016653, + 0.14046599827302514, + 0.1631112869587745, + 0.20052689359333514, + 0.17289292763136657, + 0.18185383581863152, + 0.19712411276027228, + 0.19268104932762184, + 0.15636174582751244, + 0.18138781783982133, + 0.11506241495645032, + 0.17941726248481382, + 0.1795920461836476, + 0.19381116940477022, + 0.19112163946563915, + 0.17280219734837649, + 0.16872940280544144, + 0.19097906655450664, + 0.19583504576106894, + 0.1374665302026543, + 0.20733873885242687, + 0.1904177857675568, + 0.21646750157223787, + 0.19856354226281353, + 0.208835661234966, + 0.19612692523218964, + 0.18721269506949248, + 0.2198269720819149, + 0.22076977089731356, + 0.2112525755309089, + 0.21177922778500285, + 0.22809095037084182, + 0.21261369480481956, + 0.21077930755754834, + 0.2144386134405248, + 0.18530649205010094, + 0.1750187852013466, + 0.1568775969384406, + 0.21670574204584223, + 0.22146628712603333, + 0.15968198438170011, + 0.18690363844017896, + 0.18830057110781712, + 0.17032355519595893, + 0.17337117061147486, + 0.15844887880459635, + 0.1319585454481103, + 0.16482356611742838, + 0.18334983543806588, + 0.17819140641773398, + 0.17925037356348403, + 0.19517655929210723, + 0.19428345669114797, + 0.18589418807423694, + 0.181905965305, + 0.1575775545560975, + 0.14951042719174726, + 0.1710743663203263, + 0.1670198988019873, + 0.20450920962087174, + 0.1710082796569146, + 0.19448516713035185, + 0.19125662294374415, + 0.20363123687084161, + 0.21578146532744738, + 0.180406236020863, + 0.15785708939264104, + 0.2482971008867558, + 0.21067152086712054, + 0.15736339567556792, + 0.20168782881630584, + 0.16195446733582153, + 0.21116004228609062, + 0.1778998328220946, + 0.1847212802474072, + 0.170290159160402, + 0.1617132386698006, + 0.1864898909598255, + 0.164266356874867, + 0.2155732467166363, + 0.1902176237885858, + 0.1814338047577488, + 0.17326172381343088, + 0.13802612102421205, + 0.14590933079606055, + 0.15505986107654013, + 0.24194000045823566, + 0.19093025622275248, + 0.14859596705446032, + 0.139145564654831, + 0.14347384639839045, + 0.21374881796481718, + 0.1997044700216238, + 0.22839490843870514, + 0.19337786937725346, + 0.1959857679426036, + 0.16221472302980922, + 0.15253352457710695, + 0.21648315543304456, + 0.22773296427912632, + 0.1464142941157321, + 0.14859407618407336, + 0.15163004188915064, + 0.18433495448223575, + 0.1684822786448065, + 0.18208005439808803, + 0.16461009204001023, + 0.19620383423265322, + 0.16182229026275252, + 0.14473083770550912, + 0.18316064903243945, + 0.1528025444412074, + 0.1969191999739611, + 0.1672756360841051, + 0.1892810280342141, + 0.18389624017123182, + 0.228681018177936, + 0.1552082900858097, + 0.17550395515673547, + 0.13862161849580135, + 0.14166779479320513, + 0.1376617405880239, + 0.13434026375348584, + 0.13041320283660665, + 0.17146212051269968, + 0.1959615111752819, + 0.19748467265425418, + 0.1611748127671079, + 0.23331072789928792, + 0.1938743535105406, + 0.15101891299973216, + 0.152097471697864, + 0.17657330157737408, + 0.2453448937918422, + 0.21747296820997444, + 0.14191250624506094, + 0.20349197163086796, + 0.19441016760204277, + 0.1228337579683726, + 0.1401350254675305, + 0.1397505279760856, + 0.19347284623473257, + 0.1839871207917317, + 0.11264875490121726, + 0.14569936060166574, + 0.26381644958539086, + 0.2712045621868274, + 0.18195646642981483, + 0.1485132427844615, + 0.19820188721317256, + 0.1572777491254758, + 0.17853907725231358, + 0.1755207315046603, + 0.19379183071710776, + 0.15889581024658997, + 0.19807788217395536, + 0.25589012144138834, + 0.1703298784099822, + 0.3101219624152767, + 0.20756127843037134, + 0.18608519855785882, + 0.18329671806610862, + 0.16845225251218804, + 0.12226318867198183, + 0.2089902311839013, + 0.14114086028577322, + 0.1682135963883387, + 0.1664071530499434, + 0.18717162482020727, + 0.1797471551197782, + 0.1816926636959218, + 0.21663345905512407, + 0.17984629627321147, + 0.1551850432130814, + 0.19031309994959075, + 0.18477539559989126, + 0.2578937457217044, + 0.19165581821658334, + 0.22709270975483437, + 0.16865194083152865, + 0.17569312658603578, + 0.22615164071210983, + 0.15526233698203076, + 0.22002783189921352, + 0.11348913897909996, + 0.18233681944698188, + 0.14480418543098078, + 0.277585157958009, + 0.20695648741782655, + 0.2061488578341504, + 0.20645334884263275, + 0.18740180391007627, + 0.10536420295091636, + 0.21442379486790789, + 0.20110176669556581, + 0.18046952325154253, + 0.18627455555273498, + 0.17406804969742898, + 0.21911679986038246, + 0.19268082568351735, + 0.19538393968875586, + 0.19969456286933068, + 0.2679184084514889, + 0.2051256639433557, + 0.18103312719249362, + 0.15284964952661714, + 0.15601608159728478, + 0.18220529674979896, + 0.15313591564345388, + 0.16925623229012043, + 0.19387441118334758, + 0.20417644892434283, + 0.12800184780926713, + 0.1791137575641985, + 0.1615155102292601, + 0.18103993700283588, + 0.19114189334254172, + 0.12866106939590163, + 0.1810396667377483, + 0.2108976577883292, + 0.19146141152756563, + 0.2100150012201171, + 0.195666072998692, + 0.11066772666738756, + 0.19779009452416904, + 0.19693210294623598, + 0.18182564364099124, + 0.19769578737881438, + 0.09637391987763932, + 0.26764218747141455, + 0.30446446799682964, + 0.276242692904721, + 0.25286069876456896, + 0.2603338619191649, + 0.2155885480047911, + 0.2171183317733866, + 0.24675236666787856, + 0.18868724496922976, + 0.20407076600134386, + 0.2119926849386165, + 0.1977987608584445, + 0.19749051837183182, + 0.20498380345654135, + 0.21434693713661712, + 0.1992828345796491, + 0.20756680415885137, + 0.18617662314694344, + 0.2671792620297316, + 0.26425626316134443, + 0.26793586498896005, + 0.2663253302318262, + 0.2948895539895203, + null, + 0.28190453872224563, + 0.33276625507327123, + null, + 0.07697944311120833, + 0.3982145889307656, + 0.07532189507976907, + 0.2556802653867343, + 0.24169190929312023, + 0.30515880072765605, + 0.36759301275291273, + 0.35892059536005005, + 0.37532408616510854, + 0.3733213391183945, + 0.3395109869564121, + 0.2937202798620642, + 0.22447903043015088, + 0.3791320066900833, + 0.18574955101437604, + 0.29746060183921974, + 0.2647369612909682, + null, + 0.25598352480866293, + null, + null, + 0.26071465172400105, + null, + 0.2327510721510598, + 0.20897446436404857, + null, + 0.30210191289295096, + null, + null, + null, + null, + null, + null, + null, + 0.25840323505801943, + null, + null, + null, + null, + 0.22009711427089065, + null, + 0.2621419025824258, + null, + 0.21174425784860085, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.2757079510119619, + 0.1596560387079658, + 0.24621238706992044, + 0.2582666374281978, + 0.1883175943675518, + 0.23697036341992464, + null, + 0.29963870237079027, + null, + 0.23992433614827022, + null, + null, + 0.24083536174711656, + 0.27191623156504563, + 0.23057949866637667, + 0.2372630376300417, + null, + null, + 0.2061132556747008, + null, + null, + 0.24516163028524446, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.23902287898014196, + 0.2367437751777455, + 0.25082122841447335, + 0.22273833244727878, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4032831995326907, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.20596341155157177, + 0.31166900861249935, + 0.2939791579030284, + 0.4443841464841782, + 0.36809501224478136, + 0.23737796352415597, + 0.3482634857125766, + 0.40196242731870024, + 0.3373857982846439, + 0.3473560692021163, + 0.39387727596187816, + 0.36967644850098647, + 0.26939005486089407, + 0.4072421392845433, + 0.3489199631786439, + 0.35359938624511805, + 0.38542605595474316, + 0.32053418232230635, + 0.4531285166082457, + 0.33688636485401263, + 0.3432830654339461, + 0.3272562883428359, + 0.3035656879911973, + 0.45724084346580396, + 0.3276200627745534, + 0.3561427284754092, + 0.2560117901813217, + 0.3651305639025781, + 0.35939150496132294, + 0.3795991945964458, + 0.3598846895117996, + 0.39026690377693585, + 0.4806625600179594, + 0.2736770399793426, + 0.6461584462471308, + 0.27410478160974017, + 0.20424129653690062, + 0.16165296320288125, + 0.32272539720252813, + 0.2167208756155665, + 0.26273926713205253, + 0.4531778079928414, + 0.29445324538473916, + 0.39207483422859807, + 0.28553513727525537, + 0.38595324505524514, + 0.29288528273751974, + 0.20144440713075107, + 0.4569253565158486, + 0.42331325836609585, + 0.379452085325689, + 0.35718020595678324, + 0.36828742932854625, + 0.34653130192815623, + 0.30894238211949415, + 0.37192297625695026, + 0.295219513939478, + 0.4137438301767917, + 0.39844170371295273, + 0.4200114816019427, + 0.2618396542127296, + 0.1964048785510467, + 0.3912144317818145, + 0.4106927721436892, + 0.2701164698418626, + 0.3356347929804215, + 0.33404977299981153, + 0.4371373856608021, + 0.4597767097886163, + 0.3800268132352417, + 0.3351257100420981, + 0.6757290561642809, + 0.3387234905854204, + 0.324990859351862, + 0.32395998564369494, + 0.3997987814315664, + 0.24859782647635237, + 0.6081306715497186, + 0.37646557543458214, + 0.3759411279700652, + 0.36350366415756197, + 0.39665847885892735, + 0.4681384282756416, + 0.41368538292376333, + 0.3028049431065893, + 0.26009442425108426, + 0.43504950195461933, + 0.38880357505869884, + 0.8002181485196034, + 0.4412224035304813, + 0.3375936900017678, + 0.32752004603080304, + 0.39025125653892684, + 0.31554989570837444, + 0.35594899630296356, + 0.5028975870250635, + 0.3146032875640386, + 0.31257874973084937, + 0.33276213277050837, + 0.2733768654610315, + 0.24220517709285058, + 0.3715479270101119, + 0.32820970470979066, + 0.3218994659075155, + 0.3680411128926598, + 0.1823741975110631, + 0.22324916284253518, + 0.24360020117954062, + 0.3750114974343165, + 0.31977406165756334, + 0.1873269835355793, + 0.26458890059262924, + 0.23949646945565314, + 0.31734440324605195, + 0.30759699200192214, + 0.4301524399086537, + 0.33828439934352444, + 0.3309831334545468, + 0.28035886210825256, + 0.34602193097313805, + 0.2764895022741886, + 0.3579701662880905, + 0.3317075067279533, + 0.21554883184776297, + 0.3114628871234857, + 0.2529070528654455, + 0.26924531983795175, + 0.25415418502562886, + 0.27499131843538477, + 0.5223349101193798, + 0.3165824539697752, + 0.21619819892279896, + 0.3522570870822998, + 0.5639617243164785, + 0.23771876388920576, + 0.2970189696977115, + 0.2693850541924695, + 0.27923503616136947, + 0.26231778936946104, + 0.23689384768493327, + 0.2263500639122292, + 0.4208934338733438, + 0.48854190731917124, + 0.2651991729735585, + 0.3517939048520881, + 0.4172655403343026, + 0.3823851414395783, + 0.21486937195901196, + 0.3654297758576896, + 0.28389829683647544, + 0.3277778933058627, + 0.2659522397106114, + 0.3271930694239052, + 0.4131951193647908, + 0.35564195255165393, + 0.3511688337318715, + 0.22170902392902786, + 0.2610321902191782, + 0.2407539320774382, + 0.3470807094985632, + 0.3112956131147154, + 0.28972812008082954, + 0.326417648559885, + 0.32643625138637195, + 0.3092553091313735, + 0.30968320048706705, + 0.2914754967447273, + 0.4143730720059262, + 0.21197510087825056, + 0.305779484734117, + 0.2758093250449337, + 0.22809151658309448, + 0.45450890701764546, + 0.24980944452349169, + 0.20574444969961195, + 0.36830341815722345, + 0.2151581892961229, + 0.31504871445187077, + 0.3033711443382587, + 0.25907367520180824, + 0.29979833340533296, + 0.3043749045763713, + 0.2653929117344394, + 0.34857375454841194, + 0.258311016817976, + 0.30847605664706873, + 0.36622514787072713, + 0.2228621174394303, + 0.2455356722523705, + 0.33743061809183933, + 0.21963713979983998, + 0.2845339685037795, + 0.39243637819212907, + 0.32511225277270595, + 0.25407325509762585, + 0.24722262918189847, + 0.2277821540285221, + 0.22604621552407364, + 0.227237579038627, + 0.39890018375473507, + 0.2605930336296522, + 0.29040731733621056, + 0.25788298610000354, + 0.25283723131158564, + 0.39127589904945437, + 0.27876734925172525, + 0.30385221922444006, + 0.32190306488251197, + 0.22641724482040146, + 0.43551336174409366, + 0.24259845669832902, + 0.36865572980790795, + 0.29091732821366106, + 0.3402849308813653, + 0.8401288533150701, + 0.36913632081705144, + 0.36388637242013994, + 0.35589992736707854, + 0.4303336358477263, + 0.5263193357130661, + 0.5950672392656627, + 0.6084022084047561, + 0.3777575071062858, + 0.3672483144014453, + 0.44577545119264356, + 0.7166920887392104, + 0.2766344386825425, + 0.30927134124917294, + 0.5237532422200747, + 0.4990670326046357, + 0.5107439055467222, + 0.4084290475402138, + 0.28440600021051077, + 0.34541320518435215, + 0.5822563638822911, + 0.39982515705240695, + 0.29353715555879234, + 0.6032383756365227, + 0.5313040861931702, + 0.43187481513089443, + 0.4643573197967088, + 0.693772612847507, + 0.43787977451570914, + 0.4731260989552583, + 0.42969494838407374, + 0.754695070956538, + 0.2915954717066698, + 0.6280381404066016, + 0.366935484552875, + 0.5331062402819161, + 0.40544446455858163, + 0.37932318265352905, + 0.6910409846138884, + 0.48726749158785093, + 0.26999569783313254, + 0.35273248490487424, + 0.629301367733433, + 0.3247161012061301, + 0.4323931278203677, + 0.4224129263893969, + 0.32117594156289636, + 0.4417683638394181, + 0.44286929311071005, + 0.3849667015097245, + 0.3790981819690705, + 0.45561151250742926, + 0.3200647473044442, + 0.29250755690377506, + 0.6167694918983546, + 0.3440525010920366, + 0.43501104845797883, + 0.5446446353826082, + 0.30406390702893527, + 0.6506639175432308, + 0.302470909605474, + 0.3057174850409848, + 0.5457125704303559, + 0.45753474150164397, + 0.5730438316065417, + 0.7220367397048776, + 0.6477595204620403, + 0.3816857237317046, + 0.4487234634299418, + 0.5240820775317435, + 0.2649405747836699, + 0.3034462202237159, + 0.5485996268378793, + 0.5003842064162811, + 0.7053302025920163, + 0.42899257148306164, + 0.33397046083688264, + 0.44113946258914105, + 0.352528573358462, + 0.45856625869529927, + 0.6875601921275265, + 0.6104813691037456, + 0.4578305218511022, + 0.5078676268770409, + 0.47587223955715197, + 0.500955512396706, + 0.7275896793986661, + 0.2837115586935696, + 0.4225849649725094, + 0.2555577307775121, + 0.5802521328289708, + 0.2989751752075143, + 0.7271619513349037, + 0.3830049664314718, + 0.3972396111774385, + 0.6783485607911386, + 0.49800662279775454, + 0.29882650912810194, + 0.5479617457370105, + 0.3776517654115839, + 0.5056177574427471, + 0.49208963639613484, + 0.3093286556056372, + 0.4803740683735261, + 0.36317026492866705, + 0.44266461747082214, + 0.32998171302266055, + 0.23986875597790336, + 0.37383461818689295, + 0.4938381613864741, + 0.5543147536164267, + 0.6478232297597406, + 0.36560387155385243, + 0.2712133385740364, + 0.43390854703284326, + 0.4399781459352882, + 0.4090614266047274, + 0.471189238260255, + 0.4396771577128376, + 0.3430472022187943, + 0.30397452963483845, + 0.48100528095319767, + 0.7585656918192114, + 0.5148034136003874, + 0.5853680152099953, + 0.5796835177802914, + 0.7606789782933507, + 0.5397982238990592, + 0.2889561215359693, + 0.565468991785872, + 0.3499604452874826, + 0.4908959174656084, + 0.4189767803557882, + 0.370560917635152, + 0.4762933976689646, + 0.7698789876712829, + 0.2159320814149534, + 0.4422522371141488, + 0.45136508148121945, + 0.6073594183969493, + 0.36500175766300524, + 0.23538600223819003, + 0.39505492993399466, + 0.7129490844598985, + 0.36219098754980106, + 0.41939113913105064, + 0.49242425489065406, + 0.4258514225701415, + 0.3771319465422963, + 0.47414265340202993, + 0.6013039044084386, + 0.44580757166665747, + 0.3745125322495928, + 0.5234898372908979, + 0.4404441933105542, + 0.5509896424982005, + 0.513007503661115, + 0.5081499290973429, + 0.5285086058017708, + 0.2762772072178326, + 0.25279199845767103, + 0.47148957850897477, + 0.2834828045652587, + 0.6834709185268304, + 0.2705004036007188, + 0.5456701432146353, + 0.5340005554483689, + 0.6753025027675776, + 0.275537124251429, + 0.305700311886447, + 0.30459217580465675, + 0.29639058701245247, + 0.2619054339030657, + 0.24957382701083336, + 0.5979987116984244, + 0.3165551242375742, + 0.6718877225070526, + 0.5635520388551264, + 0.41135987120263934, + 0.5420485387705626, + 0.3686795765622139, + 0.5187498969920248, + 0.5116738659976261, + 0.4484884047592156, + 0.37037835271354075, + 0.445464418542785, + 0.43640924146041404, + 0.502920990575136, + 0.3721184927870023, + 0.2938213532105104, + 0.5964042771917981, + 0.4511963552573364, + 0.645914216190324, + 0.6057723033172591, + 0.4794266219696693, + 0.2813415625313139, + 0.2970564729756127, + 0.32232151416228133, + 0.41001191477732013, + 0.7392208229473383, + 0.45679413313311545, + 0.34844732996402855, + 0.7287047521457569, + 0.4056883707644792, + 0.6123254843053283, + 0.5155312343138524, + 0.3773703130385766, + 0.47976186019407446, + 0.43229429683931264, + 0.4225766428453864, + 0.5691694348693341, + 0.5127006734433189, + 0.5005406481637749, + 0.40081240202308926, + 0.43351797869126435, + 0.40647112096793186, + 0.45565362854509855, + 0.760387413639979, + 0.46629776976698195, + 0.44125699128609036, + 0.41789651194937355, + 0.24609460840009462, + 0.6555994347132326, + 0.37425454586970003, + 0.2866358155807052, + 0.49608303922767194, + 0.3876116766923561, + 0.2961102890218003, + 0.47333191533167845, + 0.4046099806150482, + 0.768836209431387, + 0.3011786032605098, + 0.3576170065923236, + 0.28781051163689514, + 0.412395699322832, + 0.3650899786815465, + 0.2660717230901193, + 0.3613105302447648, + 0.2512836498689177, + 0.3881863394095669, + 0.7324375923833502, + 0.2707036212802841, + 0.38223123182588226, + 0.47420100512338564, + 0.3963111672449142, + 0.35924787438094963, + 0.45616713482348925, + 0.3427906604415203, + 0.5016232953746687, + 0.49410172746447145, + 0.17298053378269446, + 0.6108485651026281, + 0.667885631514838, + 0.4772296647820116, + 0.4543035223880473, + 0.49719559722848417, + 0.35134220799537974, + 0.32279825904726295, + 0.40217425197098283, + 0.7707712165121623, + 0.31802200601843744, + 0.5838705352219403, + 0.3039023222099487, + 0.4378010068894979, + 0.44713813613529463, + 0.4610358591724481, + 0.35430633056193334, + 0.27698689942074095, + 0.2825341943109902, + 0.26041721648515914, + 0.47638946189367315, + 0.4090524101417076, + 0.2341992159268462, + 0.3826169577418686, + 0.4305842910150971, + 0.26454714122775697, + 0.5345077841238528, + 0.43968853340489517, + 0.3526683574962825, + 0.43279855382562205, + 0.25987221272043826, + 0.3998550039368998, + 0.4530937802200469, + 0.5569070697323387, + 0.42147179762671405, + 0.7698789371705238, + 0.27570770337599326, + 0.5414530495409013, + 0.7272290392984627, + 0.6168743702611473, + 0.19973246859184743, + 0.4640302849946875, + 0.4007875804046907, + 0.30059120916565507, + 0.32537435781870316, + 0.32612473407298237, + 0.43787357031088614, + 0.48205549759090155, + 0.31531809505187447, + 0.5912827316968725, + 0.7544005858927434, + 0.41407135099309295, + 0.38222767985234435, + 0.41175160675897454, + 0.45795311561719576, + 0.4840659661238051, + 0.47777309882205676, + 0.4635185565838797, + 0.27270975143646, + 0.6290016984027031, + 0.4510072587177668, + 0.5346164794888364, + 0.6964954993908574, + 0.28221885489312837, + 0.5782664779659202, + 0.3665359802289674, + 0.4248851653614656, + 0.3340433907424567, + 0.4956444236231327, + 0.4245687888842212, + 0.29494255077301146, + 0.3838298764360888, + 0.3952289837820721, + 0.5296849355593198, + 0.5701436991320108, + 0.4125878967918645, + 0.4367974650432475, + 0.38837500282058374, + 0.7241963819616059, + 0.40051113237990466, + 0.47131926169501775, + 0.43761323976284944, + 0.4476663244321057, + 0.5871650312092834, + 0.24849937893166932, + 0.5486198784534854, + 0.35126477125167244, + 0.5116415352411322, + 0.3931779015750175, + 0.2965649600349598, + 0.29670514617226834, + 0.439336331588447, + 0.2759037429472055, + 0.4885759451397498, + 0.7144137741602756, + 0.7168905693521133, + 0.3480077738511364, + 0.2911126796037075, + 0.6876574542578064, + 0.4263141424458813, + 0.35602144616061976, + 0.48554864843835166, + 0.3391030800721543, + 0.5801424774471027, + 0.41882402460496393, + 0.5457769378342475, + 0.5696473350498504, + 0.39565809995866447, + 0.49037255933317075, + 0.643440327892943, + 0.4740775716180724, + 0.32404524761211184, + 0.37492441848979574, + 0.27257363867602935, + 0.5159198976109534, + 0.31826804562015115, + 0.49365371346238957, + 0.4702911197345319, + 0.3134037638771025, + 0.7074019189829137, + 0.29481922720491227, + 0.7119051785684495, + 0.40420368642994353, + 0.517375471546784, + 0.7319314920392748, + 0.4525902729425162, + 0.5366211053984442, + 0.33847215261917357, + 0.6932884273493182, + 0.3804207070764958, + 0.27604707623694746, + 0.5352379943759489, + 0.49502385223825934, + 0.6205983762811235, + 0.531307854658655, + 0.3589450332852847, + 0.4629981593143788, + 0.24066873423597626, + 0.4147874986612561, + 0.5482854920648643, + 0.48212473871064215, + 0.3176706454465577, + 0.5145067255781957, + 0.5639123919542208, + 0.5528182844619419, + 0.4086701070493449, + 0.2910838748430419, + 0.48932594665639706, + 0.44409524879524825, + 0.5399291956010904, + 0.46040979391474085, + 0.2964488998752932, + 0.37891627390313215, + 0.47758316705748, + 0.3736493678924084, + 0.7166692475249062, + 0.4905053519896345, + 0.4241355804417619, + 0.4714616735070166, + 0.3602574806340216, + 0.40590446232399396, + 0.5241175169376205, + 0.5306263057460088, + 0.7005497493070743, + 0.5140482604348896, + 0.5596780756029516, + 0.5051433895568274, + 0.4478303093147103, + 0.5565840654999478, + 0.3504720027030457, + 0.6079143666537999, + 0.5825753439034465, + 0.7176671462678033, + 0.49214492452077996, + 0.5279857149639895, + 0.5256281463517648, + 0.3587969223262963, + 0.5209190262180471, + 0.2982508269004936, + 0.6617955998006929, + 0.45569548713084856, + 0.7272297195919794, + 0.7535845221757944, + 0.4283569337129611, + 0.4227230014796611, + 0.4748584437012697, + 0.3095556371041639, + 0.47000038118076665, + 0.7498386884606779, + 0.4720752864564449, + 0.4781605281000188, + 0.5723685688476301, + 0.46636752789572006, + 0.31500195072438375, + 0.5917314068979177, + 0.4043990237356661, + 0.3637534653309247, + 0.7256919486007882, + 0.38210297285776984, + 0.4065528493112682, + 0.36698637500652626, + 0.46728347494555095, + 0.5451593606870725, + 0.5961049855869341, + 0.6651209729042513, + 0.608701205507081, + 0.573496066717449, + 0.43293324160990687, + 0.591582867369872, + 0.5316018848463132, + 0.773398582369128, + 0.7283763313694284, + 0.305107833462934, + 0.3383756118880989, + 0.281548809368562, + 0.49673214361321694, + 0.3959998354448636, + 0.780339043872293, + 0.35312535735480666, + 0.3995422477334479, + 0.5402901843178377, + 0.464159140293082, + 0.539403021294214, + 0.606349211933518, + 0.4458549631330046, + 0.6095153213089365, + 0.35581798665257813, + 0.6633971287056676, + 0.7720171635632502, + 0.7564563419590279, + 0.5632003575251209, + 0.40558162583692986, + 0.5273425219986413, + 0.730715541905261, + 0.6742577577655013, + 0.6089185672130367, + 0.4603900882360753, + 0.40824302949441177, + 0.47727122000584377, + 0.7281728599584698, + 0.3927755011914771, + 0.3995189884440651, + 0.28423470049254507, + 0.5813970187759353, + 0.26076314282899704, + 0.37500116770925307, + 0.6543259561181171, + 0.527066412421553, + 0.6699140326238067, + 0.52168293298835, + 0.4949881505954633, + 0.6536486336448166, + 0.6208568161571545, + 0.5512864939481147, + 0.36091683466055635, + 0.6329110791944719, + 0.502725099515835, + 0.38242837305775707, + 0.3217517717945592, + 0.4625817702675327, + 0.6285728374209819, + 0.4803981690332817, + 0.5958352501966009, + 0.5584928528437118, + 0.40053666761221707, + 0.38207233746941377, + 0.6742301361489093, + 0.4721889393809618, + 0.5715877059721605, + 0.7150861918861349, + 0.3975600021552958, + 0.3292095207884476, + 0.3271450824557492, + 0.3359507101292841, + 0.7031577076104623, + 0.7283369091123918, + 0.5558174904082401, + 0.40259092160367993, + 0.43553222678132086, + 0.36415240779585006, + 0.7732862254389108, + 0.32187739452250164, + 0.46424990071451294, + 0.3116896282783685, + 0.4806492159147384, + 0.34214301785862783, + 0.5471653805571175, + 0.4558254814526719, + 0.26981843958899676, + 0.682971430333451, + 0.4944662410694567, + 0.3963960010606684, + 0.3205527074171243, + 0.34002875411347816, + 0.3871303395055791, + 0.4008968577119002, + 0.35206784516543765, + 0.3478192418042774, + 0.3875670741056071, + 0.35915239529877535, + 0.39129461286975026, + 0.3128323215050493, + 0.4226319850991648, + 0.20141037423831415, + 0.23959019734059256, + 0.292165712519044, + 0.36631335180808905, + 0.41312824397690445, + 0.39219107731941383, + 0.29623605868399033, + 0.29010909108052385, + 0.2868308629320737, + 0.6495022471315358, + 0.2859044717053591, + 0.2695498125494071, + 0.7726001036896841, + 0.33721327435467946, + 0.37917526906221, + 0.31457476846135174, + 0.4298809348338168, + 0.2555106773559552, + 0.36866925249937765, + 0.33037685827332397, + 0.45595690606894285, + 0.22970834928373648, + 0.3921256069009932, + 0.3776194423357154, + 0.4841975332289066, + 0.3451762279297219, + 0.4573145440566568, + 0.2799377026064543, + 0.42318583259887743, + 0.38342126563026907, + 0.3230389468003563, + 0.45029870920613824, + 0.32813015267926365, + 0.45963080646384286, + 0.3481552554634189, + 0.3297954303876961, + 0.8029188459797274, + 0.26738248517035745, + 0.3949116661728859, + 0.33926495993975175, + 0.3747717622867749, + 0.27785832094478247, + 0.3431099446322308, + 0.4471976829064112, + 0.5347505217329174, + 0.33671512817784804, + 0.3630333224873774, + 0.24143142597464778, + 0.3913761865988925, + 0.3855787396110748, + 0.3946425780207901, + 0.4246821223472103, + 0.34285659409743724, + 0.3068517727004739, + 0.4539587553405824, + 0.36874672242584516, + 0.43724772556828573, + 0.25339540628745166, + 0.30846121213137667, + 0.46070356603717555, + 0.44696120921219823, + 0.37718630082264387, + 0.3304747965311375, + 0.3282927351254337, + 0.8189569149463777, + 0.3035710731672455, + 0.29686111157286066, + 0.32438641777664173, + 0.2656301250676127, + 0.3059537543629182, + 0.30194931986019535, + 0.2524053014645019, + 0.3450163213510799, + 0.34366133096521834, + 0.3465966386537107, + 0.30461677113128516, + 0.4041125478417331, + 0.46021128620880847, + 0.29358269722289126, + 0.3457431055019673, + 0.5355161944881405, + 0.2816107425929074, + 0.3327876093245865, + 0.47961469476574653, + 0.42626129119776157, + 0.2841166534834851, + 0.23403999303209758, + 0.28787837488658796, + 0.411789361664191, + 0.4234497935669201, + 0.2421318602896918, + 0.3214994667513306, + 0.308381723971609, + 0.4051799628520866, + 0.2886228382427895, + 0.2491868086660259, + 0.457287073963448, + 0.4885657379592838, + 0.22557922021006332, + 0.4073235514645357, + 0.23664802712257108, + 0.2882388783121065, + 0.2812069187936926, + 0.7718429016992329, + 0.40645033146157833, + 0.4181242634872189, + 0.23161866974383116, + 0.3354326826069311, + 0.3130001759357716, + 0.6799880619908483, + 0.36831049970517976, + 0.3097601395625372, + 0.43006873484596386, + 0.38712631427755784, + 0.3640766759436659, + 0.42323409367034975, + 0.46098649516883744, + 0.2288499597155206, + 0.2718938300362406, + 0.41337190995156115, + 0.3586855340120526, + 0.3743749882924052, + 0.3655873990851481, + 0.29399988177279285, + 0.24926854461217374, + 0.24361604595872066, + 0.3719312295586371, + 0.3278840376017483, + 0.5784578351775668, + 0.2753489161636438, + 0.2498718560015388, + 0.24571604400436406, + 0.502738905744717, + 0.31381843459143494, + 0.2168880871107414, + 0.26467674955498327, + 0.3393668747423945, + 0.35103808078739035, + 0.25075003528692663, + 0.5207429996639998, + 0.40223667926598716, + 0.24349391670693146, + 0.3559043147675459, + 0.40349432109953587, + 0.33333031335274316, + 0.3144171710903673, + 0.4510581336357925, + 0.2820416327698189, + 0.6343227275840552, + 0.35186976640207596, + 0.4404668701569374, + 0.38429857009522905, + 0.3446875724127084, + 0.3750787038122295, + 0.3004215358384029, + 0.2262238033621276, + 0.3666145031083938, + 0.5734878560070652, + 0.30836255342386215, + 0.23160562026109546, + 0.47355245457136597, + 0.3926111634550126, + 0.3939013290310168, + 0.5862073160221913, + 0.319636282388369, + 0.37847686662585217, + 0.2783917007453841, + 0.3407842053870781, + 0.4153089077121978, + 0.28690987162082465, + 0.4313272917092951, + 0.37619992721798184, + 0.3852802945688359, + 0.19897671222488206, + 0.3598416206414551, + 0.27187051882751745, + 0.2553408719906253, + 0.2556259522511322, + 0.23953292113194605, + 0.395811608449587, + 0.4984838540378894, + 0.41383968855029124, + 0.44127633951886175, + 0.4097996642992794, + 0.24373546613202246, + 0.3047829382427412, + 0.44421326960854696, + 0.3288226302830816, + 0.35570817494825946, + 0.43621072357590457, + 0.34521181391526945, + 0.333808967298225, + 0.48215608192071224, + 0.3678030284049559, + 0.2881250772253333, + 0.48195781738652393, + 0.2968215296259699, + 0.40492234102139435, + 0.2610833904926279, + 0.46022373097706715, + 0.5814338143541242, + 0.7813888301717231, + 0.39034742464743105, + 0.3435782048028845, + 0.3946406089429068, + 0.33951762665358465, + 0.3226000267364567, + 0.3216230966476747, + 0.3753628156706549, + 0.3123825263203482, + 0.2671493742771849, + 0.4187633456097355, + 0.33946908736086806, + 0.3189883753468142, + 0.2495274833147897, + 0.23261348093346432, + 0.3699041626537223, + 0.3790722375330636, + 0.34208990980053855, + 0.34674362094091987, + 0.36155071592840476, + 0.27202859541972974, + 0.1990597707317539, + 0.3271772605498059, + 0.4113851707159744, + 0.26492724025699194, + 0.36397117106152266, + 0.2692760820431513, + 0.28859516013411507, + 0.18729638795049025, + 0.4928519355211825, + 0.2730325730204621, + 0.2556740268732654, + 0.3588987040309015, + 0.3116694466835311, + 0.27405914988789204, + 0.5754763210464484, + 0.429234937548667, + 0.1990413153345209, + 0.40810143154196565, + 0.36849413328560815, + 0.3571563294356886, + 0.2888582174067073, + 0.22690585429404825, + 0.3078490573222458, + 0.469034505387819, + 0.21497734347457945, + 0.37384595518926556, + 0.19746872156716253, + 0.2600681773036991, + 0.26842092186789146, + 0.23463547370956309, + 0.356821983310706, + 0.373694569856951, + 0.30299934310384874, + 0.3253056059632127, + 0.35405520141546515, + 0.2317237382240234, + 0.3473688739509007, + 0.34376437456702547, + 0.4788127143172417, + 0.4821366804976421, + 0.3720473945495414, + 0.37671629890023306, + 0.3536622970076177, + 0.39354068898837935, + 0.424884058847438, + 0.5112883142750668, + 0.3741812006759494, + 0.2534466147727893, + 0.4854911669278999, + 0.3137874678364162, + 0.42029071745163293, + 0.4156396821841681, + 0.30328838506580685, + 0.4610993618374371, + 0.21410694084943802, + 0.25373343143054883, + 0.23224524415405193, + 0.20781437355879773, + 0.3267708545546746, + 0.33002672370813874, + 0.2670978278287346, + 0.3724427114559238, + 0.5148571296291536, + 0.354204373770489, + 0.48269456545220374, + 0.7685658308085236, + 0.5249636752322804, + 0.22113037675366254, + 0.39533809948392096, + 0.41496959636383207, + 0.37723646405817884, + 0.2246523053730904, + 0.5582774379598996, + 0.4300657899553627, + 0.3239007795427919, + 0.42637781687191545, + 0.321729848201482, + 0.4286150510419721, + 0.48655452012490463, + 0.35605871695393815, + 0.5045309242568783, + 0.4264686806691702, + 0.32041315552025207, + 0.43629748101754806, + 0.4748489697443866, + 0.25165497429789246, + 0.27602672280619056, + 0.21778017056253124, + 0.5606468827794582, + 0.48290324616588437, + 0.8097945174154436, + 0.5001649967123212, + 0.3684303407584837, + 0.26141967228232904, + 0.44653610361348695, + 0.422758746022606, + 0.5872179479503996, + 0.3046439110082363, + 0.5068611202512368, + 0.2706748718988273, + 0.30893650556347363, + 0.22594975573178327, + 0.2213145173454945, + 0.374772257099206, + 0.1902831515349753, + 0.37196947522709867, + 0.43025391403217417, + 0.342077774255181, + 0.22909167762357313, + 0.23821260852436074, + 0.377698168078771, + 0.2270092920608979, + 0.4808211936758845, + 0.39747194066700736, + 0.42398646829534714, + 0.3813797106157956, + 0.2900738157162903, + 0.5353202051603803, + 0.4419651179444064, + 0.30958403870013873, + 0.5723508553403257, + 0.26456851374617074, + 0.39056995174601394, + 0.29837313736606685, + 0.24045268003319106, + 0.24959760317444823, + 0.30877386836704174, + 0.41822434995427576, + 0.2776313700035387, + 0.6719236157696258, + 0.41932853445540447, + 0.24089739542760993, + 0.30453319518249616, + 0.24579168380782998, + 0.5263420765518406, + 0.2338367108350665, + 0.3084992582704413, + 0.32579491362156776, + 0.35778623478757676, + 0.23969920672113987, + 0.36991815268418243, + 0.25314690060754436, + 0.38941912553380686, + 0.7572404905483486, + 0.38939642441303535, + 0.3160907432768188, + 0.3326917398128089, + 0.3197437878795187, + 0.20846686171040724, + 0.31104122382606103, + 0.39487354346990017, + 0.7391007068262702, + 0.5224866466999399, + 0.40445295870399356, + 0.32609759249240255, + 0.31325138872967767, + 0.3552341596692916, + 0.2286673439890984, + 0.3927014067335508, + 0.4810104816488346, + 0.2774047970548084, + 0.36547517618554437, + 0.3445417095956399, + 0.4292086012645961, + 0.3209693473770878, + 0.49918242237487437, + 0.3401988763421349, + 0.27034147315860807, + 0.393258688198092, + 0.39954573511985203, + 0.5157428458584482, + 0.5101304678760372, + 0.22466936614305752, + 0.7070127943447356, + 0.7035270559039172, + 0.19435972397242315, + 0.2614958292029158, + 0.33748331468992815, + 0.4214955013929342, + 0.23105412997593622, + 0.338153543976503, + 0.441352616315119, + 0.4564534859877257, + 0.42846000553859614, + 0.3393907011964461, + 0.3239223957174247, + 0.322654136558091, + 0.298946252081339, + 0.41291211051038945, + 0.35401740655445757, + 0.3551074503869573, + 0.4762882949705628, + 0.37881083428417633, + 0.5079064497306112, + 0.423487762005547, + 0.5917682233357779, + 0.42358364747702054, + 0.39067489593950766, + 0.47099361089956543, + 0.4303837247947923, + 0.3122275517738692, + 0.606457692268784, + 0.5836464717267426, + 0.41649769906691997, + 0.3491855250547051, + 0.348882905599431, + 0.5070933538362853, + 0.3119326623644438, + 0.39343949559747043, + 0.3750399220517633, + 0.45401721481749757, + 0.41317832002988825, + 0.36468190574380693, + 0.33454467514416714, + 0.5075234977249796, + 0.4676537752430781, + 0.4424887956977896, + 0.4592524572451527, + 0.35582211742680436, + 0.22596048737013644, + 0.20485025070587357, + 0.2814158487505106, + 0.49796511180860564, + 0.43869992915693, + 0.22072170577319844, + 0.4185673346757312, + 0.3270733306391141, + 0.3178690009892735, + 0.31946260357048034, + 0.3577024199435954, + 0.3141874782538718, + 0.42288972669509983, + 0.6091150379838661, + 0.29254032763117954, + 0.4034471224798413, + 0.5115549533394023, + 0.3206404432042526, + 0.40293337564527987, + 0.5717928496585578, + 0.43670017848484066, + 0.430694952921047, + 0.39316227707216544, + 0.3730189647623583, + 0.22923716330523483, + 0.3477156236257467, + 0.4684528219609335, + 0.47417987145780355, + 0.4545059430115727, + 0.2899182138785116, + 0.4401218142147162, + 0.4304949615672866, + 0.4857744617748019, + 0.3892053475811074, + 0.4534293730237891, + 0.3740642892270512, + 0.4415914528926015, + 0.42962343302828954, + 0.39541885977003904, + 0.2650742371509101, + 0.43366102904304754, + 0.4488378122263776, + 0.2342439592501863, + 0.3378232026350678, + 0.42567449748756725, + 0.37448504385216824, + 0.3422579004503282, + 0.47135231197121574, + 0.4774049992562648, + 0.4155016223118502, + 0.4362804288951198, + 0.31215223426001254, + 0.43355572845610524, + 0.3537284532061516, + 0.34909700723444703, + 0.37311274517736825, + 0.4286205170998315, + 0.3670577855851993, + 0.39655697776852, + 0.3779419433240051, + 0.7112671217840213, + 0.5320433210319145, + 0.48968094716669985, + 0.5238634245645808, + 0.45614128015201105, + 0.4021542543999722, + 0.5271986411705364, + 0.33988078727764176, + 0.3715836545074331, + 0.45931196724188583, + 0.3087171425445684, + 0.4253888217710114, + 0.4861767599728791, + 0.4328777129281322, + 0.4384553657523713, + 0.44215110797497914, + 0.3096271562664668, + 0.5081779409046159, + 0.4819364355867185, + 0.3546069569904602, + 0.40826633430575304, + 0.36056033615831024, + 0.32669846255935925, + 0.42620033619809533, + 0.4202772198850825, + 0.39965767522895335, + 0.43777706663092625, + 0.36822027134802876, + 0.46405812493261084, + 0.34849292629658934, + 0.37495399745794433, + 0.42280676291694963, + 0.417301244083669, + 0.3951091473138294, + 0.4602522536513039, + 0.3861597967104618, + 0.3314913497849932, + 0.3400854437177904, + 0.412702518357977, + 0.3936908221797518, + 0.42957096845935516, + 0.45334227627101253, + 0.40985480215167525, + 0.40223796916657323, + 0.3401412632266895, + 0.38909046997970237, + 0.5102079671946242, + 0.3701301734911682, + 0.19595817409065544, + 0.3332546446247852, + 0.40100472311741825, + 0.29409809638983675, + 0.2894053640493705, + 0.4354248620609054, + 0.4370108543935121, + 0.25460137115147713, + 0.3027370307214868, + 0.3341062040714634, + 0.3192481067943757, + 0.4936786227564436, + 0.27019524601124867, + 0.2704905679165199, + 0.2901909095563497, + 0.29569077846131125, + 0.21314060528648804, + 0.27892057291013944, + 0.2521235775620645, + 0.28112574521775535, + 0.24204699196370402, + 0.3810071541615614, + 0.26815436012015126, + 0.3105460200003365, + 0.31943954194204477, + 0.3715109970092844, + 0.2749799469748474, + 0.37251944504875395, + 0.3218860144989845, + 0.3225446572581127, + 0.3252282653869552, + 0.3769947503961239, + 0.3481454881671495, + 0.5640732520704893, + 0.2914342065914582, + 0.36033711694591875, + 0.5107644322863882, + 0.4616377828563106, + 0.35889115488804096, + 0.44063939534051083, + 0.3973178118950621, + 0.4351068985188826, + 0.45726581365867436, + 0.34308846008216737, + 0.5654010746300663, + 0.4107940046713231, + 0.4055240115863975, + 0.34266022340581015, + 0.41065885660123924, + 0.3606028048555389, + 0.4368897164215397, + 0.4394754203788012, + 0.4311200632248685, + 0.4505351206989379, + 0.5648157196689453, + 0.42288171768922217, + 0.38275338486395805, + 0.45072610839771293, + 0.516214654821409, + 0.5053749135555925, + 0.42984358649253324, + 0.5317388661606447, + 0.5271503167172965, + 0.370131038372368, + 0.3651222394634551, + 0.41524839443802347, + 0.3317452143505963, + 0.6005277255647536, + 0.4749466411670031, + 0.40145429295721424, + 0.33953960710912795, + 0.4507691656002364, + 0.38579711139077333, + 0.4974332790731003, + 0.4246806754321921, + 0.38044011543611006, + 0.4900754387400531, + 0.27350249774215163, + 0.39269446485335296, + 0.5049307790074614, + 0.3129686341973065, + 0.393666601248793, + 0.2969796337124988, + 0.3331404221157448, + 0.35748129065385564, + 0.365998613418223, + 0.4327431363892046, + 0.31332308546303, + 0.30534296379275394, + 0.39864925746553265, + 0.3672985715270509, + 0.35547854653789923, + 0.423538607099151, + 0.32206983441417364, + 0.3999654442679311, + 0.4853697624637504, + 0.5464814895083784, + 0.4360897357512527, + 0.3346332504188715, + 0.44717101889221156, + 0.4210186279129018, + 0.36733641753517976, + 0.3959981181246265, + 0.2964766996509991, + 0.3040709858738092, + 0.2712653410097642, + 0.35005182988363015, + 0.257748439715431, + 0.6378551597100528, + 0.4100477848747324, + 0.3867466445293942, + 0.2885379185692067, + 0.28799305200629316, + 0.31045485514388804, + 0.30948272599637977, + 0.27905936723999164, + 0.28671256650049, + 0.2794331848389445, + 0.3517546793408131, + 0.3539629798733218, + 0.4151428984523791, + 0.3834777072444494, + 0.34419420692182445, + 0.48312724868341544, + 0.4776085688263738, + 0.3523040483850777, + 0.3943227466048838, + 0.40223116277968113, + 0.3681626773210971, + 0.3340263569354288, + 0.4971773538073875, + 0.4290129776353312, + 0.5628253733958429, + 0.4276652918114527, + 0.421269099967014, + 0.4287611214487051, + 0.4351527901998494, + 0.31344302473871766, + 0.55567864095771, + 0.3979461213820411, + 0.22142355800746621, + 0.4364750906604859, + 0.5417578372751775, + 0.3477010701063533, + 0.41403212588237237, + 0.72377406136509, + 0.4170196581515662, + 0.4817526609082211, + 0.26217755649601554, + 0.4513008460127709, + 0.24691862729274663, + 0.2649588935034181, + 0.28569912170921186, + 0.23907084083496577, + 0.42928746307278454, + 0.169842038532672, + 0.2802953299200943, + 0.42489665231123946, + 0.3761563239928194, + 0.3824072021601471, + 0.3797211328043396, + 0.301156669480521, + 0.3564926565407442, + 0.4983871221137972, + 0.5117614422091415, + 0.3282782125411451, + 0.4231277772958529, + 0.3778189774396583, + 0.7096130960144434, + 0.35536310774455665, + 0.5430849773259866, + 0.7263616958280243, + 0.7022505799118436, + 0.414920917209865, + 0.34212979002784627, + 0.27359768796294603, + 0.42114639394974407, + 0.46389194137936485, + 0.4757772867748438, + 0.5441887194680407, + 0.4023489269947461, + 0.5354316860649454, + 0.38986554193528944, + 0.3571815119328786, + 0.6818943826639775, + 0.6679907722336409, + 0.42948494439862417, + 0.39003966662250583, + 0.4384787380714523, + 0.3893392024618925, + 0.42371920071468794, + 0.6295884726978657, + 0.6458817598108175, + 0.3951014210775819, + 0.4103396776165762, + 0.41187380252005246, + 0.3341878435786344, + 0.47119341184011043, + 0.3022552364131233, + 0.4516311207219486, + 0.45093717105086273, + 0.3285089516236968, + 0.40978839055555555, + 0.3548844911554305, + 0.29886883075821685, + 0.4801757933712204, + 0.4051571733017842, + 0.36764438238029823, + 0.3445208220601984, + 0.4505761370412515, + 0.3861060164880522, + 0.3062811634564605, + 0.3234006379082532, + 0.22270096234260467, + 0.21240301890944352, + 0.2414750364683986, + 0.39371005377240975, + 0.42338499624497494, + 0.28888707363647825, + 0.3077132980821317, + 0.2877948684425393, + 0.3876792286332616, + 0.3863369807449004, + 0.4641740089138798, + 0.30552804767772634, + 0.3725634120671442, + 0.4739902305900791, + 0.3195552261114413, + 0.4861691549154722, + 0.43121464078585653, + 0.3789141811012661, + 0.37787097093504357, + 0.3208169532001218, + 0.20370639561604403, + 0.2589914510208359, + 0.2537650380465818, + 0.48430965287503297, + 0.3811446416295319, + 0.39568178672897303, + 0.4257573594979189, + 0.4504011438373177, + 0.2923761254217394, + 0.366749620612078, + 0.33191892362799397, + 0.6347612104863638, + 0.4095564816192332, + 0.43864523838687525, + 0.3383128737792256, + 0.424433914239153, + 0.45592681432983445, + 0.4012622801276501, + 0.3733253021296988, + 0.44970383281996656, + 0.4160294841378348, + 0.5162737384457596, + 0.38325996343087027, + 0.447771423699255, + 0.41058048749952514, + 0.3915194136707349, + 0.36416994849773215, + 0.4381510280458368, + 0.3835379964615324, + 0.45168063519708657, + 0.3100055917504381, + 0.29001117488239403, + 0.2905655247446782, + 0.36628026781680056, + 0.41389685912563984, + 0.3253255617410921, + 0.37495989225152265, + 0.42411948218924606, + 0.38984440495239986, + 0.4177876908663289, + 0.3991376701314253, + 0.25410558758337587, + 0.3787589196225708, + 0.21506381730322957, + 0.38511322769360407, + 0.19763394855486033, + 0.633297225827948, + 0.2992725854585016, + 0.4048602244037448, + 0.5126416614323605, + 0.43154841136807814, + 0.2413935914899223, + 0.374150598501811, + 0.30231345779564045, + 0.7585390157981916, + 0.25760342434952577, + 0.4693400519769007, + 0.4742209681179945, + 0.47321430538080783, + 0.6181714046006068, + 0.3763126781554175, + 0.5085580048798098, + 0.42354435505735416, + 0.33565032158613206, + 0.2772349942679527, + 0.4219604263881217, + 0.2769165855276183, + 0.697097902907423 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.3872275645210765, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.3976655214885236, + 0.7411442705416643, + 0.7272729957920885, + 0.8087860099416412, + 0.7072111810321825, + 0.6280413203610381, + 0.5814127125443819, + 0.641165207698184, + 0.3878456202453565, + 0.6764966519937279, + 0.28693665909120797, + 0.4434366827882158, + 0.29112889061407526, + 0.291599393670733, + 0.27103903063338747, + 0.2572724514235754, + 0.3044223198684044, + 0.4992316725088859, + 0.3466621679820383, + 0.3563475902008752, + 0.30371461219351853, + 0.6536498821437469, + 0.2670779784293059, + 0.5731386598848154, + 0.3422669811984753, + 0.1541298840114837, + 0.35807072354621433, + 0.4011899639043606, + 0.42359892630789847, + 0.45587890112114954, + 0.25450664989860033, + 0.36545786670519176, + 0.3040007114220027, + 0.3168649881716107, + 0.39283424990014537, + 0.3219991496155363, + 0.275881725210877, + 0.4771653944762118, + 0.4917055979377558, + 0.6182087253541343, + 0.7766961885370192, + 0.7710680149811363, + 0.7589612381095315, + 0.63368016679957, + 0.3593531893607536, + 0.31566259657300005, + 0.7661621946511161, + 0.2837138351357955, + 0.29919160747420653, + 0.421978575293799, + 0.31571490536346736, + 0.7671869052020681, + 0.7857549613406245, + 0.36274609768146576, + 0.28864082945810526, + 0.31660101147049546, + 0.30603173721081084, + 0.3531828715940201, + 0.34279146191834287, + 0.4652349906381414, + 0.40344533632693186, + 0.4006937150514917, + 0.3568406003100784, + 0.4618889771144276, + 0.3930069432329704, + 0.3311072527016958, + 0.31686030383181696, + 0.4758100487892978, + 0.32607557887932304, + 0.40225923760804105, + 0.44061977501341854, + 0.38536614822151966, + 0.3614359475740723, + 0.34891826104439266, + 0.3738964653194126, + 0.35587163487978374, + 0.252353782785226, + 0.24611536609930168, + 0.34789970691305067, + 0.2982051070522413, + 0.40373698570093053, + 0.2712756724084297, + 0.30811640157549036, + 0.2552208402256212, + 0.37008032136251273, + 0.4545479942744024, + 0.3715356540808788, + 0.25125345966096047, + 0.3192124761829807, + 0.3033425630690697, + 0.28680977914630756, + 0.30580984873860934, + 0.4476298200604215, + 0.46968489156321763, + 0.22773317370195076, + 0.5128343052963914, + 0.3074958112625487, + 0.47856587564770064, + 0.2566856348410677, + 0.40977973736037915, + 0.39134593305461784, + 0.4690220649925519, + 0.1992671792735049, + 0.35538521876416196, + 0.2616859200315021, + 0.2879729493201891, + 0.4136443425481654, + 0.2839412939553844, + 0.2935345642159053, + 0.8004670835073953, + 0.7953087626068156, + 0.7997780888437324, + 0.8155139865520605, + 0.5704528009573195, + 0.674540280771383, + 0.6961725524426507, + 0.7467481352664849, + 0.6864891381541163, + 0.5759759784023142, + 0.6457320030749601, + 0.5654694263638761, + 0.7635507299719014, + 0.6467157235804484, + 0.4002450101326072, + 0.30377198283943724, + 0.24561193182879704, + 0.2936316961774756, + 0.49971917046710984, + 0.27066290252580677, + 0.2539611075849027, + 0.3261590614490797, + 0.36206643477867556, + 0.2530046361202191, + 0.3350889716195974, + 0.36613194066312477, + 0.3888054815604352, + 0.3430655795388302, + 0.283409038362669, + 0.32050883386856155, + 0.35234748709858843, + 0.27928666575176736, + 0.28746866644928054, + 0.23737768380013138, + 0.5778281420541724, + 0.28038794959267477, + 0.5172433301171312, + 0.3389867654265239, + 0.405734613680674, + 0.219733099259669, + 0.5398151834644587, + 0.2685318235367539, + 0.4613658345069045, + 0.29053991664209083, + 0.21800252641460488, + 0.31328224050675263, + 0.2843575863317076, + 0.23788076773803593, + 0.1518734623247402, + 0.27934671866849464, + 0.2932062791635911, + 0.44163450347226124, + 0.40643502532289916, + 0.4610517782077706, + 0.4418196600348644, + 0.3038895122690497, + 0.41304648105427516, + 0.35477978311141467, + 0.38702411554899246, + 0.37890291862089204, + 0.3743121497288429, + 0.23643407562791757, + 0.14438785752155245, + 0.3544743926154429, + 0.270395677451901, + 0.21977199261217253, + 0.2237568788008318, + 0.3830639902837026, + 0.4507408909371943, + 0.5446704833877372, + 0.4105891560996219, + 0.21200893731152584, + 0.4440292090402704, + 0.4352947550272819, + 0.4651803458998379, + 0.2616801515330067, + 0.4807273445694452, + 0.38493825427641604, + 0.46647223230810847, + 0.21818227379943292, + 0.34566385187726356, + 0.3110399590517214, + 0.23039881837559487, + 0.46235788266902444, + 0.3892907450415829, + 0.49914430441498353, + 0.276824166328491, + 0.5837703046460303, + 0.27145743993042776, + 0.1870949601452594, + 0.3672987675797648, + 0.254283333688787, + 0.2470247663456476, + 0.18870226867795148, + 0.3093728834160841, + 0.5906028951050483, + 0.2154457912306113, + 0.21948323076396933, + 0.3959042994927673, + 0.31514235407775043, + 0.3560498768816064, + 0.419059823192951, + 0.30334546080037367, + 0.23392368995788973, + 0.31793785297368526, + 0.29671646976993715, + 0.48557244527373344, + 0.48519501564328504, + 0.5299856990582886, + 0.5074007393426755, + 0.24743438753866587, + 0.3053300165001544, + 0.40101992416038373, + 0.2364329309649549, + 0.30162735949858277, + 0.2642602772362084, + 0.13948670959879708, + 0.41945157238386904, + 0.29203073798599705, + 0.2255330899296006, + 0.4945367927322885, + 0.2185210116736714, + 0.17954983307092087, + 0.3646480809259747, + 0.356137553853697, + 0.2572232126323544, + 0.5668041089837383, + 0.29240700838116235, + 0.2745750850471061, + 0.22020573390212456, + 0.39539528477722236, + 0.26432925523753537, + 0.42811403803801523, + 0.24811165266510182, + 0.3678861063998304, + 0.34030991214861633, + 0.34090003883000863, + 0.3004889976056589, + 0.2963383393928709, + 0.3060193579747412, + 0.3042545339628458, + 0.42015239053647274, + 0.25472047586255053, + 0.34948558816766223, + 0.2682772027901861, + 0.3040976739222602, + 0.34988743968787545, + 0.3530465931571666, + 0.3208924350901477, + 0.39580866837360235, + 0.25846979138190457, + 0.27895273688694566, + 0.4719178059759977, + 0.23651240998843187, + 0.30109096571777044, + 0.27995687185272794, + 0.21591706814191228, + 0.2795487437382491, + 0.2496988739731168, + 0.24265317510749598, + 0.26117864935355856, + 0.23570968032688072, + 0.2883612894469733, + 0.2881544901609722, + 0.32934373392829414, + 0.3403586080377183, + 0.42760889692122916, + 0.20776086076932435, + 0.37137369205618603, + 0.44041850917059716, + 0.2340760983402914, + 0.23605454499063988, + 0.5376524372856069, + 0.21296292720713295, + 0.2176138864044498, + 0.3836230580651672, + 0.37108903852823566, + 0.4039783109556305, + 0.3293410465679762, + 0.45905535904373507, + 0.3936351004865655, + 0.38873858469769035, + 0.32046403948261964, + 0.3942549363148766, + 0.3727350642703805, + 0.35037358467234775, + 0.3623013036526231, + 0.4415480321872908, + 0.3897130045058868, + 0.643937638542214, + 0.26128174192378195, + 0.537224917427009, + 0.6219640700103103, + 0.6276316779150438, + 0.624524870087337, + 0.44183640287517795, + 0.43098237951874346, + 0.3651049620697797, + 0.6532252614477679, + 0.3948457744928705, + 0.44808067127583673, + 0.28514649958397925, + 0.4237152537090523, + 0.3481784515222254, + 0.33589900317651433, + 0.37588106606912386, + 0.27926733859618497, + 0.3997352029557198, + 0.2502610494967961, + 0.2895344507314027, + 0.31245713239294265, + 0.3175897174420951, + 0.3623899192347481, + 0.23462410717148918, + 0.30158423049527383, + 0.32147487254188223, + 0.3295050785113226, + 0.28102357358670793, + 0.32765733596075647, + 0.5093562576110389, + 0.3982911662591348, + 0.34128636503973425, + 0.4105665263909895, + 0.4041195496818498, + 0.25727475454756044, + 0.2856992520033681, + 0.33503346558473746, + 0.3639573215395291, + 0.2840591593768956, + 0.39081649743160946, + 0.4919710794007079, + 0.4852320570675413, + 0.3425358345494503, + 0.2711170858050483, + 0.426151461948495, + 0.2390619185208272, + 0.2888332386178553, + 0.41348334998500885, + 0.28752436076320786, + 0.8138906015519913, + 0.465853635598489, + 0.367177957992898, + 0.34242652411177454, + 0.40660649544092725, + 0.41304710596643734, + 0.34771195334558924, + 0.45178487708568177, + 0.30327154036664716, + 0.44153623422375654, + 0.34245971614744103, + 0.42348629797896553, + 0.4122054902655725, + 0.39597501142582536, + 0.37392707379819706, + 0.2963974933927293, + 0.30558171977482823, + 0.4598400213736201, + 0.4823700373073104, + 0.36184319430632245, + 0.5414407720985219, + 0.3832027554490256, + 0.5068951770494091, + 0.34223405766161413, + 0.37589314604475066, + 0.3911179102564107, + 0.5218225520612052, + 0.5174453205681832, + 0.460425229604704, + 0.48624901542099125, + 0.4598698411246467, + 0.46944582107368266, + 0.3525157785552433, + 0.3411722982722852, + 0.3072790219991362, + 0.23734732087031973, + 0.3360507420503659, + 0.30238076840223854, + 0.2816179001232884, + 0.3053138596949209, + 0.40945277340246156, + 0.30421894132028743, + 0.26653266520331576, + 0.2955879037559338, + 0.49850993418488637, + 0.5409651365709045, + 0.4724725821639463, + 0.5102506907662805, + 0.37541494643458845, + 0.52173924494478, + 0.3148436568433386, + 0.5850851876732454, + 0.39622272525239144, + 0.5056687372164724, + 0.39865204594666087, + 0.4975783998741929, + 0.49713324253114294, + 0.4478272493459151, + 0.41021166382660407, + 0.45037521911313105, + 0.45926475099896813, + 0.4808400494700398, + 0.26472143632989475, + 0.3232746823443498, + 0.2730620416751186, + 0.31237423659779895, + 0.46048254258556076, + 0.5149487045398415, + 0.2665169434081925, + 0.5328755670770721, + 0.4399966492864012, + 0.415554263651756, + 0.3347374848275217, + 0.4112683325398904, + 0.41779225580182544, + 0.3174984001350095, + 0.32526360113706765, + 0.840815821866426, + 0.3865464412088428, + 0.40083727454400947, + 0.4377723503741567, + 0.44105083262653316, + 0.41163033348415284, + 0.3634547996182655, + 0.5488755063043432, + 0.4103251185572136, + 0.3312042217662734, + 0.38608669784897204, + 0.3861355434637084, + 0.39452543446927774, + 0.37339143281516485, + 0.3828701203596828, + 0.4214128700602883, + 0.4128933873271614, + 0.25314178513567726, + 0.40620279670938997, + 0.25682924483375613, + 0.3516565087622497, + 0.1943262598055026, + 0.4039701273757627, + 0.3241533528552358, + 0.24441576844174884, + 0.4312715904036934, + 0.17673445341789237, + 0.23424884027313986, + 0.520664223518711, + 0.45103923674601754, + 0.4177328083316987, + 0.41905353600556605, + 0.36937127364508615, + 0.37942556897316876, + 0.5570311672372, + 0.31017546789769734, + 0.5287351692926616, + 0.3153907381722145, + 0.23465086762331935, + 0.1863999071679973, + 0.353626563771031, + 0.438870868191821, + 0.48657759035317527, + 0.5173505825348663, + 0.4029362588282348, + 0.4563637631516445, + 0.32121870383416545, + 0.43564638528390437, + 0.34749128373640253, + 0.26587647152061555, + 0.31727904224340026, + 0.3290980533717621, + 0.39548810580047694, + 0.31092216061184785, + 0.34458249346547193, + 0.36699058018500175, + 0.3715796534111719, + 0.4297202631506619, + 0.35680298519382087, + 0.3193678250310922, + 0.44946731145312435, + 0.4093998238064824, + 0.4932375580389926, + 0.4491800321111392, + 0.522621446000417, + 0.5358008803776511, + 0.6476361224171879, + 0.6330407222552422, + 0.4590994297456198, + 0.5048582853153865, + 0.3904684014652959, + 0.5014331574422812, + 0.37091236306487285, + 0.42000586224335856, + 0.3535857593775519, + 0.5569581851744181, + 0.36081319866217776, + 0.5133080015055412, + 0.6303900798389277, + 0.6053149897091814, + 0.5402561287331354, + 0.6223790233796718, + 0.5159989627962264, + 0.43648417079470475, + 0.3775715931628284, + 0.3563375435064413, + 0.24587648368929005, + 0.32197387496099616, + 0.4109760383830077, + 0.49141444295982506, + 0.36013339275217826, + 0.28312086567406786, + 0.19794426411160645, + 0.4073545024298286, + 0.40004863553346304, + 0.3766808475441805, + 0.264968813254875, + 0.4810119097554713, + 0.35875874690109266, + 0.40360703596172187, + 0.31014193395527007, + 0.4108418030880756, + 0.4368821470807389, + 0.2911510140198979, + 0.3373958219420715, + 0.39281686354583495, + 0.32689768429352933, + 0.32468061125105446, + 0.3251873209507188, + 0.3268128744632887, + 0.2896662025884128, + 0.34803898136368405, + 0.35709437572650016, + 0.361567372682443, + 0.2288230731147629, + 0.5263555011733302, + 0.35758762290896856, + 0.45572980666044133, + 0.47194082668569526, + 0.3225517607764677, + 0.4249291488180534, + 0.4326022639458515, + 0.43449247877430824, + 0.5367572976633745, + 0.36234172022176747, + 0.3930730735218179, + 0.43181517401560293, + 0.31470797295691844, + 0.4209996783020095, + 0.3804488735250381, + 0.28574864702628244, + 0.4011959522753209, + 0.41971190797918717, + 0.37630952725865136, + 0.25073046394491233, + 0.362627601155292, + 0.22251675212446945, + 0.33564572193945946, + 0.5797158698477123, + 0.4950236782274824, + 0.5099027421409214, + 0.5113446138822334, + 0.5899416431005263, + 0.49875887767469707, + 0.5456735602994673, + 0.5074718452809246, + 0.5519196748618501, + 0.3709662807527605, + 0.35287396306464847, + 0.3678696006976716, + 0.28583402025969, + 0.24480530550287813, + 0.32175401469988524, + 0.31231281236879044, + 0.2773085318987763, + 0.23648908877159502, + 0.34895062913173075, + 0.2517818507310006, + 0.4088418897006546, + 0.28618424691637295, + 0.4930400230712506, + 0.5296735968695113, + 0.4925204889315717, + 0.2527324854906104, + 0.33264139996486286, + 0.549929909674595, + 0.2650841711908344, + 0.2165253288350256, + 0.22971410188922148, + 0.8180417405518918, + 0.48560140835992927, + 0.3854683811972556, + 0.5207070089595228, + 0.47752019143602087, + 0.4555622380196013, + 0.37902008527668135, + 0.4322552464534698, + 0.4756807470379217, + 0.4061810393380376, + 0.5244469357422065, + 0.4806232746495714, + 0.2114902239185758, + 0.33410723039361007, + 0.22834834902467124, + 0.31355412194923565, + 0.2498280416611218, + 0.4556711386477128, + 0.36794621207755007, + 0.29032011835955823, + 0.4147295510384719, + 0.47284974387046513, + 0.2935747704613725, + 0.414892604771897, + 0.2406599495367464, + 0.40336612527884563, + 0.45669228620032715, + 0.2948558581171195, + 0.29426241602118597, + 0.22903373367733626, + 0.5034729201778687, + 0.22887603625296316, + 0.2961464921874968, + 0.47220882982435747, + 0.3630670643744075, + 0.3845434047387248, + 0.2807322216608577, + 0.29561230387660176, + 0.3672804304375797, + 0.3400991417473356, + 0.4115302430820732, + 0.2926250563014035, + 0.44969531333965984, + 0.23863657936541477, + 0.4582593981682212, + 0.387271767014527, + 0.350399493560379, + 0.43450839904381106, + 0.3490051219742019, + 0.44887851334616363, + 0.29795523602464147, + 0.46060318325920163, + 0.3455900164316053, + 0.2739498940069502, + 0.3977014583042208, + 0.2738482870526575, + 0.46525748594416483, + 0.20530243352944397, + 0.6579436544793414, + 0.6634146248100643, + 0.5546066816356163, + 0.555913891340696, + 0.5952466277895794, + 0.5940542142704717, + 0.24479942767388782, + 0.32845764275060363, + 0.29555380520560376, + 0.3452469551586866, + 0.29645107623344313, + 0.3130801564704293, + 0.33156103851750823, + 0.36552565230375966, + 0.3862418977082654, + 0.2954109105438618, + 0.3325325536088305, + 0.2780236341510484, + 0.4501338749159313, + 0.24077892764642195, + 0.38350224657462983, + 0.39373263278745235, + 0.3344659234537029, + 0.4486892188994322, + 0.3562584972615147, + 0.4740583326089423, + 0.23510509700243676, + 0.4397280243886708, + 0.3417605781738501, + 0.3254292680877955, + 0.28751959255003046, + 0.33922468786391635, + 0.311573957639185, + 0.33838878410898177, + 0.4794694139434017, + 0.43129165289567495, + 0.515232510357212, + 0.31758300800884254, + 0.3765468887312488, + 0.3549812760584856, + 0.34939970512493373, + 0.43615070113018767, + 0.3309128628193097, + 0.20622883813520695, + 0.3872335865269873, + 0.33547450489386316, + 0.3305433983434564, + 0.4035920075910452, + 0.313703236502733, + 0.2516492764982782, + 0.23340073907598324, + 0.3082204329802528, + 0.3286259613952004, + 0.27279321603862633, + 0.19922204632060134, + 0.29460120269963097, + 0.4748949983485575, + 0.5806694171524618, + 0.5057132913179144, + 0.47002822347467654, + 0.5059150064716067, + 0.44577042283214613, + 0.5158859729711391, + 0.46288623418466024, + 0.4533899369556863, + 0.3190184190693752, + 0.36012193740916426, + 0.43141649641987373, + 0.5717682175619092, + 0.15451664244997795, + 0.28204766445876656, + 0.2671817577375794, + 0.25035634326951706, + 0.2660873030638968, + 0.3660221422900475, + 0.2720008805390745, + 0.3432113083806387, + 0.7232565761563424, + 0.60151344758004, + 0.750082560428902, + 0.7658681951456752, + 0.6290223112768493, + 0.7083982349412657, + 0.7286647773826528, + 0.33140637170014975, + 0.3287381202184064, + 0.3561827650438471, + 0.46976169664057044, + 0.30201889579016933, + 0.34379400001235244, + 0.33256779581316515, + 0.33353220827676605, + 0.34903065049599274, + 0.3266255523057217, + 0.24348920136991417, + 0.4103831520899467, + 0.3539048155467292, + 0.33024138792237334, + 0.388075045283729, + 0.287811778623671, + 0.2946384384851649, + 0.3037258386706315, + 0.28864842993821443, + 0.2315987565125054, + 0.31509166419948104, + 0.23841530076216833, + 0.4903074814017111, + 0.44836827984764405, + 0.3213546291415345, + 0.2557950870298427, + 0.18935019929815386, + 0.48113036257041875, + 0.5501140247981053, + 0.6272988641810695, + 0.284109270781619, + 0.33792448019311666, + 0.22915014240231682, + 0.4255015161134053, + 0.30623770592871336, + 0.267719736346754, + 0.44964576454199306, + 0.4649395559168266, + 0.43397240239456364, + 0.29821599378108427, + 0.4068768491991941, + 0.3992416563699796, + 0.4503004579273626, + 0.41521064485760334, + 0.4552032642587244, + 0.34875553244085056, + 0.3940948778477379, + 0.4022733201219927, + 0.4077457281148136, + 0.23778469761629503, + 0.32611313280235826, + 0.3702950630578096, + 0.37848594247206396, + 0.5272244676657675, + 0.184778043092658, + 0.5341924358410369, + 0.693962445989488, + 0.5247713296766369, + 0.3567737482275236, + 0.2779019642838302, + 0.31122858453098307, + 0.23611709008947307, + 0.36648906233869816, + 0.44097386834471086, + 0.41285199137565654, + 0.3129275642737334, + 0.254987457676672, + 0.43724065004809, + 0.4767395117187014, + 0.5136068514230804, + 0.5785981515917847, + 0.42237413957083747, + 0.49851469303367696, + 0.32274506167434847, + 0.42871360043470985, + 0.24258075378202945, + 0.2447143177870946, + 0.3584185273164093, + 0.33366951289126595, + 0.4783151704427483, + 0.44582383659977165, + 0.5879454016573212, + 0.6423290025624158, + 0.7400402531322902, + 0.5790814368135205, + 0.7521995574910323, + 0.37692992108282813, + 0.35566936735619786, + 0.3588144851770202, + 0.41948852721389185, + 0.28122308509612637, + 0.4588736797069253, + 0.4127943858144339, + 0.4591912508172702, + 0.4401232671992812, + 0.3722881111564792, + 0.3852397364285769, + 0.39909680353814997, + 0.3345791439129585, + 0.39468149349167925, + 0.36560979892790146, + 0.394586043291443, + 0.44980683066855637, + 0.4600104964113993, + 0.44843873944918133, + 0.34169179457389987, + 0.38281126285693856, + 0.43071929105295303, + 0.3018562357496489, + 0.39899763953573636, + 0.3650601174296332, + 0.38924865559689975, + 0.34984008256949484, + 0.3975229656536275, + 0.3202680836085166, + 0.38985544152947244, + 0.3937192544143108, + 0.43178604474073323, + 0.2973043096252089, + 0.4498018279721987, + 0.41208782243767544, + 0.356609989484031, + 0.4886100124537749, + 0.4569586014182664, + 0.7314044554356353, + 0.20503021385913014, + 0.2721503791108632, + 0.326898733741167, + 0.31594300845355167, + 0.23940112841877093, + 0.20123699250408333, + 0.26605616706219554, + 0.6614852895286967, + 0.5360369596044561, + 0.6221704863311209, + 0.30577594019592114, + 0.3029792441639252, + 0.24489665295476667, + 0.33657370233988854, + 0.3729036224491345, + 0.20075138146551572, + 0.23783241276923706, + 0.3707456573591556, + 0.3398951159279847, + 0.401614291228932, + 0.3405187858840392, + 0.379638338068056, + 0.32933689437178815, + 0.2738132568626356, + 0.6825111073184067, + 0.4410999890699911, + 0.6824337143276169, + 0.5589179054861595, + 0.212196218291455, + 0.46732275519733013, + 0.44168433538758545, + 0.4632545604971889, + 0.431916109734847, + 0.3478683699810894, + 0.42721632585083485, + 0.4382663367828763, + 0.4645484098223293, + 0.4267049910491339, + 0.30299824026042566, + 0.3794245172106957, + 0.23078625347738596, + 0.25119828339365574, + 0.6217674567596327, + 0.43094338854117337, + 0.5571632551656815, + 0.45863783368275285, + 0.417271405045804, + 0.3972850354859646, + 0.36801769236475274, + 0.45577056045063546, + 0.3672893980420273, + 0.21392538301314637, + 0.3391651323727168, + 0.4222991013328265, + 0.22736120795170656, + 0.7490918489424108, + 0.3927476172844113, + 0.5518785502776652, + 0.7131084435001341, + 0.6734678019326579, + 0.3414410246525856, + 0.7935627529458101, + 0.46419539825094686, + 0.6462824153906954, + 0.3582672321007404, + 0.4121536465816863, + 0.3295207690909744, + 0.3125302277884027, + 0.4236597666092889, + 0.45739766438437507, + 0.38934065915331406, + 0.3103455763731363, + 0.3384191335705102, + 0.33943895423192094, + 0.36458963621903817, + 0.46446368644465574, + 0.22341954137854125, + 0.4461497547595447, + 0.3870244259152598, + 0.3654244553459715, + 0.44682625236663764, + 0.5162584378600572, + 0.27538237722154046, + 0.3599879495974438, + 0.3875640596704029, + 0.4909553218711682, + 0.4596228108069909, + 0.3716464353352707, + 0.25061457367800927, + 0.455712490384015, + 0.515340277912848, + 0.5017801115390631, + 0.3810333783807891, + 0.5588561234615768, + 0.5674650768083523, + 0.3898389482394629, + 0.45231601042778685, + 0.4936785476103599, + 0.4547586135384174, + 0.40811742541737706, + 0.30394589366695285, + 0.2531403786099415, + 0.2766721498401329, + 0.32491239924590215, + 0.26420244284084343, + 0.24350133952795117, + 0.28608809546049246, + 0.3114731791085618, + 0.32930114388594556, + 0.2681966229573469, + 0.28862754564694815, + 0.27995067940088014, + 0.2973426985139275, + 0.3609986826248688, + 0.2818186096912846, + 0.25650053515594334, + 0.3808541753669292, + 0.4769240394789731, + 0.6215793231621111, + 0.25555129938624577, + 0.47982228864939264, + 0.49324460565853767, + 0.49786977831174384, + 0.5071524724130861, + 0.3801227575914767, + 0.41417683255921267, + 0.3404449481875963, + 0.43603570489100757, + 0.3687814572654811, + 0.6098399694966763, + 0.41376088613831113, + 0.4488300113843384, + 0.242312804381988, + 0.22539044658595195, + 0.36562463481076557, + 0.4437544617044048, + 0.32379493147125343, + 0.539246808757302, + 0.30093444822282556, + 0.5955540183618764, + 0.4555300821862748, + 0.46927774953187834, + 0.46980313222638237, + 0.4577029197929807, + 0.376458678016438, + 0.4854197076023829, + 0.4783633453569454, + 0.4799004416203157, + 0.47865562664666034, + 0.39264183717799983, + 0.3036016881742342, + 0.23866665750019558, + 0.33745919241838795, + 0.21337340114337564, + 0.3555354416392005, + 0.38880962570805816, + 0.3957504877679929, + 0.3792370393896236, + 0.40753637716954233, + 0.3775351943195205, + 0.46544094539581987, + 0.42331055611346396, + 0.30434373235693324, + 0.4323043524535827, + 0.4992826992198969, + 0.31028555874963637, + 0.2651798619753589, + 0.3120459019886651, + 0.3486021102773114, + 0.4206086428084667, + 0.22963746949366878, + 0.39648312645980227, + 0.23245739612485403, + 0.37413541639128317, + 0.43587889461570023, + 0.4497384257232802, + 0.20856043551536674, + 0.27711742584866333, + 0.2956213861970448, + 0.20189464560557485, + 0.19206152641535298, + 0.2987466272973289, + 0.4014560132453613, + 0.24379267136410923, + 0.31164455897350524, + 0.40472214787310756, + 0.28477588000143056, + 0.2943027858213595, + 0.3982400256856634, + 0.45082930909929075, + 0.18548494394011555, + 0.4227578524402589, + 0.34573369777781116, + 0.29094666785104256, + 0.2647035744854017, + 0.41767171612481635, + 0.36557325885877273, + 0.18463315136913208, + 0.4382025456769042, + 0.37618134284236726, + 0.34691123481430514, + 0.45006894028304506, + 0.3129522775192849, + 0.2592178327820836, + 0.3101530516137147, + 0.22224036738860364, + 0.332023841570314, + 0.30961185124597385, + 0.3297732154501518, + 0.35913412075349244, + 0.36209081535861315, + 0.4415446842319621, + 0.37605821175730675, + 0.29887772016328396, + 0.3424658930762259, + 0.38531042362530277, + 0.36522866443459717, + 0.3331488765601262, + 0.4331805516749334, + 0.37209467382216416, + 0.48045782950841137, + 0.5156651302116723, + 0.4757403788176842, + 0.41490531606824993, + 0.30642454312137324, + 0.22128605468117615, + 0.35036138124791094, + 0.41714056411485223, + 0.3570204316980235, + 0.40942724521928503, + 0.4059916061823788, + 0.2773779603657326, + 0.4192249645496781, + 0.28781111097122536, + 0.23640732737736142, + 0.3648718372807607, + 0.383739423118871, + 0.41542820884714265, + 0.4223935009242307, + 0.26043701600199776, + 0.5377277425320252, + 0.44174636237940385, + 0.44183202647907116, + 0.4802920253138422, + 0.549408737166625, + 0.3888678766548841, + 0.27957075717904106, + 0.35524704322391293, + 0.3601749605910111, + 0.3536753557736666, + 0.33456850166756097, + 0.29398883354468464, + 0.30949753076897113, + 0.3627119442716134, + 0.25653188237278557 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.040421911474594556, + "valid_tracks": 4, + "per_track_pearson": [ + 0.043988671423222346, + 0.03819689527765276, + 0.04099247750006644, + 0.038509601697436675 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.16644348238907047, + "valid_tracks": 626, + "per_track_pearson": [ + 0.19831799048962967, + 0.1917843170786137, + 0.17866366622872892, + 0.19186057426622705, + 0.19723807248222655, + 0.19636072363090773, + 0.14167018801678932, + 0.166659821495055, + 0.20959774967437472, + 0.1905682918723681, + 0.1930316844451493, + 0.1947076791662904, + 0.19861125999738644, + 0.19478173741891003, + 0.18709080835971195, + 0.19039298897650964, + 0.19784366327651293, + 0.18668169687068714, + 0.1975397215374358, + 0.20589274484752407, + 0.20843507521924795, + 0.1903394959756641, + 0.19208567861718936, + 0.1531938233480093, + 0.1897239751791126, + 0.18340344324507377, + 0.17331091794955053, + 0.19711111139078177, + 0.18468437332711593, + 0.1753059205561976, + 0.17764170700346912, + 0.17597485513362, + 0.15040226294557738, + 0.20311396432301637, + 0.1483835030232424, + 0.19740350233343776, + 0.1414776906929419, + 0.1952113512048878, + 0.18895187419314727, + 0.2113853720598248, + 0.20505749863112607, + 0.17393116219884122, + 0.17684133987780645, + 0.16992434274046372, + 0.17283656749119877, + 0.15856983542122657, + 0.17563317295933686, + 0.18548073067874943, + 0.17478946914392815, + 0.16883911304988417, + 0.1840338821694419, + 0.15811341899365836, + 0.16604432308674144, + 0.1833434409511708, + 0.1770444438890449, + 0.16497537547438298, + 0.16934953481569134, + 0.1756755863937964, + 0.1940908486610027, + 0.19364507118245022, + 0.17615622326135216, + 0.19271913633235704, + 0.19529753407667497, + 0.1839692097024448, + 0.19148723950706395, + 0.18408360744393942, + 0.19593535248791963, + 0.18407937757064016, + 0.18833225479022367, + 0.17148126797275132, + 0.18061799653410304, + 0.1925932640808893, + 0.18306397693714108, + 0.17841561855426125, + 0.19373870509813412, + 0.1907549968306469, + 0.18316083506854816, + 0.17188748545339172, + 0.17919571288698793, + 0.1736637745779699, + 0.18269627215763823, + 0.1792530026397191, + 0.17727095664413964, + 0.17652045165995608, + 0.1794922877650703, + 0.19800420187202733, + 0.17955914834721529, + 0.18258835362481124, + 0.1856208435297152, + 0.17951292370301294, + 0.17972993295296882, + 0.17597296922968644, + 0.1803463336563356, + 0.1938812006299097, + 0.17837817072520487, + 0.19673962333356182, + 0.19334238046776483, + 0.19350466548066783, + 0.19125655331547006, + 0.18495141077676547, + 0.1920128998873029, + 0.18169826780284934, + 0.1911975582649802, + 0.2027422257621487, + 0.19406858952952286, + 0.16982976846866013, + 0.15390154882354234, + 0.17778643339144132, + 0.19945992668385218, + 0.19521319739295065, + 0.19181544690398458, + 0.19638650943453295, + 0.19524078508692772, + 0.19763286551560447, + 0.19854834454908393, + 0.19333476980426484, + 0.18555722677816197, + 0.18324774544873906, + 0.19478309081110268, + 0.19076600139518032, + 0.1855000541138186, + 0.1958402982999053, + 0.19472773966468068, + 0.19483595147248953, + 0.16159568203575567, + null, + 0.19788671097120558, + 0.18157217336967016, + 0.19621672641675808, + 0.19117192891141296, + 0.16366160294220566, + 0.20091777548996687, + 0.19503527659946834, + 0.16889813679885127, + 0.18198206101655123, + 0.19627915475137633, + null, + 0.19378962705034566, + 0.19991283114258762, + 0.15799301126348647, + 0.16853182293907298, + 0.18048931544389105, + 0.18229065353346843, + 0.1892509413706553, + 0.18617139534809712, + 0.18611511622327098, + 0.17736853080875792, + 0.19075503239074373, + 0.17320875719088225, + 0.17385642934070322, + 0.16543587712263302, + 0.1851950051505131, + 0.17660994865575777, + 0.18264533507133462, + 0.16790755079995914, + 0.16590599939278458, + 0.1755354378716252, + 0.1614149422485201, + 0.17840466236355607, + 0.1715291649187136, + 0.18031611130104863, + 0.18095529695669432, + 0.19731938441604902, + 0.19626260933321443, + 0.18665251602991634, + 0.18777929499694468, + 0.1764171153411306, + 0.18464692359223125, + 0.1570383025388149, + 0.1815522527270918, + 0.1741729139935207, + 0.1835399186091503, + 0.182397502896064, + 0.18591553929799473, + 0.18236151762989497, + 0.14490215424496458, + 0.1904202652462775, + 0.18089350457157363, + 0.18124624134060496, + 0.18774010725955814, + 0.17189329334170994, + 0.16944541169612334, + 0.1648443306124491, + 0.1968212752667746, + 0.18776083468376048, + 0.12460417381637232, + 0.1979265100154292, + 0.19997858112216363, + 0.19064318868443994, + 0.1931378367005137, + 0.18493165518416643, + 0.17390063723034882, + 0.18521562681145173, + 0.18933641641503415, + 0.19256851894784444, + 0.19368219645720944, + 0.19183100143703422, + 0.18459290034097925, + 0.20682680026899058, + 0.1926396335098571, + null, + 0.18472918368794158, + 0.18833796837592104, + 0.18538998579735538, + 0.18918453509145722, + 0.1818315737604654, + 0.178741918854859, + 0.18413854711772112, + null, + null, + 0.17843211798260927, + 0.1823950730436229, + null, + 0.2044601819826942, + 0.17320210748717915, + 0.1668919091386345, + 0.18589835947100317, + 0.19374519526280154, + 0.1892164970876516, + 0.18536415577768522, + null, + 0.18676219581804437, + 0.17419407078370286, + null, + 0.1674730769374932, + null, + 0.1771434903666371, + 0.18787681454284785, + 0.20173287324914102, + 0.17135688692739476, + null, + 0.1683701286917603, + 0.1688110702909937, + 0.1809511742946934, + null, + 0.1924749984273734, + null, + 0.18941696376738165, + null, + 0.18451354187921584, + 0.19157668551903723, + null, + 0.18057175205636586, + 0.18970927331010784, + null, + null, + 0.20147139985548718, + 0.19282180890211753, + 0.2035525977232577, + 0.19325174703200543, + null, + 0.18852805976000175, + null, + 0.19577434064608285, + 0.18279756144360249, + null, + 0.15434085017113777, + 0.19387532717504663, + 0.18649475336732102, + null, + 0.19142943973284388, + 0.19553515048734993, + null, + 0.18240179314813998, + 0.18166832850255024, + 0.18389212074735575, + null, + null, + null, + null, + 0.19160727013641154, + null, + null, + 0.19957202265092328, + 0.18426041839766372, + 0.1800968694622581, + null, + 0.19574646124760836, + 0.18802931875548065, + null, + null, + 0.1780249082505482, + 0.1876195689719482, + 0.17555812989011996, + 0.19302880046962492, + 0.19536539407913872, + 0.1926774440324335, + 0.18670621310875696, + 0.19495442995214177, + 0.18622437737647599, + 0.19106291670421735, + 0.17946431667048, + 0.19500916320331937, + null, + 0.1950612551377892, + 0.19027645849094366, + 0.1821641895406713, + 0.1858600001177454, + 0.18507999167441133, + 0.1783711513121435, + 0.18695230671900276, + 0.17752302062321468, + 0.1815562358628648, + 0.18433422390394488, + 0.1808792910511419, + 0.17608735099794265, + null, + 0.19303719453197105, + 0.1888567873973912, + 0.19038412078348338, + null, + 0.1942914273768442, + 0.16176641039044115, + null, + 0.201767711945539, + null, + 0.18532297356810476, + 0.1796570764354977, + 0.19173138284112462, + 0.17450651109737053, + 0.1785836151088818, + 0.19425249591517368, + 0.19065926940133635, + null, + null, + 0.19466264059893287, + 0.18767740899887217, + null, + 0.18971977563509215, + null, + 0.1908801444319977, + 0.18103053000446875, + 0.18257537052105954, + 0.18971070389330566, + null, + null, + null, + null, + 0.20081024714431128, + null, + 0.18643850860862413, + 0.19203863604963495, + null, + null, + 0.188469859822444, + 0.19777249519622814, + 0.19298776324556446, + null, + 0.18492247295041533, + 0.18255868940527373, + 0.18968262826000523, + null, + 0.19240499102984615, + null, + null, + 0.1911420175120309, + null, + 0.18821102904890966, + null, + null, + 0.18625797042242787, + 0.1933214158844085, + 0.1841220095888464, + 0.1893702417730721, + 0.1904069001489017, + null, + 0.19027136535620898, + 0.16550294869512003, + 0.15953286082186255, + 0.12094320779755956, + 0.12751972600417633, + 0.16307807644794733, + 0.14466005826003903, + 0.12149906798234883, + 0.14338038411703602, + 0.1579215861193211, + 0.15852492873739282, + 0.16724415503084275, + 0.160323211934284, + 0.1359536594272252, + 0.16066227913696862, + 0.13168985665238414, + 0.13475429714329631, + 0.1577409463016962, + 0.16254514003758846, + 0.1721011455871839, + 0.16669305679341376, + 0.16170464930777448, + 0.14520916836302178, + 0.13485523519198409, + 0.10088020025568294, + 0.1691840945551052, + 0.14813134998733501, + 0.16135571214625277, + 0.16225718566110875, + 0.1451451997159408, + 0.14678347209700962, + 0.14237843447319895, + 0.12922062577359697, + 0.10701112340364173, + 0.13483415144415348, + 0.09775876186884004, + 0.1342901419462867, + 0.09166241796720419, + 0.16171957380681415, + 0.16646749602803995, + 0.14744097407606752, + 0.14203076810764126, + 0.1107696715536744, + 0.1162171711474044, + 0.12628789454808292, + 0.14083257018757114, + 0.10528194043375046, + 0.12210749920128795, + 0.11338186072698818, + 0.12564927454246871, + 0.11179019585020246, + 0.11394136510693746, + 0.1165433644322789, + 0.12974049405751473, + 0.13087983585273696, + 0.10844133379196295, + 0.12890268256596368, + 0.1433972161701249, + 0.15550953406906964, + 0.15165666048556076, + 0.16226867397965775, + 0.1494599987476577, + 0.16986122566425982, + 0.1545261066275224, + 0.15474446963633148, + 0.166188319770224, + 0.1754853522684625, + 0.1500416057378473, + 0.15776935465781244, + 0.1591431551928453, + 0.13787600100332242, + 0.17373370443775335, + 0.1617019128105191, + 0.15429275054382097, + 0.1550183512785279, + 0.15518741959545512, + 0.15011470921094738, + 0.14894410797608906, + 0.15816824987005304, + 0.14157980594518488, + 0.15776129391779503, + 0.16099411883898374, + 0.1571366002116465, + 0.1628168447094452, + 0.1514863355002764, + 0.16084992850550536, + 0.15548870514445964, + 0.1533850814533808, + 0.15994876604063327, + 0.1434971708411373, + 0.15842249068179473, + 0.16067277727245047, + 0.16003918514699242, + 0.16687256246530754, + 0.15861963863460676, + 0.15432465320206698, + 0.16328393593382312, + 0.15536221214204313, + 0.16886288092124302, + 0.1722629907691208, + 0.1760923458292073, + 0.1537765027772652, + 0.15403392845483882, + 0.15940512114082483, + 0.1609895001932843, + 0.15898759048593356, + 0.1682876769910025, + 0.10230319997650592, + 0.15663447337726236, + 0.15195990150240718, + 0.14865019738415577, + 0.1421476370094138, + 0.150798340562725, + 0.1567630072557572, + 0.1394928294138735, + 0.15401644466991277, + 0.16033565138137734, + 0.15470535575165847, + 0.15779269902195087, + 0.12990953776745445, + 0.1392480629855765, + 0.15978581696769792, + 0.1567632554868985, + 0.1632406260700506, + 0.17031016668378585, + 0.11671130769871763, + null, + 0.15747057470725206, + 0.1333944026545829, + 0.1455320112440981, + 0.132856475746164, + 0.1403192116999784, + 0.16006451142856443, + 0.16006098427419835, + 0.15580314802367545, + 0.1501272142877578, + 0.1468270131009508, + null, + 0.15657650152567848, + 0.1468463664845111, + 0.13663562057922193, + 0.14683413426870293, + 0.144176480617926, + 0.14124612894401564, + 0.15691981823849993, + 0.15289749868895042, + 0.14811216641891234, + 0.14167662433342695, + 0.15348638766225012, + 0.15561781140541653, + 0.15622321482266527, + 0.13585282231595192, + 0.15032667820308337, + 0.16789317491641406, + 0.13025562795383566, + 0.11991028607258, + 0.13563856135486657, + 0.12461238081418811, + 0.16148645581052992, + 0.15390930015566884, + 0.1526597733753362, + 0.13752661159358856, + 0.1422111564341705, + 0.16569539989129217, + 0.1379833722356209, + 0.15305050332835698, + 0.14504369908280007, + 0.17145594576562923, + 0.1394576125970682, + 0.15324574610244787, + 0.14248440284391758, + 0.1505499320497631, + 0.16991398016705722, + 0.1492710517498338, + 0.1570017312873873, + 0.15322384840342626, + 0.1369020215497776, + 0.15421925173636053, + 0.15140354134866932, + 0.14809705506928067, + 0.13977584015748673, + 0.1576137748549573, + 0.15400636013450802, + 0.14056102364325845, + 0.18202024715454784, + 0.14678036448911183, + 0.14621553379160074, + 0.15883585557010943, + 0.14254249688295445, + 0.16115503879828744, + 0.15523098990208375, + 0.16190576617005167, + 0.14918987965518019, + 0.14527004068787344, + 0.14299301756296426, + 0.14026540101104445, + 0.1322551512662599, + 0.13518373758183855, + 0.15693241249022347, + 0.1634571635448689, + 0.15841350036754953, + null, + 0.16507434273707483, + 0.15323706108961271, + 0.13198867490565178, + 0.1493423517176986, + 0.16294272359352155, + 0.1518921713094191, + 0.13276399717867576, + null, + null, + 0.14378433870860055, + 0.14465954669224648, + null, + 0.15076201373172904, + 0.1226798840029646, + 0.14980011759861914, + 0.13337982916842675, + 0.13705879562874218, + 0.14071684613456606, + 0.16781593615582463, + null, + 0.15424698248117052, + 0.1593667166457693, + null, + 0.12585562874081874, + null, + 0.14256482972715134, + 0.1526898544342958, + 0.14035385966699795, + 0.1446815642761676, + null, + 0.14099271077869568, + 0.14615486570613326, + 0.15618273782052994, + null, + 0.12497347291202984, + null, + 0.1478823236187781, + null, + 0.15075571339529636, + 0.14624783971501204, + null, + 0.14110250349759804, + 0.16977981024039832, + null, + null, + 0.16291777101987187, + 0.1494331495700925, + 0.15008626308383352, + 0.13570191791078814, + null, + 0.13144431902620934, + null, + 0.15045919976545372, + 0.15033769241738523, + null, + 0.1213895985571195, + 0.13132210060201246, + 0.1428150199362992, + null, + 0.16235739211923383, + 0.139449209824817, + null, + 0.17427428803860068, + 0.16632905452972235, + 0.16991596930977657, + null, + null, + null, + null, + 0.17526156611032667, + null, + null, + 0.15881017047675702, + 0.1472280764429322, + 0.17076825146574176, + null, + 0.17570525149671654, + 0.17413775552511174, + null, + null, + 0.15142136344120652, + 0.14401929184930531, + 0.1494638228907775, + 0.14878468272490541, + 0.14924123234313882, + 0.15942237577589524, + 0.11041678976732462, + 0.1393828392868761, + 0.12513792539421645, + 0.14518882806096428, + 0.15446104960366933, + 0.1320042044348644, + null, + 0.1343151929338886, + 0.13887643897609195, + 0.16166053762499996, + 0.1622597887218541, + 0.14580089915617855, + 0.16581540603608647, + 0.14499239231860386, + 0.15801958828971388, + 0.15962388465179897, + 0.13552282095909393, + 0.1342879093445931, + 0.1339006741065294, + null, + 0.13706419501892297, + 0.16906544748645158, + 0.17475715143131038, + null, + 0.1397494466634862, + 0.14225410233123786, + null, + 0.15033792218276482, + null, + 0.1529733041646185, + 0.1471717027955202, + 0.15781254738835443, + 0.1451000786005516, + 0.1606626371131727, + 0.14352531966912388, + 0.1281642897830354, + null, + null, + 0.1501690577411088, + 0.1608259505855279, + null, + 0.14547067461283875, + null, + 0.1528156201347026, + 0.1425595496810254, + 0.1399109103275546, + 0.1641800446856283, + null, + null, + null, + null, + 0.14395454298372137, + null, + 0.1496892179904156, + 0.15888929859680342, + null, + null, + 0.14888228746761095, + 0.1294009342847469, + 0.12703201173584208, + null, + 0.1322402777356673, + 0.13967079647320213, + 0.1438169760504569, + null, + 0.14621362515736044, + null, + null, + 0.13931280808261673, + null, + 0.13638267512351948, + null, + null, + 0.13621894906471382, + 0.1555861426868818, + 0.1658159999027394, + 0.17709974950843735, + 0.16291495429730016, + null, + 0.14796071592160445 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753 + ] + }, + "junctions": { + "mean_pearson": 0.11819718243413396, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2902658914420824, + 0.23461437922486775, + 0.29521381878521474, + 0.22283628402593508, + 0.2736630856977953, + 0.2730546871376611, + 0.03783573824122103, + 0.12119436838368598, + 0.22320000105962845, + 0.1674478520562881, + 0.18800445629798246, + 0.12212774188298334, + 0.06564245402513225, + 0.09393550551689253, + 0.19856861600428066, + 0.1804495132149428, + 0.10261884855679751, + 0.061435475650794304, + 0.22706850738895176, + 0.20804828787132057, + 0.27267492126916276, + 0.23374461236682334, + 0.23849664220773292, + 0.14839366776716434, + 0.18111433603715443, + 0.22383184064020048, + 0.1567169258183633, + 0.24263704338144934, + 0.14785527893454917, + 0.15010103908083489, + 0.22567051985707248, + 0.30915609169809977, + 0.224682604982667, + 0.31607827330222876, + 0.2428974074002046, + 0.28425765016428794, + 0.28568380301264745, + 0.2044918804745652, + 0.21257228904876108, + 0.12278536763619, + 0.14024509470735816, + 0.1947581098793242, + 0.19960511948230306, + 0.18930000954757686, + 0.18587632212934582, + 0.14306617330704982, + 0.15304288414515158, + 0.19255679873047352, + 0.17135398708395624, + 0.16879292298075504, + 0.21252114048966092, + 0.2377263158787338, + 0.13986461306339493, + 0.2735650584390085, + 0.1630209163283301, + 0.23025260102128983, + 0.1607156971941222, + 0.08318461763370992, + 0.1781697615488755, + 0.1312661291120525, + 0.04709165457877786, + 0.1462682179283785, + 0.17003259813077284, + 0.09915929406757826, + 0.06190642361851705, + 0.23423021618257803, + 0.20103999379112938, + 0.2163318975114487, + 0.13724296081258128, + 0.14075126696168036, + 0.18690670851017419, + 0.21606812470602246, + 0.1655865167688358, + 0.06363105122017595, + 0.4060035726852439, + 0.33519259460492135, + 0.18024829490273975, + 0.11481487176081306, + 0.12966724898062046, + 0.24833045559729053, + 0.1722256137123989, + 0.1842621744708491, + 0.18161110877640851, + 0.14883419012580018, + 0.1623881098501496, + 0.24156830432742557, + 0.258471539758443, + 0.2629922351122241, + 0.24695815515280378, + 0.20073127778466268, + 0.24158864966210267, + 0.19638325050733665, + 0.10407475661484124, + 0.19475015436350862, + 0.20427312878946133, + 0.33446781457183444, + 0.23795593983480876, + 0.17256911902548433, + 0.20025749763742723, + 0.20251802199919924, + 0.12794565562223664, + 0.1849313554499684, + 0.3798911455492873, + 0.2035997584907655, + 0.18275006447009026, + 0.13101197599664804, + 0.07538041695926306, + 0.23822488822590715, + 0.33506617904326386, + 0.30402774088595425, + 0.11015582087772229, + 0.3915891914412481, + 0.30611227389522383, + 0.3805681969464538, + 0.30171522774960446, + 0.12092490551319497, + 0.07951831225940074, + 0.14216061273836852, + 0.11132149026187367, + 0.23603282019893837, + 0.2091440920393534, + 0.14681556866095719, + 0.12821301624444742, + 0.20667546854306698, + 0.07431052057292076, + null, + 0.30046981525768524, + 0.178536238457799, + 0.06313850504552822, + 0.17998250011188208, + 0.23418809005488359, + 0.1754801257554113, + 0.12695276154387233, + 0.21963236102969802, + 0.2875290811594995, + 0.12869622730281255, + null, + 0.2979129347259227, + 0.29783167282511785, + 0.23430912136395904, + 0.23201862296080247, + 0.21623610566037943, + 0.1934604839768065, + 0.17828801491348936, + 0.1793756102785928, + 0.21147086055798667, + 0.2382312971293661, + 0.19366955803839866, + 0.13432220096321995, + 0.18896099158369342, + 0.14194669805171, + 0.22809664436069535, + 0.16361047671117085, + 0.25312265410048485, + 0.38390855858546324, + 0.29553266074796297, + 0.37488128998774795, + 0.19177119928916034, + 0.16888760533296068, + 0.2084580037088189, + 0.16452153462067312, + 0.24642144273389494, + 0.23653473150638898, + 0.2607352631866807, + 0.10391082743999644, + 0.31240860420819616, + 0.2478509172019855, + 0.15247730557726716, + 0.05854624539178005, + 0.17496549361540423, + 0.20986878097900302, + 0.19634406595216364, + 0.2705346517960266, + 0.08976821312806385, + 0.13529883997120848, + 0.18749342801158614, + 0.14520472546441165, + 0.16795952213749943, + 0.13319061578931693, + 0.20331713058641213, + 0.07737117384737507, + 0.11356647241330094, + 0.044888405605995885, + 0.21802561945155333, + 0.22737727815397796, + 0.22666947702799833, + 0.28877292574296365, + 0.20832601598646333, + 0.285249946172617, + 0.2161345182636188, + 0.2937664670173916, + 0.33512265533375507, + 0.18035857148111406, + 0.12358809689410448, + 0.14083329419185514, + 0.11543396604304736, + 0.09564843873396688, + 0.15814604791381345, + 0.30531997783873344, + 0.38577346350324393, + null, + 0.17616077769615518, + 0.18888774503220682, + 0.2254640182068243, + 0.27264793660049885, + 0.10084030234706709, + 0.1467652655308685, + 0.07821641076197768, + null, + null, + 0.21966736161627817, + 0.2738901312153522, + null, + 0.13119655314793055, + 0.12481384494401694, + 0.07566787550370975, + 0.1700198817670209, + 0.10744367362294947, + 0.17506390671484662, + 0.06375110140906415, + null, + 0.3057973792408819, + 0.1448245046503137, + null, + 0.11464445064322788, + null, + 0.20261958339235192, + 0.2540385069109749, + 0.16082117647369534, + 0.1653649097157571, + null, + 0.08276269043507166, + 0.04522639724909839, + 0.20104036123991936, + null, + 0.1517351452187132, + null, + 0.09077241850477677, + null, + 0.11267935852660736, + 0.1391798547437332, + null, + 0.17521031861218986, + 0.19427098391751768, + null, + null, + 0.2341833137455263, + 0.26175500592774986, + 0.04650559118373123, + 0.08048969284827492, + null, + 0.21117157583421758, + null, + 0.19960618496769253, + 0.1952754099339412, + null, + 0.16834531512671805, + 0.26732220589073824, + 0.23756942846329057, + null, + 0.2392974678873182, + 0.24739547556268915, + null, + 0.2723732435192056, + 0.18917202290579468, + 0.2679031412228103, + null, + null, + null, + null, + 0.1364590691724613, + null, + null, + 0.16189854872262036, + 0.2501784758902929, + 0.2347528472761586, + null, + 0.11079759230703039, + 0.28406583728388984, + null, + null, + 0.16455647472826326, + 0.09022316002932418, + 0.22726715628266872, + 0.1204034878718086, + 0.1226803293623104, + 0.09477805166896688, + 0.13661014583475245, + 0.14910514700790534, + 0.03291897284814954, + 0.09940978635898559, + 0.33869092623220476, + 0.17944027210392668, + null, + 0.18789283525097494, + 0.18949907464023022, + 0.04078108600226243, + 0.15081637616533, + 0.09233337965599799, + 0.06574399920789706, + 0.10424069855354869, + 0.1600641008814888, + 0.2360786522577667, + 0.16632772251427413, + 0.21919791762489987, + 0.17776384680093682, + null, + 0.15110715376693232, + 0.19239777969347366, + 0.20176090691742715, + null, + 0.25289313288639886, + 0.18019711940228403, + null, + 0.17581916519787283, + null, + 0.1657981178957035, + 0.14778984723470506, + 0.21929051491076354, + 0.11424997449374384, + 0.17663279032624138, + 0.18814319794224724, + 0.23048556471408796, + null, + null, + 0.09196764362469859, + 0.11344566473665683, + null, + 0.19909491994433956, + null, + 0.19826738307621888, + 0.08985107623229861, + 0.07189490509044112, + 0.1779100782666768, + null, + null, + null, + null, + 0.019894135338419147, + null, + 0.07468434948840343, + 0.2277671321095792, + null, + null, + 0.1521839171267367, + 0.2373728270504875, + 0.16786662981703335, + null, + 0.22519062168046977, + 0.2172339701739609, + 0.06138236969921056, + null, + 0.20045594671569958, + null, + null, + 0.25465191404246007, + null, + 0.16478132757447525, + null, + null, + 0.17999103044752118, + 0.2065979379613223, + 0.18150014380420892, + 0.20488922660339798, + 0.16148530960181268, + null, + 0.13466080971097366, + 0.008789468491418291, + 0.06588108674447536, + 0.1550497384139887, + 0.0528051079202445, + 0.15023396925421834, + 0.05456974109542695, + -0.031551643876968724, + 0.011295186228405033, + 0.011825639846405363, + 0.06708976298994226, + 0.05337206893306165, + 0.06351564159413037, + 0.04089132749097856, + -0.0022636399807489417, + 0.09347126076462922, + 0.0603026977440796, + 0.0051047510945245525, + -0.031826880543757735, + 0.08885266360159692, + 0.040913565495078526, + 0.021210937305415654, + 0.1559049436506531, + 0.09616012377978496, + 0.07986241272919879, + 0.029498337320117193, + 0.0010402524195321618, + 0.06430968693320824, + 0.04989237547760234, + -0.021409897168586135, + 0.005412337431957856, + 0.044541281456393696, + 0.037856271008671546, + 0.0203703120954007, + 0.10911975678795374, + 0.02069236980034915, + 0.09094139105784824, + 0.05328689776605196, + 0.02813497719333625, + 0.03788459520392296, + 0.10301181498265997, + 0.03339323801019781, + 0.024717275666651346, + 0.011877984493564855, + 0.0006016298068552481, + 0.04284565645236985, + -0.014733264971192366, + 0.03190561754020212, + 0.12123243414703117, + 0.03210502218497117, + 0.05514262426901867, + 0.04049574322051008, + 0.058716360901255614, + 0.031037454051248047, + 0.03570227167724271, + 0.012069448134443663, + 0.053522673548208444, + 0.08593571546933851, + 0.07893799578690117, + 0.0982507224108907, + 0.03592928221336278, + 0.0003624189147828933, + 0.002478228095524609, + 0.03770455476873961, + 0.10971254022546037, + 0.08526674189493672, + 0.07780215274716204, + 0.06944857947195249, + 0.02448608054946696, + 0.04362550770222404, + 0.01261300415909968, + 0.08222178914579929, + 0.030084782727250733, + 0.058314372548305626, + 0.07422827645393314, + 0.023513779395563202, + 0.04026365141783864, + 0.013209834142149232, + 0.06207127591873982, + -0.016725975547740128, + 0.03775560614617522, + 0.023887131214631072, + 0.0728380767171579, + 0.071339106209365, + 0.016855729848396155, + 0.06342098008611799, + 0.0633088632438891, + 0.13101793089501454, + 0.042399272233159195, + 0.042370735519070245, + 0.008929963799280603, + 0.03834206341492904, + 0.02601545387877105, + 0.010223704501971317, + 0.017924790338941714, + 0.04158839936675773, + 0.05099569107378621, + 0.04178126604267644, + -0.011324149457176127, + 0.10673488042829113, + 0.008195218062751532, + 0.04497938217854739, + 0.06171149427529073, + 0.055259535646068836, + 0.10148146613040183, + 0.06574476246109917, + 0.035019655538183446, + -0.0077091801592042266, + 0.09210117418483489, + 0.04576832381172274, + 0.070832668949911, + 0.02009094244350934, + 0.06194262344578234, + 0.042652857185528956, + 0.0780757951241332, + 0.06602997573722469, + 0.03038454862761969, + 0.04133907901624571, + 0.01953815297024596, + 0.04776053859401494, + 0.06876819669214168, + 0.05203808791305409, + 0.023882235007430575, + 0.03752133312985066, + 0.08602151949501607, + 0.07580677846405703, + null, + 0.07022815446857067, + 0.06596059207549373, + -0.01087406889419873, + 0.06599281905415491, + 0.08863263958096808, + 0.0006223919015612164, + -0.008547417525700313, + 0.05006464639203788, + 0.03307652241545164, + 0.06046596283868293, + null, + 0.06859590247044144, + 0.06204802970635478, + -0.0184533692865297, + 0.01922235131716414, + 0.12370288832268915, + -0.007861824742072804, + 0.09747877822651659, + 0.05239341694552306, + 0.0010313010349639845, + 0.0457750121074979, + 0.008014242860373609, + -0.014906070616138013, + 0.049596845873897295, + 0.10710333815946703, + 0.05640364774691101, + -0.0045038535102068, + 0.016328680144160796, + 0.03257918292194778, + 0.05002952012937762, + 0.038114483806417084, + -0.014979085660140634, + 0.09681622633602342, + 0.09010504580085167, + 0.020958504263238815, + 0.05031945557667239, + 0.08405531605706461, + 0.08567538698421528, + 0.11098130471630485, + 0.07507526388661802, + 0.14639263839926467, + 0.029632571771892283, + 0.05082422390900129, + 0.20863847955418427, + 0.04031294807055757, + 0.003347849273132562, + 0.05389128195881587, + 0.083671341251755, + 0.023184234702493433, + 0.05219210240381874, + 0.07364869248199893, + 0.0454132231787843, + 0.0448326162694319, + 0.13515548502817706, + 0.007022133268755858, + 0.009830936934423131, + 0.3012579589028359, + 0.10394088496763972, + -0.018029376668490155, + 0.12441586296851335, + 0.046247772482867044, + -0.015944561647327445, + 0.08699841364208961, + 0.09380870299234637, + 0.20648413311838065, + 0.28009917542429846, + 0.06739522191860123, + 0.1711953810782445, + -0.011733365173459517, + -0.033155854665342405, + -0.026284961889463065, + 0.02135796204813457, + 0.07271951304844103, + 0.05403358824063568, + null, + 0.07223981205533511, + 0.07139186248928703, + 0.036075122880603, + 0.0444936165874379, + 0.15633464660680596, + 0.023350166662957503, + 0.068648990588453, + null, + null, + 0.06314419512265511, + 0.11792400616742021, + null, + -0.019389777625320157, + -0.010406103108132735, + 0.004442486447629215, + 0.04744649367944903, + -0.034014712839106426, + -0.0032177547009635053, + 0.028211017595009116, + null, + 0.0011044934245067924, + -0.001489878421264753, + null, + -0.030213295166591724, + null, + 0.05904281514084167, + 0.058064751969070856, + 0.0385876174527074, + 0.00024204910976168283, + null, + -0.024156355339872402, + 0.02781211916876546, + 0.01581082548357238, + null, + 0.011944238824365579, + null, + 0.1494040965075453, + null, + -0.029159148821950313, + 0.14224210174132407, + null, + 0.06590158275033242, + 0.051952656906622194, + null, + null, + 0.08752792160650033, + 0.0694858838346083, + 0.1269153214116636, + 0.08314108870944209, + null, + 0.012700662278579826, + null, + -0.00757234054063663, + 0.011390474266762003, + null, + 0.1575566534074691, + 0.01003142682251729, + 0.08501751906146676, + null, + 0.04018146720650186, + 0.017638040313896854, + null, + 0.07575859571501198, + 0.09923169641607381, + 0.12910939193492196, + null, + null, + null, + null, + 0.0717081012566274, + null, + null, + 0.04641626565150954, + 0.08622867706067605, + 0.058257099737117776, + null, + 0.01842847434518605, + 0.07063027549620259, + null, + null, + 0.01890068314860329, + 0.04998141745401549, + 0.07595403421052534, + 0.0074260130259238225, + 0.00437370559716815, + -0.02147361090098413, + 0.19054169161318374, + 0.023559753675429094, + 0.043337116524867506, + 0.0015763513538801842, + 0.049902371729888206, + 0.015259264605920688, + null, + 0.024110380324863914, + 0.04016316072535872, + 0.0604445946452271, + 0.03441392089084522, + 0.00562979162383074, + -0.044766061639403616, + 0.047298763388714415, + -0.015026628843491723, + 0.05024009792194072, + 0.04405590228202297, + 0.014295415835569845, + 0.04414155485415944, + null, + 0.02364953438435782, + 0.021641466940641743, + 0.06997875299503542, + null, + 0.0005455527230018109, + -0.025575589367190676, + null, + 0.05400806955013162, + null, + 0.020692670099955318, + 0.05262142763455807, + 0.024936190272892466, + 0.03197272123339553, + 0.0829022285249569, + 0.08822125062691172, + 0.09382949434956749, + null, + null, + 0.08795164722642643, + -0.005668408893389517, + null, + 0.014133033265808372, + null, + 0.13306352130999533, + 0.10122618337449991, + 0.1038749821317352, + 0.03404763748902796, + null, + null, + null, + null, + 0.03418418906677969, + null, + 0.05664818267770683, + 0.1402122495012327, + null, + null, + -0.0037899954832037066, + 0.14604151402934726, + 0.03484562943278646, + null, + 0.04791365649675741, + 0.011049221508263964, + 0.15242354171040579, + null, + 0.09065611175252886, + null, + null, + 0.028580352407696533, + null, + 0.04789915594934698, + null, + null, + -0.019662869033888628, + 0.0036671895750194994, + 0.039081542001949104, + 0.024034936444132744, + 0.016227249457711798, + null, + 0.09484244396942973 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9536423841059603, + 0.9407894736842105, + 0.9537366548042705, + 0.9525316455696202, + 0.9701986754966887, + 0.9525423728813559, + 0.9683098591549296, + 0.9548611111111112, + 0.9519230769230769, + 0.950354609929078, + 0.9494949494949495, + 0.9355932203389831, + 0.948051948051948, + 0.94375, + 0.9633333333333334, + 0.944078947368421, + 0.951048951048951, + 0.9519230769230769, + 0.9522292993630573, + 0.9509803921568627, + 0.9496402877697842, + 0.9483870967741935, + 0.948051948051948, + 0.9405204460966543, + 0.9537366548042705, + 0.9539473684210527, + 0.9615384615384616, + 0.9559322033898305, + 0.9527027027027027, + 0.9482758620689655, + 0.9486301369863014, + 0.9676375404530745, + 0.9365671641791045, + 0.9604863221884499, + 0.945054945054945, + 0.9495268138801262, + 0.9427480916030534, + 0.9404388714733543, + 0.9300699300699301, + 0.9463087248322147, + 0.9453924914675768, + 0.9553264604810997, + 0.9427480916030534, + 0.9518518518518518, + 0.9508196721311475, + 0.9459459459459459, + 0.948339483394834, + 0.9354838709677419, + 0.9361702127659575, + 0.95578231292517, + 0.9416058394160584, + 0.9595588235294118, + 0.9393939393939394, + 0.9523809523809523, + 0.9328358208955224, + 0.9455782312925171, + 0.9480968858131488, + 0.9503311258278145, + 0.9554794520547946, + 0.9471830985915493, + 0.9517241379310345, + 0.9573770491803278, + 0.9503311258278145, + 0.9522184300341296, + 0.9362416107382551, + 0.9525316455696202, + 0.9358108108108109, + 0.9563758389261745, + 0.9385665529010239, + 0.9575163398692811, + 0.9572953736654805, + 0.9451612903225807, + 0.9494949494949495, + 0.9602649006622517, + 0.942652329749104, + 0.9559322033898305, + 0.9517684887459807, + 0.9464882943143813, + 0.9551724137931035, + 0.95578231292517, + 0.9403973509933775, + 0.9588607594936709, + 0.9452054794520548, + 0.941747572815534, + 0.9543859649122807, + 0.9473684210526315, + 0.9452054794520548, + 0.9673202614379085, + 0.95, + 0.9493670886075949, + 0.9512195121951219, + 0.9553264604810997, + 0.9543859649122807, + 0.9523809523809523, + 0.9425675675675675, + 0.9618055555555556, + 0.9457627118644067, + 0.9377162629757786, + 0.9470198675496688, + 0.939873417721519, + 0.9466666666666667, + 0.9475409836065574, + 0.9395017793594306, + 0.9488817891373802, + 0.9590443686006825, + 0.9566666666666667, + 0.9325842696629213, + 0.9498327759197325, + 0.946843853820598, + 0.9459459459459459, + 0.95, + 0.9395973154362416, + 0.9577922077922078, + 0.9559322033898305, + 0.9489795918367347, + 0.9391891891891891, + 0.9448275862068966, + 0.9459459459459459, + 0.9422382671480144, + 0.9590443686006825, + 0.9464285714285714, + 0.9559322033898305, + 0.9411764705882353, + 0.9473684210526315, + 0.9433198380566802, + 1.0, + 0.9550173010380623, + 0.9407665505226481, + 0.9431818181818182, + 0.9514925373134329, + 0.9528985507246377, + 0.9517241379310345, + 0.9442508710801394, + 0.9360902255639098, + 0.9508771929824561, + 0.939799331103679, + 0.9919354838709677, + 0.9559322033898305, + 0.9514925373134329, + 0.9539473684210527, + 0.9525547445255474, + 0.9328358208955224, + 0.9522058823529411, + 0.9482758620689655, + 0.9368421052631579, + 0.9465648854961832, + 0.9488188976377953, + 0.9479166666666666, + 0.9264705882352942, + 0.9469964664310954, + 0.9407407407407408, + 0.9574468085106383, + 0.9413919413919414, + 0.9642857142857143, + 0.9517241379310345, + 0.9415807560137457, + 0.9545454545454546, + 0.9530685920577617, + 0.950530035335689, + 0.9416058394160584, + 0.9393939393939394, + 0.9583333333333334, + 0.9467084639498433, + 0.9407665505226481, + 0.939622641509434, + 0.9496402877697842, + 0.9446366782006921, + 0.9668874172185431, + 0.9433962264150944, + 0.9508771929824561, + 0.9597069597069597, + 0.9419795221843004, + 0.9548872180451128, + 0.9623287671232876, + 0.9423076923076923, + 0.953307392996109, + 0.9311594202898551, + 0.9442508710801394, + 0.9558823529411765, + 0.9440559440559441, + 0.9477351916376306, + 0.9440298507462687, + 0.9520295202952029, + 0.942652329749104, + 0.9577464788732394, + 0.959409594095941, + 0.9417808219178082, + 0.936026936026936, + 0.931899641577061, + 0.9508196721311475, + 0.9409722222222222, + 0.9522058823529411, + 0.9357142857142857, + 0.9401408450704225, + 0.9337979094076655, + 0.9293286219081273, + 0.9301470588235294, + 0.9479166666666666, + 0.9477124183006536, + 0.9566666666666667, + 0.9922480620155039, + 0.9572368421052632, + 0.9491525423728814, + 0.9501661129568106, + 0.9391891891891891, + 0.9614147909967846, + 0.9611307420494699, + 0.9375, + 1.0, + 0.9924242424242424, + 0.9636963696369637, + 0.9512987012987013, + 0.9923076923076923, + 0.9551724137931035, + 0.9304029304029304, + 0.9343065693430657, + 0.9419795221843004, + 0.9536423841059603, + 0.9517241379310345, + 0.9496644295302014, + 1.0, + 0.9565217391304348, + 0.9381443298969072, + 0.9924812030075187, + 0.93359375, + 1.0, + 0.9490909090909091, + 0.9473684210526315, + 0.9436619718309859, + 0.9528619528619529, + 1.0, + 0.9492753623188406, + 0.9366197183098591, + 0.9368421052631579, + 1.0, + 0.9461279461279462, + 1.0, + 0.935064935064935, + 1.0, + 0.9566666666666667, + 0.9530201342281879, + 1.0, + 0.929368029739777, + 0.9442622950819672, + 1.0, + 1.0, + 0.950354609929078, + 0.9575163398692811, + 0.9509803921568627, + 0.9461279461279462, + 1.0, + 0.9399293286219081, + 0.9922480620155039, + 0.948051948051948, + 0.9484536082474226, + 0.9930555555555556, + 0.95, + 0.9421221864951769, + 0.9418181818181818, + 0.9927536231884058, + 0.9321428571428572, + 0.9565217391304348, + 1.0, + 0.9540983606557377, + 0.9403508771929825, + 0.9559322033898305, + 1.0, + 1.0, + 0.9922480620155039, + 1.0, + 0.9401993355481728, + 1.0, + 0.984251968503937, + 0.9489795918367347, + 0.9432624113475178, + 0.9519230769230769, + 1.0, + 0.9419795221843004, + 0.9452054794520548, + 1.0, + 0.9923076923076923, + 0.9407665505226481, + 0.9456521739130435, + 0.9498327759197325, + 0.9493243243243243, + 0.931740614334471, + 0.9387755102040817, + 0.9590443686006825, + 0.9411764705882353, + 0.9355932203389831, + 0.9252669039145908, + 0.945859872611465, + 0.9372693726937269, + 1.0, + 0.9508771929824561, + 0.956081081081081, + 0.9428571428571428, + 0.9562289562289562, + 0.9591836734693877, + 0.9583333333333334, + 0.9438596491228071, + 0.9448275862068966, + 0.9508196721311475, + 0.9493243243243243, + 0.9492753623188406, + 0.9419795221843004, + 1.0, + 0.9453376205787781, + 0.9493243243243243, + 0.9583333333333334, + 1.0, + 0.952076677316294, + 0.946236559139785, + 1.0, + 0.9347079037800687, + 1.0, + 0.9455128205128205, + 0.9479166666666666, + 0.9537953795379538, + 0.9347826086956522, + 0.9489795918367347, + 0.934931506849315, + 0.9405594405594405, + 0.9929078014184397, + 1.0, + 0.9411764705882353, + 0.9319727891156463, + 1.0, + 0.9386281588447654, + 1.0, + 0.946843853820598, + 0.9395973154362416, + 0.9418181818181818, + 0.9401993355481728, + 1.0, + 1.0, + 0.992, + 1.0, + 0.9452054794520548, + 1.0, + 0.9342560553633218, + 0.9619047619047619, + 1.0, + 1.0, + 0.954225352112676, + 0.946843853820598, + 0.9407407407407408, + 1.0, + 0.9417808219178082, + 0.9461279461279462, + 0.925531914893617, + 0.9919354838709677, + 0.9681528662420382, + 1.0, + 0.9922480620155039, + 0.9556313993174061, + 1.0, + 0.9514563106796117, + 0.9921259842519685, + 0.9922480620155039, + 0.9420849420849421, + 0.9512987012987013, + 0.9415807560137457, + 0.9484536082474226, + 0.9436619718309859, + 1.0, + 0.9403973509933775, + 0.9774436090225563, + 0.9859649122807017, + 0.9927536231884058, + 0.9857142857142858, + 0.9873417721518988, + 0.9822064056939501, + 0.9860627177700348, + 0.9811320754716981, + 0.9829351535836177, + 0.9856115107913669, + 0.9930795847750865, + 0.9855595667870036, + 0.9863013698630136, + 0.9863013698630136, + 0.9925650557620818, + 0.989010989010989, + 0.9788732394366197, + 0.9814814814814815, + 0.9857142857142858, + 0.9829351535836177, + 0.9755244755244755, + 0.9858657243816255, + 0.9732441471571907, + 0.9849624060150376, + 0.9857142857142858, + 0.9745454545454545, + 0.9867549668874173, + 0.9825783972125436, + 0.982078853046595, + 0.9802371541501976, + 0.9859649122807017, + 0.9852941176470589, + 0.9879032258064516, + 0.9822695035460993, + 0.9816849816849816, + 0.9787234042553191, + 0.9920318725099602, + 0.9680851063829787, + 0.9825174825174825, + 0.9784172661870504, + 0.9753521126760564, + 0.9817518248175182, + 0.9929328621908127, + 0.9882352941176471, + 0.9828178694158075, + 0.9820143884892086, + 0.9924812030075187, + 0.9847328244274809, + 0.9924528301886792, + 0.9884615384615385, + 0.9879518072289156, + 0.9888475836431226, + 0.9924242424242424, + 0.9822064056939501, + 0.9849624060150376, + 0.9814126394052045, + 0.9856115107913669, + 0.989247311827957, + 0.9893238434163701, + 0.9782608695652174, + 0.9895470383275261, + 0.9700374531835206, + 0.9692832764505119, + 0.989010989010989, + 0.9891696750902527, + 0.9889705882352942, + 0.9857142857142858, + 0.9922779922779923, + 0.9847908745247148, + 0.985663082437276, + 0.9817518248175182, + 0.9710144927536232, + 0.9830508474576272, + 0.985663082437276, + 0.978494623655914, + 0.9782608695652174, + 0.9758620689655172, + 0.9858657243816255, + 0.9820143884892086, + 0.9773584905660377, + 0.9757785467128027, + 0.9828767123287672, + 0.9809885931558935, + 0.9814814814814815, + 0.9858156028368794, + 0.9852941176470589, + 0.9809160305343512, + 0.9791666666666666, + 0.9817518248175182, + 0.978021978021978, + 0.9825783972125436, + 0.9822064056939501, + 0.9758620689655172, + 0.9888475836431226, + 0.9894736842105263, + 0.992831541218638, + 0.9792387543252595, + 0.966542750929368, + 0.9863013698630136, + 0.982078853046595, + 0.9731800766283525, + 0.9781021897810219, + 0.9855072463768116, + 0.975, + 0.9818840579710145, + 0.9895833333333334, + 0.9886792452830189, + 0.9888888888888889, + 0.9789473684210527, + 0.9893238434163701, + 0.9859154929577465, + 0.9887640449438202, + 0.9768211920529801, + 0.978494623655914, + 0.985663082437276, + 0.9769230769230769, + 0.9858657243816255, + 0.9807692307692307, + 0.9841897233201581, + 0.988929889298893, + 0.9859649122807017, + 0.989247311827957, + 0.9853479853479854, + 0.9858657243816255, + 0.9814126394052045, + 1.0, + 0.9822695035460993, + 0.9821428571428571, + 0.9858156028368794, + 0.9881422924901185, + 0.9675090252707581, + 0.986013986013986, + 0.9858657243816255, + 0.9849624060150376, + 0.9860627177700348, + 0.9847328244274809, + 0.9906542056074766, + 0.9825783972125436, + 0.9816176470588235, + 0.9790209790209791, + 0.9825783972125436, + 0.9851301115241635, + 0.9822695035460993, + 0.986013986013986, + 0.9858156028368794, + 0.9851301115241635, + 0.9925373134328358, + 0.9850187265917603, + 0.9819494584837545, + 0.9926739926739927, + 0.9807692307692307, + 0.981549815498155, + 0.985663082437276, + 0.981549815498155, + 0.9886792452830189, + 0.9827586206896551, + 0.9816849816849816, + 0.981203007518797, + 0.9898648648648649, + 0.9888888888888889, + 0.9782608695652174, + 0.9853479853479854, + 0.9838187702265372, + 0.986013986013986, + 0.9925093632958801, + 0.9881889763779528, + 0.9897260273972602, + 0.978494623655914, + 0.9877551020408163, + 0.9925650557620818, + 0.9891304347826086, + 0.9814814814814815, + 0.9922480620155039, + 0.9858657243816255, + 0.985239852398524, + 0.9878048780487805, + 0.9858156028368794, + 0.9857142857142858, + 0.9895104895104895, + 0.9927007299270073, + 0.9860627177700348, + 0.9888475836431226, + 0.9963235294117647, + 0.978021978021978, + 0.9788732394366197, + 0.9835390946502057, + 0.9776951672862454, + 0.9820143884892086, + 0.9788732394366197, + 0.9857142857142858, + 0.9896907216494846, + 0.992619926199262, + 0.9787985865724381, + 0.985239852398524, + 0.985663082437276, + 0.9820143884892086, + 0.9859154929577465, + 0.9807692307692307, + 0.9870967741935484, + 0.9795221843003413, + 0.9914529914529915, + 0.9822695035460993, + 0.9831649831649831, + 0.9813432835820896, + 0.9706959706959707, + 0.9828767123287672, + 0.9795918367346939, + 0.9752650176678446, + 0.9920634920634921, + 0.9916666666666667, + 0.987460815047022, + 0.9837133550488599, + 1.0, + 0.9824561403508771, + 0.9757785467128027, + 0.9846153846153847, + 0.9795221843003413, + 0.9790940766550522, + 0.9786476868327402, + 0.9858657243816255, + 1.0, + 0.978494623655914, + 0.9851851851851852, + 0.9923076923076923, + 0.9888059701492538, + 0.984, + 0.9796610169491525, + 0.9786476868327402, + 0.9831081081081081, + 0.9824561403508771, + 0.9925925925925926, + 0.9743589743589743, + 0.9894736842105263, + 0.9779411764705882, + 1.0, + 0.9822064056939501, + 1.0, + 0.9823321554770318, + 0.9920634920634921, + 0.9700374531835206, + 0.9854545454545455, + 0.9924812030075187, + 0.9745454545454545, + 0.982078853046595, + 1.0, + 1.0, + 0.9793103448275862, + 0.9823321554770318, + 0.9868421052631579, + 0.9831649831649831, + 1.0, + 0.980544747081712, + 1.0, + 0.9787234042553191, + 0.9813432835820896, + 1.0, + 0.981549815498155, + 0.9790209790209791, + 0.9851301115241635, + 0.9922480620155039, + 0.9855072463768116, + 0.9788732394366197, + 1.0, + 0.9863481228668942, + 0.9866220735785953, + 0.989247311827957, + 0.9852941176470589, + 0.9838709677419355, + 0.9920634920634921, + 0.9819819819819819, + 0.9826989619377162, + 0.9908256880733946, + 0.9908256880733946, + 0.9858156028368794, + 0.9863945578231292, + 0.9797297297297297, + 0.9922480620155039, + 0.986013986013986, + 0.9834437086092715, + 0.9915966386554622, + 1.0, + 0.9792387543252595, + 0.9828178694158075, + 0.9781818181818182, + 0.9735973597359736, + 0.9824561403508771, + 0.9744525547445255, + 0.9852941176470589, + 0.9897959183673469, + 0.9924528301886792, + 0.9773584905660377, + 0.976271186440678, + 0.9858156028368794, + 0.9917355371900827, + 0.96875, + 0.982078853046595, + 0.9829931972789115, + 0.9754385964912281, + 0.9825174825174825, + 0.986013986013986, + 0.9828767123287672, + 0.9743589743589743, + 0.9832214765100671, + 0.9823321554770318, + 0.9888888888888889, + 0.9890909090909091, + 1.0, + 0.9821428571428571, + 0.9821428571428571, + 0.9866666666666667, + 0.991869918699187, + 0.9826989619377162, + 0.9866220735785953, + 1.0, + 0.9672131147540983, + 1.0, + 0.9833887043189369, + 0.9857651245551602, + 0.9797297297297297, + 0.9803921568627451, + 0.9828767123287672, + 0.9738805970149254, + 0.9847908745247148, + 1.0, + 0.991869918699187, + 0.9859154929577465, + 0.978494623655914, + 0.9922480620155039, + 0.9893238434163701, + 1.0, + 0.9829351535836177, + 0.9826989619377162, + 0.9790209790209791, + 0.9816176470588235, + 0.9854014598540146, + 0.9921259842519685, + 1.0, + 0.9813084112149533, + 0.9894366197183099, + 0.991304347826087, + 0.9769230769230769, + 0.98989898989899, + 1.0, + 0.9923076923076923, + 0.982078853046595, + 0.9824561403508771, + 0.975, + 1.0, + 0.982078853046595, + 0.9825783972125436, + 0.9863481228668942, + 1.0, + 0.9805194805194806, + 0.991869918699187, + 0.9917355371900827, + 0.9818181818181818, + 1.0, + 0.9773584905660377, + 0.981651376146789, + 0.992, + 0.977859778597786, + 0.9688581314878892, + 0.9802631578947368, + 0.9790209790209791, + 0.9830508474576272, + 0.9915966386554622, + 0.9864406779661017 + ] + }, + "contact_maps": { + "mean_pearson": 0.06340988360421722, + "valid_tracks": 28, + "per_track_pearson": [ + 0.07055954824965306, + 0.0418391550801568, + -0.009086868990666956, + 0.041098115356746685, + 0.004703400863468791, + 0.04654886133669468, + 0.1048494983572494, + 0.07824843130034398, + 0.163696420920301, + 0.130058577101617, + 0.017508397662956855, + 0.010179232145272343, + 0.15482098230686275, + 0.028625269416162842, + 0.061192551387141, + 0.04650804552764992, + 0.047605377815484885, + 0.05524000247939563, + 0.08476781045654883, + 0.12901119214597445, + 0.01590472862149021, + -0.03868973396068581, + 0.002377071011704869, + 0.13511983977243444, + -0.014079493075419905, + 0.14991796726982912, + 0.09138800990873494, + 0.12556435045098052 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.27297327346382555, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..d9ecd3f95886891fde03395ee57ee7d8d29fb8aa --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S007/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5115438656826252, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6235245449531164, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.01236848782916468, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7862374337171356, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.27502152584489153, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.22434091660572703, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.26823243401441055, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.18452347424679816, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.20324228597800556, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2161136381307941, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5702911795385012, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.15275213525354572, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.21124736618139076, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..45bdf927f635baa77ab98cc8fcefbb60196aa0ff --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S008/train_single.py @@ -0,0 +1,155 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/README.md new file mode 100644 index 0000000000000000000000000000000000000000..221c9768c38845d18415808c96c7f58ef93762a8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/README.md @@ -0,0 +1,90 @@ +# SeqFunc bridge03 at step 22,000 with calibrated splice probabilities + +One model with **13,281,416 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and four dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. This connection was attached with zero output while preserving every parent parameter and optimizer moment. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The initial compact model trained for 2,000 updates with balanced splice loss, then a zero-output local path was attached. At local01 step 18,000 a zero-output projection from local pooled features into the global trunk was attached, preserving all parent weights, optimizer moments and RNG states. This candidate trained the next 4,000 intervals with the same learning rates as the local01 control. The only scientific change is the added connection. + +This checkpoint records **22,000 updates**, 0 complete training permutations and +22,000 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup updates. The splice objective is `balanced_splice`; +other family losses follow the reference. Recorded accumulated training-loop wall time is +56.3 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +## Measured validation results + +Family-mean Pearson: **0.284329** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.473191 | +| dnase | 0.597237 | +| procap | 0.349566 | +| cage | 0.193042 | +| rna_seq | 0.210687 | +| chip_tf | 0.403283 | +| chip_histone | 0.387228 | +| splice_sites | 0.165330 | +| splice_usage | 0.166443 | +| junctions | 0.118197 | +| contact_maps | 0.063410 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5115 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6235 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0124 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.6878 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2599 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2001 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2216 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4021 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.2032 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2161 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5703 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1528 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2112 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..83531f9752357e6ee4a2d57f36e60487c3a97749 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/package_manifest.json @@ -0,0 +1,34 @@ +{ + "parameters": 13281416, + "source_checkpoint_sha256": "1f1bfe3f3ee43c0febfc85184bccda7d302344014af7331a9ba94a89f5972e3f", + "files": { + "README.md": "5d54487acd18813a515980873500defe09b3024cc3d9a438438702ca4f294eb6", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "58cb6a3db4d24356332b46e5c78bfa51e5b99dde77d59e197b2addc2ed4bb44b", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "64faa2391c4ef0abfd8b826692186777b725f1127ed04d921f8545970435e02b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "f6e13a3c7aaf110d6a7bfbbae719d08e1be4abb8d1c576a35ecbeb91465d73d7", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "ec77e3de38f4a9990992a3c27b2935a4692e24c07ef68790b1034ed21a0b8fd9", + "training_provenance.json": "4381e253305f50628dd9c60c3f1ea02c85b242379dded59d8acb38233d33ace8", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "validation_tracks.json": "9a74425d3471cd66e96eef79c572a75ae69566b9111c2625fa87e1f48908011c", + "validation_variants.json": "c2f97b5c81fbc06827a26f575fa63486dee35c427277f1d1d9f39729fbc8b24b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..050d412dc0c9c5085da6aa705840e140ec1034f3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/predict.py @@ -0,0 +1,131 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..c432425931d84585d994ee839351ff8bd59c8426 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/model.py @@ -0,0 +1,450 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + + def __post_init__(self) -> None: + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..30ffd5e8aad11fba7e5362c55cb47a05839ad65b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_config.yaml @@ -0,0 +1,30 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..7998d384cab5bbfefbe3c3a4a6770e379abb660e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/training_provenance.json @@ -0,0 +1,84 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "splice_context_mode": "offset", + "local_to_trunk": true + }, + "epoch": 0, + "offset": 22000, + "global_step": 22000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/bridge03", + "config": "/workspace/research/experiments/bridge03/config.yaml", + "resume": "/workspace/research/experiments/bridge03/start.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 22000, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 3378.2578727870714, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "local_to_global_trunk", + "parent_sha256": "cd64f7ccf50d6335cac129447dd451d5b829fc7d3857f780985b47aab7101842", + "step": 18000, + "initial_outputs_unchanged": true, + "computation": "average aligned local features into 1-kb tokens, zero linear projection, add before transformer" + }, + "interventions": [ + { + "kind": "local_to_global_trunk", + "parent_sha256": "cd64f7ccf50d6335cac129447dd451d5b829fc7d3857f780985b47aab7101842", + "step": 18000, + "initial_outputs_unchanged": true, + "computation": "average aligned local features into 1-kb tokens, zero linear projection, add before transformer" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..9299ea3dcd6715dbf2d47971db6739b34bd8ff72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.47319080337453506, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6013569800083834, + 0.611230826606813, + 0.3715117945868288, + 0.5511424457436707, + 0.3629125711945762, + 0.35654874664890407, + 0.36989550633852447, + 0.4451403188891816, + 0.4947862642058731, + 0.5487417968153977, + 0.4774852488126404, + 0.41718372668167625, + 0.2771671525279424, + 0.4122614102308198, + 0.5987590328673628, + 0.3767810748988027, + 0.4488039383182791, + 0.39130416252459943, + 0.4435090578658996, + 0.4298864166820064, + 0.3973432022754257, + 0.3554749768779291, + 0.40861399186658753, + 0.42548698014389397, + 0.3707584577037519, + 0.3872132348892331, + 0.37283706711057535, + 0.3986515152361764, + 0.40455104366511874, + 0.3923066605898128, + 0.39141428506747583, + 0.39637208561282855, + 0.38375193312457767, + 0.2956162453517259, + 0.2707746425137148, + 0.45799869189698456, + 0.38641390142512944, + 0.5066500236423994, + 0.40366001627091774, + 0.37660099848628786, + 0.4572512841355684, + 0.4181203415320314, + 0.39404308164317814, + 0.3507138608420332, + 0.45565305972080206, + 0.40184392431118315, + 0.3167246937828745, + 0.3673306786328815, + 0.3574736644435944, + 0.3673296167848582, + 0.396687430845057, + 0.32150547295418175, + 0.36939499418081473, + 0.5108054593493221, + 0.34433082849927765, + 0.6079266511064885, + 0.7186509219901122, + 0.4310927297712832, + 0.5223282010372912, + 0.5154275846996749, + 0.6272784430785567, + 0.39121889290288936, + 0.6001879926507501, + 0.5247495992133304, + 0.5515549111245756, + 0.39667024568334003, + 0.4658759544967921, + 0.5514952894121927, + 0.7009484843709354, + 0.6313365483181583, + 0.4246587501443308, + 0.3565040133893579, + 0.3107802241412696, + 0.43051099540251525, + 0.4102547716876519, + 0.35768142507088035, + 0.4032784479579138, + 0.4321671457459026, + 0.4742687939135322, + 0.37396404968046915, + 0.4332545753673459, + 0.46714822386209137, + 0.4118492414916431, + 0.3865321575640476, + 0.383946248458017, + 0.3816749414612892, + 0.3416790004727847, + 0.36361445838940115, + 0.3500184534620256, + 0.40721863366147504, + 0.4020352585517828, + 0.39700860942740923, + 0.39113497032494526, + 0.41366239134494975, + 0.4772287489897358, + 0.30033088130719665, + 0.3383490827593295, + 0.31143092806558553, + 0.3778968124153396, + 0.4069560958668667, + 0.341861941040783, + 0.3697940377472689, + 0.3978033847241105, + 0.3698657879295328, + 0.40839660875640604, + 0.4238325148536602, + 0.4412739771825903, + 0.4156108454041656, + 0.4174085462806695, + 0.42613168444285493, + 0.45423136217540483, + 0.4170674866066782, + 0.40363817539378155, + 0.34893054584432914, + 0.35957748239549736, + 0.34656049725483623, + 0.36566289239636535, + 0.38980052534925397, + 0.38611041766618065, + 0.37413957744168813, + 0.5818546380753078, + 0.563705243521356, + 0.6320165947326114, + 0.6470620662935201, + 0.6175543342383838, + 0.5105765661712526, + 0.7827605746421785, + 0.616767870800456, + 0.6768357825582676, + 0.6000625205649728, + 0.6501728878353985, + 0.4869939789230486, + 0.5210728599984886, + 0.4921864960123308, + 0.6254844461382348, + 0.625155330767958, + 0.6118997088125304, + 0.45467764461857374, + 0.36739901904774885, + 0.7261627967362708, + 0.6920300637085525, + 0.7072323069791313, + 0.701714804551732, + 0.5234255364996532, + 0.7770716894879003, + 0.6897072248445472, + 0.6258463812198516, + 0.634514381019221, + 0.7738630794166514, + 0.7216249736957288, + 0.5457252855143914, + 0.5397438566590256, + 0.6286178652069961, + 0.6582516097547308, + 0.6630960538524545, + 0.5334665889293448, + 0.6544289752094904, + 0.5113792613428266, + 0.5323458470644185, + 0.6075548880651679, + 0.48834282899917464, + 0.6816124634671773, + 0.7040198126527397, + 0.5480678727553623, + 0.5327971510655569, + 0.48812922922221164, + 0.6582668828985359 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5972369929990613, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5051464287081975, + 0.9040064855485875, + 0.7599811893591262, + 0.4582454837933557, + 0.4109466902566876, + 0.4282630957048445, + 0.9273840097753159, + 0.6844059007426798, + 0.36179562333821996, + 0.2991670393230131, + 0.8596535558019618, + 0.6255074544944994, + 0.3835047011415663, + 0.4475894437042361, + 0.42776923204165274, + 0.7792464508875344, + 0.7621453968695244, + 0.8376713047993213, + 0.8350132800042932, + 0.8704146439884979, + 0.5957969832543013, + 0.5636376786924457, + 0.8852082197808491, + 0.4437107050837591, + 0.4667850479693932, + 0.7561056007461083, + 0.4088639237357805, + 0.6043946958625348, + 0.441650329571576, + 0.4509137079181166, + 0.5606220938593567, + 0.38999492057278295, + 0.8251877934266737, + 0.6096487582481648, + 0.8676531667546116, + 0.8741266199797021, + 0.3935250488964507, + 0.8105418627116354, + 0.8649676597872599, + 0.8454100659201653, + 0.8054168946425659, + 0.6552472662005258, + 0.6903586013830724, + 0.5941764682151254, + 0.4176149449829323, + 0.6186939017212982, + 0.7156464377785566, + 0.48030916764969556, + 0.5477641871527885, + 0.3647482031464813, + 0.3242449193254172, + 0.5135392352608715, + 0.4423417009864615, + 0.8176287365262694, + 0.3750288324687304, + 0.36804940895333277, + 0.38746978962430406, + 0.4178972037414227, + 0.3819109734539283, + 0.4890929571106253, + 0.36497796032909285, + 0.5790922178718865, + 0.4254310778727347, + 0.41174308083837663, + 0.43794310806046866, + 0.43438309371094, + 0.49828787718542183, + 0.4145770307030255, + 0.3341421657411775, + 0.36514191508802035, + 0.47767111774696464, + 0.5802995322875978, + 0.4222640292641831, + 0.5688609945051096, + 0.28741977706449945, + 0.7918288677141055, + 0.882469864869311, + 0.6108961582875497, + 0.38534732227108115, + 0.39782713145030535, + 0.44307138960852704, + 0.420605836856261, + 0.44837179589245446, + 0.4208249971577822, + 0.4625959455470755, + 0.4427998426757078, + 0.529318240942104, + 0.45660677533363137, + 0.5944018177367659, + 0.38336093131765275, + 0.5775486063549733, + 0.37546356357072763, + 0.6963015784915666, + 0.41902209540161983, + 0.6844451491141329, + 0.4545061710880882, + 0.6900918630189271, + 0.6428855550251171, + 0.293290834356576, + 0.3008796951404523, + 0.4126632650125963, + 0.3324968932579727, + 0.31077711447169704, + 0.6589489448396658, + 0.6308945006525427, + 0.38473305837399613, + 0.4045527490178858, + 0.5474814678906212, + 0.8278231861446294, + 0.5131966266032941, + 0.4801410748049326, + 0.5362578354459829, + 0.448644057040946, + 0.4114133228883751, + 0.755352503683002, + 0.4837671099803786, + 0.45505089204739885, + 0.5156146869702245, + 0.5651765907728481, + 0.4103925714153132, + 0.46014608673376345, + 0.46339305106014644, + 0.49231513037860936, + 0.4698185612252677, + 0.5615131813856594, + 0.48954577806080457, + 0.5386860834954825, + 0.6477177673394762, + 0.6705915045152536, + 0.5467838696385618, + 0.426599420947635, + 0.5529368612616599, + 0.5803306875182386, + 0.7651446309087153, + 0.39893844677282575, + 0.2743336068438853, + 0.5294721007034671, + 0.4359647461447, + 0.5672857701576846, + 0.5444082964186903, + 0.5499912851457619, + 0.5063472046332705, + 0.41080357615963864, + 0.7554115634397447, + 0.5583453732596939, + 0.7414336257705796, + 0.3536230260772644, + 0.3249113742826109, + 0.37039597149822867, + 0.29423287934749665, + 0.88132164269304, + 0.28431343057004727, + 0.8807689000329385, + 0.6583149424846829, + 0.4451226130093942, + 0.30109285591893054, + 0.31333738774845815, + 0.3330717852363559, + 0.6050946760578954, + 0.648044156042965, + 0.6701066659949098, + 0.5936444055192259, + 0.48388760797934244, + 0.41592817721766284, + 0.6422585664210787, + 0.5276855471139255, + 0.4732631534557596, + 0.5577892459322403, + 0.4185388347652587, + 0.42768424367024294, + 0.33851642107277624, + 0.8163129560013446, + 0.3817916183884341, + 0.41314913537824083, + 0.611175842116569, + 0.666912040667777, + 0.4197770844139869, + 0.3062577325540888, + 0.8667889010851785, + 0.3365176477705042, + 0.7377808207526263, + 0.38109029315600396, + 0.5592363040089487, + 0.6774969816395758, + 0.4320291351140382, + 0.5215552976508007, + 0.5249196656935252, + 0.8009589898539154, + 0.39389106354042475, + 0.5680090424161645, + 0.5033003335480184, + 0.6813565194267117, + 0.8841505204189964, + 0.3506400083925581, + 0.5325796479312582, + 0.4043093767670347, + 0.5785724563427923, + 0.6728914435525277, + 0.5341613937447999, + 0.39854396560277844, + 0.8564889204424364, + 0.8652151754258923, + 0.7397854661872288, + 0.8193662991615421, + 0.8298077991630294, + 0.8177391213313542, + 0.8283264169989877, + 0.3188909479179335, + 0.32877560999580585, + 0.4620541334955505, + 0.4149822659272492, + 0.3948729336334369, + 0.5563044485609598, + 0.8469084551090093, + 0.8358264980735723, + 0.6295026799099597, + 0.8091105357000382, + 0.8059829508556513, + 0.8430430196823874, + 0.8913996453225206, + 0.43459661165513114, + 0.48746767119188505, + 0.6012632013997647, + 0.5026447324691392, + 0.799817654423142, + 0.5147789093087264, + 0.8921409336842042, + 0.8668716960065588, + 0.8600227186024813, + 0.887652780096815, + 0.8803558092006152, + 0.8541092476886623, + 0.8129857038387429, + 0.846345321319629, + 0.8905593228324823, + 0.38094692496207555, + 0.9103627241671178, + 0.8642856701590852, + 0.8037080681896945, + 0.5381983311450295, + 0.859435687149562, + 0.46015357302901805, + 0.7375771800279055, + 0.8613547450367912, + 0.5155028674006044, + 0.502354220313448, + 0.6063524220351131, + 0.48228687914777546, + 0.4760801733454178, + 0.7006585705784053, + 0.8672315883091891, + 0.6525923471644431, + 0.7882485508835309, + 0.5580066933923811, + 0.8473988596265536, + 0.7891803935967545, + 0.8210927720996523, + 0.8892529884044479, + 0.8201358099387646, + 0.8850031507160516, + 0.8541049896479801, + 0.47977083890399985, + 0.8646450033021486, + 0.4852748216758936, + 0.46854127571905346, + 0.8676179803741704, + 0.893574462976209, + 0.8646620071272132, + 0.5065920085440799, + 0.5526710469698969, + 0.6220205823454538, + 0.7798607193634521, + 0.8717324754528205, + 0.6591096976105665, + 0.8883877911929138, + 0.9008138101555477, + 0.5365514546963454, + 0.2722332113247191, + 0.4684903468825709, + 0.8867388934593328, + 0.3664076369113273, + 0.5493782130940472, + 0.8652720466341627, + 0.8153334238124229, + 0.5481068119649307, + 0.8526252929441862, + 0.8811922053126112, + 0.8753638559991634, + 0.8626250994782892, + 0.850578090406638, + 0.8808970538598544, + 0.8378385714436191, + 0.9125621801517521, + 0.768394542664312, + 0.8179519849912746, + 0.8972138453831112, + 0.49051167546900615, + 0.49740247811967714, + 0.4703383810229919, + 0.46031192818372796, + 0.895503977384706, + 0.7800539995521896, + 0.8345923373631745, + 0.772414543119302, + 0.7441019241144452 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.34956550154905325, + "valid_tracks": 12, + "per_track_pearson": [ + 0.342583975020021, + 0.3844080157361664, + 0.29193647947308327, + 0.3828237890532391, + 0.37945738757018704, + 0.381257042275856, + 0.31919838821547963, + 0.3690742289657949, + 0.25954457337855685, + 0.3622746990207944, + 0.35421949895828836, + 0.36800794092117173 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.1930421274386517, + "valid_tracks": 546, + "per_track_pearson": [ + 0.19955618236466635, + 0.2245851065309552, + 0.1906140466539765, + 0.11951416133946903, + 0.0716866516781868, + 0.20025783154930754, + 0.10037635762241008, + 0.1290160018314467, + 0.20829654159378191, + 0.18292753761212366, + 0.23189861145548754, + 0.1293719330711211, + 0.13300535977251393, + 0.1119465372521382, + 0.08104786075745787, + 0.15598731738622787, + 0.19037613022233976, + 0.22864570215507782, + 0.22357622681218245, + 0.21472609149540284, + 0.18210673807964625, + 0.17031013750146534, + 0.2848138823306632, + 0.15294705019829297, + 0.16193296521188438, + 0.19718722927556412, + 0.219204721982228, + 0.2227202508038051, + 0.2441571893806723, + 0.22360630264566606, + 0.14062909172566893, + 0.23492936246369364, + 0.16877043942280848, + 0.11739686111775059, + 0.24415030938101737, + 0.2348478616097727, + 0.21415928375288093, + 0.24245726404614198, + 0.2053248672442707, + 0.2787880612253706, + 0.285961912752194, + 0.24728958610731439, + 0.21593821565400537, + 0.21190599742150745, + 0.1223972532814035, + 0.17640524857307918, + 0.18910935502665063, + 0.1880088699300437, + 0.11594512941425876, + 0.14800237401984495, + 0.20780401929029935, + 0.24429556581660122, + 0.25341979159673284, + 0.17342939582674777, + 0.2622484163779034, + 0.28891412424029267, + 0.1570630463925638, + 0.2675609976918749, + 0.20682275179151005, + 0.2290441286062082, + 0.2693270075763393, + 0.1639657568677222, + 0.2403702534166539, + 0.18225634324556567, + 0.25423228258622793, + 0.26328641570140526, + 0.27889333346678663, + 0.16187118135793882, + 0.11642732700216267, + 0.10168264688455979, + 0.21642774683829125, + 0.2526590736579851, + 0.2083910526237493, + 0.22113146778662718, + 0.2399838725370858, + 0.20165659037002168, + 0.2416771199727865, + 0.2193273515044356, + 0.24353355232011278, + 0.24962053390215755, + 0.20582012536330974, + 0.12428454812010258, + 0.22064660984707377, + 0.2554397684879941, + 0.12383663241130934, + 0.24047680967864232, + 0.19696293133800763, + 0.25954330111492685, + 0.1646362641604071, + 0.248473427888448, + 0.17820410538443254, + 0.20680377428438632, + 0.2286706953421968, + 0.10708699098803119, + 0.23800897740504456, + 0.20750676045035568, + 0.21857938108542418, + 0.23186613913832116, + 0.2149646462225257, + 0.13172744340793444, + 0.20581948056690919, + 0.22398507696229347, + 0.2252351568745013, + 0.3069988078546611, + 0.2609979962253995, + 0.27215494314249716, + 0.1784716044138763, + 0.23194510661697834, + 0.1825735102567508, + 0.1838100830224675, + 0.20076291017660391, + 0.20187855083963427, + 0.22560725842707285, + 0.22343612827199444, + 0.1758400003375151, + 0.22128539346918824, + 0.22728320496149237, + 0.2330519488477772, + 0.21657221763956533, + 0.25148018125348315, + 0.2555094041264715, + 0.25183577099560106, + 0.15513301241221117, + 0.17094543723450298, + 0.19715695401570896, + 0.2444161114138086, + 0.21894036327244218, + 0.24127217729770853, + 0.31250343670032726, + 0.2281041432419601, + 0.2529502823535138, + 0.1787240442624391, + 0.24145395585756999, + 0.24430226006019642, + 0.28917584617849645, + 0.21126221124051564, + 0.25914341311574585, + 0.18142444526737173, + 0.10938153394964355, + 0.22352646997898176, + 0.269758533272295, + 0.2277954384825295, + 0.2396377646024378, + 0.19467360387255214, + 0.1967196566602217, + 0.20977289051661838, + 0.21746376908686005, + 0.2569152089717521, + 0.18938931460634598, + 0.18461558623629612, + 0.21868509891315718, + 0.16327357924567087, + 0.20432392176753209, + 0.23074344965755433, + 0.20571438463735295, + 0.1434080826334781, + 0.23437116626980645, + 0.26626433271877353, + 0.17205608527557542, + 0.13949471972950242, + 0.13854184120732296, + 0.19648330106152118, + 0.18934609621553128, + 0.21681719351473444, + 0.19043645532908088, + 0.15322090448533512, + 0.21480955090218962, + 0.14669399688260568, + 0.14213634067898812, + 0.23704016754245244, + 0.2173980025534816, + 0.17030858732263826, + 0.19108415002232412, + 0.29457669931368624, + 0.2189601328703026, + 0.13516430564307283, + 0.13757918027108978, + 0.1731134405076223, + 0.22639035373641483, + 0.26735095489658195, + 0.2532353416806349, + 0.2559319853674132, + 0.11925594043881045, + 0.2539828242386957, + 0.2067763505350997, + 0.17015366270549406, + 0.17369873830098648, + 0.14705560774908408, + 0.27114328905692453, + 0.22755827292619055, + 0.18772844615217213, + 0.11819317566683311, + 0.12207927922301387, + 0.1979710265534275, + 0.1942355428537205, + 0.179058434359934, + 0.20576985380615978, + 0.20780967772876976, + 0.15661604292754736, + 0.19761761717375464, + 0.11439102371594093, + 0.09172182285580623, + 0.10794562835667958, + 0.09844199617482663, + 0.12207397518894834, + 0.2024037787964843, + 0.19824996677514606, + 0.17097143824082228, + 0.1043553256741376, + 0.1953569559853853, + 0.25541336997753067, + 0.2175752506044569, + 0.22263040664792957, + 0.22152635108910243, + 0.2060019450341039, + 0.20224320979718538, + 0.238336676383899, + 0.2715927879817231, + 0.24163357620210454, + 0.23189378387302312, + 0.21050946448414087, + 0.15280564332654661, + 0.21527593226421704, + 0.21552022405464866, + 0.2251058712650847, + 0.27924014056764596, + 0.20643188461278014, + 0.17662020144158266, + 0.20119456142269115, + 0.21466291991824338, + 0.15966947006622842, + 0.16110431181711435, + 0.2612575727660771, + 0.2055300221795156, + 0.17364458162206528, + 0.20686104655279364, + 0.27197083191510446, + 0.14013007621162973, + 0.2149533122424987, + 0.21270512815860365, + 0.10055703109546674, + 0.14241734333777592, + 0.12612836512019504, + 0.20478549930262918, + 0.15352583706024817, + 0.18837850340996026, + 0.08467824683098811, + 0.2592552942006553, + 0.22157731568349606, + 0.22762131864846716, + 0.24346883035670572, + 0.15404978292659008, + 0.2701643171315074, + 0.12214413795266454, + 0.24504491594863834, + 0.19062132828906744, + 0.3083329188371513, + 0.22502380211145376, + 0.22922875465815074, + 0.33394881420820727, + 0.19179240133829875, + 0.21624488215851526, + 0.11541580325088342, + 0.23132427439483821, + 0.214086391475855, + 0.14931679256637584, + 0.11676441484666773, + 0.11960837828823204, + 0.14063277001538624, + 0.2235938525227398, + 0.2738007568155193, + 0.12730627350677323, + 0.10967376758419962, + 0.17123943984457055, + 0.23331538203003133, + 0.18066746689465368, + 0.12493377053823768, + 0.060626055660009975, + 0.1886127195963779, + 0.1127968158577476, + 0.12603841745859826, + 0.2009681734401198, + 0.1721756478827082, + 0.21031606659027002, + 0.12192612905823681, + 0.13033808670638075, + 0.112967935979915, + 0.10261511815382761, + 0.14509570045485368, + 0.1793511212054257, + 0.20566691894935918, + 0.20484299227447061, + 0.19476868635409497, + 0.17194358089500858, + 0.16561291429595273, + 0.2618059167331183, + 0.14209794756609123, + 0.148682272989394, + 0.19901002219126912, + 0.199686964635623, + 0.20075433484920283, + 0.21966909892346448, + 0.21024669204088284, + 0.12777133207356992, + 0.20342777102397716, + 0.1566188230061999, + 0.10881955060798904, + 0.21935099690299603, + 0.22705839112668472, + 0.20902241216748263, + 0.21649676559265021, + 0.1844738796568263, + 0.25475853000757986, + 0.2555679568318227, + 0.22426987199096812, + 0.20625214705744518, + 0.19678496161853362, + 0.11378162974721312, + 0.16851887397888043, + 0.1794292927324039, + 0.19362804966356748, + 0.10494977497839451, + 0.14029018763073925, + 0.17960076454089793, + 0.23225806571767704, + 0.233494924228106, + 0.16420083093165472, + 0.23523792303128405, + 0.25768327190118495, + 0.15620734123847094, + 0.23307464515353332, + 0.18666951177963714, + 0.20034454754019096, + 0.2407410481794789, + 0.15172753963916943, + 0.22210533457220052, + 0.17375982097472945, + 0.23362005123736518, + 0.22862028080970975, + 0.24935531176831602, + 0.14743629081189527, + 0.10491735502222808, + 0.09677217813160675, + 0.19780637955008618, + 0.23212643668801977, + 0.19730979773470322, + 0.20232138660561097, + 0.22508662932845544, + 0.1897000047762809, + 0.20838729793563093, + 0.1970385205456463, + 0.21811927733941863, + 0.2352480210902977, + 0.19410717470310596, + 0.1189734532416716, + 0.2080173704055007, + 0.2292118331502473, + 0.12334352298874772, + 0.2163273038389398, + 0.20605070740501907, + 0.24215672399611823, + 0.14927802076562657, + 0.23706021207574515, + 0.1715612713049819, + 0.18627184005201453, + 0.21161046596990407, + 0.10496554039893093, + 0.2147092832884699, + 0.1853337576362698, + 0.1918873892702912, + 0.22871068690716143, + 0.19421375670596072, + 0.12272406126738189, + 0.18953609578029199, + 0.2172783819633657, + 0.20134881363396576, + 0.27941653708768055, + 0.23358685592148376, + 0.25531462531335, + 0.1632142051259845, + 0.2123336988092235, + 0.1674364037599529, + 0.1630164128931391, + 0.1810279959887571, + 0.18258332030623586, + 0.20850012888284364, + 0.21250427876083683, + 0.16218625012532525, + 0.2016781976780125, + 0.20065655608866947, + 0.21302797030413764, + 0.19473153979992913, + 0.23629712053266183, + 0.22621991228750843, + 0.2240139154463825, + 0.14655341801508978, + 0.16374173493175817, + 0.19239369761353511, + 0.21859393925453247, + 0.1971370880008696, + 0.2285387628149081, + 0.2900852075180928, + 0.2063792957287801, + 0.2343402512927579, + 0.16850836967713792, + 0.2076232348549899, + 0.201951880705826, + 0.2580870323222759, + 0.19958033749117485, + 0.24031215377050708, + 0.16456130533946295, + 0.10340666535876968, + 0.20608440024060196, + 0.23940797919729506, + 0.19968963674619064, + 0.23238891271536782, + 0.17207355637722102, + 0.17473275398823412, + 0.18258570065767812, + 0.18968622119411174, + 0.2170389832343234, + 0.17712191931338844, + 0.16301564533708093, + 0.18251457967063867, + 0.1387876320742056, + 0.17313595060050904, + 0.19260091985389693, + 0.1798956734315381, + 0.12227081444669484, + 0.20249063391365618, + 0.22682698961280418, + 0.14832904003720193, + 0.12892554061519954, + 0.12474500581917501, + 0.16753084589845157, + 0.16642508194830832, + 0.2077916621658573, + 0.17827902021485317, + 0.13599086017698916, + 0.19775649820145771, + 0.1454189439426982, + 0.13140169716904124, + 0.21835067018717585, + 0.1997999715434767, + 0.16947308794143306, + 0.16320093138381422, + 0.2749078621764768, + 0.2046673901516236, + 0.12147220735963989, + 0.12952226949870124, + 0.17929510759939477, + 0.18094673445655726, + 0.2644995811803432, + 0.2320883303586097, + 0.243015780576851, + 0.1038970953813028, + 0.2338640887573672, + 0.1959086771130411, + 0.17807689669360646, + 0.1494378244957304, + 0.1383410467176855, + 0.2273335616884033, + 0.20629804962355866, + 0.1795540467029413, + 0.13077631014904922, + 0.09642531231287918, + 0.14361762037053938, + 0.1950530871851025, + 0.1651629441744796, + 0.19854373282630458, + 0.19198474946311822, + 0.14960335690477974, + 0.1623561953309846, + 0.09133150575983846, + 0.06796169360456275, + 0.08966105895981427, + 0.08578318492434432, + 0.08503408355196798, + 0.19150913542242312, + 0.1550005799577552, + 0.13576421075062123, + 0.09549847164886092, + 0.18042650506656002, + 0.24968001632266265, + 0.22042201769957373, + 0.21829483221999305, + 0.21798487166886774, + 0.2054288735921829, + 0.2004250641775354, + 0.22990929179678193, + 0.26431086969676126, + 0.23740435753391906, + 0.2251821052454753, + 0.2093256739309808, + 0.14436875922429873, + 0.21472511225645538, + 0.20364162842713399, + 0.22080772908208804, + 0.26646300500864495, + 0.18738833547214462, + 0.1717388180437895, + 0.170876901083507, + 0.19356766323301192, + 0.11533845323253158, + 0.10630562174127979, + 0.244295021788761, + 0.20010240134702367, + 0.1685937116429196, + 0.19241547633858896, + 0.24983106654931073, + 0.14192165761756045, + 0.197109878275414, + 0.1905447702272679, + 0.09063581202804785, + 0.11438744072178621, + 0.11279196745579735, + 0.2004464301879699, + 0.12436890271130757, + 0.18001225104356305, + 0.08164948799503802, + 0.24882058706442794, + 0.2176918178298283, + 0.21123938301555756, + 0.23459326274031234, + 0.1612442217341263, + 0.23457697986721, + 0.11465443090398364, + 0.21236737999764738, + 0.18950088224694445, + 0.28954152336082994, + 0.21785846933931405, + 0.22755544912908476, + 0.3189009165846459, + 0.18917542760114403, + 0.19112308000662492, + 0.10554739598671549, + 0.20922997340654872, + 0.19586396319860885, + 0.15963747010154228, + 0.10666468583134212, + 0.10817336276834513, + 0.12989475271851386, + 0.22073215896115628, + 0.26270337259704996, + 0.1189918650410344, + 0.09150785338667813 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.21068735878499603, + "valid_tracks": 613, + "per_track_pearson": [ + 0.2396121765083079, + 0.21742617378297072, + 0.1944567561852619, + 0.254575053872246, + 0.23684763819818888, + 0.19535534380279307, + 0.22150627243633472, + 0.2367186297400461, + 0.2433750676995738, + 0.26769853996768667, + 0.18358557052842428, + 0.2297276275925011, + 0.15200581387533388, + 0.2795054582754713, + 0.2679380292372057, + 0.2767658563931556, + 0.2684881806686551, + 0.23572991964052437, + 0.21572088145498813, + 0.2452829304031348, + 0.20615403998670326, + 0.22586807046619314, + 0.2312258900009004, + 0.22639425233225888, + 0.16398057072939187, + 0.21394235009659757, + 0.16834604491942048, + 0.22094684036348278, + 0.2297318506671091, + 0.2382664267822145, + 0.282093980760062, + 0.21824816842758085, + 0.2980225758085653, + 0.19369530211423946, + 0.30112750744964956, + 0.16105423554419265, + 0.23865509265207965, + 0.23226581513267025, + 0.20721583934470458, + 0.23332193820083144, + 0.25998874580978554, + 0.20883027935329718, + 0.23483883128481073, + 0.23147560089067842, + 0.2594532928340572, + 0.1931830854508425, + 0.22776281937048637, + 0.27661515694708855, + 0.2518365315488076, + 0.21299805647334466, + 0.23810849523606784, + 0.23775063710614092, + 0.2327442468915728, + 0.25648832480562644, + 0.17232725711663466, + 0.24676513170241396, + 0.2399492667391458, + 0.24233208166004017, + 0.2398456194331544, + 0.21888488699445308, + 0.19363327561891383, + 0.2197264727613775, + 0.21967774427839767, + 0.19296804062558154, + 0.2398228662689071, + 0.21202176408134799, + 0.24326636193248322, + 0.2276154601960646, + 0.23587188027974534, + 0.22696137601723868, + 0.2268604291025291, + 0.23999693908879957, + 0.26166643864182254, + 0.22987937658134236, + 0.24423419798657708, + 0.2580626719438884, + 0.24237900162529127, + 0.24268873311819386, + 0.28560593958505337, + 0.23688032232891254, + 0.22388578383581353, + 0.20838984444022549, + 0.275448047083708, + 0.2556824027837078, + 0.18333253950484757, + 0.2041759956016737, + 0.24907237965010373, + 0.2561179220420642, + 0.23944771002761514, + 0.2550131960249841, + 0.18070976203383124, + 0.19654134576498722, + 0.22718084012261158, + 0.2312000354156662, + 0.24182485155566963, + 0.24120432898888633, + 0.22533058965167574, + 0.24064158534797417, + 0.2368558375102774, + 0.20890667634578078, + 0.21491418749918528, + 0.24650354800844246, + 0.22818480934926258, + 0.2246922352444411, + 0.2413418449743371, + 0.2712404747622371, + 0.23100784464485552, + 0.2833208367561689, + 0.26264140019435095, + 0.24015098968322632, + 0.20768413896127005, + 0.32194412084059254, + 0.29546219913163785, + 0.1821737199553992, + 0.23285778089682252, + 0.23655721451746697, + 0.28936148797365857, + 0.25665826813456777, + 0.2559447151471447, + 0.2333183769595399, + 0.1796537171549368, + 0.24046227451132815, + 0.22964150280847279, + 0.2673902348007867, + 0.26197450560995683, + 0.23915466286622994, + 0.252008749320876, + 0.17772643214523284, + 0.2201441335091361, + 0.20977221698840856, + 0.3049221692291915, + 0.250447980301304, + 0.21166412680704774, + 0.16138561473451332, + 0.18184371307022465, + 0.23233064129971778, + 0.26919789199826155, + 0.2860277684856349, + 0.276692213051496, + 0.271507581406338, + 0.24009842220841893, + 0.194841338167427, + 0.2885492434870452, + 0.29887982978716804, + 0.20960534417637003, + 0.21362776043940976, + 0.18262286043520948, + 0.22662755279441635, + 0.1707560726356683, + 0.24829822745635857, + 0.1870056160156746, + 0.25203087741725183, + 0.2048022856632488, + 0.20739205030262217, + 0.2664608907177063, + 0.1946522145008973, + 0.2322629117760048, + 0.2703644990948379, + 0.20863642437508462, + 0.2693442843927822, + 0.30580198596520913, + 0.2285395245166197, + 0.2563225553022507, + 0.23868846437540917, + 0.23677481235409886, + 0.23271524447290873, + 0.23781601107491457, + 0.21528687922868847, + 0.2257624071125744, + 0.2556584685932397, + 0.19223862346705495, + 0.20083319645106373, + 0.24283631735561187, + 0.2491218384600011, + 0.2123444531165577, + 0.20402801721174701, + 0.2148556638787894, + 0.2706074093104324, + 0.2703990316004476, + 0.1884731220812994, + 0.23851781435715202, + 0.1967824688574875, + 0.196867582190135, + 0.18461444272206487, + 0.1974398985239025, + 0.24374787551908708, + 0.214119682910095, + 0.12151660611755885, + 0.1772349342447161, + 0.3045530922195927, + 0.3330761009951277, + 0.18843976472568602, + 0.17238046416207317, + 0.23342211327961576, + 0.18283849993141846, + 0.20338472080518116, + 0.20622876172739796, + 0.21200420056523495, + 0.17141839207349277, + 0.22870845921249466, + 0.29506648687047166, + 0.17561278279353268, + 0.3490942778420528, + 0.2534191923053743, + 0.20037642547523593, + 0.21195315810070503, + 0.21007168994448486, + 0.17190136928954503, + 0.20099208639618668, + 0.18281030286826808, + 0.17313816955276187, + 0.17244965779786803, + 0.17117933719298442, + 0.18653720606159763, + 0.192521441277054, + 0.22529466700934933, + 0.17375644512176383, + 0.20108448367591877, + 0.1790061038321579, + 0.1946794607887882, + 0.3273645418040744, + 0.2002949474044493, + 0.2852787904202679, + 0.2501894560621998, + 0.2533703368637356, + 0.2887573660780674, + 0.2522001556724362, + 0.27617541911277627, + 0.20709241900468198, + 0.19056574466802445, + 0.21985571710244656, + 0.32198004916898393, + 0.20697530255325863, + 0.24524134663911898, + 0.21406913272487407, + 0.2248152865180252, + 0.14908726782446471, + 0.2418601886408755, + 0.2417359844129213, + 0.24006048900071816, + 0.2434246674189741, + 0.18262849096157655, + 0.22897952506688754, + 0.19100740212689465, + 0.18841555369796706, + 0.25061308533155396, + 0.30286748503933314, + 0.2261625206690557, + 0.22606323181235918, + 0.2050862316355344, + 0.2071110392001243, + 0.22900196669281955, + 0.17672143522770808, + 0.19582809129715484, + 0.23046793093520151, + 0.2574757622680812, + 0.20693637081268618, + 0.21288306218997868, + 0.2068925127060573, + 0.19310843885473342, + 0.2548624210724129, + 0.22860102642608884, + 0.23694563574252922, + 0.26461461708732903, + 0.23320637548470363, + 0.21659852264832424, + 0.20367958397660357, + 0.1840719495241105, + 0.20836599342321216, + 0.23067636592492444, + 0.22311442979349927, + 0.13863875346552393, + 0.18963945510777586, + 0.16947804383676757, + 0.21205036875133312, + 0.189808732650034, + 0.13725770133359128, + 0.14759642363761702, + 0.20221053086186525, + 0.20912703815141837, + 0.2173380615581059, + 0.18087536086269024, + 0.1527787599226931, + 0.1430811234217776, + 0.1833549995454915, + 0.1797256781248388, + 0.17954418037326025, + 0.19699757318726205, + 0.19075582476129876, + 0.20254800475122725, + 0.18296503795236554, + 0.1506383722070428, + 0.19866509609990443, + 0.15869156606611734, + 0.15024901895806045, + 0.1129645445908297, + 0.1785972050478994, + 0.16774912953219062, + 0.20045420891626753, + 0.17908277539656628, + 0.1829265735305624, + 0.2136096771906913, + 0.18948986057194298, + 0.16130454429989052, + 0.16727003896080456, + 0.2673356126431994, + 0.14305846966230235, + 0.19427774271755477, + 0.1881398475567588, + 0.1805135763782955, + 0.18377701781970773, + 0.19996268755228083, + 0.19323164197465414, + 0.20491250803614985, + 0.19016172612127363, + 0.1773994258016653, + 0.14046599827302514, + 0.1631112869587745, + 0.20052689359333514, + 0.17289292763136657, + 0.18185383581863152, + 0.19712411276027228, + 0.19268104932762184, + 0.15636174582751244, + 0.18138781783982133, + 0.11506241495645032, + 0.17941726248481382, + 0.1795920461836476, + 0.19381116940477022, + 0.19112163946563915, + 0.17280219734837649, + 0.16872940280544144, + 0.19097906655450664, + 0.19583504576106894, + 0.1374665302026543, + 0.20733873885242687, + 0.1904177857675568, + 0.21646750157223787, + 0.19856354226281353, + 0.208835661234966, + 0.19612692523218964, + 0.18721269506949248, + 0.2198269720819149, + 0.22076977089731356, + 0.2112525755309089, + 0.21177922778500285, + 0.22809095037084182, + 0.21261369480481956, + 0.21077930755754834, + 0.2144386134405248, + 0.18530649205010094, + 0.1750187852013466, + 0.1568775969384406, + 0.21670574204584223, + 0.22146628712603333, + 0.15968198438170011, + 0.18690363844017896, + 0.18830057110781712, + 0.17032355519595893, + 0.17337117061147486, + 0.15844887880459635, + 0.1319585454481103, + 0.16482356611742838, + 0.18334983543806588, + 0.17819140641773398, + 0.17925037356348403, + 0.19517655929210723, + 0.19428345669114797, + 0.18589418807423694, + 0.181905965305, + 0.1575775545560975, + 0.14951042719174726, + 0.1710743663203263, + 0.1670198988019873, + 0.20450920962087174, + 0.1710082796569146, + 0.19448516713035185, + 0.19125662294374415, + 0.20363123687084161, + 0.21578146532744738, + 0.180406236020863, + 0.15785708939264104, + 0.2482971008867558, + 0.21067152086712054, + 0.15736339567556792, + 0.20168782881630584, + 0.16195446733582153, + 0.21116004228609062, + 0.1778998328220946, + 0.1847212802474072, + 0.170290159160402, + 0.1617132386698006, + 0.1864898909598255, + 0.164266356874867, + 0.2155732467166363, + 0.1902176237885858, + 0.1814338047577488, + 0.17326172381343088, + 0.13802612102421205, + 0.14590933079606055, + 0.15505986107654013, + 0.24194000045823566, + 0.19093025622275248, + 0.14859596705446032, + 0.139145564654831, + 0.14347384639839045, + 0.21374881796481718, + 0.1997044700216238, + 0.22839490843870514, + 0.19337786937725346, + 0.1959857679426036, + 0.16221472302980922, + 0.15253352457710695, + 0.21648315543304456, + 0.22773296427912632, + 0.1464142941157321, + 0.14859407618407336, + 0.15163004188915064, + 0.18433495448223575, + 0.1684822786448065, + 0.18208005439808803, + 0.16461009204001023, + 0.19620383423265322, + 0.16182229026275252, + 0.14473083770550912, + 0.18316064903243945, + 0.1528025444412074, + 0.1969191999739611, + 0.1672756360841051, + 0.1892810280342141, + 0.18389624017123182, + 0.228681018177936, + 0.1552082900858097, + 0.17550395515673547, + 0.13862161849580135, + 0.14166779479320513, + 0.1376617405880239, + 0.13434026375348584, + 0.13041320283660665, + 0.17146212051269968, + 0.1959615111752819, + 0.19748467265425418, + 0.1611748127671079, + 0.23331072789928792, + 0.1938743535105406, + 0.15101891299973216, + 0.152097471697864, + 0.17657330157737408, + 0.2453448937918422, + 0.21747296820997444, + 0.14191250624506094, + 0.20349197163086796, + 0.19441016760204277, + 0.1228337579683726, + 0.1401350254675305, + 0.1397505279760856, + 0.19347284623473257, + 0.1839871207917317, + 0.11264875490121726, + 0.14569936060166574, + 0.26381644958539086, + 0.2712045621868274, + 0.18195646642981483, + 0.1485132427844615, + 0.19820188721317256, + 0.1572777491254758, + 0.17853907725231358, + 0.1755207315046603, + 0.19379183071710776, + 0.15889581024658997, + 0.19807788217395536, + 0.25589012144138834, + 0.1703298784099822, + 0.3101219624152767, + 0.20756127843037134, + 0.18608519855785882, + 0.18329671806610862, + 0.16845225251218804, + 0.12226318867198183, + 0.2089902311839013, + 0.14114086028577322, + 0.1682135963883387, + 0.1664071530499434, + 0.18717162482020727, + 0.1797471551197782, + 0.1816926636959218, + 0.21663345905512407, + 0.17984629627321147, + 0.1551850432130814, + 0.19031309994959075, + 0.18477539559989126, + 0.2578937457217044, + 0.19165581821658334, + 0.22709270975483437, + 0.16865194083152865, + 0.17569312658603578, + 0.22615164071210983, + 0.15526233698203076, + 0.22002783189921352, + 0.11348913897909996, + 0.18233681944698188, + 0.14480418543098078, + 0.277585157958009, + 0.20695648741782655, + 0.2061488578341504, + 0.20645334884263275, + 0.18740180391007627, + 0.10536420295091636, + 0.21442379486790789, + 0.20110176669556581, + 0.18046952325154253, + 0.18627455555273498, + 0.17406804969742898, + 0.21911679986038246, + 0.19268082568351735, + 0.19538393968875586, + 0.19969456286933068, + 0.2679184084514889, + 0.2051256639433557, + 0.18103312719249362, + 0.15284964952661714, + 0.15601608159728478, + 0.18220529674979896, + 0.15313591564345388, + 0.16925623229012043, + 0.19387441118334758, + 0.20417644892434283, + 0.12800184780926713, + 0.1791137575641985, + 0.1615155102292601, + 0.18103993700283588, + 0.19114189334254172, + 0.12866106939590163, + 0.1810396667377483, + 0.2108976577883292, + 0.19146141152756563, + 0.2100150012201171, + 0.195666072998692, + 0.11066772666738756, + 0.19779009452416904, + 0.19693210294623598, + 0.18182564364099124, + 0.19769578737881438, + 0.09637391987763932, + 0.26764218747141455, + 0.30446446799682964, + 0.276242692904721, + 0.25286069876456896, + 0.2603338619191649, + 0.2155885480047911, + 0.2171183317733866, + 0.24675236666787856, + 0.18868724496922976, + 0.20407076600134386, + 0.2119926849386165, + 0.1977987608584445, + 0.19749051837183182, + 0.20498380345654135, + 0.21434693713661712, + 0.1992828345796491, + 0.20756680415885137, + 0.18617662314694344, + 0.2671792620297316, + 0.26425626316134443, + 0.26793586498896005, + 0.2663253302318262, + 0.2948895539895203, + null, + 0.28190453872224563, + 0.33276625507327123, + null, + 0.07697944311120833, + 0.3982145889307656, + 0.07532189507976907, + 0.2556802653867343, + 0.24169190929312023, + 0.30515880072765605, + 0.36759301275291273, + 0.35892059536005005, + 0.37532408616510854, + 0.3733213391183945, + 0.3395109869564121, + 0.2937202798620642, + 0.22447903043015088, + 0.3791320066900833, + 0.18574955101437604, + 0.29746060183921974, + 0.2647369612909682, + null, + 0.25598352480866293, + null, + null, + 0.26071465172400105, + null, + 0.2327510721510598, + 0.20897446436404857, + null, + 0.30210191289295096, + null, + null, + null, + null, + null, + null, + null, + 0.25840323505801943, + null, + null, + null, + null, + 0.22009711427089065, + null, + 0.2621419025824258, + null, + 0.21174425784860085, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.2757079510119619, + 0.1596560387079658, + 0.24621238706992044, + 0.2582666374281978, + 0.1883175943675518, + 0.23697036341992464, + null, + 0.29963870237079027, + null, + 0.23992433614827022, + null, + null, + 0.24083536174711656, + 0.27191623156504563, + 0.23057949866637667, + 0.2372630376300417, + null, + null, + 0.2061132556747008, + null, + null, + 0.24516163028524446, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.23902287898014196, + 0.2367437751777455, + 0.25082122841447335, + 0.22273833244727878, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4032831995326907, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.20596341155157177, + 0.31166900861249935, + 0.2939791579030284, + 0.4443841464841782, + 0.36809501224478136, + 0.23737796352415597, + 0.3482634857125766, + 0.40196242731870024, + 0.3373857982846439, + 0.3473560692021163, + 0.39387727596187816, + 0.36967644850098647, + 0.26939005486089407, + 0.4072421392845433, + 0.3489199631786439, + 0.35359938624511805, + 0.38542605595474316, + 0.32053418232230635, + 0.4531285166082457, + 0.33688636485401263, + 0.3432830654339461, + 0.3272562883428359, + 0.3035656879911973, + 0.45724084346580396, + 0.3276200627745534, + 0.3561427284754092, + 0.2560117901813217, + 0.3651305639025781, + 0.35939150496132294, + 0.3795991945964458, + 0.3598846895117996, + 0.39026690377693585, + 0.4806625600179594, + 0.2736770399793426, + 0.6461584462471308, + 0.27410478160974017, + 0.20424129653690062, + 0.16165296320288125, + 0.32272539720252813, + 0.2167208756155665, + 0.26273926713205253, + 0.4531778079928414, + 0.29445324538473916, + 0.39207483422859807, + 0.28553513727525537, + 0.38595324505524514, + 0.29288528273751974, + 0.20144440713075107, + 0.4569253565158486, + 0.42331325836609585, + 0.379452085325689, + 0.35718020595678324, + 0.36828742932854625, + 0.34653130192815623, + 0.30894238211949415, + 0.37192297625695026, + 0.295219513939478, + 0.4137438301767917, + 0.39844170371295273, + 0.4200114816019427, + 0.2618396542127296, + 0.1964048785510467, + 0.3912144317818145, + 0.4106927721436892, + 0.2701164698418626, + 0.3356347929804215, + 0.33404977299981153, + 0.4371373856608021, + 0.4597767097886163, + 0.3800268132352417, + 0.3351257100420981, + 0.6757290561642809, + 0.3387234905854204, + 0.324990859351862, + 0.32395998564369494, + 0.3997987814315664, + 0.24859782647635237, + 0.6081306715497186, + 0.37646557543458214, + 0.3759411279700652, + 0.36350366415756197, + 0.39665847885892735, + 0.4681384282756416, + 0.41368538292376333, + 0.3028049431065893, + 0.26009442425108426, + 0.43504950195461933, + 0.38880357505869884, + 0.8002181485196034, + 0.4412224035304813, + 0.3375936900017678, + 0.32752004603080304, + 0.39025125653892684, + 0.31554989570837444, + 0.35594899630296356, + 0.5028975870250635, + 0.3146032875640386, + 0.31257874973084937, + 0.33276213277050837, + 0.2733768654610315, + 0.24220517709285058, + 0.3715479270101119, + 0.32820970470979066, + 0.3218994659075155, + 0.3680411128926598, + 0.1823741975110631, + 0.22324916284253518, + 0.24360020117954062, + 0.3750114974343165, + 0.31977406165756334, + 0.1873269835355793, + 0.26458890059262924, + 0.23949646945565314, + 0.31734440324605195, + 0.30759699200192214, + 0.4301524399086537, + 0.33828439934352444, + 0.3309831334545468, + 0.28035886210825256, + 0.34602193097313805, + 0.2764895022741886, + 0.3579701662880905, + 0.3317075067279533, + 0.21554883184776297, + 0.3114628871234857, + 0.2529070528654455, + 0.26924531983795175, + 0.25415418502562886, + 0.27499131843538477, + 0.5223349101193798, + 0.3165824539697752, + 0.21619819892279896, + 0.3522570870822998, + 0.5639617243164785, + 0.23771876388920576, + 0.2970189696977115, + 0.2693850541924695, + 0.27923503616136947, + 0.26231778936946104, + 0.23689384768493327, + 0.2263500639122292, + 0.4208934338733438, + 0.48854190731917124, + 0.2651991729735585, + 0.3517939048520881, + 0.4172655403343026, + 0.3823851414395783, + 0.21486937195901196, + 0.3654297758576896, + 0.28389829683647544, + 0.3277778933058627, + 0.2659522397106114, + 0.3271930694239052, + 0.4131951193647908, + 0.35564195255165393, + 0.3511688337318715, + 0.22170902392902786, + 0.2610321902191782, + 0.2407539320774382, + 0.3470807094985632, + 0.3112956131147154, + 0.28972812008082954, + 0.326417648559885, + 0.32643625138637195, + 0.3092553091313735, + 0.30968320048706705, + 0.2914754967447273, + 0.4143730720059262, + 0.21197510087825056, + 0.305779484734117, + 0.2758093250449337, + 0.22809151658309448, + 0.45450890701764546, + 0.24980944452349169, + 0.20574444969961195, + 0.36830341815722345, + 0.2151581892961229, + 0.31504871445187077, + 0.3033711443382587, + 0.25907367520180824, + 0.29979833340533296, + 0.3043749045763713, + 0.2653929117344394, + 0.34857375454841194, + 0.258311016817976, + 0.30847605664706873, + 0.36622514787072713, + 0.2228621174394303, + 0.2455356722523705, + 0.33743061809183933, + 0.21963713979983998, + 0.2845339685037795, + 0.39243637819212907, + 0.32511225277270595, + 0.25407325509762585, + 0.24722262918189847, + 0.2277821540285221, + 0.22604621552407364, + 0.227237579038627, + 0.39890018375473507, + 0.2605930336296522, + 0.29040731733621056, + 0.25788298610000354, + 0.25283723131158564, + 0.39127589904945437, + 0.27876734925172525, + 0.30385221922444006, + 0.32190306488251197, + 0.22641724482040146, + 0.43551336174409366, + 0.24259845669832902, + 0.36865572980790795, + 0.29091732821366106, + 0.3402849308813653, + 0.8401288533150701, + 0.36913632081705144, + 0.36388637242013994, + 0.35589992736707854, + 0.4303336358477263, + 0.5263193357130661, + 0.5950672392656627, + 0.6084022084047561, + 0.3777575071062858, + 0.3672483144014453, + 0.44577545119264356, + 0.7166920887392104, + 0.2766344386825425, + 0.30927134124917294, + 0.5237532422200747, + 0.4990670326046357, + 0.5107439055467222, + 0.4084290475402138, + 0.28440600021051077, + 0.34541320518435215, + 0.5822563638822911, + 0.39982515705240695, + 0.29353715555879234, + 0.6032383756365227, + 0.5313040861931702, + 0.43187481513089443, + 0.4643573197967088, + 0.693772612847507, + 0.43787977451570914, + 0.4731260989552583, + 0.42969494838407374, + 0.754695070956538, + 0.2915954717066698, + 0.6280381404066016, + 0.366935484552875, + 0.5331062402819161, + 0.40544446455858163, + 0.37932318265352905, + 0.6910409846138884, + 0.48726749158785093, + 0.26999569783313254, + 0.35273248490487424, + 0.629301367733433, + 0.3247161012061301, + 0.4323931278203677, + 0.4224129263893969, + 0.32117594156289636, + 0.4417683638394181, + 0.44286929311071005, + 0.3849667015097245, + 0.3790981819690705, + 0.45561151250742926, + 0.3200647473044442, + 0.29250755690377506, + 0.6167694918983546, + 0.3440525010920366, + 0.43501104845797883, + 0.5446446353826082, + 0.30406390702893527, + 0.6506639175432308, + 0.302470909605474, + 0.3057174850409848, + 0.5457125704303559, + 0.45753474150164397, + 0.5730438316065417, + 0.7220367397048776, + 0.6477595204620403, + 0.3816857237317046, + 0.4487234634299418, + 0.5240820775317435, + 0.2649405747836699, + 0.3034462202237159, + 0.5485996268378793, + 0.5003842064162811, + 0.7053302025920163, + 0.42899257148306164, + 0.33397046083688264, + 0.44113946258914105, + 0.352528573358462, + 0.45856625869529927, + 0.6875601921275265, + 0.6104813691037456, + 0.4578305218511022, + 0.5078676268770409, + 0.47587223955715197, + 0.500955512396706, + 0.7275896793986661, + 0.2837115586935696, + 0.4225849649725094, + 0.2555577307775121, + 0.5802521328289708, + 0.2989751752075143, + 0.7271619513349037, + 0.3830049664314718, + 0.3972396111774385, + 0.6783485607911386, + 0.49800662279775454, + 0.29882650912810194, + 0.5479617457370105, + 0.3776517654115839, + 0.5056177574427471, + 0.49208963639613484, + 0.3093286556056372, + 0.4803740683735261, + 0.36317026492866705, + 0.44266461747082214, + 0.32998171302266055, + 0.23986875597790336, + 0.37383461818689295, + 0.4938381613864741, + 0.5543147536164267, + 0.6478232297597406, + 0.36560387155385243, + 0.2712133385740364, + 0.43390854703284326, + 0.4399781459352882, + 0.4090614266047274, + 0.471189238260255, + 0.4396771577128376, + 0.3430472022187943, + 0.30397452963483845, + 0.48100528095319767, + 0.7585656918192114, + 0.5148034136003874, + 0.5853680152099953, + 0.5796835177802914, + 0.7606789782933507, + 0.5397982238990592, + 0.2889561215359693, + 0.565468991785872, + 0.3499604452874826, + 0.4908959174656084, + 0.4189767803557882, + 0.370560917635152, + 0.4762933976689646, + 0.7698789876712829, + 0.2159320814149534, + 0.4422522371141488, + 0.45136508148121945, + 0.6073594183969493, + 0.36500175766300524, + 0.23538600223819003, + 0.39505492993399466, + 0.7129490844598985, + 0.36219098754980106, + 0.41939113913105064, + 0.49242425489065406, + 0.4258514225701415, + 0.3771319465422963, + 0.47414265340202993, + 0.6013039044084386, + 0.44580757166665747, + 0.3745125322495928, + 0.5234898372908979, + 0.4404441933105542, + 0.5509896424982005, + 0.513007503661115, + 0.5081499290973429, + 0.5285086058017708, + 0.2762772072178326, + 0.25279199845767103, + 0.47148957850897477, + 0.2834828045652587, + 0.6834709185268304, + 0.2705004036007188, + 0.5456701432146353, + 0.5340005554483689, + 0.6753025027675776, + 0.275537124251429, + 0.305700311886447, + 0.30459217580465675, + 0.29639058701245247, + 0.2619054339030657, + 0.24957382701083336, + 0.5979987116984244, + 0.3165551242375742, + 0.6718877225070526, + 0.5635520388551264, + 0.41135987120263934, + 0.5420485387705626, + 0.3686795765622139, + 0.5187498969920248, + 0.5116738659976261, + 0.4484884047592156, + 0.37037835271354075, + 0.445464418542785, + 0.43640924146041404, + 0.502920990575136, + 0.3721184927870023, + 0.2938213532105104, + 0.5964042771917981, + 0.4511963552573364, + 0.645914216190324, + 0.6057723033172591, + 0.4794266219696693, + 0.2813415625313139, + 0.2970564729756127, + 0.32232151416228133, + 0.41001191477732013, + 0.7392208229473383, + 0.45679413313311545, + 0.34844732996402855, + 0.7287047521457569, + 0.4056883707644792, + 0.6123254843053283, + 0.5155312343138524, + 0.3773703130385766, + 0.47976186019407446, + 0.43229429683931264, + 0.4225766428453864, + 0.5691694348693341, + 0.5127006734433189, + 0.5005406481637749, + 0.40081240202308926, + 0.43351797869126435, + 0.40647112096793186, + 0.45565362854509855, + 0.760387413639979, + 0.46629776976698195, + 0.44125699128609036, + 0.41789651194937355, + 0.24609460840009462, + 0.6555994347132326, + 0.37425454586970003, + 0.2866358155807052, + 0.49608303922767194, + 0.3876116766923561, + 0.2961102890218003, + 0.47333191533167845, + 0.4046099806150482, + 0.768836209431387, + 0.3011786032605098, + 0.3576170065923236, + 0.28781051163689514, + 0.412395699322832, + 0.3650899786815465, + 0.2660717230901193, + 0.3613105302447648, + 0.2512836498689177, + 0.3881863394095669, + 0.7324375923833502, + 0.2707036212802841, + 0.38223123182588226, + 0.47420100512338564, + 0.3963111672449142, + 0.35924787438094963, + 0.45616713482348925, + 0.3427906604415203, + 0.5016232953746687, + 0.49410172746447145, + 0.17298053378269446, + 0.6108485651026281, + 0.667885631514838, + 0.4772296647820116, + 0.4543035223880473, + 0.49719559722848417, + 0.35134220799537974, + 0.32279825904726295, + 0.40217425197098283, + 0.7707712165121623, + 0.31802200601843744, + 0.5838705352219403, + 0.3039023222099487, + 0.4378010068894979, + 0.44713813613529463, + 0.4610358591724481, + 0.35430633056193334, + 0.27698689942074095, + 0.2825341943109902, + 0.26041721648515914, + 0.47638946189367315, + 0.4090524101417076, + 0.2341992159268462, + 0.3826169577418686, + 0.4305842910150971, + 0.26454714122775697, + 0.5345077841238528, + 0.43968853340489517, + 0.3526683574962825, + 0.43279855382562205, + 0.25987221272043826, + 0.3998550039368998, + 0.4530937802200469, + 0.5569070697323387, + 0.42147179762671405, + 0.7698789371705238, + 0.27570770337599326, + 0.5414530495409013, + 0.7272290392984627, + 0.6168743702611473, + 0.19973246859184743, + 0.4640302849946875, + 0.4007875804046907, + 0.30059120916565507, + 0.32537435781870316, + 0.32612473407298237, + 0.43787357031088614, + 0.48205549759090155, + 0.31531809505187447, + 0.5912827316968725, + 0.7544005858927434, + 0.41407135099309295, + 0.38222767985234435, + 0.41175160675897454, + 0.45795311561719576, + 0.4840659661238051, + 0.47777309882205676, + 0.4635185565838797, + 0.27270975143646, + 0.6290016984027031, + 0.4510072587177668, + 0.5346164794888364, + 0.6964954993908574, + 0.28221885489312837, + 0.5782664779659202, + 0.3665359802289674, + 0.4248851653614656, + 0.3340433907424567, + 0.4956444236231327, + 0.4245687888842212, + 0.29494255077301146, + 0.3838298764360888, + 0.3952289837820721, + 0.5296849355593198, + 0.5701436991320108, + 0.4125878967918645, + 0.4367974650432475, + 0.38837500282058374, + 0.7241963819616059, + 0.40051113237990466, + 0.47131926169501775, + 0.43761323976284944, + 0.4476663244321057, + 0.5871650312092834, + 0.24849937893166932, + 0.5486198784534854, + 0.35126477125167244, + 0.5116415352411322, + 0.3931779015750175, + 0.2965649600349598, + 0.29670514617226834, + 0.439336331588447, + 0.2759037429472055, + 0.4885759451397498, + 0.7144137741602756, + 0.7168905693521133, + 0.3480077738511364, + 0.2911126796037075, + 0.6876574542578064, + 0.4263141424458813, + 0.35602144616061976, + 0.48554864843835166, + 0.3391030800721543, + 0.5801424774471027, + 0.41882402460496393, + 0.5457769378342475, + 0.5696473350498504, + 0.39565809995866447, + 0.49037255933317075, + 0.643440327892943, + 0.4740775716180724, + 0.32404524761211184, + 0.37492441848979574, + 0.27257363867602935, + 0.5159198976109534, + 0.31826804562015115, + 0.49365371346238957, + 0.4702911197345319, + 0.3134037638771025, + 0.7074019189829137, + 0.29481922720491227, + 0.7119051785684495, + 0.40420368642994353, + 0.517375471546784, + 0.7319314920392748, + 0.4525902729425162, + 0.5366211053984442, + 0.33847215261917357, + 0.6932884273493182, + 0.3804207070764958, + 0.27604707623694746, + 0.5352379943759489, + 0.49502385223825934, + 0.6205983762811235, + 0.531307854658655, + 0.3589450332852847, + 0.4629981593143788, + 0.24066873423597626, + 0.4147874986612561, + 0.5482854920648643, + 0.48212473871064215, + 0.3176706454465577, + 0.5145067255781957, + 0.5639123919542208, + 0.5528182844619419, + 0.4086701070493449, + 0.2910838748430419, + 0.48932594665639706, + 0.44409524879524825, + 0.5399291956010904, + 0.46040979391474085, + 0.2964488998752932, + 0.37891627390313215, + 0.47758316705748, + 0.3736493678924084, + 0.7166692475249062, + 0.4905053519896345, + 0.4241355804417619, + 0.4714616735070166, + 0.3602574806340216, + 0.40590446232399396, + 0.5241175169376205, + 0.5306263057460088, + 0.7005497493070743, + 0.5140482604348896, + 0.5596780756029516, + 0.5051433895568274, + 0.4478303093147103, + 0.5565840654999478, + 0.3504720027030457, + 0.6079143666537999, + 0.5825753439034465, + 0.7176671462678033, + 0.49214492452077996, + 0.5279857149639895, + 0.5256281463517648, + 0.3587969223262963, + 0.5209190262180471, + 0.2982508269004936, + 0.6617955998006929, + 0.45569548713084856, + 0.7272297195919794, + 0.7535845221757944, + 0.4283569337129611, + 0.4227230014796611, + 0.4748584437012697, + 0.3095556371041639, + 0.47000038118076665, + 0.7498386884606779, + 0.4720752864564449, + 0.4781605281000188, + 0.5723685688476301, + 0.46636752789572006, + 0.31500195072438375, + 0.5917314068979177, + 0.4043990237356661, + 0.3637534653309247, + 0.7256919486007882, + 0.38210297285776984, + 0.4065528493112682, + 0.36698637500652626, + 0.46728347494555095, + 0.5451593606870725, + 0.5961049855869341, + 0.6651209729042513, + 0.608701205507081, + 0.573496066717449, + 0.43293324160990687, + 0.591582867369872, + 0.5316018848463132, + 0.773398582369128, + 0.7283763313694284, + 0.305107833462934, + 0.3383756118880989, + 0.281548809368562, + 0.49673214361321694, + 0.3959998354448636, + 0.780339043872293, + 0.35312535735480666, + 0.3995422477334479, + 0.5402901843178377, + 0.464159140293082, + 0.539403021294214, + 0.606349211933518, + 0.4458549631330046, + 0.6095153213089365, + 0.35581798665257813, + 0.6633971287056676, + 0.7720171635632502, + 0.7564563419590279, + 0.5632003575251209, + 0.40558162583692986, + 0.5273425219986413, + 0.730715541905261, + 0.6742577577655013, + 0.6089185672130367, + 0.4603900882360753, + 0.40824302949441177, + 0.47727122000584377, + 0.7281728599584698, + 0.3927755011914771, + 0.3995189884440651, + 0.28423470049254507, + 0.5813970187759353, + 0.26076314282899704, + 0.37500116770925307, + 0.6543259561181171, + 0.527066412421553, + 0.6699140326238067, + 0.52168293298835, + 0.4949881505954633, + 0.6536486336448166, + 0.6208568161571545, + 0.5512864939481147, + 0.36091683466055635, + 0.6329110791944719, + 0.502725099515835, + 0.38242837305775707, + 0.3217517717945592, + 0.4625817702675327, + 0.6285728374209819, + 0.4803981690332817, + 0.5958352501966009, + 0.5584928528437118, + 0.40053666761221707, + 0.38207233746941377, + 0.6742301361489093, + 0.4721889393809618, + 0.5715877059721605, + 0.7150861918861349, + 0.3975600021552958, + 0.3292095207884476, + 0.3271450824557492, + 0.3359507101292841, + 0.7031577076104623, + 0.7283369091123918, + 0.5558174904082401, + 0.40259092160367993, + 0.43553222678132086, + 0.36415240779585006, + 0.7732862254389108, + 0.32187739452250164, + 0.46424990071451294, + 0.3116896282783685, + 0.4806492159147384, + 0.34214301785862783, + 0.5471653805571175, + 0.4558254814526719, + 0.26981843958899676, + 0.682971430333451, + 0.4944662410694567, + 0.3963960010606684, + 0.3205527074171243, + 0.34002875411347816, + 0.3871303395055791, + 0.4008968577119002, + 0.35206784516543765, + 0.3478192418042774, + 0.3875670741056071, + 0.35915239529877535, + 0.39129461286975026, + 0.3128323215050493, + 0.4226319850991648, + 0.20141037423831415, + 0.23959019734059256, + 0.292165712519044, + 0.36631335180808905, + 0.41312824397690445, + 0.39219107731941383, + 0.29623605868399033, + 0.29010909108052385, + 0.2868308629320737, + 0.6495022471315358, + 0.2859044717053591, + 0.2695498125494071, + 0.7726001036896841, + 0.33721327435467946, + 0.37917526906221, + 0.31457476846135174, + 0.4298809348338168, + 0.2555106773559552, + 0.36866925249937765, + 0.33037685827332397, + 0.45595690606894285, + 0.22970834928373648, + 0.3921256069009932, + 0.3776194423357154, + 0.4841975332289066, + 0.3451762279297219, + 0.4573145440566568, + 0.2799377026064543, + 0.42318583259887743, + 0.38342126563026907, + 0.3230389468003563, + 0.45029870920613824, + 0.32813015267926365, + 0.45963080646384286, + 0.3481552554634189, + 0.3297954303876961, + 0.8029188459797274, + 0.26738248517035745, + 0.3949116661728859, + 0.33926495993975175, + 0.3747717622867749, + 0.27785832094478247, + 0.3431099446322308, + 0.4471976829064112, + 0.5347505217329174, + 0.33671512817784804, + 0.3630333224873774, + 0.24143142597464778, + 0.3913761865988925, + 0.3855787396110748, + 0.3946425780207901, + 0.4246821223472103, + 0.34285659409743724, + 0.3068517727004739, + 0.4539587553405824, + 0.36874672242584516, + 0.43724772556828573, + 0.25339540628745166, + 0.30846121213137667, + 0.46070356603717555, + 0.44696120921219823, + 0.37718630082264387, + 0.3304747965311375, + 0.3282927351254337, + 0.8189569149463777, + 0.3035710731672455, + 0.29686111157286066, + 0.32438641777664173, + 0.2656301250676127, + 0.3059537543629182, + 0.30194931986019535, + 0.2524053014645019, + 0.3450163213510799, + 0.34366133096521834, + 0.3465966386537107, + 0.30461677113128516, + 0.4041125478417331, + 0.46021128620880847, + 0.29358269722289126, + 0.3457431055019673, + 0.5355161944881405, + 0.2816107425929074, + 0.3327876093245865, + 0.47961469476574653, + 0.42626129119776157, + 0.2841166534834851, + 0.23403999303209758, + 0.28787837488658796, + 0.411789361664191, + 0.4234497935669201, + 0.2421318602896918, + 0.3214994667513306, + 0.308381723971609, + 0.4051799628520866, + 0.2886228382427895, + 0.2491868086660259, + 0.457287073963448, + 0.4885657379592838, + 0.22557922021006332, + 0.4073235514645357, + 0.23664802712257108, + 0.2882388783121065, + 0.2812069187936926, + 0.7718429016992329, + 0.40645033146157833, + 0.4181242634872189, + 0.23161866974383116, + 0.3354326826069311, + 0.3130001759357716, + 0.6799880619908483, + 0.36831049970517976, + 0.3097601395625372, + 0.43006873484596386, + 0.38712631427755784, + 0.3640766759436659, + 0.42323409367034975, + 0.46098649516883744, + 0.2288499597155206, + 0.2718938300362406, + 0.41337190995156115, + 0.3586855340120526, + 0.3743749882924052, + 0.3655873990851481, + 0.29399988177279285, + 0.24926854461217374, + 0.24361604595872066, + 0.3719312295586371, + 0.3278840376017483, + 0.5784578351775668, + 0.2753489161636438, + 0.2498718560015388, + 0.24571604400436406, + 0.502738905744717, + 0.31381843459143494, + 0.2168880871107414, + 0.26467674955498327, + 0.3393668747423945, + 0.35103808078739035, + 0.25075003528692663, + 0.5207429996639998, + 0.40223667926598716, + 0.24349391670693146, + 0.3559043147675459, + 0.40349432109953587, + 0.33333031335274316, + 0.3144171710903673, + 0.4510581336357925, + 0.2820416327698189, + 0.6343227275840552, + 0.35186976640207596, + 0.4404668701569374, + 0.38429857009522905, + 0.3446875724127084, + 0.3750787038122295, + 0.3004215358384029, + 0.2262238033621276, + 0.3666145031083938, + 0.5734878560070652, + 0.30836255342386215, + 0.23160562026109546, + 0.47355245457136597, + 0.3926111634550126, + 0.3939013290310168, + 0.5862073160221913, + 0.319636282388369, + 0.37847686662585217, + 0.2783917007453841, + 0.3407842053870781, + 0.4153089077121978, + 0.28690987162082465, + 0.4313272917092951, + 0.37619992721798184, + 0.3852802945688359, + 0.19897671222488206, + 0.3598416206414551, + 0.27187051882751745, + 0.2553408719906253, + 0.2556259522511322, + 0.23953292113194605, + 0.395811608449587, + 0.4984838540378894, + 0.41383968855029124, + 0.44127633951886175, + 0.4097996642992794, + 0.24373546613202246, + 0.3047829382427412, + 0.44421326960854696, + 0.3288226302830816, + 0.35570817494825946, + 0.43621072357590457, + 0.34521181391526945, + 0.333808967298225, + 0.48215608192071224, + 0.3678030284049559, + 0.2881250772253333, + 0.48195781738652393, + 0.2968215296259699, + 0.40492234102139435, + 0.2610833904926279, + 0.46022373097706715, + 0.5814338143541242, + 0.7813888301717231, + 0.39034742464743105, + 0.3435782048028845, + 0.3946406089429068, + 0.33951762665358465, + 0.3226000267364567, + 0.3216230966476747, + 0.3753628156706549, + 0.3123825263203482, + 0.2671493742771849, + 0.4187633456097355, + 0.33946908736086806, + 0.3189883753468142, + 0.2495274833147897, + 0.23261348093346432, + 0.3699041626537223, + 0.3790722375330636, + 0.34208990980053855, + 0.34674362094091987, + 0.36155071592840476, + 0.27202859541972974, + 0.1990597707317539, + 0.3271772605498059, + 0.4113851707159744, + 0.26492724025699194, + 0.36397117106152266, + 0.2692760820431513, + 0.28859516013411507, + 0.18729638795049025, + 0.4928519355211825, + 0.2730325730204621, + 0.2556740268732654, + 0.3588987040309015, + 0.3116694466835311, + 0.27405914988789204, + 0.5754763210464484, + 0.429234937548667, + 0.1990413153345209, + 0.40810143154196565, + 0.36849413328560815, + 0.3571563294356886, + 0.2888582174067073, + 0.22690585429404825, + 0.3078490573222458, + 0.469034505387819, + 0.21497734347457945, + 0.37384595518926556, + 0.19746872156716253, + 0.2600681773036991, + 0.26842092186789146, + 0.23463547370956309, + 0.356821983310706, + 0.373694569856951, + 0.30299934310384874, + 0.3253056059632127, + 0.35405520141546515, + 0.2317237382240234, + 0.3473688739509007, + 0.34376437456702547, + 0.4788127143172417, + 0.4821366804976421, + 0.3720473945495414, + 0.37671629890023306, + 0.3536622970076177, + 0.39354068898837935, + 0.424884058847438, + 0.5112883142750668, + 0.3741812006759494, + 0.2534466147727893, + 0.4854911669278999, + 0.3137874678364162, + 0.42029071745163293, + 0.4156396821841681, + 0.30328838506580685, + 0.4610993618374371, + 0.21410694084943802, + 0.25373343143054883, + 0.23224524415405193, + 0.20781437355879773, + 0.3267708545546746, + 0.33002672370813874, + 0.2670978278287346, + 0.3724427114559238, + 0.5148571296291536, + 0.354204373770489, + 0.48269456545220374, + 0.7685658308085236, + 0.5249636752322804, + 0.22113037675366254, + 0.39533809948392096, + 0.41496959636383207, + 0.37723646405817884, + 0.2246523053730904, + 0.5582774379598996, + 0.4300657899553627, + 0.3239007795427919, + 0.42637781687191545, + 0.321729848201482, + 0.4286150510419721, + 0.48655452012490463, + 0.35605871695393815, + 0.5045309242568783, + 0.4264686806691702, + 0.32041315552025207, + 0.43629748101754806, + 0.4748489697443866, + 0.25165497429789246, + 0.27602672280619056, + 0.21778017056253124, + 0.5606468827794582, + 0.48290324616588437, + 0.8097945174154436, + 0.5001649967123212, + 0.3684303407584837, + 0.26141967228232904, + 0.44653610361348695, + 0.422758746022606, + 0.5872179479503996, + 0.3046439110082363, + 0.5068611202512368, + 0.2706748718988273, + 0.30893650556347363, + 0.22594975573178327, + 0.2213145173454945, + 0.374772257099206, + 0.1902831515349753, + 0.37196947522709867, + 0.43025391403217417, + 0.342077774255181, + 0.22909167762357313, + 0.23821260852436074, + 0.377698168078771, + 0.2270092920608979, + 0.4808211936758845, + 0.39747194066700736, + 0.42398646829534714, + 0.3813797106157956, + 0.2900738157162903, + 0.5353202051603803, + 0.4419651179444064, + 0.30958403870013873, + 0.5723508553403257, + 0.26456851374617074, + 0.39056995174601394, + 0.29837313736606685, + 0.24045268003319106, + 0.24959760317444823, + 0.30877386836704174, + 0.41822434995427576, + 0.2776313700035387, + 0.6719236157696258, + 0.41932853445540447, + 0.24089739542760993, + 0.30453319518249616, + 0.24579168380782998, + 0.5263420765518406, + 0.2338367108350665, + 0.3084992582704413, + 0.32579491362156776, + 0.35778623478757676, + 0.23969920672113987, + 0.36991815268418243, + 0.25314690060754436, + 0.38941912553380686, + 0.7572404905483486, + 0.38939642441303535, + 0.3160907432768188, + 0.3326917398128089, + 0.3197437878795187, + 0.20846686171040724, + 0.31104122382606103, + 0.39487354346990017, + 0.7391007068262702, + 0.5224866466999399, + 0.40445295870399356, + 0.32609759249240255, + 0.31325138872967767, + 0.3552341596692916, + 0.2286673439890984, + 0.3927014067335508, + 0.4810104816488346, + 0.2774047970548084, + 0.36547517618554437, + 0.3445417095956399, + 0.4292086012645961, + 0.3209693473770878, + 0.49918242237487437, + 0.3401988763421349, + 0.27034147315860807, + 0.393258688198092, + 0.39954573511985203, + 0.5157428458584482, + 0.5101304678760372, + 0.22466936614305752, + 0.7070127943447356, + 0.7035270559039172, + 0.19435972397242315, + 0.2614958292029158, + 0.33748331468992815, + 0.4214955013929342, + 0.23105412997593622, + 0.338153543976503, + 0.441352616315119, + 0.4564534859877257, + 0.42846000553859614, + 0.3393907011964461, + 0.3239223957174247, + 0.322654136558091, + 0.298946252081339, + 0.41291211051038945, + 0.35401740655445757, + 0.3551074503869573, + 0.4762882949705628, + 0.37881083428417633, + 0.5079064497306112, + 0.423487762005547, + 0.5917682233357779, + 0.42358364747702054, + 0.39067489593950766, + 0.47099361089956543, + 0.4303837247947923, + 0.3122275517738692, + 0.606457692268784, + 0.5836464717267426, + 0.41649769906691997, + 0.3491855250547051, + 0.348882905599431, + 0.5070933538362853, + 0.3119326623644438, + 0.39343949559747043, + 0.3750399220517633, + 0.45401721481749757, + 0.41317832002988825, + 0.36468190574380693, + 0.33454467514416714, + 0.5075234977249796, + 0.4676537752430781, + 0.4424887956977896, + 0.4592524572451527, + 0.35582211742680436, + 0.22596048737013644, + 0.20485025070587357, + 0.2814158487505106, + 0.49796511180860564, + 0.43869992915693, + 0.22072170577319844, + 0.4185673346757312, + 0.3270733306391141, + 0.3178690009892735, + 0.31946260357048034, + 0.3577024199435954, + 0.3141874782538718, + 0.42288972669509983, + 0.6091150379838661, + 0.29254032763117954, + 0.4034471224798413, + 0.5115549533394023, + 0.3206404432042526, + 0.40293337564527987, + 0.5717928496585578, + 0.43670017848484066, + 0.430694952921047, + 0.39316227707216544, + 0.3730189647623583, + 0.22923716330523483, + 0.3477156236257467, + 0.4684528219609335, + 0.47417987145780355, + 0.4545059430115727, + 0.2899182138785116, + 0.4401218142147162, + 0.4304949615672866, + 0.4857744617748019, + 0.3892053475811074, + 0.4534293730237891, + 0.3740642892270512, + 0.4415914528926015, + 0.42962343302828954, + 0.39541885977003904, + 0.2650742371509101, + 0.43366102904304754, + 0.4488378122263776, + 0.2342439592501863, + 0.3378232026350678, + 0.42567449748756725, + 0.37448504385216824, + 0.3422579004503282, + 0.47135231197121574, + 0.4774049992562648, + 0.4155016223118502, + 0.4362804288951198, + 0.31215223426001254, + 0.43355572845610524, + 0.3537284532061516, + 0.34909700723444703, + 0.37311274517736825, + 0.4286205170998315, + 0.3670577855851993, + 0.39655697776852, + 0.3779419433240051, + 0.7112671217840213, + 0.5320433210319145, + 0.48968094716669985, + 0.5238634245645808, + 0.45614128015201105, + 0.4021542543999722, + 0.5271986411705364, + 0.33988078727764176, + 0.3715836545074331, + 0.45931196724188583, + 0.3087171425445684, + 0.4253888217710114, + 0.4861767599728791, + 0.4328777129281322, + 0.4384553657523713, + 0.44215110797497914, + 0.3096271562664668, + 0.5081779409046159, + 0.4819364355867185, + 0.3546069569904602, + 0.40826633430575304, + 0.36056033615831024, + 0.32669846255935925, + 0.42620033619809533, + 0.4202772198850825, + 0.39965767522895335, + 0.43777706663092625, + 0.36822027134802876, + 0.46405812493261084, + 0.34849292629658934, + 0.37495399745794433, + 0.42280676291694963, + 0.417301244083669, + 0.3951091473138294, + 0.4602522536513039, + 0.3861597967104618, + 0.3314913497849932, + 0.3400854437177904, + 0.412702518357977, + 0.3936908221797518, + 0.42957096845935516, + 0.45334227627101253, + 0.40985480215167525, + 0.40223796916657323, + 0.3401412632266895, + 0.38909046997970237, + 0.5102079671946242, + 0.3701301734911682, + 0.19595817409065544, + 0.3332546446247852, + 0.40100472311741825, + 0.29409809638983675, + 0.2894053640493705, + 0.4354248620609054, + 0.4370108543935121, + 0.25460137115147713, + 0.3027370307214868, + 0.3341062040714634, + 0.3192481067943757, + 0.4936786227564436, + 0.27019524601124867, + 0.2704905679165199, + 0.2901909095563497, + 0.29569077846131125, + 0.21314060528648804, + 0.27892057291013944, + 0.2521235775620645, + 0.28112574521775535, + 0.24204699196370402, + 0.3810071541615614, + 0.26815436012015126, + 0.3105460200003365, + 0.31943954194204477, + 0.3715109970092844, + 0.2749799469748474, + 0.37251944504875395, + 0.3218860144989845, + 0.3225446572581127, + 0.3252282653869552, + 0.3769947503961239, + 0.3481454881671495, + 0.5640732520704893, + 0.2914342065914582, + 0.36033711694591875, + 0.5107644322863882, + 0.4616377828563106, + 0.35889115488804096, + 0.44063939534051083, + 0.3973178118950621, + 0.4351068985188826, + 0.45726581365867436, + 0.34308846008216737, + 0.5654010746300663, + 0.4107940046713231, + 0.4055240115863975, + 0.34266022340581015, + 0.41065885660123924, + 0.3606028048555389, + 0.4368897164215397, + 0.4394754203788012, + 0.4311200632248685, + 0.4505351206989379, + 0.5648157196689453, + 0.42288171768922217, + 0.38275338486395805, + 0.45072610839771293, + 0.516214654821409, + 0.5053749135555925, + 0.42984358649253324, + 0.5317388661606447, + 0.5271503167172965, + 0.370131038372368, + 0.3651222394634551, + 0.41524839443802347, + 0.3317452143505963, + 0.6005277255647536, + 0.4749466411670031, + 0.40145429295721424, + 0.33953960710912795, + 0.4507691656002364, + 0.38579711139077333, + 0.4974332790731003, + 0.4246806754321921, + 0.38044011543611006, + 0.4900754387400531, + 0.27350249774215163, + 0.39269446485335296, + 0.5049307790074614, + 0.3129686341973065, + 0.393666601248793, + 0.2969796337124988, + 0.3331404221157448, + 0.35748129065385564, + 0.365998613418223, + 0.4327431363892046, + 0.31332308546303, + 0.30534296379275394, + 0.39864925746553265, + 0.3672985715270509, + 0.35547854653789923, + 0.423538607099151, + 0.32206983441417364, + 0.3999654442679311, + 0.4853697624637504, + 0.5464814895083784, + 0.4360897357512527, + 0.3346332504188715, + 0.44717101889221156, + 0.4210186279129018, + 0.36733641753517976, + 0.3959981181246265, + 0.2964766996509991, + 0.3040709858738092, + 0.2712653410097642, + 0.35005182988363015, + 0.257748439715431, + 0.6378551597100528, + 0.4100477848747324, + 0.3867466445293942, + 0.2885379185692067, + 0.28799305200629316, + 0.31045485514388804, + 0.30948272599637977, + 0.27905936723999164, + 0.28671256650049, + 0.2794331848389445, + 0.3517546793408131, + 0.3539629798733218, + 0.4151428984523791, + 0.3834777072444494, + 0.34419420692182445, + 0.48312724868341544, + 0.4776085688263738, + 0.3523040483850777, + 0.3943227466048838, + 0.40223116277968113, + 0.3681626773210971, + 0.3340263569354288, + 0.4971773538073875, + 0.4290129776353312, + 0.5628253733958429, + 0.4276652918114527, + 0.421269099967014, + 0.4287611214487051, + 0.4351527901998494, + 0.31344302473871766, + 0.55567864095771, + 0.3979461213820411, + 0.22142355800746621, + 0.4364750906604859, + 0.5417578372751775, + 0.3477010701063533, + 0.41403212588237237, + 0.72377406136509, + 0.4170196581515662, + 0.4817526609082211, + 0.26217755649601554, + 0.4513008460127709, + 0.24691862729274663, + 0.2649588935034181, + 0.28569912170921186, + 0.23907084083496577, + 0.42928746307278454, + 0.169842038532672, + 0.2802953299200943, + 0.42489665231123946, + 0.3761563239928194, + 0.3824072021601471, + 0.3797211328043396, + 0.301156669480521, + 0.3564926565407442, + 0.4983871221137972, + 0.5117614422091415, + 0.3282782125411451, + 0.4231277772958529, + 0.3778189774396583, + 0.7096130960144434, + 0.35536310774455665, + 0.5430849773259866, + 0.7263616958280243, + 0.7022505799118436, + 0.414920917209865, + 0.34212979002784627, + 0.27359768796294603, + 0.42114639394974407, + 0.46389194137936485, + 0.4757772867748438, + 0.5441887194680407, + 0.4023489269947461, + 0.5354316860649454, + 0.38986554193528944, + 0.3571815119328786, + 0.6818943826639775, + 0.6679907722336409, + 0.42948494439862417, + 0.39003966662250583, + 0.4384787380714523, + 0.3893392024618925, + 0.42371920071468794, + 0.6295884726978657, + 0.6458817598108175, + 0.3951014210775819, + 0.4103396776165762, + 0.41187380252005246, + 0.3341878435786344, + 0.47119341184011043, + 0.3022552364131233, + 0.4516311207219486, + 0.45093717105086273, + 0.3285089516236968, + 0.40978839055555555, + 0.3548844911554305, + 0.29886883075821685, + 0.4801757933712204, + 0.4051571733017842, + 0.36764438238029823, + 0.3445208220601984, + 0.4505761370412515, + 0.3861060164880522, + 0.3062811634564605, + 0.3234006379082532, + 0.22270096234260467, + 0.21240301890944352, + 0.2414750364683986, + 0.39371005377240975, + 0.42338499624497494, + 0.28888707363647825, + 0.3077132980821317, + 0.2877948684425393, + 0.3876792286332616, + 0.3863369807449004, + 0.4641740089138798, + 0.30552804767772634, + 0.3725634120671442, + 0.4739902305900791, + 0.3195552261114413, + 0.4861691549154722, + 0.43121464078585653, + 0.3789141811012661, + 0.37787097093504357, + 0.3208169532001218, + 0.20370639561604403, + 0.2589914510208359, + 0.2537650380465818, + 0.48430965287503297, + 0.3811446416295319, + 0.39568178672897303, + 0.4257573594979189, + 0.4504011438373177, + 0.2923761254217394, + 0.366749620612078, + 0.33191892362799397, + 0.6347612104863638, + 0.4095564816192332, + 0.43864523838687525, + 0.3383128737792256, + 0.424433914239153, + 0.45592681432983445, + 0.4012622801276501, + 0.3733253021296988, + 0.44970383281996656, + 0.4160294841378348, + 0.5162737384457596, + 0.38325996343087027, + 0.447771423699255, + 0.41058048749952514, + 0.3915194136707349, + 0.36416994849773215, + 0.4381510280458368, + 0.3835379964615324, + 0.45168063519708657, + 0.3100055917504381, + 0.29001117488239403, + 0.2905655247446782, + 0.36628026781680056, + 0.41389685912563984, + 0.3253255617410921, + 0.37495989225152265, + 0.42411948218924606, + 0.38984440495239986, + 0.4177876908663289, + 0.3991376701314253, + 0.25410558758337587, + 0.3787589196225708, + 0.21506381730322957, + 0.38511322769360407, + 0.19763394855486033, + 0.633297225827948, + 0.2992725854585016, + 0.4048602244037448, + 0.5126416614323605, + 0.43154841136807814, + 0.2413935914899223, + 0.374150598501811, + 0.30231345779564045, + 0.7585390157981916, + 0.25760342434952577, + 0.4693400519769007, + 0.4742209681179945, + 0.47321430538080783, + 0.6181714046006068, + 0.3763126781554175, + 0.5085580048798098, + 0.42354435505735416, + 0.33565032158613206, + 0.2772349942679527, + 0.4219604263881217, + 0.2769165855276183, + 0.697097902907423 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.3872275645210765, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.3976655214885236, + 0.7411442705416643, + 0.7272729957920885, + 0.8087860099416412, + 0.7072111810321825, + 0.6280413203610381, + 0.5814127125443819, + 0.641165207698184, + 0.3878456202453565, + 0.6764966519937279, + 0.28693665909120797, + 0.4434366827882158, + 0.29112889061407526, + 0.291599393670733, + 0.27103903063338747, + 0.2572724514235754, + 0.3044223198684044, + 0.4992316725088859, + 0.3466621679820383, + 0.3563475902008752, + 0.30371461219351853, + 0.6536498821437469, + 0.2670779784293059, + 0.5731386598848154, + 0.3422669811984753, + 0.1541298840114837, + 0.35807072354621433, + 0.4011899639043606, + 0.42359892630789847, + 0.45587890112114954, + 0.25450664989860033, + 0.36545786670519176, + 0.3040007114220027, + 0.3168649881716107, + 0.39283424990014537, + 0.3219991496155363, + 0.275881725210877, + 0.4771653944762118, + 0.4917055979377558, + 0.6182087253541343, + 0.7766961885370192, + 0.7710680149811363, + 0.7589612381095315, + 0.63368016679957, + 0.3593531893607536, + 0.31566259657300005, + 0.7661621946511161, + 0.2837138351357955, + 0.29919160747420653, + 0.421978575293799, + 0.31571490536346736, + 0.7671869052020681, + 0.7857549613406245, + 0.36274609768146576, + 0.28864082945810526, + 0.31660101147049546, + 0.30603173721081084, + 0.3531828715940201, + 0.34279146191834287, + 0.4652349906381414, + 0.40344533632693186, + 0.4006937150514917, + 0.3568406003100784, + 0.4618889771144276, + 0.3930069432329704, + 0.3311072527016958, + 0.31686030383181696, + 0.4758100487892978, + 0.32607557887932304, + 0.40225923760804105, + 0.44061977501341854, + 0.38536614822151966, + 0.3614359475740723, + 0.34891826104439266, + 0.3738964653194126, + 0.35587163487978374, + 0.252353782785226, + 0.24611536609930168, + 0.34789970691305067, + 0.2982051070522413, + 0.40373698570093053, + 0.2712756724084297, + 0.30811640157549036, + 0.2552208402256212, + 0.37008032136251273, + 0.4545479942744024, + 0.3715356540808788, + 0.25125345966096047, + 0.3192124761829807, + 0.3033425630690697, + 0.28680977914630756, + 0.30580984873860934, + 0.4476298200604215, + 0.46968489156321763, + 0.22773317370195076, + 0.5128343052963914, + 0.3074958112625487, + 0.47856587564770064, + 0.2566856348410677, + 0.40977973736037915, + 0.39134593305461784, + 0.4690220649925519, + 0.1992671792735049, + 0.35538521876416196, + 0.2616859200315021, + 0.2879729493201891, + 0.4136443425481654, + 0.2839412939553844, + 0.2935345642159053, + 0.8004670835073953, + 0.7953087626068156, + 0.7997780888437324, + 0.8155139865520605, + 0.5704528009573195, + 0.674540280771383, + 0.6961725524426507, + 0.7467481352664849, + 0.6864891381541163, + 0.5759759784023142, + 0.6457320030749601, + 0.5654694263638761, + 0.7635507299719014, + 0.6467157235804484, + 0.4002450101326072, + 0.30377198283943724, + 0.24561193182879704, + 0.2936316961774756, + 0.49971917046710984, + 0.27066290252580677, + 0.2539611075849027, + 0.3261590614490797, + 0.36206643477867556, + 0.2530046361202191, + 0.3350889716195974, + 0.36613194066312477, + 0.3888054815604352, + 0.3430655795388302, + 0.283409038362669, + 0.32050883386856155, + 0.35234748709858843, + 0.27928666575176736, + 0.28746866644928054, + 0.23737768380013138, + 0.5778281420541724, + 0.28038794959267477, + 0.5172433301171312, + 0.3389867654265239, + 0.405734613680674, + 0.219733099259669, + 0.5398151834644587, + 0.2685318235367539, + 0.4613658345069045, + 0.29053991664209083, + 0.21800252641460488, + 0.31328224050675263, + 0.2843575863317076, + 0.23788076773803593, + 0.1518734623247402, + 0.27934671866849464, + 0.2932062791635911, + 0.44163450347226124, + 0.40643502532289916, + 0.4610517782077706, + 0.4418196600348644, + 0.3038895122690497, + 0.41304648105427516, + 0.35477978311141467, + 0.38702411554899246, + 0.37890291862089204, + 0.3743121497288429, + 0.23643407562791757, + 0.14438785752155245, + 0.3544743926154429, + 0.270395677451901, + 0.21977199261217253, + 0.2237568788008318, + 0.3830639902837026, + 0.4507408909371943, + 0.5446704833877372, + 0.4105891560996219, + 0.21200893731152584, + 0.4440292090402704, + 0.4352947550272819, + 0.4651803458998379, + 0.2616801515330067, + 0.4807273445694452, + 0.38493825427641604, + 0.46647223230810847, + 0.21818227379943292, + 0.34566385187726356, + 0.3110399590517214, + 0.23039881837559487, + 0.46235788266902444, + 0.3892907450415829, + 0.49914430441498353, + 0.276824166328491, + 0.5837703046460303, + 0.27145743993042776, + 0.1870949601452594, + 0.3672987675797648, + 0.254283333688787, + 0.2470247663456476, + 0.18870226867795148, + 0.3093728834160841, + 0.5906028951050483, + 0.2154457912306113, + 0.21948323076396933, + 0.3959042994927673, + 0.31514235407775043, + 0.3560498768816064, + 0.419059823192951, + 0.30334546080037367, + 0.23392368995788973, + 0.31793785297368526, + 0.29671646976993715, + 0.48557244527373344, + 0.48519501564328504, + 0.5299856990582886, + 0.5074007393426755, + 0.24743438753866587, + 0.3053300165001544, + 0.40101992416038373, + 0.2364329309649549, + 0.30162735949858277, + 0.2642602772362084, + 0.13948670959879708, + 0.41945157238386904, + 0.29203073798599705, + 0.2255330899296006, + 0.4945367927322885, + 0.2185210116736714, + 0.17954983307092087, + 0.3646480809259747, + 0.356137553853697, + 0.2572232126323544, + 0.5668041089837383, + 0.29240700838116235, + 0.2745750850471061, + 0.22020573390212456, + 0.39539528477722236, + 0.26432925523753537, + 0.42811403803801523, + 0.24811165266510182, + 0.3678861063998304, + 0.34030991214861633, + 0.34090003883000863, + 0.3004889976056589, + 0.2963383393928709, + 0.3060193579747412, + 0.3042545339628458, + 0.42015239053647274, + 0.25472047586255053, + 0.34948558816766223, + 0.2682772027901861, + 0.3040976739222602, + 0.34988743968787545, + 0.3530465931571666, + 0.3208924350901477, + 0.39580866837360235, + 0.25846979138190457, + 0.27895273688694566, + 0.4719178059759977, + 0.23651240998843187, + 0.30109096571777044, + 0.27995687185272794, + 0.21591706814191228, + 0.2795487437382491, + 0.2496988739731168, + 0.24265317510749598, + 0.26117864935355856, + 0.23570968032688072, + 0.2883612894469733, + 0.2881544901609722, + 0.32934373392829414, + 0.3403586080377183, + 0.42760889692122916, + 0.20776086076932435, + 0.37137369205618603, + 0.44041850917059716, + 0.2340760983402914, + 0.23605454499063988, + 0.5376524372856069, + 0.21296292720713295, + 0.2176138864044498, + 0.3836230580651672, + 0.37108903852823566, + 0.4039783109556305, + 0.3293410465679762, + 0.45905535904373507, + 0.3936351004865655, + 0.38873858469769035, + 0.32046403948261964, + 0.3942549363148766, + 0.3727350642703805, + 0.35037358467234775, + 0.3623013036526231, + 0.4415480321872908, + 0.3897130045058868, + 0.643937638542214, + 0.26128174192378195, + 0.537224917427009, + 0.6219640700103103, + 0.6276316779150438, + 0.624524870087337, + 0.44183640287517795, + 0.43098237951874346, + 0.3651049620697797, + 0.6532252614477679, + 0.3948457744928705, + 0.44808067127583673, + 0.28514649958397925, + 0.4237152537090523, + 0.3481784515222254, + 0.33589900317651433, + 0.37588106606912386, + 0.27926733859618497, + 0.3997352029557198, + 0.2502610494967961, + 0.2895344507314027, + 0.31245713239294265, + 0.3175897174420951, + 0.3623899192347481, + 0.23462410717148918, + 0.30158423049527383, + 0.32147487254188223, + 0.3295050785113226, + 0.28102357358670793, + 0.32765733596075647, + 0.5093562576110389, + 0.3982911662591348, + 0.34128636503973425, + 0.4105665263909895, + 0.4041195496818498, + 0.25727475454756044, + 0.2856992520033681, + 0.33503346558473746, + 0.3639573215395291, + 0.2840591593768956, + 0.39081649743160946, + 0.4919710794007079, + 0.4852320570675413, + 0.3425358345494503, + 0.2711170858050483, + 0.426151461948495, + 0.2390619185208272, + 0.2888332386178553, + 0.41348334998500885, + 0.28752436076320786, + 0.8138906015519913, + 0.465853635598489, + 0.367177957992898, + 0.34242652411177454, + 0.40660649544092725, + 0.41304710596643734, + 0.34771195334558924, + 0.45178487708568177, + 0.30327154036664716, + 0.44153623422375654, + 0.34245971614744103, + 0.42348629797896553, + 0.4122054902655725, + 0.39597501142582536, + 0.37392707379819706, + 0.2963974933927293, + 0.30558171977482823, + 0.4598400213736201, + 0.4823700373073104, + 0.36184319430632245, + 0.5414407720985219, + 0.3832027554490256, + 0.5068951770494091, + 0.34223405766161413, + 0.37589314604475066, + 0.3911179102564107, + 0.5218225520612052, + 0.5174453205681832, + 0.460425229604704, + 0.48624901542099125, + 0.4598698411246467, + 0.46944582107368266, + 0.3525157785552433, + 0.3411722982722852, + 0.3072790219991362, + 0.23734732087031973, + 0.3360507420503659, + 0.30238076840223854, + 0.2816179001232884, + 0.3053138596949209, + 0.40945277340246156, + 0.30421894132028743, + 0.26653266520331576, + 0.2955879037559338, + 0.49850993418488637, + 0.5409651365709045, + 0.4724725821639463, + 0.5102506907662805, + 0.37541494643458845, + 0.52173924494478, + 0.3148436568433386, + 0.5850851876732454, + 0.39622272525239144, + 0.5056687372164724, + 0.39865204594666087, + 0.4975783998741929, + 0.49713324253114294, + 0.4478272493459151, + 0.41021166382660407, + 0.45037521911313105, + 0.45926475099896813, + 0.4808400494700398, + 0.26472143632989475, + 0.3232746823443498, + 0.2730620416751186, + 0.31237423659779895, + 0.46048254258556076, + 0.5149487045398415, + 0.2665169434081925, + 0.5328755670770721, + 0.4399966492864012, + 0.415554263651756, + 0.3347374848275217, + 0.4112683325398904, + 0.41779225580182544, + 0.3174984001350095, + 0.32526360113706765, + 0.840815821866426, + 0.3865464412088428, + 0.40083727454400947, + 0.4377723503741567, + 0.44105083262653316, + 0.41163033348415284, + 0.3634547996182655, + 0.5488755063043432, + 0.4103251185572136, + 0.3312042217662734, + 0.38608669784897204, + 0.3861355434637084, + 0.39452543446927774, + 0.37339143281516485, + 0.3828701203596828, + 0.4214128700602883, + 0.4128933873271614, + 0.25314178513567726, + 0.40620279670938997, + 0.25682924483375613, + 0.3516565087622497, + 0.1943262598055026, + 0.4039701273757627, + 0.3241533528552358, + 0.24441576844174884, + 0.4312715904036934, + 0.17673445341789237, + 0.23424884027313986, + 0.520664223518711, + 0.45103923674601754, + 0.4177328083316987, + 0.41905353600556605, + 0.36937127364508615, + 0.37942556897316876, + 0.5570311672372, + 0.31017546789769734, + 0.5287351692926616, + 0.3153907381722145, + 0.23465086762331935, + 0.1863999071679973, + 0.353626563771031, + 0.438870868191821, + 0.48657759035317527, + 0.5173505825348663, + 0.4029362588282348, + 0.4563637631516445, + 0.32121870383416545, + 0.43564638528390437, + 0.34749128373640253, + 0.26587647152061555, + 0.31727904224340026, + 0.3290980533717621, + 0.39548810580047694, + 0.31092216061184785, + 0.34458249346547193, + 0.36699058018500175, + 0.3715796534111719, + 0.4297202631506619, + 0.35680298519382087, + 0.3193678250310922, + 0.44946731145312435, + 0.4093998238064824, + 0.4932375580389926, + 0.4491800321111392, + 0.522621446000417, + 0.5358008803776511, + 0.6476361224171879, + 0.6330407222552422, + 0.4590994297456198, + 0.5048582853153865, + 0.3904684014652959, + 0.5014331574422812, + 0.37091236306487285, + 0.42000586224335856, + 0.3535857593775519, + 0.5569581851744181, + 0.36081319866217776, + 0.5133080015055412, + 0.6303900798389277, + 0.6053149897091814, + 0.5402561287331354, + 0.6223790233796718, + 0.5159989627962264, + 0.43648417079470475, + 0.3775715931628284, + 0.3563375435064413, + 0.24587648368929005, + 0.32197387496099616, + 0.4109760383830077, + 0.49141444295982506, + 0.36013339275217826, + 0.28312086567406786, + 0.19794426411160645, + 0.4073545024298286, + 0.40004863553346304, + 0.3766808475441805, + 0.264968813254875, + 0.4810119097554713, + 0.35875874690109266, + 0.40360703596172187, + 0.31014193395527007, + 0.4108418030880756, + 0.4368821470807389, + 0.2911510140198979, + 0.3373958219420715, + 0.39281686354583495, + 0.32689768429352933, + 0.32468061125105446, + 0.3251873209507188, + 0.3268128744632887, + 0.2896662025884128, + 0.34803898136368405, + 0.35709437572650016, + 0.361567372682443, + 0.2288230731147629, + 0.5263555011733302, + 0.35758762290896856, + 0.45572980666044133, + 0.47194082668569526, + 0.3225517607764677, + 0.4249291488180534, + 0.4326022639458515, + 0.43449247877430824, + 0.5367572976633745, + 0.36234172022176747, + 0.3930730735218179, + 0.43181517401560293, + 0.31470797295691844, + 0.4209996783020095, + 0.3804488735250381, + 0.28574864702628244, + 0.4011959522753209, + 0.41971190797918717, + 0.37630952725865136, + 0.25073046394491233, + 0.362627601155292, + 0.22251675212446945, + 0.33564572193945946, + 0.5797158698477123, + 0.4950236782274824, + 0.5099027421409214, + 0.5113446138822334, + 0.5899416431005263, + 0.49875887767469707, + 0.5456735602994673, + 0.5074718452809246, + 0.5519196748618501, + 0.3709662807527605, + 0.35287396306464847, + 0.3678696006976716, + 0.28583402025969, + 0.24480530550287813, + 0.32175401469988524, + 0.31231281236879044, + 0.2773085318987763, + 0.23648908877159502, + 0.34895062913173075, + 0.2517818507310006, + 0.4088418897006546, + 0.28618424691637295, + 0.4930400230712506, + 0.5296735968695113, + 0.4925204889315717, + 0.2527324854906104, + 0.33264139996486286, + 0.549929909674595, + 0.2650841711908344, + 0.2165253288350256, + 0.22971410188922148, + 0.8180417405518918, + 0.48560140835992927, + 0.3854683811972556, + 0.5207070089595228, + 0.47752019143602087, + 0.4555622380196013, + 0.37902008527668135, + 0.4322552464534698, + 0.4756807470379217, + 0.4061810393380376, + 0.5244469357422065, + 0.4806232746495714, + 0.2114902239185758, + 0.33410723039361007, + 0.22834834902467124, + 0.31355412194923565, + 0.2498280416611218, + 0.4556711386477128, + 0.36794621207755007, + 0.29032011835955823, + 0.4147295510384719, + 0.47284974387046513, + 0.2935747704613725, + 0.414892604771897, + 0.2406599495367464, + 0.40336612527884563, + 0.45669228620032715, + 0.2948558581171195, + 0.29426241602118597, + 0.22903373367733626, + 0.5034729201778687, + 0.22887603625296316, + 0.2961464921874968, + 0.47220882982435747, + 0.3630670643744075, + 0.3845434047387248, + 0.2807322216608577, + 0.29561230387660176, + 0.3672804304375797, + 0.3400991417473356, + 0.4115302430820732, + 0.2926250563014035, + 0.44969531333965984, + 0.23863657936541477, + 0.4582593981682212, + 0.387271767014527, + 0.350399493560379, + 0.43450839904381106, + 0.3490051219742019, + 0.44887851334616363, + 0.29795523602464147, + 0.46060318325920163, + 0.3455900164316053, + 0.2739498940069502, + 0.3977014583042208, + 0.2738482870526575, + 0.46525748594416483, + 0.20530243352944397, + 0.6579436544793414, + 0.6634146248100643, + 0.5546066816356163, + 0.555913891340696, + 0.5952466277895794, + 0.5940542142704717, + 0.24479942767388782, + 0.32845764275060363, + 0.29555380520560376, + 0.3452469551586866, + 0.29645107623344313, + 0.3130801564704293, + 0.33156103851750823, + 0.36552565230375966, + 0.3862418977082654, + 0.2954109105438618, + 0.3325325536088305, + 0.2780236341510484, + 0.4501338749159313, + 0.24077892764642195, + 0.38350224657462983, + 0.39373263278745235, + 0.3344659234537029, + 0.4486892188994322, + 0.3562584972615147, + 0.4740583326089423, + 0.23510509700243676, + 0.4397280243886708, + 0.3417605781738501, + 0.3254292680877955, + 0.28751959255003046, + 0.33922468786391635, + 0.311573957639185, + 0.33838878410898177, + 0.4794694139434017, + 0.43129165289567495, + 0.515232510357212, + 0.31758300800884254, + 0.3765468887312488, + 0.3549812760584856, + 0.34939970512493373, + 0.43615070113018767, + 0.3309128628193097, + 0.20622883813520695, + 0.3872335865269873, + 0.33547450489386316, + 0.3305433983434564, + 0.4035920075910452, + 0.313703236502733, + 0.2516492764982782, + 0.23340073907598324, + 0.3082204329802528, + 0.3286259613952004, + 0.27279321603862633, + 0.19922204632060134, + 0.29460120269963097, + 0.4748949983485575, + 0.5806694171524618, + 0.5057132913179144, + 0.47002822347467654, + 0.5059150064716067, + 0.44577042283214613, + 0.5158859729711391, + 0.46288623418466024, + 0.4533899369556863, + 0.3190184190693752, + 0.36012193740916426, + 0.43141649641987373, + 0.5717682175619092, + 0.15451664244997795, + 0.28204766445876656, + 0.2671817577375794, + 0.25035634326951706, + 0.2660873030638968, + 0.3660221422900475, + 0.2720008805390745, + 0.3432113083806387, + 0.7232565761563424, + 0.60151344758004, + 0.750082560428902, + 0.7658681951456752, + 0.6290223112768493, + 0.7083982349412657, + 0.7286647773826528, + 0.33140637170014975, + 0.3287381202184064, + 0.3561827650438471, + 0.46976169664057044, + 0.30201889579016933, + 0.34379400001235244, + 0.33256779581316515, + 0.33353220827676605, + 0.34903065049599274, + 0.3266255523057217, + 0.24348920136991417, + 0.4103831520899467, + 0.3539048155467292, + 0.33024138792237334, + 0.388075045283729, + 0.287811778623671, + 0.2946384384851649, + 0.3037258386706315, + 0.28864842993821443, + 0.2315987565125054, + 0.31509166419948104, + 0.23841530076216833, + 0.4903074814017111, + 0.44836827984764405, + 0.3213546291415345, + 0.2557950870298427, + 0.18935019929815386, + 0.48113036257041875, + 0.5501140247981053, + 0.6272988641810695, + 0.284109270781619, + 0.33792448019311666, + 0.22915014240231682, + 0.4255015161134053, + 0.30623770592871336, + 0.267719736346754, + 0.44964576454199306, + 0.4649395559168266, + 0.43397240239456364, + 0.29821599378108427, + 0.4068768491991941, + 0.3992416563699796, + 0.4503004579273626, + 0.41521064485760334, + 0.4552032642587244, + 0.34875553244085056, + 0.3940948778477379, + 0.4022733201219927, + 0.4077457281148136, + 0.23778469761629503, + 0.32611313280235826, + 0.3702950630578096, + 0.37848594247206396, + 0.5272244676657675, + 0.184778043092658, + 0.5341924358410369, + 0.693962445989488, + 0.5247713296766369, + 0.3567737482275236, + 0.2779019642838302, + 0.31122858453098307, + 0.23611709008947307, + 0.36648906233869816, + 0.44097386834471086, + 0.41285199137565654, + 0.3129275642737334, + 0.254987457676672, + 0.43724065004809, + 0.4767395117187014, + 0.5136068514230804, + 0.5785981515917847, + 0.42237413957083747, + 0.49851469303367696, + 0.32274506167434847, + 0.42871360043470985, + 0.24258075378202945, + 0.2447143177870946, + 0.3584185273164093, + 0.33366951289126595, + 0.4783151704427483, + 0.44582383659977165, + 0.5879454016573212, + 0.6423290025624158, + 0.7400402531322902, + 0.5790814368135205, + 0.7521995574910323, + 0.37692992108282813, + 0.35566936735619786, + 0.3588144851770202, + 0.41948852721389185, + 0.28122308509612637, + 0.4588736797069253, + 0.4127943858144339, + 0.4591912508172702, + 0.4401232671992812, + 0.3722881111564792, + 0.3852397364285769, + 0.39909680353814997, + 0.3345791439129585, + 0.39468149349167925, + 0.36560979892790146, + 0.394586043291443, + 0.44980683066855637, + 0.4600104964113993, + 0.44843873944918133, + 0.34169179457389987, + 0.38281126285693856, + 0.43071929105295303, + 0.3018562357496489, + 0.39899763953573636, + 0.3650601174296332, + 0.38924865559689975, + 0.34984008256949484, + 0.3975229656536275, + 0.3202680836085166, + 0.38985544152947244, + 0.3937192544143108, + 0.43178604474073323, + 0.2973043096252089, + 0.4498018279721987, + 0.41208782243767544, + 0.356609989484031, + 0.4886100124537749, + 0.4569586014182664, + 0.7314044554356353, + 0.20503021385913014, + 0.2721503791108632, + 0.326898733741167, + 0.31594300845355167, + 0.23940112841877093, + 0.20123699250408333, + 0.26605616706219554, + 0.6614852895286967, + 0.5360369596044561, + 0.6221704863311209, + 0.30577594019592114, + 0.3029792441639252, + 0.24489665295476667, + 0.33657370233988854, + 0.3729036224491345, + 0.20075138146551572, + 0.23783241276923706, + 0.3707456573591556, + 0.3398951159279847, + 0.401614291228932, + 0.3405187858840392, + 0.379638338068056, + 0.32933689437178815, + 0.2738132568626356, + 0.6825111073184067, + 0.4410999890699911, + 0.6824337143276169, + 0.5589179054861595, + 0.212196218291455, + 0.46732275519733013, + 0.44168433538758545, + 0.4632545604971889, + 0.431916109734847, + 0.3478683699810894, + 0.42721632585083485, + 0.4382663367828763, + 0.4645484098223293, + 0.4267049910491339, + 0.30299824026042566, + 0.3794245172106957, + 0.23078625347738596, + 0.25119828339365574, + 0.6217674567596327, + 0.43094338854117337, + 0.5571632551656815, + 0.45863783368275285, + 0.417271405045804, + 0.3972850354859646, + 0.36801769236475274, + 0.45577056045063546, + 0.3672893980420273, + 0.21392538301314637, + 0.3391651323727168, + 0.4222991013328265, + 0.22736120795170656, + 0.7490918489424108, + 0.3927476172844113, + 0.5518785502776652, + 0.7131084435001341, + 0.6734678019326579, + 0.3414410246525856, + 0.7935627529458101, + 0.46419539825094686, + 0.6462824153906954, + 0.3582672321007404, + 0.4121536465816863, + 0.3295207690909744, + 0.3125302277884027, + 0.4236597666092889, + 0.45739766438437507, + 0.38934065915331406, + 0.3103455763731363, + 0.3384191335705102, + 0.33943895423192094, + 0.36458963621903817, + 0.46446368644465574, + 0.22341954137854125, + 0.4461497547595447, + 0.3870244259152598, + 0.3654244553459715, + 0.44682625236663764, + 0.5162584378600572, + 0.27538237722154046, + 0.3599879495974438, + 0.3875640596704029, + 0.4909553218711682, + 0.4596228108069909, + 0.3716464353352707, + 0.25061457367800927, + 0.455712490384015, + 0.515340277912848, + 0.5017801115390631, + 0.3810333783807891, + 0.5588561234615768, + 0.5674650768083523, + 0.3898389482394629, + 0.45231601042778685, + 0.4936785476103599, + 0.4547586135384174, + 0.40811742541737706, + 0.30394589366695285, + 0.2531403786099415, + 0.2766721498401329, + 0.32491239924590215, + 0.26420244284084343, + 0.24350133952795117, + 0.28608809546049246, + 0.3114731791085618, + 0.32930114388594556, + 0.2681966229573469, + 0.28862754564694815, + 0.27995067940088014, + 0.2973426985139275, + 0.3609986826248688, + 0.2818186096912846, + 0.25650053515594334, + 0.3808541753669292, + 0.4769240394789731, + 0.6215793231621111, + 0.25555129938624577, + 0.47982228864939264, + 0.49324460565853767, + 0.49786977831174384, + 0.5071524724130861, + 0.3801227575914767, + 0.41417683255921267, + 0.3404449481875963, + 0.43603570489100757, + 0.3687814572654811, + 0.6098399694966763, + 0.41376088613831113, + 0.4488300113843384, + 0.242312804381988, + 0.22539044658595195, + 0.36562463481076557, + 0.4437544617044048, + 0.32379493147125343, + 0.539246808757302, + 0.30093444822282556, + 0.5955540183618764, + 0.4555300821862748, + 0.46927774953187834, + 0.46980313222638237, + 0.4577029197929807, + 0.376458678016438, + 0.4854197076023829, + 0.4783633453569454, + 0.4799004416203157, + 0.47865562664666034, + 0.39264183717799983, + 0.3036016881742342, + 0.23866665750019558, + 0.33745919241838795, + 0.21337340114337564, + 0.3555354416392005, + 0.38880962570805816, + 0.3957504877679929, + 0.3792370393896236, + 0.40753637716954233, + 0.3775351943195205, + 0.46544094539581987, + 0.42331055611346396, + 0.30434373235693324, + 0.4323043524535827, + 0.4992826992198969, + 0.31028555874963637, + 0.2651798619753589, + 0.3120459019886651, + 0.3486021102773114, + 0.4206086428084667, + 0.22963746949366878, + 0.39648312645980227, + 0.23245739612485403, + 0.37413541639128317, + 0.43587889461570023, + 0.4497384257232802, + 0.20856043551536674, + 0.27711742584866333, + 0.2956213861970448, + 0.20189464560557485, + 0.19206152641535298, + 0.2987466272973289, + 0.4014560132453613, + 0.24379267136410923, + 0.31164455897350524, + 0.40472214787310756, + 0.28477588000143056, + 0.2943027858213595, + 0.3982400256856634, + 0.45082930909929075, + 0.18548494394011555, + 0.4227578524402589, + 0.34573369777781116, + 0.29094666785104256, + 0.2647035744854017, + 0.41767171612481635, + 0.36557325885877273, + 0.18463315136913208, + 0.4382025456769042, + 0.37618134284236726, + 0.34691123481430514, + 0.45006894028304506, + 0.3129522775192849, + 0.2592178327820836, + 0.3101530516137147, + 0.22224036738860364, + 0.332023841570314, + 0.30961185124597385, + 0.3297732154501518, + 0.35913412075349244, + 0.36209081535861315, + 0.4415446842319621, + 0.37605821175730675, + 0.29887772016328396, + 0.3424658930762259, + 0.38531042362530277, + 0.36522866443459717, + 0.3331488765601262, + 0.4331805516749334, + 0.37209467382216416, + 0.48045782950841137, + 0.5156651302116723, + 0.4757403788176842, + 0.41490531606824993, + 0.30642454312137324, + 0.22128605468117615, + 0.35036138124791094, + 0.41714056411485223, + 0.3570204316980235, + 0.40942724521928503, + 0.4059916061823788, + 0.2773779603657326, + 0.4192249645496781, + 0.28781111097122536, + 0.23640732737736142, + 0.3648718372807607, + 0.383739423118871, + 0.41542820884714265, + 0.4223935009242307, + 0.26043701600199776, + 0.5377277425320252, + 0.44174636237940385, + 0.44183202647907116, + 0.4802920253138422, + 0.549408737166625, + 0.3888678766548841, + 0.27957075717904106, + 0.35524704322391293, + 0.3601749605910111, + 0.3536753557736666, + 0.33456850166756097, + 0.29398883354468464, + 0.30949753076897113, + 0.3627119442716134, + 0.25653188237278557 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.16533045465237003, + "valid_tracks": 4, + "per_track_pearson": [ + 0.2137139086732069, + 0.14132573787389827, + 0.18109895147234764, + 0.12518322059002726 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.16644348238907047, + "valid_tracks": 626, + "per_track_pearson": [ + 0.19831799048962967, + 0.1917843170786137, + 0.17866366622872892, + 0.19186057426622705, + 0.19723807248222655, + 0.19636072363090773, + 0.14167018801678932, + 0.166659821495055, + 0.20959774967437472, + 0.1905682918723681, + 0.1930316844451493, + 0.1947076791662904, + 0.19861125999738644, + 0.19478173741891003, + 0.18709080835971195, + 0.19039298897650964, + 0.19784366327651293, + 0.18668169687068714, + 0.1975397215374358, + 0.20589274484752407, + 0.20843507521924795, + 0.1903394959756641, + 0.19208567861718936, + 0.1531938233480093, + 0.1897239751791126, + 0.18340344324507377, + 0.17331091794955053, + 0.19711111139078177, + 0.18468437332711593, + 0.1753059205561976, + 0.17764170700346912, + 0.17597485513362, + 0.15040226294557738, + 0.20311396432301637, + 0.1483835030232424, + 0.19740350233343776, + 0.1414776906929419, + 0.1952113512048878, + 0.18895187419314727, + 0.2113853720598248, + 0.20505749863112607, + 0.17393116219884122, + 0.17684133987780645, + 0.16992434274046372, + 0.17283656749119877, + 0.15856983542122657, + 0.17563317295933686, + 0.18548073067874943, + 0.17478946914392815, + 0.16883911304988417, + 0.1840338821694419, + 0.15811341899365836, + 0.16604432308674144, + 0.1833434409511708, + 0.1770444438890449, + 0.16497537547438298, + 0.16934953481569134, + 0.1756755863937964, + 0.1940908486610027, + 0.19364507118245022, + 0.17615622326135216, + 0.19271913633235704, + 0.19529753407667497, + 0.1839692097024448, + 0.19148723950706395, + 0.18408360744393942, + 0.19593535248791963, + 0.18407937757064016, + 0.18833225479022367, + 0.17148126797275132, + 0.18061799653410304, + 0.1925932640808893, + 0.18306397693714108, + 0.17841561855426125, + 0.19373870509813412, + 0.1907549968306469, + 0.18316083506854816, + 0.17188748545339172, + 0.17919571288698793, + 0.1736637745779699, + 0.18269627215763823, + 0.1792530026397191, + 0.17727095664413964, + 0.17652045165995608, + 0.1794922877650703, + 0.19800420187202733, + 0.17955914834721529, + 0.18258835362481124, + 0.1856208435297152, + 0.17951292370301294, + 0.17972993295296882, + 0.17597296922968644, + 0.1803463336563356, + 0.1938812006299097, + 0.17837817072520487, + 0.19673962333356182, + 0.19334238046776483, + 0.19350466548066783, + 0.19125655331547006, + 0.18495141077676547, + 0.1920128998873029, + 0.18169826780284934, + 0.1911975582649802, + 0.2027422257621487, + 0.19406858952952286, + 0.16982976846866013, + 0.15390154882354234, + 0.17778643339144132, + 0.19945992668385218, + 0.19521319739295065, + 0.19181544690398458, + 0.19638650943453295, + 0.19524078508692772, + 0.19763286551560447, + 0.19854834454908393, + 0.19333476980426484, + 0.18555722677816197, + 0.18324774544873906, + 0.19478309081110268, + 0.19076600139518032, + 0.1855000541138186, + 0.1958402982999053, + 0.19472773966468068, + 0.19483595147248953, + 0.16159568203575567, + null, + 0.19788671097120558, + 0.18157217336967016, + 0.19621672641675808, + 0.19117192891141296, + 0.16366160294220566, + 0.20091777548996687, + 0.19503527659946834, + 0.16889813679885127, + 0.18198206101655123, + 0.19627915475137633, + null, + 0.19378962705034566, + 0.19991283114258762, + 0.15799301126348647, + 0.16853182293907298, + 0.18048931544389105, + 0.18229065353346843, + 0.1892509413706553, + 0.18617139534809712, + 0.18611511622327098, + 0.17736853080875792, + 0.19075503239074373, + 0.17320875719088225, + 0.17385642934070322, + 0.16543587712263302, + 0.1851950051505131, + 0.17660994865575777, + 0.18264533507133462, + 0.16790755079995914, + 0.16590599939278458, + 0.1755354378716252, + 0.1614149422485201, + 0.17840466236355607, + 0.1715291649187136, + 0.18031611130104863, + 0.18095529695669432, + 0.19731938441604902, + 0.19626260933321443, + 0.18665251602991634, + 0.18777929499694468, + 0.1764171153411306, + 0.18464692359223125, + 0.1570383025388149, + 0.1815522527270918, + 0.1741729139935207, + 0.1835399186091503, + 0.182397502896064, + 0.18591553929799473, + 0.18236151762989497, + 0.14490215424496458, + 0.1904202652462775, + 0.18089350457157363, + 0.18124624134060496, + 0.18774010725955814, + 0.17189329334170994, + 0.16944541169612334, + 0.1648443306124491, + 0.1968212752667746, + 0.18776083468376048, + 0.12460417381637232, + 0.1979265100154292, + 0.19997858112216363, + 0.19064318868443994, + 0.1931378367005137, + 0.18493165518416643, + 0.17390063723034882, + 0.18521562681145173, + 0.18933641641503415, + 0.19256851894784444, + 0.19368219645720944, + 0.19183100143703422, + 0.18459290034097925, + 0.20682680026899058, + 0.1926396335098571, + null, + 0.18472918368794158, + 0.18833796837592104, + 0.18538998579735538, + 0.18918453509145722, + 0.1818315737604654, + 0.178741918854859, + 0.18413854711772112, + null, + null, + 0.17843211798260927, + 0.1823950730436229, + null, + 0.2044601819826942, + 0.17320210748717915, + 0.1668919091386345, + 0.18589835947100317, + 0.19374519526280154, + 0.1892164970876516, + 0.18536415577768522, + null, + 0.18676219581804437, + 0.17419407078370286, + null, + 0.1674730769374932, + null, + 0.1771434903666371, + 0.18787681454284785, + 0.20173287324914102, + 0.17135688692739476, + null, + 0.1683701286917603, + 0.1688110702909937, + 0.1809511742946934, + null, + 0.1924749984273734, + null, + 0.18941696376738165, + null, + 0.18451354187921584, + 0.19157668551903723, + null, + 0.18057175205636586, + 0.18970927331010784, + null, + null, + 0.20147139985548718, + 0.19282180890211753, + 0.2035525977232577, + 0.19325174703200543, + null, + 0.18852805976000175, + null, + 0.19577434064608285, + 0.18279756144360249, + null, + 0.15434085017113777, + 0.19387532717504663, + 0.18649475336732102, + null, + 0.19142943973284388, + 0.19553515048734993, + null, + 0.18240179314813998, + 0.18166832850255024, + 0.18389212074735575, + null, + null, + null, + null, + 0.19160727013641154, + null, + null, + 0.19957202265092328, + 0.18426041839766372, + 0.1800968694622581, + null, + 0.19574646124760836, + 0.18802931875548065, + null, + null, + 0.1780249082505482, + 0.1876195689719482, + 0.17555812989011996, + 0.19302880046962492, + 0.19536539407913872, + 0.1926774440324335, + 0.18670621310875696, + 0.19495442995214177, + 0.18622437737647599, + 0.19106291670421735, + 0.17946431667048, + 0.19500916320331937, + null, + 0.1950612551377892, + 0.19027645849094366, + 0.1821641895406713, + 0.1858600001177454, + 0.18507999167441133, + 0.1783711513121435, + 0.18695230671900276, + 0.17752302062321468, + 0.1815562358628648, + 0.18433422390394488, + 0.1808792910511419, + 0.17608735099794265, + null, + 0.19303719453197105, + 0.1888567873973912, + 0.19038412078348338, + null, + 0.1942914273768442, + 0.16176641039044115, + null, + 0.201767711945539, + null, + 0.18532297356810476, + 0.1796570764354977, + 0.19173138284112462, + 0.17450651109737053, + 0.1785836151088818, + 0.19425249591517368, + 0.19065926940133635, + null, + null, + 0.19466264059893287, + 0.18767740899887217, + null, + 0.18971977563509215, + null, + 0.1908801444319977, + 0.18103053000446875, + 0.18257537052105954, + 0.18971070389330566, + null, + null, + null, + null, + 0.20081024714431128, + null, + 0.18643850860862413, + 0.19203863604963495, + null, + null, + 0.188469859822444, + 0.19777249519622814, + 0.19298776324556446, + null, + 0.18492247295041533, + 0.18255868940527373, + 0.18968262826000523, + null, + 0.19240499102984615, + null, + null, + 0.1911420175120309, + null, + 0.18821102904890966, + null, + null, + 0.18625797042242787, + 0.1933214158844085, + 0.1841220095888464, + 0.1893702417730721, + 0.1904069001489017, + null, + 0.19027136535620898, + 0.16550294869512003, + 0.15953286082186255, + 0.12094320779755956, + 0.12751972600417633, + 0.16307807644794733, + 0.14466005826003903, + 0.12149906798234883, + 0.14338038411703602, + 0.1579215861193211, + 0.15852492873739282, + 0.16724415503084275, + 0.160323211934284, + 0.1359536594272252, + 0.16066227913696862, + 0.13168985665238414, + 0.13475429714329631, + 0.1577409463016962, + 0.16254514003758846, + 0.1721011455871839, + 0.16669305679341376, + 0.16170464930777448, + 0.14520916836302178, + 0.13485523519198409, + 0.10088020025568294, + 0.1691840945551052, + 0.14813134998733501, + 0.16135571214625277, + 0.16225718566110875, + 0.1451451997159408, + 0.14678347209700962, + 0.14237843447319895, + 0.12922062577359697, + 0.10701112340364173, + 0.13483415144415348, + 0.09775876186884004, + 0.1342901419462867, + 0.09166241796720419, + 0.16171957380681415, + 0.16646749602803995, + 0.14744097407606752, + 0.14203076810764126, + 0.1107696715536744, + 0.1162171711474044, + 0.12628789454808292, + 0.14083257018757114, + 0.10528194043375046, + 0.12210749920128795, + 0.11338186072698818, + 0.12564927454246871, + 0.11179019585020246, + 0.11394136510693746, + 0.1165433644322789, + 0.12974049405751473, + 0.13087983585273696, + 0.10844133379196295, + 0.12890268256596368, + 0.1433972161701249, + 0.15550953406906964, + 0.15165666048556076, + 0.16226867397965775, + 0.1494599987476577, + 0.16986122566425982, + 0.1545261066275224, + 0.15474446963633148, + 0.166188319770224, + 0.1754853522684625, + 0.1500416057378473, + 0.15776935465781244, + 0.1591431551928453, + 0.13787600100332242, + 0.17373370443775335, + 0.1617019128105191, + 0.15429275054382097, + 0.1550183512785279, + 0.15518741959545512, + 0.15011470921094738, + 0.14894410797608906, + 0.15816824987005304, + 0.14157980594518488, + 0.15776129391779503, + 0.16099411883898374, + 0.1571366002116465, + 0.1628168447094452, + 0.1514863355002764, + 0.16084992850550536, + 0.15548870514445964, + 0.1533850814533808, + 0.15994876604063327, + 0.1434971708411373, + 0.15842249068179473, + 0.16067277727245047, + 0.16003918514699242, + 0.16687256246530754, + 0.15861963863460676, + 0.15432465320206698, + 0.16328393593382312, + 0.15536221214204313, + 0.16886288092124302, + 0.1722629907691208, + 0.1760923458292073, + 0.1537765027772652, + 0.15403392845483882, + 0.15940512114082483, + 0.1609895001932843, + 0.15898759048593356, + 0.1682876769910025, + 0.10230319997650592, + 0.15663447337726236, + 0.15195990150240718, + 0.14865019738415577, + 0.1421476370094138, + 0.150798340562725, + 0.1567630072557572, + 0.1394928294138735, + 0.15401644466991277, + 0.16033565138137734, + 0.15470535575165847, + 0.15779269902195087, + 0.12990953776745445, + 0.1392480629855765, + 0.15978581696769792, + 0.1567632554868985, + 0.1632406260700506, + 0.17031016668378585, + 0.11671130769871763, + null, + 0.15747057470725206, + 0.1333944026545829, + 0.1455320112440981, + 0.132856475746164, + 0.1403192116999784, + 0.16006451142856443, + 0.16006098427419835, + 0.15580314802367545, + 0.1501272142877578, + 0.1468270131009508, + null, + 0.15657650152567848, + 0.1468463664845111, + 0.13663562057922193, + 0.14683413426870293, + 0.144176480617926, + 0.14124612894401564, + 0.15691981823849993, + 0.15289749868895042, + 0.14811216641891234, + 0.14167662433342695, + 0.15348638766225012, + 0.15561781140541653, + 0.15622321482266527, + 0.13585282231595192, + 0.15032667820308337, + 0.16789317491641406, + 0.13025562795383566, + 0.11991028607258, + 0.13563856135486657, + 0.12461238081418811, + 0.16148645581052992, + 0.15390930015566884, + 0.1526597733753362, + 0.13752661159358856, + 0.1422111564341705, + 0.16569539989129217, + 0.1379833722356209, + 0.15305050332835698, + 0.14504369908280007, + 0.17145594576562923, + 0.1394576125970682, + 0.15324574610244787, + 0.14248440284391758, + 0.1505499320497631, + 0.16991398016705722, + 0.1492710517498338, + 0.1570017312873873, + 0.15322384840342626, + 0.1369020215497776, + 0.15421925173636053, + 0.15140354134866932, + 0.14809705506928067, + 0.13977584015748673, + 0.1576137748549573, + 0.15400636013450802, + 0.14056102364325845, + 0.18202024715454784, + 0.14678036448911183, + 0.14621553379160074, + 0.15883585557010943, + 0.14254249688295445, + 0.16115503879828744, + 0.15523098990208375, + 0.16190576617005167, + 0.14918987965518019, + 0.14527004068787344, + 0.14299301756296426, + 0.14026540101104445, + 0.1322551512662599, + 0.13518373758183855, + 0.15693241249022347, + 0.1634571635448689, + 0.15841350036754953, + null, + 0.16507434273707483, + 0.15323706108961271, + 0.13198867490565178, + 0.1493423517176986, + 0.16294272359352155, + 0.1518921713094191, + 0.13276399717867576, + null, + null, + 0.14378433870860055, + 0.14465954669224648, + null, + 0.15076201373172904, + 0.1226798840029646, + 0.14980011759861914, + 0.13337982916842675, + 0.13705879562874218, + 0.14071684613456606, + 0.16781593615582463, + null, + 0.15424698248117052, + 0.1593667166457693, + null, + 0.12585562874081874, + null, + 0.14256482972715134, + 0.1526898544342958, + 0.14035385966699795, + 0.1446815642761676, + null, + 0.14099271077869568, + 0.14615486570613326, + 0.15618273782052994, + null, + 0.12497347291202984, + null, + 0.1478823236187781, + null, + 0.15075571339529636, + 0.14624783971501204, + null, + 0.14110250349759804, + 0.16977981024039832, + null, + null, + 0.16291777101987187, + 0.1494331495700925, + 0.15008626308383352, + 0.13570191791078814, + null, + 0.13144431902620934, + null, + 0.15045919976545372, + 0.15033769241738523, + null, + 0.1213895985571195, + 0.13132210060201246, + 0.1428150199362992, + null, + 0.16235739211923383, + 0.139449209824817, + null, + 0.17427428803860068, + 0.16632905452972235, + 0.16991596930977657, + null, + null, + null, + null, + 0.17526156611032667, + null, + null, + 0.15881017047675702, + 0.1472280764429322, + 0.17076825146574176, + null, + 0.17570525149671654, + 0.17413775552511174, + null, + null, + 0.15142136344120652, + 0.14401929184930531, + 0.1494638228907775, + 0.14878468272490541, + 0.14924123234313882, + 0.15942237577589524, + 0.11041678976732462, + 0.1393828392868761, + 0.12513792539421645, + 0.14518882806096428, + 0.15446104960366933, + 0.1320042044348644, + null, + 0.1343151929338886, + 0.13887643897609195, + 0.16166053762499996, + 0.1622597887218541, + 0.14580089915617855, + 0.16581540603608647, + 0.14499239231860386, + 0.15801958828971388, + 0.15962388465179897, + 0.13552282095909393, + 0.1342879093445931, + 0.1339006741065294, + null, + 0.13706419501892297, + 0.16906544748645158, + 0.17475715143131038, + null, + 0.1397494466634862, + 0.14225410233123786, + null, + 0.15033792218276482, + null, + 0.1529733041646185, + 0.1471717027955202, + 0.15781254738835443, + 0.1451000786005516, + 0.1606626371131727, + 0.14352531966912388, + 0.1281642897830354, + null, + null, + 0.1501690577411088, + 0.1608259505855279, + null, + 0.14547067461283875, + null, + 0.1528156201347026, + 0.1425595496810254, + 0.1399109103275546, + 0.1641800446856283, + null, + null, + null, + null, + 0.14395454298372137, + null, + 0.1496892179904156, + 0.15888929859680342, + null, + null, + 0.14888228746761095, + 0.1294009342847469, + 0.12703201173584208, + null, + 0.1322402777356673, + 0.13967079647320213, + 0.1438169760504569, + null, + 0.14621362515736044, + null, + null, + 0.13931280808261673, + null, + 0.13638267512351948, + null, + null, + 0.13621894906471382, + 0.1555861426868818, + 0.1658159999027394, + 0.17709974950843735, + 0.16291495429730016, + null, + 0.14796071592160445 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753, + 0.7480558701396753 + ] + }, + "junctions": { + "mean_pearson": 0.11819718243413396, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2902658914420824, + 0.23461437922486775, + 0.29521381878521474, + 0.22283628402593508, + 0.2736630856977953, + 0.2730546871376611, + 0.03783573824122103, + 0.12119436838368598, + 0.22320000105962845, + 0.1674478520562881, + 0.18800445629798246, + 0.12212774188298334, + 0.06564245402513225, + 0.09393550551689253, + 0.19856861600428066, + 0.1804495132149428, + 0.10261884855679751, + 0.061435475650794304, + 0.22706850738895176, + 0.20804828787132057, + 0.27267492126916276, + 0.23374461236682334, + 0.23849664220773292, + 0.14839366776716434, + 0.18111433603715443, + 0.22383184064020048, + 0.1567169258183633, + 0.24263704338144934, + 0.14785527893454917, + 0.15010103908083489, + 0.22567051985707248, + 0.30915609169809977, + 0.224682604982667, + 0.31607827330222876, + 0.2428974074002046, + 0.28425765016428794, + 0.28568380301264745, + 0.2044918804745652, + 0.21257228904876108, + 0.12278536763619, + 0.14024509470735816, + 0.1947581098793242, + 0.19960511948230306, + 0.18930000954757686, + 0.18587632212934582, + 0.14306617330704982, + 0.15304288414515158, + 0.19255679873047352, + 0.17135398708395624, + 0.16879292298075504, + 0.21252114048966092, + 0.2377263158787338, + 0.13986461306339493, + 0.2735650584390085, + 0.1630209163283301, + 0.23025260102128983, + 0.1607156971941222, + 0.08318461763370992, + 0.1781697615488755, + 0.1312661291120525, + 0.04709165457877786, + 0.1462682179283785, + 0.17003259813077284, + 0.09915929406757826, + 0.06190642361851705, + 0.23423021618257803, + 0.20103999379112938, + 0.2163318975114487, + 0.13724296081258128, + 0.14075126696168036, + 0.18690670851017419, + 0.21606812470602246, + 0.1655865167688358, + 0.06363105122017595, + 0.4060035726852439, + 0.33519259460492135, + 0.18024829490273975, + 0.11481487176081306, + 0.12966724898062046, + 0.24833045559729053, + 0.1722256137123989, + 0.1842621744708491, + 0.18161110877640851, + 0.14883419012580018, + 0.1623881098501496, + 0.24156830432742557, + 0.258471539758443, + 0.2629922351122241, + 0.24695815515280378, + 0.20073127778466268, + 0.24158864966210267, + 0.19638325050733665, + 0.10407475661484124, + 0.19475015436350862, + 0.20427312878946133, + 0.33446781457183444, + 0.23795593983480876, + 0.17256911902548433, + 0.20025749763742723, + 0.20251802199919924, + 0.12794565562223664, + 0.1849313554499684, + 0.3798911455492873, + 0.2035997584907655, + 0.18275006447009026, + 0.13101197599664804, + 0.07538041695926306, + 0.23822488822590715, + 0.33506617904326386, + 0.30402774088595425, + 0.11015582087772229, + 0.3915891914412481, + 0.30611227389522383, + 0.3805681969464538, + 0.30171522774960446, + 0.12092490551319497, + 0.07951831225940074, + 0.14216061273836852, + 0.11132149026187367, + 0.23603282019893837, + 0.2091440920393534, + 0.14681556866095719, + 0.12821301624444742, + 0.20667546854306698, + 0.07431052057292076, + null, + 0.30046981525768524, + 0.178536238457799, + 0.06313850504552822, + 0.17998250011188208, + 0.23418809005488359, + 0.1754801257554113, + 0.12695276154387233, + 0.21963236102969802, + 0.2875290811594995, + 0.12869622730281255, + null, + 0.2979129347259227, + 0.29783167282511785, + 0.23430912136395904, + 0.23201862296080247, + 0.21623610566037943, + 0.1934604839768065, + 0.17828801491348936, + 0.1793756102785928, + 0.21147086055798667, + 0.2382312971293661, + 0.19366955803839866, + 0.13432220096321995, + 0.18896099158369342, + 0.14194669805171, + 0.22809664436069535, + 0.16361047671117085, + 0.25312265410048485, + 0.38390855858546324, + 0.29553266074796297, + 0.37488128998774795, + 0.19177119928916034, + 0.16888760533296068, + 0.2084580037088189, + 0.16452153462067312, + 0.24642144273389494, + 0.23653473150638898, + 0.2607352631866807, + 0.10391082743999644, + 0.31240860420819616, + 0.2478509172019855, + 0.15247730557726716, + 0.05854624539178005, + 0.17496549361540423, + 0.20986878097900302, + 0.19634406595216364, + 0.2705346517960266, + 0.08976821312806385, + 0.13529883997120848, + 0.18749342801158614, + 0.14520472546441165, + 0.16795952213749943, + 0.13319061578931693, + 0.20331713058641213, + 0.07737117384737507, + 0.11356647241330094, + 0.044888405605995885, + 0.21802561945155333, + 0.22737727815397796, + 0.22666947702799833, + 0.28877292574296365, + 0.20832601598646333, + 0.285249946172617, + 0.2161345182636188, + 0.2937664670173916, + 0.33512265533375507, + 0.18035857148111406, + 0.12358809689410448, + 0.14083329419185514, + 0.11543396604304736, + 0.09564843873396688, + 0.15814604791381345, + 0.30531997783873344, + 0.38577346350324393, + null, + 0.17616077769615518, + 0.18888774503220682, + 0.2254640182068243, + 0.27264793660049885, + 0.10084030234706709, + 0.1467652655308685, + 0.07821641076197768, + null, + null, + 0.21966736161627817, + 0.2738901312153522, + null, + 0.13119655314793055, + 0.12481384494401694, + 0.07566787550370975, + 0.1700198817670209, + 0.10744367362294947, + 0.17506390671484662, + 0.06375110140906415, + null, + 0.3057973792408819, + 0.1448245046503137, + null, + 0.11464445064322788, + null, + 0.20261958339235192, + 0.2540385069109749, + 0.16082117647369534, + 0.1653649097157571, + null, + 0.08276269043507166, + 0.04522639724909839, + 0.20104036123991936, + null, + 0.1517351452187132, + null, + 0.09077241850477677, + null, + 0.11267935852660736, + 0.1391798547437332, + null, + 0.17521031861218986, + 0.19427098391751768, + null, + null, + 0.2341833137455263, + 0.26175500592774986, + 0.04650559118373123, + 0.08048969284827492, + null, + 0.21117157583421758, + null, + 0.19960618496769253, + 0.1952754099339412, + null, + 0.16834531512671805, + 0.26732220589073824, + 0.23756942846329057, + null, + 0.2392974678873182, + 0.24739547556268915, + null, + 0.2723732435192056, + 0.18917202290579468, + 0.2679031412228103, + null, + null, + null, + null, + 0.1364590691724613, + null, + null, + 0.16189854872262036, + 0.2501784758902929, + 0.2347528472761586, + null, + 0.11079759230703039, + 0.28406583728388984, + null, + null, + 0.16455647472826326, + 0.09022316002932418, + 0.22726715628266872, + 0.1204034878718086, + 0.1226803293623104, + 0.09477805166896688, + 0.13661014583475245, + 0.14910514700790534, + 0.03291897284814954, + 0.09940978635898559, + 0.33869092623220476, + 0.17944027210392668, + null, + 0.18789283525097494, + 0.18949907464023022, + 0.04078108600226243, + 0.15081637616533, + 0.09233337965599799, + 0.06574399920789706, + 0.10424069855354869, + 0.1600641008814888, + 0.2360786522577667, + 0.16632772251427413, + 0.21919791762489987, + 0.17776384680093682, + null, + 0.15110715376693232, + 0.19239777969347366, + 0.20176090691742715, + null, + 0.25289313288639886, + 0.18019711940228403, + null, + 0.17581916519787283, + null, + 0.1657981178957035, + 0.14778984723470506, + 0.21929051491076354, + 0.11424997449374384, + 0.17663279032624138, + 0.18814319794224724, + 0.23048556471408796, + null, + null, + 0.09196764362469859, + 0.11344566473665683, + null, + 0.19909491994433956, + null, + 0.19826738307621888, + 0.08985107623229861, + 0.07189490509044112, + 0.1779100782666768, + null, + null, + null, + null, + 0.019894135338419147, + null, + 0.07468434948840343, + 0.2277671321095792, + null, + null, + 0.1521839171267367, + 0.2373728270504875, + 0.16786662981703335, + null, + 0.22519062168046977, + 0.2172339701739609, + 0.06138236969921056, + null, + 0.20045594671569958, + null, + null, + 0.25465191404246007, + null, + 0.16478132757447525, + null, + null, + 0.17999103044752118, + 0.2065979379613223, + 0.18150014380420892, + 0.20488922660339798, + 0.16148530960181268, + null, + 0.13466080971097366, + 0.008789468491418291, + 0.06588108674447536, + 0.1550497384139887, + 0.0528051079202445, + 0.15023396925421834, + 0.05456974109542695, + -0.031551643876968724, + 0.011295186228405033, + 0.011825639846405363, + 0.06708976298994226, + 0.05337206893306165, + 0.06351564159413037, + 0.04089132749097856, + -0.0022636399807489417, + 0.09347126076462922, + 0.0603026977440796, + 0.0051047510945245525, + -0.031826880543757735, + 0.08885266360159692, + 0.040913565495078526, + 0.021210937305415654, + 0.1559049436506531, + 0.09616012377978496, + 0.07986241272919879, + 0.029498337320117193, + 0.0010402524195321618, + 0.06430968693320824, + 0.04989237547760234, + -0.021409897168586135, + 0.005412337431957856, + 0.044541281456393696, + 0.037856271008671546, + 0.0203703120954007, + 0.10911975678795374, + 0.02069236980034915, + 0.09094139105784824, + 0.05328689776605196, + 0.02813497719333625, + 0.03788459520392296, + 0.10301181498265997, + 0.03339323801019781, + 0.024717275666651346, + 0.011877984493564855, + 0.0006016298068552481, + 0.04284565645236985, + -0.014733264971192366, + 0.03190561754020212, + 0.12123243414703117, + 0.03210502218497117, + 0.05514262426901867, + 0.04049574322051008, + 0.058716360901255614, + 0.031037454051248047, + 0.03570227167724271, + 0.012069448134443663, + 0.053522673548208444, + 0.08593571546933851, + 0.07893799578690117, + 0.0982507224108907, + 0.03592928221336278, + 0.0003624189147828933, + 0.002478228095524609, + 0.03770455476873961, + 0.10971254022546037, + 0.08526674189493672, + 0.07780215274716204, + 0.06944857947195249, + 0.02448608054946696, + 0.04362550770222404, + 0.01261300415909968, + 0.08222178914579929, + 0.030084782727250733, + 0.058314372548305626, + 0.07422827645393314, + 0.023513779395563202, + 0.04026365141783864, + 0.013209834142149232, + 0.06207127591873982, + -0.016725975547740128, + 0.03775560614617522, + 0.023887131214631072, + 0.0728380767171579, + 0.071339106209365, + 0.016855729848396155, + 0.06342098008611799, + 0.0633088632438891, + 0.13101793089501454, + 0.042399272233159195, + 0.042370735519070245, + 0.008929963799280603, + 0.03834206341492904, + 0.02601545387877105, + 0.010223704501971317, + 0.017924790338941714, + 0.04158839936675773, + 0.05099569107378621, + 0.04178126604267644, + -0.011324149457176127, + 0.10673488042829113, + 0.008195218062751532, + 0.04497938217854739, + 0.06171149427529073, + 0.055259535646068836, + 0.10148146613040183, + 0.06574476246109917, + 0.035019655538183446, + -0.0077091801592042266, + 0.09210117418483489, + 0.04576832381172274, + 0.070832668949911, + 0.02009094244350934, + 0.06194262344578234, + 0.042652857185528956, + 0.0780757951241332, + 0.06602997573722469, + 0.03038454862761969, + 0.04133907901624571, + 0.01953815297024596, + 0.04776053859401494, + 0.06876819669214168, + 0.05203808791305409, + 0.023882235007430575, + 0.03752133312985066, + 0.08602151949501607, + 0.07580677846405703, + null, + 0.07022815446857067, + 0.06596059207549373, + -0.01087406889419873, + 0.06599281905415491, + 0.08863263958096808, + 0.0006223919015612164, + -0.008547417525700313, + 0.05006464639203788, + 0.03307652241545164, + 0.06046596283868293, + null, + 0.06859590247044144, + 0.06204802970635478, + -0.0184533692865297, + 0.01922235131716414, + 0.12370288832268915, + -0.007861824742072804, + 0.09747877822651659, + 0.05239341694552306, + 0.0010313010349639845, + 0.0457750121074979, + 0.008014242860373609, + -0.014906070616138013, + 0.049596845873897295, + 0.10710333815946703, + 0.05640364774691101, + -0.0045038535102068, + 0.016328680144160796, + 0.03257918292194778, + 0.05002952012937762, + 0.038114483806417084, + -0.014979085660140634, + 0.09681622633602342, + 0.09010504580085167, + 0.020958504263238815, + 0.05031945557667239, + 0.08405531605706461, + 0.08567538698421528, + 0.11098130471630485, + 0.07507526388661802, + 0.14639263839926467, + 0.029632571771892283, + 0.05082422390900129, + 0.20863847955418427, + 0.04031294807055757, + 0.003347849273132562, + 0.05389128195881587, + 0.083671341251755, + 0.023184234702493433, + 0.05219210240381874, + 0.07364869248199893, + 0.0454132231787843, + 0.0448326162694319, + 0.13515548502817706, + 0.007022133268755858, + 0.009830936934423131, + 0.3012579589028359, + 0.10394088496763972, + -0.018029376668490155, + 0.12441586296851335, + 0.046247772482867044, + -0.015944561647327445, + 0.08699841364208961, + 0.09380870299234637, + 0.20648413311838065, + 0.28009917542429846, + 0.06739522191860123, + 0.1711953810782445, + -0.011733365173459517, + -0.033155854665342405, + -0.026284961889463065, + 0.02135796204813457, + 0.07271951304844103, + 0.05403358824063568, + null, + 0.07223981205533511, + 0.07139186248928703, + 0.036075122880603, + 0.0444936165874379, + 0.15633464660680596, + 0.023350166662957503, + 0.068648990588453, + null, + null, + 0.06314419512265511, + 0.11792400616742021, + null, + -0.019389777625320157, + -0.010406103108132735, + 0.004442486447629215, + 0.04744649367944903, + -0.034014712839106426, + -0.0032177547009635053, + 0.028211017595009116, + null, + 0.0011044934245067924, + -0.001489878421264753, + null, + -0.030213295166591724, + null, + 0.05904281514084167, + 0.058064751969070856, + 0.0385876174527074, + 0.00024204910976168283, + null, + -0.024156355339872402, + 0.02781211916876546, + 0.01581082548357238, + null, + 0.011944238824365579, + null, + 0.1494040965075453, + null, + -0.029159148821950313, + 0.14224210174132407, + null, + 0.06590158275033242, + 0.051952656906622194, + null, + null, + 0.08752792160650033, + 0.0694858838346083, + 0.1269153214116636, + 0.08314108870944209, + null, + 0.012700662278579826, + null, + -0.00757234054063663, + 0.011390474266762003, + null, + 0.1575566534074691, + 0.01003142682251729, + 0.08501751906146676, + null, + 0.04018146720650186, + 0.017638040313896854, + null, + 0.07575859571501198, + 0.09923169641607381, + 0.12910939193492196, + null, + null, + null, + null, + 0.0717081012566274, + null, + null, + 0.04641626565150954, + 0.08622867706067605, + 0.058257099737117776, + null, + 0.01842847434518605, + 0.07063027549620259, + null, + null, + 0.01890068314860329, + 0.04998141745401549, + 0.07595403421052534, + 0.0074260130259238225, + 0.00437370559716815, + -0.02147361090098413, + 0.19054169161318374, + 0.023559753675429094, + 0.043337116524867506, + 0.0015763513538801842, + 0.049902371729888206, + 0.015259264605920688, + null, + 0.024110380324863914, + 0.04016316072535872, + 0.0604445946452271, + 0.03441392089084522, + 0.00562979162383074, + -0.044766061639403616, + 0.047298763388714415, + -0.015026628843491723, + 0.05024009792194072, + 0.04405590228202297, + 0.014295415835569845, + 0.04414155485415944, + null, + 0.02364953438435782, + 0.021641466940641743, + 0.06997875299503542, + null, + 0.0005455527230018109, + -0.025575589367190676, + null, + 0.05400806955013162, + null, + 0.020692670099955318, + 0.05262142763455807, + 0.024936190272892466, + 0.03197272123339553, + 0.0829022285249569, + 0.08822125062691172, + 0.09382949434956749, + null, + null, + 0.08795164722642643, + -0.005668408893389517, + null, + 0.014133033265808372, + null, + 0.13306352130999533, + 0.10122618337449991, + 0.1038749821317352, + 0.03404763748902796, + null, + null, + null, + null, + 0.03418418906677969, + null, + 0.05664818267770683, + 0.1402122495012327, + null, + null, + -0.0037899954832037066, + 0.14604151402934726, + 0.03484562943278646, + null, + 0.04791365649675741, + 0.011049221508263964, + 0.15242354171040579, + null, + 0.09065611175252886, + null, + null, + 0.028580352407696533, + null, + 0.04789915594934698, + null, + null, + -0.019662869033888628, + 0.0036671895750194994, + 0.039081542001949104, + 0.024034936444132744, + 0.016227249457711798, + null, + 0.09484244396942973 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9536423841059603, + 0.9407894736842105, + 0.9537366548042705, + 0.9525316455696202, + 0.9701986754966887, + 0.9525423728813559, + 0.9683098591549296, + 0.9548611111111112, + 0.9519230769230769, + 0.950354609929078, + 0.9494949494949495, + 0.9355932203389831, + 0.948051948051948, + 0.94375, + 0.9633333333333334, + 0.944078947368421, + 0.951048951048951, + 0.9519230769230769, + 0.9522292993630573, + 0.9509803921568627, + 0.9496402877697842, + 0.9483870967741935, + 0.948051948051948, + 0.9405204460966543, + 0.9537366548042705, + 0.9539473684210527, + 0.9615384615384616, + 0.9559322033898305, + 0.9527027027027027, + 0.9482758620689655, + 0.9486301369863014, + 0.9676375404530745, + 0.9365671641791045, + 0.9604863221884499, + 0.945054945054945, + 0.9495268138801262, + 0.9427480916030534, + 0.9404388714733543, + 0.9300699300699301, + 0.9463087248322147, + 0.9453924914675768, + 0.9553264604810997, + 0.9427480916030534, + 0.9518518518518518, + 0.9508196721311475, + 0.9459459459459459, + 0.948339483394834, + 0.9354838709677419, + 0.9361702127659575, + 0.95578231292517, + 0.9416058394160584, + 0.9595588235294118, + 0.9393939393939394, + 0.9523809523809523, + 0.9328358208955224, + 0.9455782312925171, + 0.9480968858131488, + 0.9503311258278145, + 0.9554794520547946, + 0.9471830985915493, + 0.9517241379310345, + 0.9573770491803278, + 0.9503311258278145, + 0.9522184300341296, + 0.9362416107382551, + 0.9525316455696202, + 0.9358108108108109, + 0.9563758389261745, + 0.9385665529010239, + 0.9575163398692811, + 0.9572953736654805, + 0.9451612903225807, + 0.9494949494949495, + 0.9602649006622517, + 0.942652329749104, + 0.9559322033898305, + 0.9517684887459807, + 0.9464882943143813, + 0.9551724137931035, + 0.95578231292517, + 0.9403973509933775, + 0.9588607594936709, + 0.9452054794520548, + 0.941747572815534, + 0.9543859649122807, + 0.9473684210526315, + 0.9452054794520548, + 0.9673202614379085, + 0.95, + 0.9493670886075949, + 0.9512195121951219, + 0.9553264604810997, + 0.9543859649122807, + 0.9523809523809523, + 0.9425675675675675, + 0.9618055555555556, + 0.9457627118644067, + 0.9377162629757786, + 0.9470198675496688, + 0.939873417721519, + 0.9466666666666667, + 0.9475409836065574, + 0.9395017793594306, + 0.9488817891373802, + 0.9590443686006825, + 0.9566666666666667, + 0.9325842696629213, + 0.9498327759197325, + 0.946843853820598, + 0.9459459459459459, + 0.95, + 0.9395973154362416, + 0.9577922077922078, + 0.9559322033898305, + 0.9489795918367347, + 0.9391891891891891, + 0.9448275862068966, + 0.9459459459459459, + 0.9422382671480144, + 0.9590443686006825, + 0.9464285714285714, + 0.9559322033898305, + 0.9411764705882353, + 0.9473684210526315, + 0.9433198380566802, + 1.0, + 0.9550173010380623, + 0.9407665505226481, + 0.9431818181818182, + 0.9514925373134329, + 0.9528985507246377, + 0.9517241379310345, + 0.9442508710801394, + 0.9360902255639098, + 0.9508771929824561, + 0.939799331103679, + 0.9919354838709677, + 0.9559322033898305, + 0.9514925373134329, + 0.9539473684210527, + 0.9525547445255474, + 0.9328358208955224, + 0.9522058823529411, + 0.9482758620689655, + 0.9368421052631579, + 0.9465648854961832, + 0.9488188976377953, + 0.9479166666666666, + 0.9264705882352942, + 0.9469964664310954, + 0.9407407407407408, + 0.9574468085106383, + 0.9413919413919414, + 0.9642857142857143, + 0.9517241379310345, + 0.9415807560137457, + 0.9545454545454546, + 0.9530685920577617, + 0.950530035335689, + 0.9416058394160584, + 0.9393939393939394, + 0.9583333333333334, + 0.9467084639498433, + 0.9407665505226481, + 0.939622641509434, + 0.9496402877697842, + 0.9446366782006921, + 0.9668874172185431, + 0.9433962264150944, + 0.9508771929824561, + 0.9597069597069597, + 0.9419795221843004, + 0.9548872180451128, + 0.9623287671232876, + 0.9423076923076923, + 0.953307392996109, + 0.9311594202898551, + 0.9442508710801394, + 0.9558823529411765, + 0.9440559440559441, + 0.9477351916376306, + 0.9440298507462687, + 0.9520295202952029, + 0.942652329749104, + 0.9577464788732394, + 0.959409594095941, + 0.9417808219178082, + 0.936026936026936, + 0.931899641577061, + 0.9508196721311475, + 0.9409722222222222, + 0.9522058823529411, + 0.9357142857142857, + 0.9401408450704225, + 0.9337979094076655, + 0.9293286219081273, + 0.9301470588235294, + 0.9479166666666666, + 0.9477124183006536, + 0.9566666666666667, + 0.9922480620155039, + 0.9572368421052632, + 0.9491525423728814, + 0.9501661129568106, + 0.9391891891891891, + 0.9614147909967846, + 0.9611307420494699, + 0.9375, + 1.0, + 0.9924242424242424, + 0.9636963696369637, + 0.9512987012987013, + 0.9923076923076923, + 0.9551724137931035, + 0.9304029304029304, + 0.9343065693430657, + 0.9419795221843004, + 0.9536423841059603, + 0.9517241379310345, + 0.9496644295302014, + 1.0, + 0.9565217391304348, + 0.9381443298969072, + 0.9924812030075187, + 0.93359375, + 1.0, + 0.9490909090909091, + 0.9473684210526315, + 0.9436619718309859, + 0.9528619528619529, + 1.0, + 0.9492753623188406, + 0.9366197183098591, + 0.9368421052631579, + 1.0, + 0.9461279461279462, + 1.0, + 0.935064935064935, + 1.0, + 0.9566666666666667, + 0.9530201342281879, + 1.0, + 0.929368029739777, + 0.9442622950819672, + 1.0, + 1.0, + 0.950354609929078, + 0.9575163398692811, + 0.9509803921568627, + 0.9461279461279462, + 1.0, + 0.9399293286219081, + 0.9922480620155039, + 0.948051948051948, + 0.9484536082474226, + 0.9930555555555556, + 0.95, + 0.9421221864951769, + 0.9418181818181818, + 0.9927536231884058, + 0.9321428571428572, + 0.9565217391304348, + 1.0, + 0.9540983606557377, + 0.9403508771929825, + 0.9559322033898305, + 1.0, + 1.0, + 0.9922480620155039, + 1.0, + 0.9401993355481728, + 1.0, + 0.984251968503937, + 0.9489795918367347, + 0.9432624113475178, + 0.9519230769230769, + 1.0, + 0.9419795221843004, + 0.9452054794520548, + 1.0, + 0.9923076923076923, + 0.9407665505226481, + 0.9456521739130435, + 0.9498327759197325, + 0.9493243243243243, + 0.931740614334471, + 0.9387755102040817, + 0.9590443686006825, + 0.9411764705882353, + 0.9355932203389831, + 0.9252669039145908, + 0.945859872611465, + 0.9372693726937269, + 1.0, + 0.9508771929824561, + 0.956081081081081, + 0.9428571428571428, + 0.9562289562289562, + 0.9591836734693877, + 0.9583333333333334, + 0.9438596491228071, + 0.9448275862068966, + 0.9508196721311475, + 0.9493243243243243, + 0.9492753623188406, + 0.9419795221843004, + 1.0, + 0.9453376205787781, + 0.9493243243243243, + 0.9583333333333334, + 1.0, + 0.952076677316294, + 0.946236559139785, + 1.0, + 0.9347079037800687, + 1.0, + 0.9455128205128205, + 0.9479166666666666, + 0.9537953795379538, + 0.9347826086956522, + 0.9489795918367347, + 0.934931506849315, + 0.9405594405594405, + 0.9929078014184397, + 1.0, + 0.9411764705882353, + 0.9319727891156463, + 1.0, + 0.9386281588447654, + 1.0, + 0.946843853820598, + 0.9395973154362416, + 0.9418181818181818, + 0.9401993355481728, + 1.0, + 1.0, + 0.992, + 1.0, + 0.9452054794520548, + 1.0, + 0.9342560553633218, + 0.9619047619047619, + 1.0, + 1.0, + 0.954225352112676, + 0.946843853820598, + 0.9407407407407408, + 1.0, + 0.9417808219178082, + 0.9461279461279462, + 0.925531914893617, + 0.9919354838709677, + 0.9681528662420382, + 1.0, + 0.9922480620155039, + 0.9556313993174061, + 1.0, + 0.9514563106796117, + 0.9921259842519685, + 0.9922480620155039, + 0.9420849420849421, + 0.9512987012987013, + 0.9415807560137457, + 0.9484536082474226, + 0.9436619718309859, + 1.0, + 0.9403973509933775, + 0.9774436090225563, + 0.9859649122807017, + 0.9927536231884058, + 0.9857142857142858, + 0.9873417721518988, + 0.9822064056939501, + 0.9860627177700348, + 0.9811320754716981, + 0.9829351535836177, + 0.9856115107913669, + 0.9930795847750865, + 0.9855595667870036, + 0.9863013698630136, + 0.9863013698630136, + 0.9925650557620818, + 0.989010989010989, + 0.9788732394366197, + 0.9814814814814815, + 0.9857142857142858, + 0.9829351535836177, + 0.9755244755244755, + 0.9858657243816255, + 0.9732441471571907, + 0.9849624060150376, + 0.9857142857142858, + 0.9745454545454545, + 0.9867549668874173, + 0.9825783972125436, + 0.982078853046595, + 0.9802371541501976, + 0.9859649122807017, + 0.9852941176470589, + 0.9879032258064516, + 0.9822695035460993, + 0.9816849816849816, + 0.9787234042553191, + 0.9920318725099602, + 0.9680851063829787, + 0.9825174825174825, + 0.9784172661870504, + 0.9753521126760564, + 0.9817518248175182, + 0.9929328621908127, + 0.9882352941176471, + 0.9828178694158075, + 0.9820143884892086, + 0.9924812030075187, + 0.9847328244274809, + 0.9924528301886792, + 0.9884615384615385, + 0.9879518072289156, + 0.9888475836431226, + 0.9924242424242424, + 0.9822064056939501, + 0.9849624060150376, + 0.9814126394052045, + 0.9856115107913669, + 0.989247311827957, + 0.9893238434163701, + 0.9782608695652174, + 0.9895470383275261, + 0.9700374531835206, + 0.9692832764505119, + 0.989010989010989, + 0.9891696750902527, + 0.9889705882352942, + 0.9857142857142858, + 0.9922779922779923, + 0.9847908745247148, + 0.985663082437276, + 0.9817518248175182, + 0.9710144927536232, + 0.9830508474576272, + 0.985663082437276, + 0.978494623655914, + 0.9782608695652174, + 0.9758620689655172, + 0.9858657243816255, + 0.9820143884892086, + 0.9773584905660377, + 0.9757785467128027, + 0.9828767123287672, + 0.9809885931558935, + 0.9814814814814815, + 0.9858156028368794, + 0.9852941176470589, + 0.9809160305343512, + 0.9791666666666666, + 0.9817518248175182, + 0.978021978021978, + 0.9825783972125436, + 0.9822064056939501, + 0.9758620689655172, + 0.9888475836431226, + 0.9894736842105263, + 0.992831541218638, + 0.9792387543252595, + 0.966542750929368, + 0.9863013698630136, + 0.982078853046595, + 0.9731800766283525, + 0.9781021897810219, + 0.9855072463768116, + 0.975, + 0.9818840579710145, + 0.9895833333333334, + 0.9886792452830189, + 0.9888888888888889, + 0.9789473684210527, + 0.9893238434163701, + 0.9859154929577465, + 0.9887640449438202, + 0.9768211920529801, + 0.978494623655914, + 0.985663082437276, + 0.9769230769230769, + 0.9858657243816255, + 0.9807692307692307, + 0.9841897233201581, + 0.988929889298893, + 0.9859649122807017, + 0.989247311827957, + 0.9853479853479854, + 0.9858657243816255, + 0.9814126394052045, + 1.0, + 0.9822695035460993, + 0.9821428571428571, + 0.9858156028368794, + 0.9881422924901185, + 0.9675090252707581, + 0.986013986013986, + 0.9858657243816255, + 0.9849624060150376, + 0.9860627177700348, + 0.9847328244274809, + 0.9906542056074766, + 0.9825783972125436, + 0.9816176470588235, + 0.9790209790209791, + 0.9825783972125436, + 0.9851301115241635, + 0.9822695035460993, + 0.986013986013986, + 0.9858156028368794, + 0.9851301115241635, + 0.9925373134328358, + 0.9850187265917603, + 0.9819494584837545, + 0.9926739926739927, + 0.9807692307692307, + 0.981549815498155, + 0.985663082437276, + 0.981549815498155, + 0.9886792452830189, + 0.9827586206896551, + 0.9816849816849816, + 0.981203007518797, + 0.9898648648648649, + 0.9888888888888889, + 0.9782608695652174, + 0.9853479853479854, + 0.9838187702265372, + 0.986013986013986, + 0.9925093632958801, + 0.9881889763779528, + 0.9897260273972602, + 0.978494623655914, + 0.9877551020408163, + 0.9925650557620818, + 0.9891304347826086, + 0.9814814814814815, + 0.9922480620155039, + 0.9858657243816255, + 0.985239852398524, + 0.9878048780487805, + 0.9858156028368794, + 0.9857142857142858, + 0.9895104895104895, + 0.9927007299270073, + 0.9860627177700348, + 0.9888475836431226, + 0.9963235294117647, + 0.978021978021978, + 0.9788732394366197, + 0.9835390946502057, + 0.9776951672862454, + 0.9820143884892086, + 0.9788732394366197, + 0.9857142857142858, + 0.9896907216494846, + 0.992619926199262, + 0.9787985865724381, + 0.985239852398524, + 0.985663082437276, + 0.9820143884892086, + 0.9859154929577465, + 0.9807692307692307, + 0.9870967741935484, + 0.9795221843003413, + 0.9914529914529915, + 0.9822695035460993, + 0.9831649831649831, + 0.9813432835820896, + 0.9706959706959707, + 0.9828767123287672, + 0.9795918367346939, + 0.9752650176678446, + 0.9920634920634921, + 0.9916666666666667, + 0.987460815047022, + 0.9837133550488599, + 1.0, + 0.9824561403508771, + 0.9757785467128027, + 0.9846153846153847, + 0.9795221843003413, + 0.9790940766550522, + 0.9786476868327402, + 0.9858657243816255, + 1.0, + 0.978494623655914, + 0.9851851851851852, + 0.9923076923076923, + 0.9888059701492538, + 0.984, + 0.9796610169491525, + 0.9786476868327402, + 0.9831081081081081, + 0.9824561403508771, + 0.9925925925925926, + 0.9743589743589743, + 0.9894736842105263, + 0.9779411764705882, + 1.0, + 0.9822064056939501, + 1.0, + 0.9823321554770318, + 0.9920634920634921, + 0.9700374531835206, + 0.9854545454545455, + 0.9924812030075187, + 0.9745454545454545, + 0.982078853046595, + 1.0, + 1.0, + 0.9793103448275862, + 0.9823321554770318, + 0.9868421052631579, + 0.9831649831649831, + 1.0, + 0.980544747081712, + 1.0, + 0.9787234042553191, + 0.9813432835820896, + 1.0, + 0.981549815498155, + 0.9790209790209791, + 0.9851301115241635, + 0.9922480620155039, + 0.9855072463768116, + 0.9788732394366197, + 1.0, + 0.9863481228668942, + 0.9866220735785953, + 0.989247311827957, + 0.9852941176470589, + 0.9838709677419355, + 0.9920634920634921, + 0.9819819819819819, + 0.9826989619377162, + 0.9908256880733946, + 0.9908256880733946, + 0.9858156028368794, + 0.9863945578231292, + 0.9797297297297297, + 0.9922480620155039, + 0.986013986013986, + 0.9834437086092715, + 0.9915966386554622, + 1.0, + 0.9792387543252595, + 0.9828178694158075, + 0.9781818181818182, + 0.9735973597359736, + 0.9824561403508771, + 0.9744525547445255, + 0.9852941176470589, + 0.9897959183673469, + 0.9924528301886792, + 0.9773584905660377, + 0.976271186440678, + 0.9858156028368794, + 0.9917355371900827, + 0.96875, + 0.982078853046595, + 0.9829931972789115, + 0.9754385964912281, + 0.9825174825174825, + 0.986013986013986, + 0.9828767123287672, + 0.9743589743589743, + 0.9832214765100671, + 0.9823321554770318, + 0.9888888888888889, + 0.9890909090909091, + 1.0, + 0.9821428571428571, + 0.9821428571428571, + 0.9866666666666667, + 0.991869918699187, + 0.9826989619377162, + 0.9866220735785953, + 1.0, + 0.9672131147540983, + 1.0, + 0.9833887043189369, + 0.9857651245551602, + 0.9797297297297297, + 0.9803921568627451, + 0.9828767123287672, + 0.9738805970149254, + 0.9847908745247148, + 1.0, + 0.991869918699187, + 0.9859154929577465, + 0.978494623655914, + 0.9922480620155039, + 0.9893238434163701, + 1.0, + 0.9829351535836177, + 0.9826989619377162, + 0.9790209790209791, + 0.9816176470588235, + 0.9854014598540146, + 0.9921259842519685, + 1.0, + 0.9813084112149533, + 0.9894366197183099, + 0.991304347826087, + 0.9769230769230769, + 0.98989898989899, + 1.0, + 0.9923076923076923, + 0.982078853046595, + 0.9824561403508771, + 0.975, + 1.0, + 0.982078853046595, + 0.9825783972125436, + 0.9863481228668942, + 1.0, + 0.9805194805194806, + 0.991869918699187, + 0.9917355371900827, + 0.9818181818181818, + 1.0, + 0.9773584905660377, + 0.981651376146789, + 0.992, + 0.977859778597786, + 0.9688581314878892, + 0.9802631578947368, + 0.9790209790209791, + 0.9830508474576272, + 0.9915966386554622, + 0.9864406779661017 + ] + }, + "contact_maps": { + "mean_pearson": 0.06340988360421722, + "valid_tracks": 28, + "per_track_pearson": [ + 0.07055954824965306, + 0.0418391550801568, + -0.009086868990666956, + 0.041098115356746685, + 0.004703400863468791, + 0.04654886133669468, + 0.1048494983572494, + 0.07824843130034398, + 0.163696420920301, + 0.130058577101617, + 0.017508397662956855, + 0.010179232145272343, + 0.15482098230686275, + 0.028625269416162842, + 0.061192551387141, + 0.04650804552764992, + 0.047605377815484885, + 0.05524000247939563, + 0.08476781045654883, + 0.12901119214597445, + 0.01590472862149021, + -0.03868973396068581, + 0.002377071011704869, + 0.13511983977243444, + -0.014079493075419905, + 0.14991796726982912, + 0.09138800990873494, + 0.12556435045098052 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.284328595570896, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..7004cbeb3e99247cfa4d0d497d7bcb1887a3ef68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S010/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5115438656826252, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6235245449531164, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.01236848782916468, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.6877807770720009, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2599390307355395, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.2000893347799833, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.22162235430136662, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.4021481737675726, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.20324228597800556, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2161136381307941, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5702911795385012, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.15275213525354572, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.21124736618139076, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/README.md new file mode 100644 index 0000000000000000000000000000000000000000..734adead8dab2eacd2110363957a175029eaa5af --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/README.md @@ -0,0 +1,90 @@ +# SeqFunc norm05 at step 26,000 with calibrated splice probabilities + +One model with **13,256,456 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and four dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The compact model trained with balanced splice loss for 2,000 examples, then a zero-output local path was attached. At local01 step 22,000, the continuous-family loss weights changed to inverse training variances while all parameters, optimizer moments and RNG states were preserved. The next 4,000 examples and learning rates match the unchanged control. Only this objective change is being tested. + +This checkpoint records **26,000 updates**, 0 complete training permutations and +26,000 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup updates. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +67.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +## Measured validation results + +Family-mean Pearson: **0.291938** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.474164 | +| dnase | 0.597316 | +| procap | 0.358947 | +| cage | 0.199841 | +| rna_seq | 0.221150 | +| chip_tf | 0.395736 | +| chip_histone | 0.380621 | +| splice_sites | 0.182921 | +| splice_usage | 0.220738 | +| junctions | 0.179636 | +| contact_maps | 0.000247 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5481 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6387 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0945 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7158 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2692 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2151 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2115 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4455 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1890 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2085 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.3768 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1452 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1637 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/change_loss_scales.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/change_loss_scales.py new file mode 100644 index 0000000000000000000000000000000000000000..5fbb9c9c7a72f6d4c04ebb9ab7b08c0fde952cb3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/change_loss_scales.py @@ -0,0 +1,35 @@ +"""Fork our checkpoint with one objective change and no parameter/optimizer changes.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('scales', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint,map_location='cpu',weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' and s['objective'] == 'balanced_splice' +r = json.loads(a.scales.read_text()) +assert r['split'] == 'train' +with a.checkpoint.open('rb') as f: + parent_hash = hashlib.file_digest(f,'sha256').hexdigest() +intervention = {'kind':'variance_balanced_continuous_losses', 'step':s['global_step'], + 'parent_sha256':parent_hash, 'parameters_optimizer_rng_unchanged':True, + 'scales_sha256':hashlib.sha256(a.scales.read_bytes()).hexdigest()} +history = list(s.get('interventions',[])) +if s.get('intervention') and s['intervention'] not in history: history.append(s['intervention']) +history.append(intervention) +s.update(objective='variance_balanced', family_loss_scales=r['family_loss_scales'], + loss_scale_provenance={k:v for k,v in r.items() if k!='junction_moments'}, + intervention=intervention, interventions=history) +a.output.mkdir(parents=True,exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(s['config'])) +(a.output/'intervention.json').write_text(json.dumps(intervention,indent=2)+'\n') +(a.output/'loss_scales.json').write_text(a.scales.read_text()) +_atomic_save(s,a.output/'start.pt') +print(json.dumps(intervention),flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..1930db6f4802b8647fb77777e76159a1df9b2ad5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/package_manifest.json @@ -0,0 +1,33 @@ +{ + "parameters": 13256456, + "source_checkpoint_sha256": "dddc7949ea3039f7a01f22a421e606b540e359bb5c9ab12e3674b3524ad68ca3", + "files": { + "README.md": "07aae0b5c5007c5fac76a326fbbf042c9e396cff0b614217c948f538df586fce", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "change_loss_scales.py": "3898e5ea510e7bf35eca74905569c59e429f7f6e58e91e213520029ac004d01f", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "3a0ed8881ac883d3e33d5686638a288c447bbdf59825e68f09e9505d39649300", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "64faa2391c4ef0abfd8b826692186777b725f1127ed04d921f8545970435e02b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "2871dc8893879c3e14a526dea659701d72c8c4d0e2f835b0bfb857034a2c2e6d", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "70b9dd2635417e6d9e25ec3f29dd517a389a3d3f66144363ed40cc1b4854082e", + "training_config.yaml": "deab298b32f0aef33e66ddf538541ec470eee0c05a6ebc983684f2575287d791", + "training_provenance.json": "613f7ef6553ffda155514e4fe81651b5545b3e0bb755052999db7a869431eede", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "validation_tracks.json": "f20c6750a2cbbe419b32c48d50c779c2e70975cdf5c7101f0388758948631e7d", + "validation_variants.json": "907f321581499c1ffd57943ed499ae2ffb5abe346609a7b77eccd63d49e0c39a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..050d412dc0c9c5085da6aa705840e140ec1034f3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/predict.py @@ -0,0 +1,131 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..80536df3fe6ad9f1974ed7ef8952ac6fdf140b0b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/train_single.py @@ -0,0 +1,170 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice', 'variance_balanced'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage, family_loss_scales = {}, {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k:state[k] for k in ['intervention', 'interventions', 'loss_scale_provenance'] if k in state} + family_loss_scales = state.get('family_loss_scales', {}) + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + if args.objective == 'variance_balanced': + expected = set(model.dense_heads) | {'splice_usage', 'junctions', 'contact_maps'} + if set(family_loss_scales) != expected or not all(math.isfinite(v) and v > 0 for v in family_loss_scales.values()): + raise ValueError('variance_balanced requires fixed positive scales for every continuous family') + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, 'family_loss_scales': family_loss_scales, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0), + family_loss_scales=family_loss_scales) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective in ['balanced_splice', 'variance_balanced']: + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + raw_losses = dict(losses) + if args.objective == 'variance_balanced': + for name, scale in family_loss_scales.items(): + losses[name] = losses[name] * scale + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + raw_losses={k: float(v.detach()) for k, v in raw_losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f9a3298d69b138a02fed00880182d4004b23327 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_config.yaml @@ -0,0 +1,28 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..68a3c002f4c8aaf1d5dfaab4a778bd3e64efbff8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/training_provenance.json @@ -0,0 +1,126 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 0, + "offset": 26000, + "global_step": 26000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/norm05", + "config": "/workspace/research/experiments/norm05/config.yaml", + "resume": "/workspace/research/experiments/norm05/start.pt", + "objective": "variance_balanced", + "epochs": 4, + "stop_at": 26000, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 2000, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 4055.450142032001, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "variance_balanced_continuous_losses", + "step": 22000, + "parent_sha256": "35572150e0b3df21837a52ca2d6144327394821f662bb7c7599d12b3ea92bf24", + "parameters_optimizer_rng_unchanged": true, + "scales_sha256": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42" + }, + "interventions": [ + { + "kind": "variance_balanced_continuous_losses", + "step": 22000, + "parent_sha256": "35572150e0b3df21837a52ca2d6144327394821f662bb7c7599d12b3ea92bf24", + "parameters_optimizer_rng_unchanged": true, + "scales_sha256": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42" + } + ], + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..ec75479c126352d37e78c4932a9d40667d051203 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.4741637936847066, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6034180834935738, + 0.6136824518619696, + 0.37420372640000416, + 0.5512857027560426, + 0.36708245429675124, + 0.36091150472566896, + 0.37311497155686063, + 0.44868721953497054, + 0.4955927956157022, + 0.5495970919337914, + 0.48012173846966427, + 0.4198491177300601, + 0.2795925552058367, + 0.4162926638086273, + 0.5992547852530593, + 0.3788124747320483, + 0.44994202748289824, + 0.3932995109123901, + 0.44513727793275987, + 0.4321851606932069, + 0.39883783722544686, + 0.3575514933685695, + 0.4109331298421821, + 0.42700118781403223, + 0.3729973454758728, + 0.38887375162437604, + 0.37390191364693726, + 0.40066272463701236, + 0.40608646602286486, + 0.3932049275670013, + 0.39129009002115367, + 0.39593929353285595, + 0.38408079814550034, + 0.29585285806699924, + 0.2713634305000278, + 0.45992877037000857, + 0.38851400751671494, + 0.5077135581002697, + 0.40311626094468944, + 0.3763930186035515, + 0.4585003383387581, + 0.41881333387889946, + 0.3936086841460312, + 0.35046215832743144, + 0.4531661634868802, + 0.4020541340051071, + 0.319006761373122, + 0.3665067548606586, + 0.35813460177792344, + 0.3654838830777091, + 0.3976685382957182, + 0.3227182558817008, + 0.37155172639759626, + 0.5090784936855891, + 0.34563919101781576, + 0.6070136409466758, + 0.7200535816519245, + 0.4337013105621389, + 0.5232468013718174, + 0.5161081787929912, + 0.6286538628423629, + 0.39418808760156776, + 0.5995718209048748, + 0.5271315886753631, + 0.5529880554535177, + 0.3969661967776135, + 0.46806720579630573, + 0.552511243283434, + 0.7034509049511614, + 0.6337050998784737, + 0.4245395871118014, + 0.35605602124662983, + 0.31160476084569927, + 0.43247293698581624, + 0.4112391170951734, + 0.35946903824258536, + 0.4036405363461816, + 0.4328844804567418, + 0.47420329194664884, + 0.37444451978730514, + 0.4334294887272668, + 0.4640323839554979, + 0.4137077400302828, + 0.38787194129385544, + 0.38471437260518937, + 0.38370654003149496, + 0.3412552769307428, + 0.36555320444645156, + 0.35097201812106144, + 0.40824129610113086, + 0.40345333398949745, + 0.39701102386429576, + 0.3929512759517101, + 0.41440450901504344, + 0.4773027597154745, + 0.30085967788235174, + 0.3395487270831779, + 0.3131568123958017, + 0.38013561380708, + 0.40769908046186204, + 0.34366435162683884, + 0.37073013704156144, + 0.3988828106838514, + 0.3705609418679976, + 0.40984419949301504, + 0.42434170253626297, + 0.4429717011052271, + 0.4171728944793015, + 0.4184043729266664, + 0.4278393033647615, + 0.45537784164840006, + 0.41944347014185535, + 0.40490517676267557, + 0.3504452373609761, + 0.3616272542026787, + 0.3471383034023293, + 0.36658603661478906, + 0.3917139261703338, + 0.3873709197822519, + 0.3748446218302269, + 0.5826548301809941, + 0.5663043372596875, + 0.6346838205494766, + 0.645929721117152, + 0.6189899083873664, + 0.5102507192189055, + 0.7824034941967933, + 0.6183884583053316, + 0.6763101520956832, + 0.6007212018785222, + 0.6503611678378866, + 0.4880947920400586, + 0.5217684010317182, + 0.493707068101364, + 0.626022766785492, + 0.6271443609966778, + 0.6125358606342322, + 0.4573380114716204, + 0.3670806867221991, + 0.726782071604587, + 0.6933084313752956, + 0.7083605749818187, + 0.7030352327144842, + 0.523504484987358, + 0.7757908349148153, + 0.6886056678497358, + 0.6253494246185572, + 0.636221443054302, + 0.7735420656349987, + 0.7219879264188529, + 0.5449908547182648, + 0.5403836684309515, + 0.6275069997256023, + 0.6569046967098202, + 0.6628587950842347, + 0.5325077477219868, + 0.6558174696641972, + 0.512673416844664, + 0.5330044551588885, + 0.6059876480837955, + 0.4884744106171634, + 0.6817767152879256, + 0.7045197307992352, + 0.5503057547408817, + 0.5312485544728421, + 0.49035089910254054, + 0.6583804662380812 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.5973155829733069, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5052178846620393, + 0.9025956361985709, + 0.7583818872328527, + 0.45771375287122207, + 0.4138452407226156, + 0.43083795672952263, + 0.9254935359703224, + 0.681734445374352, + 0.3639439147820746, + 0.30123289045116913, + 0.8573743448644939, + 0.6255740437031325, + 0.3812919131915255, + 0.4492556270523156, + 0.42778916664301647, + 0.7787666046231433, + 0.7626052571108106, + 0.8348771596474895, + 0.8350369731813705, + 0.8694277217594899, + 0.5975023121336132, + 0.5598670034187646, + 0.8839729258605453, + 0.44625623480929943, + 0.4692511611702157, + 0.7554209513779829, + 0.4109518974132677, + 0.6056836628762133, + 0.4451624621403516, + 0.45142019018491975, + 0.5613002497930137, + 0.38982318963407103, + 0.8246684845405472, + 0.6098680152062, + 0.8667174230576379, + 0.8737104579308691, + 0.39498113924156014, + 0.8094221707172563, + 0.8630925209320394, + 0.8448151421576874, + 0.8050591427288667, + 0.6551870866861199, + 0.6902415312204407, + 0.593533813706999, + 0.4223192979267822, + 0.6193663838664718, + 0.715347474475008, + 0.4771164747483384, + 0.547246942852538, + 0.364417535123254, + 0.32136516595477926, + 0.5090891768764715, + 0.4407173380778948, + 0.8167409341678706, + 0.3762305428850942, + 0.36680804956178803, + 0.3886754079234133, + 0.4183364633892181, + 0.3830392989827241, + 0.4889111312176082, + 0.3679091062517099, + 0.5770669663202831, + 0.4272793062544557, + 0.4131545116929338, + 0.4396248273415027, + 0.43573438565784883, + 0.4985269975188323, + 0.41328196810712714, + 0.3347249355019045, + 0.3659745068731893, + 0.47638636285660924, + 0.5813973147126725, + 0.42325011018432085, + 0.566164900018231, + 0.2907100437417834, + 0.7920050925207401, + 0.881047373453955, + 0.6111187321323315, + 0.38718132817566125, + 0.40026283062745693, + 0.44213203794273487, + 0.41797840250765506, + 0.4472888899439934, + 0.4159466013638991, + 0.46675022565921515, + 0.44520881403322526, + 0.5299345897506791, + 0.457477666155003, + 0.5935694948203354, + 0.3850629160035775, + 0.5775111658609111, + 0.37601920249757725, + 0.6978257711747291, + 0.41891376693768123, + 0.6838784176441839, + 0.45409482411295365, + 0.6934280669704479, + 0.6455900917016147, + 0.2963938722453241, + 0.30229213311980285, + 0.4139726124898934, + 0.3322687413368124, + 0.3118454421000515, + 0.6620775194646731, + 0.6339755961859574, + 0.3829316935638208, + 0.40442657846526564, + 0.5515780179893258, + 0.8279939851079651, + 0.512863260889792, + 0.4839255495912404, + 0.5345705522039441, + 0.44716142835276307, + 0.40867195206386936, + 0.7547023495262883, + 0.481268024354866, + 0.4528426288246025, + 0.5103487149676466, + 0.5664925840286454, + 0.4135853676342918, + 0.4595767907724408, + 0.45782670843218454, + 0.4907377570762762, + 0.4694609867512719, + 0.5654175076143495, + 0.4878040446019347, + 0.5381153727207758, + 0.6387860840255771, + 0.6702142401181853, + 0.5417768511206436, + 0.4259329333459819, + 0.5541924418846299, + 0.583170574377761, + 0.7655412942824674, + 0.39228650470933796, + 0.27564879591193875, + 0.5306857225984587, + 0.43286560057304496, + 0.5682212306803659, + 0.5455207446707911, + 0.5512334084079505, + 0.5077176161370991, + 0.40897542639075546, + 0.7543986254369118, + 0.5602224442385552, + 0.7395923409135897, + 0.3523609069122859, + 0.3235070498940857, + 0.368633401687604, + 0.2975819573931065, + 0.8813804966393353, + 0.28597458631967215, + 0.8785812825975924, + 0.6619233378062286, + 0.44643421676200784, + 0.30274663509444183, + 0.3114836314756117, + 0.3338114273613525, + 0.6061190613317831, + 0.649374448974123, + 0.672624973236378, + 0.5961537479452178, + 0.48475875006513214, + 0.41546454551329753, + 0.6423532778338028, + 0.5301590308028574, + 0.46214383811177107, + 0.5606513254314405, + 0.41925332024281187, + 0.4291344432817607, + 0.3401679854113957, + 0.8165361170878261, + 0.3791769089974995, + 0.4154961760839811, + 0.6078207321273319, + 0.667464067118001, + 0.42486670824676476, + 0.30783734298486687, + 0.8652121211954744, + 0.33706458475149426, + 0.7377867753298065, + 0.3831856705054597, + 0.5600433150486869, + 0.680327147404837, + 0.4319383197512692, + 0.5189861469184829, + 0.5207708236902626, + 0.7998694797084351, + 0.39535926108359726, + 0.5691580687322877, + 0.49989598128744767, + 0.6726085699979053, + 0.8836330374654171, + 0.35428678418411513, + 0.5328777206292434, + 0.40805325104953855, + 0.5804198333207131, + 0.6761531130873352, + 0.5291858432439108, + 0.39803195861054136, + 0.8569882465649223, + 0.8629044106799018, + 0.740607984054059, + 0.820340114321334, + 0.830052910560239, + 0.8187849530290698, + 0.8290624984605357, + 0.3207766775756672, + 0.33125900827415233, + 0.4631087558153651, + 0.4166665951209527, + 0.39758931900381267, + 0.5568214946101062, + 0.8437044720282667, + 0.8352940853496678, + 0.6265453977263559, + 0.8092445624709568, + 0.8054501055088976, + 0.8417466384599488, + 0.8906393422915428, + 0.43788506173438113, + 0.49373544147413695, + 0.6037460699933517, + 0.507127165814596, + 0.7986718334926844, + 0.5099510227967731, + 0.8919776065061843, + 0.8663903855589018, + 0.8584935664871199, + 0.887338826419462, + 0.8785367211638765, + 0.8545949676764409, + 0.8126545288588498, + 0.845701774286984, + 0.8892591909494376, + 0.3837613326434196, + 0.9109428000242585, + 0.8635331553930774, + 0.804514089219445, + 0.5412362149822619, + 0.8580285085763774, + 0.4627147327934192, + 0.7374288553141689, + 0.8566726132169021, + 0.5164628291484827, + 0.5057420028532309, + 0.6084020699261673, + 0.4806830940221177, + 0.4744004132714909, + 0.6984922373826143, + 0.8668542098596117, + 0.6547400207783166, + 0.7893801367256288, + 0.5605580326571117, + 0.8486713881170326, + 0.7899014014619683, + 0.8224408264710713, + 0.890174966336492, + 0.8173251815163324, + 0.8832232017855446, + 0.8509956809887337, + 0.47847564921240393, + 0.8642193910702495, + 0.4845251980528097, + 0.47111557558381206, + 0.8677992984858135, + 0.8912128306765756, + 0.8650264188710347, + 0.5099209183982288, + 0.5554765203450648, + 0.6222887375335951, + 0.7800137843704417, + 0.8711177790769277, + 0.6573663821847434, + 0.8885962647286199, + 0.8998192928851928, + 0.5381225960614253, + 0.2749489478633932, + 0.4707157445529345, + 0.8855014895093332, + 0.37001736692712, + 0.5513695008500659, + 0.866383749643254, + 0.8146759560390674, + 0.5512773049143909, + 0.8491577029372318, + 0.8800307503493263, + 0.8736629557552221, + 0.8623597252615756, + 0.8536848339249575, + 0.8822715962328995, + 0.8376766269029073, + 0.9116480676858167, + 0.7663774294812006, + 0.8185492450102164, + 0.8930439695735557, + 0.4931118360309767, + 0.49951580113129157, + 0.4728085704110667, + 0.46192536006038465, + 0.8942527810919256, + 0.777506955955939, + 0.8308984522116151, + 0.771834947780646, + 0.7430372918181578 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.358946673434071, + "valid_tracks": 12, + "per_track_pearson": [ + 0.34978131476743946, + 0.3899769315850776, + 0.3023052505059016, + 0.387725377915241, + 0.38677461115370415, + 0.3885749956350349, + 0.33050435904435377, + 0.38057620717505947, + 0.2701412697608646, + 0.3741761957684188, + 0.36573690804352116, + 0.3810866598542351 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.19984051996360552, + "valid_tracks": 546, + "per_track_pearson": [ + 0.2027190607038923, + 0.22811142555769104, + 0.19452785299268402, + 0.124451584035908, + 0.07484866038182028, + 0.2057542263859211, + 0.10341946838853787, + 0.1339750232725061, + 0.2160611766802273, + 0.18613495606112812, + 0.23422869558353654, + 0.1319311865379861, + 0.13744169646761378, + 0.11680530785528331, + 0.09571151637374715, + 0.1617110228998345, + 0.19233073897507832, + 0.23103767874678008, + 0.23729772317475514, + 0.21781746531088791, + 0.18701647429040547, + 0.17678474628353286, + 0.2903621093383641, + 0.15896010342663786, + 0.1624811446369333, + 0.2014889301368929, + 0.2240327595684613, + 0.22591442802524572, + 0.2477998895536968, + 0.22646196709515418, + 0.1462430968988403, + 0.23922234632869932, + 0.17565886022026342, + 0.12409626595097287, + 0.24855673953433988, + 0.23780028764906283, + 0.22393260152577113, + 0.24828422079027246, + 0.2148944356494452, + 0.28119262479360846, + 0.29056103700931796, + 0.25099014893492005, + 0.21703150651842842, + 0.21652975680050457, + 0.12444370438903572, + 0.18102359413813682, + 0.1927765946568318, + 0.1948245856490327, + 0.12004695992123934, + 0.15636343000665848, + 0.2094235120221698, + 0.2456106332992709, + 0.254761671492776, + 0.17841923217166408, + 0.26570148438991525, + 0.2938817931111108, + 0.1608458161412843, + 0.27133522490283196, + 0.21027311041041483, + 0.23496558772353707, + 0.2726595157764793, + 0.1698131086688495, + 0.24456996224699087, + 0.18385657026561458, + 0.256721019215139, + 0.2657512019978858, + 0.2818008071792731, + 0.16966702217387164, + 0.11802082217941684, + 0.10617652660091188, + 0.21759560045853812, + 0.25629157080919623, + 0.21192749480213513, + 0.22635285742185574, + 0.2444076509055927, + 0.2046502125776802, + 0.24617111671610653, + 0.2236293538198455, + 0.24661658907064232, + 0.2543460830075347, + 0.209807409203915, + 0.1272027413670266, + 0.22650097911830183, + 0.2637676208605016, + 0.12726713386426988, + 0.24097347820279352, + 0.1969517245229533, + 0.2650307900371622, + 0.16580778406629265, + 0.25047613137106295, + 0.17920713632185203, + 0.20844355852363983, + 0.23166714979477365, + 0.11058456067484304, + 0.24251616589299052, + 0.21178166458242526, + 0.2224184070667742, + 0.2348590883314144, + 0.2172627261622947, + 0.13391336139267299, + 0.2104218451766229, + 0.2273192785488869, + 0.22728263682576574, + 0.3091476832435101, + 0.26387661466655693, + 0.27609033475756906, + 0.18045610721301278, + 0.23147162428547927, + 0.1811754675434853, + 0.1906229942660803, + 0.20531214633792652, + 0.20573178273057366, + 0.22852039229043417, + 0.22705382522533415, + 0.1813340668285151, + 0.224379968551385, + 0.23031493180012239, + 0.23645660540839414, + 0.21996452722854526, + 0.2570073371007925, + 0.25883000110554044, + 0.25478838049642977, + 0.1553093614405059, + 0.17055810251113487, + 0.1984757805709563, + 0.24660409122384427, + 0.22021174987378336, + 0.2429949873312078, + 0.3154162490071481, + 0.23179406320552554, + 0.25281920502912386, + 0.17490782860840867, + 0.2404630966860668, + 0.2506147742189151, + 0.2926659861079583, + 0.21344101632618417, + 0.2622314328638978, + 0.18621278941818695, + 0.11679248819507981, + 0.227786120158642, + 0.27342579606949, + 0.22849027764206206, + 0.2420456632363636, + 0.19835057015099003, + 0.20146315698333847, + 0.21413226912432456, + 0.22875608404192885, + 0.26543347851198124, + 0.19408036230546558, + 0.18854563888852183, + 0.22791553135729636, + 0.17146809740235966, + 0.2096868581911929, + 0.23017477544965229, + 0.21079111610250495, + 0.14991606842495073, + 0.24015506651825325, + 0.2746732304866148, + 0.1792543787713371, + 0.1480808328661372, + 0.1419949312156784, + 0.2029702939903863, + 0.19298652616739503, + 0.22133820092075127, + 0.19810198217211084, + 0.15688528867652826, + 0.21835294930134985, + 0.14998294894992348, + 0.15445069648464366, + 0.24444777915852364, + 0.2211240734326312, + 0.17313949385645147, + 0.1959166774196702, + 0.3006530346231102, + 0.22185618281106115, + 0.14668828836974143, + 0.14346388305530183, + 0.17894474966133275, + 0.2330586689851354, + 0.27260805395431803, + 0.25930313666698324, + 0.26111437058186826, + 0.12561665047070186, + 0.256689479779529, + 0.2107080822801708, + 0.17370970521072893, + 0.1748335539073423, + 0.15244932626728866, + 0.2786173220798399, + 0.2290849348866312, + 0.1902666954452736, + 0.12150904380111924, + 0.12406590146727832, + 0.2009021506285336, + 0.20046360009333694, + 0.18331513052028825, + 0.2119890567911017, + 0.2098643495555728, + 0.1585597293287941, + 0.20370479433740213, + 0.11908183608508362, + 0.09778816338838486, + 0.11717290155123355, + 0.10570290428155787, + 0.12556963630942322, + 0.20446372662242399, + 0.20020220320314286, + 0.17249722066384737, + 0.11171970772498493, + 0.19871303243867056, + 0.2617685675778197, + 0.22587314396340286, + 0.23330982157495533, + 0.23116486785783688, + 0.2154624204356676, + 0.20842506689357485, + 0.24882693916614954, + 0.27928956100659813, + 0.24735264492125922, + 0.24153955328993107, + 0.22064806973563855, + 0.16325754819117455, + 0.22495094574589197, + 0.21962600648442276, + 0.23547748470335478, + 0.28668471093091513, + 0.21566898695894976, + 0.18532701356894343, + 0.20620106579357803, + 0.22010486631642623, + 0.16098140334146188, + 0.1629347874045733, + 0.26621880343138726, + 0.21112423264560506, + 0.1786623020952055, + 0.20901468629865674, + 0.27680448621186193, + 0.14791342433506546, + 0.21939313399251803, + 0.2167650780892111, + 0.1094248984849792, + 0.14637140037208726, + 0.13542819071636053, + 0.21788515773617786, + 0.15827020281923737, + 0.19838859855015029, + 0.08604381522598777, + 0.2640578213590629, + 0.2277074392158072, + 0.23407217867420008, + 0.24751820482273842, + 0.15951546720370285, + 0.2767112364702989, + 0.12748845046533772, + 0.2483762177235267, + 0.20651519100502239, + 0.31556534849891493, + 0.2354496872830627, + 0.23981346501234926, + 0.3416915974345254, + 0.19977395470085724, + 0.2212129534138382, + 0.12608909452186418, + 0.23438023425454768, + 0.22005463296252065, + 0.15321600606813263, + 0.12819542705732637, + 0.12984696265830417, + 0.15259314702410717, + 0.22886177598562105, + 0.28067071709443875, + 0.14064068561629578, + 0.12117061198512041, + 0.1794398030625435, + 0.24408046388996202, + 0.18965529764427394, + 0.13251450214694185, + 0.0658878276529053, + 0.19837105238950337, + 0.12431329535035926, + 0.13065982079041613, + 0.2131735466502691, + 0.18175372410036852, + 0.22064913551331458, + 0.13243350232260012, + 0.13450360935387085, + 0.1203635933157049, + 0.10823322887828006, + 0.1501606567264173, + 0.18680947245375024, + 0.21318492038758916, + 0.2199274401180465, + 0.20146768217305006, + 0.18013421448887038, + 0.1703234925733483, + 0.2705519780659613, + 0.14722901946706385, + 0.15638906121869434, + 0.20646576424317514, + 0.20930286218539126, + 0.20810639362968444, + 0.23198126308619996, + 0.21762833692529443, + 0.1355506698797431, + 0.2126981918051115, + 0.16334110363977333, + 0.11347370341765256, + 0.2281236600322685, + 0.2363962217253162, + 0.2142629902395099, + 0.22558779359827807, + 0.19050950257862917, + 0.26583082691241544, + 0.26780722765602893, + 0.2318790318215234, + 0.21375416227451507, + 0.20435846544595518, + 0.11841330861330145, + 0.17469196896358996, + 0.18474374514608155, + 0.20246510348557503, + 0.11531552745877245, + 0.14799313269653186, + 0.18770853732006243, + 0.2405977064346906, + 0.241727581349111, + 0.171370927213577, + 0.2431170247325504, + 0.26715273792859556, + 0.16414450340647552, + 0.24370626224892486, + 0.1941538480716718, + 0.21061662267958606, + 0.2521089493645141, + 0.1610730909874423, + 0.23128826804832647, + 0.18278927929865155, + 0.242949634002947, + 0.2374006260343357, + 0.25906509342177514, + 0.15652430809931075, + 0.11410407848693092, + 0.10077899223522191, + 0.20439086527593667, + 0.24172657417390556, + 0.2051336054969393, + 0.21211440734314846, + 0.23437025615627072, + 0.19805773676236968, + 0.2177513796117456, + 0.20494489175199643, + 0.22629232938918173, + 0.24298202415017975, + 0.20333058675271745, + 0.12629268528409898, + 0.218617154013374, + 0.2405752022477983, + 0.1287169616346061, + 0.22582517041251904, + 0.21672389206917234, + 0.2524267452878115, + 0.1575031286587165, + 0.24646153993601494, + 0.17839908014480596, + 0.19420610888164064, + 0.22046066028569525, + 0.10903247284699488, + 0.22417653175354252, + 0.19397325193768575, + 0.19951861153791828, + 0.23739808455160819, + 0.2014412681210983, + 0.1274381381258488, + 0.20119903627278507, + 0.22640892821927366, + 0.21051649871556588, + 0.290052994955466, + 0.24310432005820373, + 0.2643720856269494, + 0.16870532390685808, + 0.21869054299130306, + 0.17301666991179637, + 0.17228037464633755, + 0.18924543756867646, + 0.19277107318864592, + 0.2173030730429791, + 0.2214462632343638, + 0.16983548152526315, + 0.20950918555050804, + 0.20881345068447177, + 0.22207453863819093, + 0.20295386638445484, + 0.24658303639521767, + 0.2355658199839568, + 0.2341969480714657, + 0.1533086173965632, + 0.17290456849019475, + 0.20029921657958027, + 0.2288699932217246, + 0.2049474830166759, + 0.23754688098572532, + 0.30066941061786145, + 0.2146715378226147, + 0.24042738174700398, + 0.17423277943722965, + 0.21408014232190942, + 0.21115289308061228, + 0.26675069480894503, + 0.20754093791542194, + 0.24719330822943067, + 0.17234680932669041, + 0.11196440998451584, + 0.21372037528787982, + 0.24800791669559158, + 0.20707738333066125, + 0.23999196583408852, + 0.18102561157641298, + 0.18485372722650625, + 0.19180167221304503, + 0.20350182092055533, + 0.22934405714725148, + 0.19113942218428762, + 0.17230423724519656, + 0.19298274461423373, + 0.14483247811844224, + 0.18172778263327888, + 0.19937326643177242, + 0.1882246906648758, + 0.12842052505746923, + 0.21084105727421712, + 0.23867353741289407, + 0.15642566471095856, + 0.1342093497009287, + 0.1369203052165152, + 0.17604250061984592, + 0.17423903301076538, + 0.215503903733439, + 0.18915452924684414, + 0.14457620835584425, + 0.20647871303454926, + 0.15268474596529996, + 0.1421828248857785, + 0.22612538868371884, + 0.20658854022466963, + 0.17892569284968798, + 0.17143564840242245, + 0.28668281866548917, + 0.21301673110140465, + 0.1300040602418844, + 0.1399541034738901, + 0.1865608020470555, + 0.18911636778789345, + 0.2746509948914441, + 0.2417384578829056, + 0.2541750686897709, + 0.10922499637950585, + 0.24414428164622937, + 0.20686172572983708, + 0.18885020140270092, + 0.1530676704615224, + 0.14766406075113298, + 0.23601328941025648, + 0.21445288374387547, + 0.19071168623485632, + 0.14045010860137358, + 0.10081374499586152, + 0.1506262028340476, + 0.20256664692645932, + 0.17182304774424617, + 0.20837931767308948, + 0.19788258176301915, + 0.15966989202112086, + 0.17118174026915112, + 0.09536022833642283, + 0.07300540228897372, + 0.09353541354900746, + 0.0915504065137007, + 0.08670617722448101, + 0.19877480516775514, + 0.16121747854627988, + 0.14409486516876052, + 0.10291928517167945, + 0.19002129297828144, + 0.2613444377429901, + 0.23302242246747637, + 0.23177057558165953, + 0.2336311241321993, + 0.2177749615191038, + 0.20773408764455178, + 0.24163393203310812, + 0.27585310162162985, + 0.24892700897389455, + 0.23773697103141844, + 0.22083518024253573, + 0.15513993466427634, + 0.2265416215015278, + 0.21206955886005535, + 0.2366914176940233, + 0.27936089722974505, + 0.1986139854097365, + 0.18232274987850614, + 0.18030246415877865, + 0.20349882486825616, + 0.11915684041893804, + 0.11116381726319056, + 0.2533056044903907, + 0.20951730006353192, + 0.1775696745486491, + 0.20195644616055825, + 0.2603683877948792, + 0.1498638369778778, + 0.20610296810546716, + 0.19977533098957762, + 0.0985932465182998, + 0.11964482494575523, + 0.1239367927039672, + 0.21417693686522962, + 0.13274184284459603, + 0.19258309423193942, + 0.08649978887727099, + 0.2590682608345037, + 0.22897386158942581, + 0.2205747123794665, + 0.24528214117219302, + 0.17184540260263118, + 0.24137593752833691, + 0.1191830224542632, + 0.22016192231543963, + 0.19621253801938737, + 0.30044182073461145, + 0.22902946482730127, + 0.23986475373414468, + 0.33062392538346447, + 0.19935924667764784, + 0.20052160424100174, + 0.11188184417660253, + 0.219635925395008, + 0.20398220927948804, + 0.16847998219633062, + 0.11486655999876197, + 0.11599512706824074, + 0.1406595848995196, + 0.23095198288756297, + 0.2742040859635238, + 0.12675318651921733, + 0.10043696775132849 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.22114960824718674, + "valid_tracks": 613, + "per_track_pearson": [ + 0.23865329395811663, + 0.23219193303308117, + 0.2010509307371053, + 0.25773170994306693, + 0.2504001016834614, + 0.19775192587203055, + 0.22632131852635534, + 0.2508826435798037, + 0.2555646571577106, + 0.2794360198766765, + 0.21094719342356397, + 0.23827480798552603, + 0.16725315294200005, + 0.2820143938959039, + 0.27146278485558206, + 0.2815508224992491, + 0.2800200578813429, + 0.24220289491866603, + 0.22156572364770674, + 0.2519429181976668, + 0.2113988547037569, + 0.23869665533627232, + 0.23514553735094051, + 0.2300129850958265, + 0.17516602336256026, + 0.2272311122311008, + 0.1817529792081264, + 0.2305574217814555, + 0.23353347759417495, + 0.24420557932190706, + 0.28587455902810144, + 0.2302049083798808, + 0.309809668487096, + 0.19323145893521054, + 0.3069439295895795, + 0.17080755309445725, + 0.24410884651268694, + 0.23480451431601387, + 0.2094221176155839, + 0.2377330792506094, + 0.2683499891564424, + 0.21981933503154963, + 0.25205640333615953, + 0.24219639623193703, + 0.2630384181552928, + 0.20045208015785257, + 0.23676358786810528, + 0.2812860057794213, + 0.25226232243222435, + 0.2287125271114349, + 0.2447195096255241, + 0.25384440082412657, + 0.23558983712852627, + 0.2637368970419644, + 0.18400331664687658, + 0.2585901649308505, + 0.24593783021158055, + 0.2535236811766849, + 0.25701957371802026, + 0.23286544601284753, + 0.2109377784562428, + 0.23295116917934955, + 0.23283612204936632, + 0.19990093383448776, + 0.24824173651394218, + 0.22116753967800512, + 0.25346314885205107, + 0.23979351558398346, + 0.2444933416679265, + 0.23560925270185074, + 0.24154476655287338, + 0.254227138429391, + 0.2683723761902719, + 0.24075016169958727, + 0.2543961408339718, + 0.26764988311682314, + 0.2549922133048797, + 0.25207418625934336, + 0.2864131493336794, + 0.2463806033076454, + 0.2389779206924906, + 0.21947487469936597, + 0.2853797949089342, + 0.26932925379641337, + 0.2035441706301691, + 0.21457915424289992, + 0.25918159990852163, + 0.2582824306702635, + 0.2485233664489805, + 0.25576874397679755, + 0.1834573604764955, + 0.21212609073363867, + 0.23893153918081134, + 0.23860442706365184, + 0.2511810183109985, + 0.25285418977559193, + 0.23846762095430948, + 0.24864011128605767, + 0.2450055820892835, + 0.21341517021061954, + 0.220462855997745, + 0.25090283414941655, + 0.2342203855636667, + 0.23657918683807994, + 0.24754132949440127, + 0.2736090026606409, + 0.24221396136982853, + 0.2866247831864859, + 0.2724236903431109, + 0.24521350376644993, + 0.22056965279971935, + 0.32959518503777563, + 0.29767482451622673, + 0.19379449646372437, + 0.23954450633280283, + 0.23505535611429137, + 0.2933079372843543, + 0.2637516015074765, + 0.2636435657717326, + 0.23997850320723121, + 0.1928187047052022, + 0.25026475624148103, + 0.2386558572778232, + 0.27819040852692417, + 0.27129848184630767, + 0.24157245891249243, + 0.25566073271254675, + 0.18564570403192043, + 0.22576091411476923, + 0.221521990032181, + 0.3119534656384159, + 0.2514254364357889, + 0.21559566903410535, + 0.17061454942703705, + 0.1902450504139758, + 0.2469070574518084, + 0.27862611733965176, + 0.29383510563038157, + 0.27676594460308357, + 0.27856128889466714, + 0.2495399038350997, + 0.21280814594277836, + 0.29627684626344236, + 0.299822128129933, + 0.21851277981999664, + 0.2230433883285187, + 0.19630300287255686, + 0.2359160289683467, + 0.17518670590635324, + 0.25192481147298845, + 0.19596844616154768, + 0.26143413543157906, + 0.2219735527265309, + 0.2146292506260562, + 0.27227698187752847, + 0.19865206648096534, + 0.2337241783627618, + 0.27277859458802695, + 0.2273158139150646, + 0.27665243739801637, + 0.31155822863589616, + 0.23297319186166482, + 0.25223013139301276, + 0.24655526186248986, + 0.2452108891046357, + 0.24232362793409515, + 0.2464048502749894, + 0.21971436936506278, + 0.2302674962313696, + 0.26478503221200866, + 0.22854499270848974, + 0.21292568068410161, + 0.26616208295995586, + 0.2537851601702916, + 0.21650366028481358, + 0.21534496509779172, + 0.2256409094554933, + 0.2849487161717274, + 0.2733521392250517, + 0.19377610796309208, + 0.23884078842197726, + 0.2151913338667912, + 0.20304568662784062, + 0.2115075539354601, + 0.20363126059466088, + 0.24780815794430835, + 0.22744323691352225, + 0.13080148168525482, + 0.1887663564251412, + 0.3052215086717508, + 0.33047086801226777, + 0.20609896525872878, + 0.1902923316993642, + 0.24240903153683138, + 0.21247331538888145, + 0.21632563734179208, + 0.2199139422477399, + 0.2117533137804664, + 0.19266515724389335, + 0.23640195832203606, + 0.29494261998029714, + 0.20941975983322492, + 0.35122983717359646, + 0.2646103084283772, + 0.2178160110342338, + 0.22732888897670273, + 0.22950938574287363, + 0.17732829658049, + 0.2169703728933031, + 0.20604313058812707, + 0.20151060433036674, + 0.20067439817149982, + 0.19931170928299177, + 0.21621827213802072, + 0.2019208019458098, + 0.2361163662648219, + 0.20475786835874488, + 0.206088339456663, + 0.2101282335147986, + 0.216716293958978, + 0.3290903916000551, + 0.23151682114548774, + 0.29029317300618174, + 0.26237672457842776, + 0.2629123138125876, + 0.2912901570002521, + 0.2636940845210775, + 0.2823281739693168, + 0.21751959571341015, + 0.22440394128938798, + 0.23584108213652547, + 0.32298190083771766, + 0.22681647592065554, + 0.2538786093187128, + 0.23514708038645124, + 0.23059871861107148, + 0.1613564857689585, + 0.2519463417348174, + 0.2532491047601257, + 0.25078092950535025, + 0.2525240214368515, + 0.20536122472433435, + 0.24878004988977084, + 0.22758810942562474, + 0.18672809125950388, + 0.2596077389580962, + 0.30722583086172167, + 0.23785841796317753, + 0.2406327260121932, + 0.21207331801785143, + 0.212339838630029, + 0.23669314203841194, + 0.19578634124430574, + 0.21140126428753314, + 0.2425816643986124, + 0.2685379365960442, + 0.21897200346152754, + 0.22791414149684677, + 0.21321692077250465, + 0.21169371041804877, + 0.25540525244973894, + 0.23228776521514674, + 0.2506704401360121, + 0.27492655280023276, + 0.2412070111247855, + 0.23473523718164452, + 0.22317896256874364, + 0.18889706790766492, + 0.21879414578251158, + 0.24286241220094218, + 0.228973041527699, + 0.1421441915501806, + 0.20928329205412216, + 0.1906595579557299, + 0.22508318616272874, + 0.20629641710056348, + 0.13854895072810255, + 0.14921279168651788, + 0.21378746176480729, + 0.21937206387802974, + 0.23312375024117724, + 0.2071005476814916, + 0.16713336063377693, + 0.16321236642695883, + 0.1909768182851314, + 0.18528035398229323, + 0.18696031350845788, + 0.2160278440219658, + 0.2030748295068588, + 0.21784349131654954, + 0.19207689766800762, + 0.15516811655586252, + 0.21538680591664083, + 0.15886659930821267, + 0.15559988494521992, + 0.12775720797188456, + 0.19752911576444634, + 0.1922956362049471, + 0.2152743024503132, + 0.19061338071812522, + 0.19256306151280903, + 0.21929167587662007, + 0.20546685041396287, + 0.1816000525759384, + 0.17968753829269662, + 0.2747985428407577, + 0.1604093476343727, + 0.21054723936676628, + 0.20324300010867827, + 0.19502143863161886, + 0.19690650576721938, + 0.21057901133898874, + 0.20562338132022326, + 0.21492005294283972, + 0.19891612755264854, + 0.1833768057616982, + 0.14311690583082784, + 0.1713639757602569, + 0.20997816474488984, + 0.17881428735311597, + 0.20474489681892755, + 0.2178677242727941, + 0.21077330184392934, + 0.15718461272821793, + 0.18646968514643772, + 0.12594965312981488, + 0.18811135407062388, + 0.18915279400715257, + 0.21055462724289395, + 0.2168180604115325, + 0.18994926523751868, + 0.19146245319450308, + 0.2140119344252766, + 0.21773268748636954, + 0.14221005070011886, + 0.21988813888538555, + 0.20602483036133595, + 0.23146016099059252, + 0.22110082960519967, + 0.22503867222097387, + 0.21387508009005202, + 0.20492981863648238, + 0.23063387534871707, + 0.22772035957618003, + 0.224241748002268, + 0.2212788320889879, + 0.23933127124561882, + 0.2287567994408448, + 0.22811810741784985, + 0.21747155787736186, + 0.1978120952249077, + 0.18875562748954344, + 0.16717630541126724, + 0.2300578628446723, + 0.23952595265251497, + 0.1809921485836259, + 0.2045648826895627, + 0.20006855821138003, + 0.17402293960815454, + 0.18392328080763207, + 0.1637666883680884, + 0.1363396111851365, + 0.1923150818454442, + 0.19891150882614286, + 0.19232920321216643, + 0.19317253737717868, + 0.20833961158304165, + 0.21022442997986313, + 0.20019150009774087, + 0.19498884075925008, + 0.16770315047408774, + 0.15724968421169377, + 0.17512473022138003, + 0.17326325088671837, + 0.22182996930214402, + 0.17589848210324358, + 0.19880966202109543, + 0.20887530341026284, + 0.20921786443066367, + 0.23323584131978758, + 0.18691760585632017, + 0.17605595517810058, + 0.2564935415548462, + 0.2120797926506184, + 0.17464086617832741, + 0.21959254773485332, + 0.1699383497880306, + 0.22049197624340386, + 0.18443030825967516, + 0.19426679916702383, + 0.1797595509413361, + 0.18601664456614803, + 0.19580284388373112, + 0.17461007895266395, + 0.2269996620711391, + 0.21017327568885327, + 0.18476440332559482, + 0.17348132673288827, + 0.14473316020159457, + 0.1505145614977663, + 0.16625929650308865, + 0.24643203101839564, + 0.193703190998247, + 0.16153904151290777, + 0.16387911601670968, + 0.16381295318075517, + 0.23281954513368436, + 0.2101189078778635, + 0.2407809211177109, + 0.1991960358945213, + 0.20695695171023756, + 0.16972841823091836, + 0.16895551720550941, + 0.22774239228648693, + 0.2289908180560159, + 0.15145684568645024, + 0.15884486150896943, + 0.16258236134880327, + 0.19970551595914648, + 0.1855267573425284, + 0.18752000406994498, + 0.17350007166596615, + 0.20072498590751947, + 0.1813626304887985, + 0.1491322943471752, + 0.18534312284350365, + 0.15645201520706087, + 0.20989805487616928, + 0.173023772815711, + 0.21249767144352122, + 0.1917679683858054, + 0.2395762073711499, + 0.15906250670218516, + 0.18416038047187347, + 0.1452039952154689, + 0.14926767772434785, + 0.1452514867559399, + 0.14209746632876538, + 0.13379470919695777, + 0.1792800661428214, + 0.2044623081204809, + 0.23246784482054372, + 0.18130064331218385, + 0.2563324878491853, + 0.19955313772033076, + 0.1593620272595887, + 0.16763381837360464, + 0.1855083158206645, + 0.2669904433018178, + 0.2200155538286749, + 0.1564974253187853, + 0.20501499891489075, + 0.2112563589179542, + 0.12941929543700903, + 0.16815311423077164, + 0.1421376778948689, + 0.20120677931291514, + 0.2072127456231355, + 0.12583549752175294, + 0.16134906115987985, + 0.26400761358357383, + 0.27498949520284455, + 0.20007602051723647, + 0.1689794043607537, + 0.20305729203065714, + 0.18796723625725686, + 0.19812254422965678, + 0.19890580096742055, + 0.19334042704624693, + 0.18666436793587937, + 0.21654746382164486, + 0.2571058349371839, + 0.2046820068770319, + 0.31505083490207353, + 0.22146986643704533, + 0.20577510382030242, + 0.20186587699605643, + 0.18628153423980515, + 0.12426157164666167, + 0.22413280725880755, + 0.16114132878733786, + 0.20007849802651104, + 0.19672559857927507, + 0.22460742506443426, + 0.2159188616843033, + 0.1957536769098848, + 0.22942769697259285, + 0.2179000539362805, + 0.16358448610595072, + 0.22703988807587333, + 0.21608527504354968, + 0.2640264533410798, + 0.22135451294513747, + 0.23563149405876901, + 0.1801521118580315, + 0.1856303862018948, + 0.2328975772221891, + 0.16493076661859482, + 0.22831556311723167, + 0.12303377431987116, + 0.21747217766405824, + 0.156808125332064, + 0.2803389726861092, + 0.2273107885717016, + 0.20942794043986482, + 0.2258289436032898, + 0.19091522548259632, + 0.10894557027251535, + 0.2293494005741575, + 0.21711204029483278, + 0.19582011262322366, + 0.20309741622075855, + 0.1995372974191387, + 0.2372188336327976, + 0.231718052614776, + 0.19705797150237256, + 0.21610347812962685, + 0.2735493625499628, + 0.22222553104545453, + 0.19529476096512546, + 0.16946281882159897, + 0.16363278586115823, + 0.1989508696456629, + 0.16955105985720234, + 0.1839342143440566, + 0.213124542334584, + 0.21422514959693414, + 0.14133661047565754, + 0.20374429458416593, + 0.17391496350318886, + 0.21222080833178256, + 0.19427572253450454, + 0.13757200765026528, + 0.20414017841857127, + 0.22646513720371578, + 0.20836153818179984, + 0.2293699650696505, + 0.21762390926209307, + 0.1254558883924743, + 0.2091591003757728, + 0.218245718516476, + 0.1868511104054248, + 0.20056711347399264, + 0.10212215946437146, + 0.2677014456277355, + 0.30631249758517026, + 0.27736521465750075, + 0.25503369898381195, + 0.26078753352829903, + 0.2140192361675287, + 0.21471844853620758, + 0.248161458486325, + 0.19097491912139833, + 0.20828876966262638, + 0.21390042340540819, + 0.2005297954852912, + 0.20129261362034176, + 0.20812102363959353, + 0.21659669325252756, + 0.2015964018272227, + 0.2099473625535643, + 0.18936403024555384, + 0.26741432371591967, + 0.2653019665822711, + 0.2653671559502992, + 0.26569631160482093, + 0.29779844984286113, + null, + 0.2840235508763768, + 0.33243090959764365, + null, + 0.07507065418095163, + 0.3971749280209828, + 0.07041144476363329, + 0.25767009957978004, + 0.2430569922584567, + 0.30984476174088593, + 0.3674260920166239, + 0.3589368784422337, + 0.37618563005748484, + 0.371899076648419, + 0.3396346604592182, + 0.2952073291128273, + 0.22606031460098217, + 0.3785864046955948, + 0.18460993738806517, + 0.2955140733065208, + 0.26313089647275384, + null, + 0.2569920556903448, + null, + null, + 0.2613624826764391, + null, + 0.2397040293328179, + 0.21305508680948257, + null, + 0.30561598323684036, + null, + null, + null, + null, + null, + null, + null, + 0.2595260698407167, + null, + null, + null, + null, + 0.22342345158815777, + null, + 0.2629432911786497, + null, + 0.21753669161753106, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.276743003916728, + 0.1601204974656048, + 0.2486577202859765, + 0.25960993352207756, + 0.19026108214703336, + 0.2383980610689289, + null, + 0.29821048355227225, + null, + 0.24068333553638752, + null, + null, + 0.23959909716176328, + 0.27089903345063043, + 0.23131842431743735, + 0.23747291501229215, + null, + null, + 0.20704034690759682, + null, + null, + 0.2466078421263764, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.2395871157621702, + 0.23704207763412138, + 0.2513726209870522, + 0.22381860816538476, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.395735861429802, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.20377183113085168, + 0.3009085984252398, + 0.28765597506809365, + 0.4328866613719765, + 0.35341808329707786, + 0.23380728229387215, + 0.33912794341429797, + 0.38768582669053653, + 0.32713877048560486, + 0.3371898683933893, + 0.39137154035105287, + 0.36137046837140696, + 0.2671652247462357, + 0.3942562786068416, + 0.3379780852411832, + 0.34190757648243086, + 0.3728456854045958, + 0.3146431836987537, + 0.4436324014588762, + 0.32887120799154773, + 0.33585027802554995, + 0.32250272622968984, + 0.2968284340305769, + 0.4493933121799935, + 0.32076730278385573, + 0.34513120871933883, + 0.24884203882076397, + 0.35555116139449017, + 0.34939304481296923, + 0.3777083794975935, + 0.35179125162846864, + 0.3792752497175114, + 0.4712176650267007, + 0.2689845026088542, + 0.6274870335951491, + 0.2739310967920271, + 0.20078318333383766, + 0.16122995346698976, + 0.3161554421547604, + 0.21423762279968628, + 0.2599546998151194, + 0.4469303673893954, + 0.2896649522956799, + 0.3829523125003093, + 0.2785354778532743, + 0.37462573527945126, + 0.28753611863139855, + 0.20134331839301725, + 0.4488267868429487, + 0.41883793222647553, + 0.3724723131414663, + 0.35483239054053883, + 0.36913465319491345, + 0.3404054731862652, + 0.30961618587818074, + 0.3597931737117437, + 0.28588092027845036, + 0.4013059365239766, + 0.3850414452735289, + 0.40699237062362237, + 0.25353439359082564, + 0.19214956053215454, + 0.38791242435202067, + 0.4027037312631727, + 0.26558156547662903, + 0.3296725422090399, + 0.3310523000383959, + 0.41572867274709474, + 0.4349381139731849, + 0.364857762163392, + 0.3240986582475679, + 0.6648019010975267, + 0.3173445475718516, + 0.3074375675374028, + 0.3126721354752177, + 0.3874738031779811, + 0.24037908545271097, + 0.5940914081853854, + 0.3571390088806006, + 0.36246470926209706, + 0.3523542656802058, + 0.3842461832212282, + 0.4528551801805542, + 0.396929740902932, + 0.28851552220939536, + 0.24726504064444343, + 0.4188529494681053, + 0.3707306290614763, + 0.7895558007340965, + 0.42540215316602975, + 0.3204808019577684, + 0.31246439795834635, + 0.3764025206166232, + 0.30096707294091896, + 0.34303529638873126, + 0.4885846625720887, + 0.30399816046983236, + 0.3017956460481667, + 0.3208834832315434, + 0.2594176305347583, + 0.23698770499277094, + 0.3650538384328428, + 0.3235756966360024, + 0.3191565207127985, + 0.36488133845964826, + 0.1796582454376359, + 0.22054362732931834, + 0.23835740019689497, + 0.37197075554490033, + 0.3167047661774401, + 0.18856986791696434, + 0.2561150092406586, + 0.23535188691622932, + 0.31479684962549126, + 0.30315885798177244, + 0.42538334956911, + 0.32777291988538787, + 0.32264244470155184, + 0.27791742706524775, + 0.3404553261056389, + 0.2776245370684724, + 0.3513034352378819, + 0.3179387575291046, + 0.2150001168294432, + 0.3042180704129722, + 0.2503114195363899, + 0.26373244284231034, + 0.2528471278842801, + 0.273007436566302, + 0.5162481464938214, + 0.3060214139416499, + 0.21583839368438126, + 0.33756369407211484, + 0.556442754341662, + 0.23264305849914182, + 0.288841740829456, + 0.2665185857703159, + 0.26971743796170733, + 0.2567104385973268, + 0.2337241871127216, + 0.22411748039662607, + 0.41796269542381115, + 0.47329471794603073, + 0.25893262453833105, + 0.33809980854979343, + 0.40988717974578875, + 0.374868980597385, + 0.21101131196350398, + 0.35196134274815455, + 0.2833168177658038, + 0.327197285899463, + 0.2601378935117025, + 0.32888608329934893, + 0.4026278969698897, + 0.3514452471055382, + 0.33831512670015185, + 0.2186906239682385, + 0.25546326883096904, + 0.23940532467209147, + 0.3436389205930204, + 0.30316907209580857, + 0.28358054865880167, + 0.32248154148873265, + 0.3202281001695834, + 0.30564459937534394, + 0.30677770128557197, + 0.28627154641493124, + 0.40602691131070545, + 0.21144195590190218, + 0.300517573546942, + 0.2737901326894964, + 0.22447426687408895, + 0.4481236937365397, + 0.24419729113071517, + 0.2046995832302489, + 0.3634419586404579, + 0.2126351554634051, + 0.30895801940439455, + 0.2970454180693974, + 0.25441224720675815, + 0.2949983748106564, + 0.30113154341223075, + 0.26146714537938165, + 0.3411694666622873, + 0.24915678941195304, + 0.29988037097695885, + 0.35847060116321167, + 0.21753682453547665, + 0.23488991791317013, + 0.3301939550235739, + 0.21616192463964096, + 0.28028223271703107, + 0.38055977840373173, + 0.32720808186767963, + 0.24523463335794207, + 0.24175358317178874, + 0.2266687977383016, + 0.2239198193471954, + 0.21873599922535, + 0.38425943323950396, + 0.2555001799336859, + 0.2844263576101404, + 0.25394680011266396, + 0.2502431927909039, + 0.38113126957856797, + 0.2752174173955431, + 0.2985364109603332, + 0.30996935327721564, + 0.22317017902553898, + 0.4286113112646723, + 0.24020773396173867, + 0.3611456278331892, + 0.28402480258572316, + 0.3335058140403475, + 0.8258817072191659, + 0.36264016001070065, + 0.35525965546595006, + 0.35323194959018817, + 0.41956435419509325, + 0.5235473168180174, + 0.5923452432613117, + 0.6018345787996506, + 0.3663836174272153, + 0.359517102135962, + 0.43570445530594976, + 0.7157559555858242, + 0.27101357386774894, + 0.3045127910807483, + 0.5175830720658908, + 0.49144660012863195, + 0.4968068427362923, + 0.39679626470674484, + 0.28001709114507156, + 0.33607031883921384, + 0.5743597881881448, + 0.38762942163260955, + 0.29054383209253787, + 0.600665127921767, + 0.525134727102753, + 0.4223995462308169, + 0.45669040022158774, + 0.6911424823808137, + 0.4259990953639461, + 0.4590267141708126, + 0.4211293496441792, + 0.7494838984145696, + 0.2827217277428194, + 0.6254129692840992, + 0.3538963351302994, + 0.5293762845754516, + 0.397590826434201, + 0.37836338661931007, + 0.6902962319161748, + 0.4783355594934477, + 0.26379294597923275, + 0.34758168141435697, + 0.6281944594273716, + 0.3150632026509769, + 0.42613646287004364, + 0.4116983282479808, + 0.3184171321656489, + 0.4305299431641291, + 0.43034340787980946, + 0.378993825737574, + 0.37221265781343343, + 0.45158917385096004, + 0.31201303418625287, + 0.2848131254725191, + 0.6146338768785371, + 0.33523217979452824, + 0.4295399493987924, + 0.5325100338862259, + 0.29501802065142846, + 0.6482760278107398, + 0.2930963489168961, + 0.29549223722112977, + 0.5403773582606839, + 0.44451250279652277, + 0.5704759145983584, + 0.7201119643292992, + 0.6434608621759726, + 0.3708954166119302, + 0.44084990792879164, + 0.5130187223219325, + 0.25843008392291117, + 0.2901071597202509, + 0.5437120624885293, + 0.48738592387530205, + 0.6942211612654251, + 0.4216571930069676, + 0.32771840636438704, + 0.43109331404627155, + 0.3458641414303818, + 0.4512373227352843, + 0.6854441329871465, + 0.605425060213924, + 0.4453258153219753, + 0.49916794549254356, + 0.463608272708817, + 0.4965689129249194, + 0.7238197516327893, + 0.2795897533764353, + 0.41717793555415056, + 0.2515721989664975, + 0.5729803919012625, + 0.29081398937770436, + 0.7232603780155061, + 0.366872719447528, + 0.39074333558702906, + 0.6746760421874386, + 0.4898573068478158, + 0.28960640869893145, + 0.5430223941738753, + 0.37002245247827736, + 0.5027781995384203, + 0.4921620666193694, + 0.30140348436837583, + 0.47958513138491443, + 0.3502657569366927, + 0.43322422664377924, + 0.3219733811752499, + 0.23312022720135486, + 0.3649357482938428, + 0.48024172845290813, + 0.5414079780033718, + 0.6455154943667823, + 0.35580814455630616, + 0.2619946727314919, + 0.4199765475600629, + 0.4282586429558077, + 0.4062543492896901, + 0.4630477908061282, + 0.4277505772469493, + 0.33367353733520094, + 0.2967809019001897, + 0.47128214439643756, + 0.7563918266596323, + 0.5103054348698906, + 0.5791840755428632, + 0.577175180775826, + 0.7531828941803862, + 0.5269098284948486, + 0.279481884724797, + 0.5596218995606079, + 0.3401162921407819, + 0.47946535408994967, + 0.40745384710184496, + 0.35996586396009655, + 0.471618888146, + 0.7654608257273012, + 0.21492990658248917, + 0.433970664900623, + 0.4371737055623029, + 0.602450079154074, + 0.35589454291566325, + 0.2280862566487398, + 0.38713017362236657, + 0.7113886524579365, + 0.34929461834618114, + 0.4114004676935235, + 0.4865835108872903, + 0.42251091282039305, + 0.37138475026775447, + 0.4632354643967992, + 0.5973702610475303, + 0.43918445357186986, + 0.36206446221590827, + 0.512233704202443, + 0.43345198462724327, + 0.5457233413501569, + 0.4979359072683981, + 0.4966958144912947, + 0.5181417873784294, + 0.26900927935451013, + 0.24634274397629802, + 0.45704331245307317, + 0.2781868815413682, + 0.6811890726177211, + 0.2645876104816581, + 0.5402974030140403, + 0.5204831586583607, + 0.6737477516283846, + 0.2696094042117807, + 0.2937366526503654, + 0.29718602943006994, + 0.28686992934954725, + 0.25384291621897437, + 0.2402656389499751, + 0.5937397992548571, + 0.30647147185318224, + 0.6699934015216235, + 0.5588086818832193, + 0.3996552859642171, + 0.5304190569034728, + 0.3618979580995518, + 0.5065736075628509, + 0.5013401969256007, + 0.4438485871948627, + 0.36158561606556056, + 0.4389273637621677, + 0.4239896433682245, + 0.4949313694911003, + 0.36175625358011115, + 0.2835094233245479, + 0.585141548598348, + 0.44268069176024166, + 0.6448300487168347, + 0.5939142296715545, + 0.470104266350366, + 0.27538573168971925, + 0.2882045187107874, + 0.31267647619231115, + 0.4005309562699266, + 0.7356903816020315, + 0.45216994988252796, + 0.3393378868748453, + 0.7266685234885586, + 0.39542995191631464, + 0.6076040240990596, + 0.512756170611634, + 0.36804158416737665, + 0.47077547061263386, + 0.4218981698646651, + 0.40722606755664265, + 0.567147819344477, + 0.5054527431207179, + 0.4887315933737094, + 0.38693542234335476, + 0.42149395078075896, + 0.3930823441493408, + 0.4431848433658299, + 0.7557035579313081, + 0.4511936960632665, + 0.43276646125441615, + 0.4072947470098096, + 0.2397675010209665, + 0.6548891185312493, + 0.36977431327567684, + 0.2786008565899276, + 0.49105786101800575, + 0.3742982866809632, + 0.28645502428660813, + 0.46796142517325956, + 0.39763978045007853, + 0.7658584750735935, + 0.2992750445059302, + 0.3485884327839027, + 0.2790501627614704, + 0.40489888074456526, + 0.35487883344082516, + 0.2604007743264687, + 0.3475982782621685, + 0.24728955977957698, + 0.3799342044442833, + 0.7252575894191589, + 0.26318137914514195, + 0.37386328823920845, + 0.4672062329819802, + 0.38820346583182413, + 0.34871048330290083, + 0.4484552330520358, + 0.33454817578670293, + 0.49612294983143473, + 0.4855262576595895, + 0.17113805856036735, + 0.6051149990688812, + 0.6614820929885142, + 0.4695327867891685, + 0.444332174163969, + 0.4926602120137547, + 0.3459487168401385, + 0.31909755968848447, + 0.4021925956816856, + 0.765288405460507, + 0.30830691242815716, + 0.5819826581569807, + 0.2962388032568395, + 0.4330297075478457, + 0.4347076871671361, + 0.4510546715782169, + 0.3464920273519271, + 0.2743739067079289, + 0.27549924752440347, + 0.2633003262864813, + 0.47529072821023544, + 0.3999099868109457, + 0.22625206250641702, + 0.37522832111855947, + 0.42311270030485504, + 0.2571831539379478, + 0.526250648203619, + 0.4312903346025751, + 0.34010008457415175, + 0.42436674862651363, + 0.25120099460794354, + 0.39522824259371536, + 0.44472298204049754, + 0.5518006358431515, + 0.41803644803601525, + 0.7662748253049594, + 0.26595341177087345, + 0.5391962289010299, + 0.7252754649683981, + 0.6141498180534983, + 0.19899603151022408, + 0.4515416268354552, + 0.39560163720326985, + 0.29110692907733, + 0.31671615154846344, + 0.3175306294198455, + 0.43091714751304333, + 0.4794832119753072, + 0.30329972289734475, + 0.5877059939257602, + 0.7482007046277523, + 0.40556557894378953, + 0.36847970283948117, + 0.40061554858395404, + 0.4422682258107713, + 0.4708656665844235, + 0.47503416548822747, + 0.4546685172816256, + 0.26507517636249023, + 0.6245912532668877, + 0.4437770272336802, + 0.5286884517519707, + 0.6934874395599014, + 0.2766559146785518, + 0.5754891498862691, + 0.3620670918048459, + 0.4160849318562502, + 0.3217418210287127, + 0.49101499089573497, + 0.4161718150994605, + 0.2879850665898632, + 0.36797632329137403, + 0.3919009353067762, + 0.5297269645253142, + 0.5601038904943931, + 0.4007325328129176, + 0.42693444112528606, + 0.3812345339483102, + 0.722046247830419, + 0.3915437154518578, + 0.4606901216022534, + 0.42999533517670546, + 0.4410483483346737, + 0.5784110728137339, + 0.24085691343182147, + 0.5434326202988286, + 0.3404458386858768, + 0.5021282645678514, + 0.3837430947185995, + 0.28950229603406796, + 0.28664198687839754, + 0.4285920860131298, + 0.2686338102257558, + 0.48364029527586916, + 0.7120750210584357, + 0.7111201047727135, + 0.3365181671357387, + 0.2802593563447458, + 0.6859888612756028, + 0.4185942478966153, + 0.34880248893776605, + 0.4703309058008586, + 0.3367347916165861, + 0.5771319791902433, + 0.41428534394948424, + 0.5402255626554732, + 0.5585369088014985, + 0.3830181152005223, + 0.4810980599784364, + 0.6407643226150102, + 0.4648887138337229, + 0.3144322971101659, + 0.3627129895405006, + 0.26456231272606456, + 0.5079304233084059, + 0.3130231116489667, + 0.4828080729677067, + 0.4600746285629643, + 0.30145595587240637, + 0.7060326192325002, + 0.28517838028057557, + 0.7091605207096445, + 0.39349362977809954, + 0.5086891406708988, + 0.7301510345767017, + 0.44371125396646377, + 0.5326209081944548, + 0.3308450804245312, + 0.6905434348383475, + 0.3704315804453714, + 0.2679155622976268, + 0.5264947456978023, + 0.4859280466763531, + 0.6167804960432838, + 0.5284891194961412, + 0.34794088395105355, + 0.4532209504660927, + 0.23331433141754004, + 0.4006139248189818, + 0.5454361218667175, + 0.47388619087902073, + 0.3094414515167458, + 0.5036530448959923, + 0.5616424717436473, + 0.5509853541273563, + 0.4049232799443637, + 0.2847467980344234, + 0.48670457671547024, + 0.43505351382698726, + 0.5346523840250776, + 0.44876790541741807, + 0.2867957142562032, + 0.36949559999837883, + 0.4686271283961243, + 0.3658392242212896, + 0.7154192329078688, + 0.4859388292646425, + 0.4152827405901536, + 0.4621762579453678, + 0.35010646377369986, + 0.4020231419282088, + 0.5187132697355107, + 0.5266262868378109, + 0.6989324724379753, + 0.5088520076162675, + 0.5556865450954243, + 0.49748266193943796, + 0.43508510588786314, + 0.5535963747514022, + 0.34201152813903823, + 0.6024793037157437, + 0.5789993953199126, + 0.7158611945461202, + 0.48783173458670603, + 0.5230806342261121, + 0.5203958344008638, + 0.3521468967559664, + 0.5183233191524488, + 0.29031466828227664, + 0.6591910966480639, + 0.44404469434921967, + 0.7235847062788403, + 0.7471474808073693, + 0.4149465432031117, + 0.41211515289428086, + 0.46903955658100277, + 0.3008240355150411, + 0.4557105373275107, + 0.7456636627887763, + 0.4667746811666746, + 0.4674847326775193, + 0.567938456919662, + 0.4541356770258857, + 0.304817260353269, + 0.588047720532902, + 0.3953140045290742, + 0.355495315375084, + 0.7207029995420452, + 0.37491236880415557, + 0.39445555255977827, + 0.3574427615764098, + 0.46337734409255704, + 0.541452074310936, + 0.5900762456950552, + 0.6637226370796434, + 0.6055200386595531, + 0.5692480561537635, + 0.4230874860801384, + 0.5885867110982079, + 0.5238489725403318, + 0.7698696938864871, + 0.7211561151134978, + 0.29343864194737623, + 0.3280426458031073, + 0.2718598387912419, + 0.4858788613512069, + 0.3886259218797611, + 0.7753104918141788, + 0.3437365329124809, + 0.3899438146306626, + 0.5353345744925508, + 0.4530967941499756, + 0.5267934943665267, + 0.6023647830858995, + 0.4319905678906123, + 0.6067871541724974, + 0.3487750945603451, + 0.6584537190855241, + 0.7648865195116, + 0.7519866406212697, + 0.5603461604620157, + 0.39952861277096247, + 0.519582860427728, + 0.7256674625097032, + 0.6715558426849537, + 0.6052794784066873, + 0.45082127384962284, + 0.40140264168857304, + 0.4712702424257094, + 0.7237860264655144, + 0.38273988699033284, + 0.38697132485949826, + 0.2775791261736672, + 0.5754003516834377, + 0.25214531794847556, + 0.3641185176266219, + 0.6506096640116951, + 0.5241372807391815, + 0.6675866230284732, + 0.5106507097869722, + 0.4908012166524083, + 0.6516878855191591, + 0.617876307774354, + 0.5481290750310416, + 0.35440950745280936, + 0.629655718535152, + 0.4932856454344081, + 0.37514698761802945, + 0.3153964853750954, + 0.4595416765871135, + 0.6247385364210595, + 0.46903084568005604, + 0.5896465937147453, + 0.5505815094483663, + 0.3921169563411774, + 0.37543414557445565, + 0.6711943962230027, + 0.4618750391522453, + 0.5622825089899404, + 0.7137371002930044, + 0.3903350506511024, + 0.31948475821967004, + 0.31367964377103685, + 0.32418001375815275, + 0.7016807644365035, + 0.725077128139236, + 0.5516668206225619, + 0.3968503289909285, + 0.42337271099721047, + 0.35643001174421446, + 0.7668066219761532, + 0.30945848047872215, + 0.45827595454952197, + 0.30241786749242616, + 0.47402422347291084, + 0.3325997504569437, + 0.5397719444969994, + 0.44205734671549235, + 0.25919490832094005, + 0.6799534796505305, + 0.48433981391540204, + 0.3861475869669826, + 0.316891691105851, + 0.3320038724753537, + 0.37654914766530095, + 0.39268770782716605, + 0.34601544395869027, + 0.3413181247037668, + 0.37880701402403477, + 0.3525706847979825, + 0.3829799743631274, + 0.30724091585693486, + 0.41137055569917585, + 0.19494377976206337, + 0.22265710339059658, + 0.2753640042491781, + 0.35830211994390354, + 0.3998653474278404, + 0.36767833626868884, + 0.28541039625837583, + 0.2782461848730632, + 0.2757201374122985, + 0.6374674492547184, + 0.2749874338975819, + 0.2564761741815829, + 0.7560541800811078, + 0.32533186303226, + 0.3582176989690383, + 0.2971487979136402, + 0.41834665760670264, + 0.2439657514570755, + 0.35303868838862545, + 0.3189917152323219, + 0.433611418558068, + 0.2183217968890465, + 0.3801029853395441, + 0.370108340062454, + 0.47759616871480487, + 0.32854151077548327, + 0.4470232811279423, + 0.2722181447894351, + 0.4075854241518113, + 0.36861976579771855, + 0.3103671551375149, + 0.44270947953008283, + 0.31382445083283633, + 0.44985390182627705, + 0.3367456445911247, + 0.3147127520787775, + 0.8005978842559872, + 0.2550137172048529, + 0.382486311068891, + 0.32956911005818085, + 0.35708075556930574, + 0.26162773617494195, + 0.32649348278533247, + 0.4350159683493557, + 0.5305885565757578, + 0.32547069867284134, + 0.35302511546671217, + 0.23033600039360178, + 0.3826833189458372, + 0.3708891004878036, + 0.3837434684538639, + 0.4113272399360047, + 0.32135508197674556, + 0.28765009144566234, + 0.4381831046642266, + 0.3553846578402094, + 0.41829055987069685, + 0.2421201441307567, + 0.29561658378148303, + 0.44925393055209234, + 0.4340069409556601, + 0.3510961480636062, + 0.312441788803561, + 0.3159099085332972, + 0.8142891773370355, + 0.29376689719437654, + 0.28748770203533364, + 0.3031482844382786, + 0.26182074296549274, + 0.2918305775635532, + 0.2940794850558325, + 0.24417584558535177, + 0.3311551079278269, + 0.3392224776056393, + 0.3343345683708203, + 0.2993710238871492, + 0.39541091272685475, + 0.45594836329966537, + 0.286058443135223, + 0.3365791090139038, + 0.5318781969750663, + 0.27769235850090696, + 0.32496075512478384, + 0.473894143641585, + 0.42189584493191534, + 0.28084816590829603, + 0.23010775122703556, + 0.2818036616219758, + 0.4019629610503947, + 0.4123724691902406, + 0.24112204343881952, + 0.3167984118158623, + 0.3059577005151353, + 0.3941280786551962, + 0.27917440468421895, + 0.25060675392394, + 0.4491125293951508, + 0.4729961168044979, + 0.2211501568205259, + 0.40476853181552963, + 0.23149153953521182, + 0.28097233598375576, + 0.276325898139701, + 0.7723636206946112, + 0.40398646341863176, + 0.4121221929201666, + 0.22866116872157283, + 0.3284980760871555, + 0.30912585803054193, + 0.6790311777649246, + 0.36336062563825416, + 0.30475193461411804, + 0.4253463254717171, + 0.3814587890578125, + 0.36732987775623605, + 0.4152747609324023, + 0.4543839279244332, + 0.22537732408353325, + 0.2683798007098848, + 0.41058281827692, + 0.35581609563628674, + 0.36496167744288144, + 0.35260801537072434, + 0.2909257677110286, + 0.24385104892008264, + 0.23742362055827682, + 0.36069819230786127, + 0.32222153926372843, + 0.5771758031915749, + 0.2717442904586757, + 0.24923674572766727, + 0.24408918304437496, + 0.4900219024390473, + 0.30182858123777234, + 0.21192320340743623, + 0.2582110488791201, + 0.33384466685728403, + 0.3486466834859171, + 0.2407381810778315, + 0.511949767466345, + 0.39137316820220536, + 0.23707921307193114, + 0.35362873352194935, + 0.392695017360891, + 0.32442361096337186, + 0.31007884277979914, + 0.4502666791099548, + 0.2777396636178005, + 0.6309637789780738, + 0.34512608361003055, + 0.4401950216052059, + 0.371829936379625, + 0.33926386526796476, + 0.3636915280541453, + 0.2960399932442238, + 0.22265917269427327, + 0.36086733361831624, + 0.5696634792061046, + 0.29624947312255284, + 0.22354292360520456, + 0.468732572181403, + 0.38542715901558927, + 0.38487575504796673, + 0.5758252040118708, + 0.3118986712767775, + 0.3674493071733285, + 0.271658338284938, + 0.3347993994959765, + 0.4007224910516273, + 0.28224589953126833, + 0.429622128341418, + 0.36871988153477614, + 0.3804179359752669, + 0.19629832772641648, + 0.3525380656711563, + 0.27028892583174235, + 0.25040749873975826, + 0.2506088342687049, + 0.23506899579507806, + 0.3903471678954859, + 0.4933642475806157, + 0.40658797583368905, + 0.4364235073620965, + 0.40427500928583987, + 0.24142927278179843, + 0.3042193033330338, + 0.43234740831182145, + 0.3216245627884873, + 0.34933994079980313, + 0.42568133591117474, + 0.338191095456718, + 0.3366315233363541, + 0.48049737913423224, + 0.3566718456715536, + 0.28213156063208294, + 0.47410903755019634, + 0.2911061720657253, + 0.39466608405530074, + 0.2564141051457285, + 0.450112445594587, + 0.5802868838588789, + 0.7782035399582012, + 0.3846424614169143, + 0.34057002525611707, + 0.3889605513973394, + 0.332315448613665, + 0.3141860923737843, + 0.31946069172034336, + 0.36884041207424423, + 0.30608174424671686, + 0.25686314363571194, + 0.41102012492958917, + 0.33591393558065813, + 0.31343190805275584, + 0.2469632115803993, + 0.22963619205372138, + 0.36748500609229107, + 0.3706365852914194, + 0.3344834360097312, + 0.3442942603397697, + 0.35612507161993073, + 0.265424821793988, + 0.19480843815821713, + 0.32276338787441794, + 0.40480786182545203, + 0.2610783941919207, + 0.3547154170624847, + 0.26559283612119194, + 0.28415655575546916, + 0.18489417281859422, + 0.4840780116723495, + 0.27033235988938203, + 0.2535664543122676, + 0.35479389844541936, + 0.3059047911261839, + 0.2696533463730536, + 0.5716017437204644, + 0.41918172522051034, + 0.19413549179986186, + 0.4027713708779262, + 0.35672492776748593, + 0.35163821727352157, + 0.2844745526715078, + 0.22326839394139733, + 0.3018585781492095, + 0.458497706732017, + 0.20847164819425382, + 0.36752330368673547, + 0.197965059470544, + 0.2580331842159767, + 0.2677931277432203, + 0.23250258019084172, + 0.35389724284385826, + 0.36220615841664594, + 0.2978919169509679, + 0.31968239008583593, + 0.34702412045805925, + 0.22679685141309178, + 0.34053681810794045, + 0.3367541255681905, + 0.47521176701388185, + 0.4733729447408814, + 0.3616860538810284, + 0.37256371423425444, + 0.3491754877987993, + 0.3872798869740764, + 0.4163627768080941, + 0.5065347580123396, + 0.36578647195844566, + 0.24777021300224272, + 0.4802340915112511, + 0.30641727893137133, + 0.4105151205309316, + 0.4048288071288818, + 0.2989066338135825, + 0.4557334618093554, + 0.21119176170601006, + 0.250123756099541, + 0.22244792150302453, + 0.20433083246504344, + 0.3222382665218673, + 0.3249753674778279, + 0.26276034525304487, + 0.36442229008590865, + 0.51217262983306, + 0.3444997286863276, + 0.47764835141412365, + 0.7675392941584575, + 0.5119468997412584, + 0.2207952577881802, + 0.3927437110028354, + 0.41098261418892473, + 0.3733308987726645, + 0.219490907128127, + 0.5541495935259977, + 0.4189395006449678, + 0.3195378714612355, + 0.4169239701860669, + 0.3149816272876019, + 0.4219805396189424, + 0.48111492945311324, + 0.3506479568177141, + 0.4994989991322676, + 0.4153520947913026, + 0.3190567174060711, + 0.4345458865999831, + 0.4664678104689674, + 0.24812767231233623, + 0.27101729415500725, + 0.21451902359028044, + 0.5578624605719439, + 0.47292028725767365, + 0.8082058261405931, + 0.49330895882835213, + 0.3553562433491546, + 0.26117018496973127, + 0.4452151604377276, + 0.4225543642522275, + 0.5855031345794882, + 0.2962358612846252, + 0.49596639374736917, + 0.26980277464093455, + 0.3059952920726175, + 0.21956723158055055, + 0.21917209594871925, + 0.3662274593419945, + 0.18874096067884963, + 0.35778435803132447, + 0.4290626288359164, + 0.33775119382157454, + 0.22461597801989858, + 0.23344210154270145, + 0.3711482764101722, + 0.2211413557282307, + 0.47043698228149117, + 0.39078425762395547, + 0.4119016496113513, + 0.37287689862612966, + 0.28435939830455775, + 0.5323827623750607, + 0.4336149250036749, + 0.3049805121131496, + 0.570250552577498, + 0.26335799123843123, + 0.3784995851709179, + 0.2888866074030455, + 0.2368709411245946, + 0.24097777994657615, + 0.3040958841778415, + 0.40947664095662484, + 0.2768658288411081, + 0.6715314390107525, + 0.41729773896830596, + 0.2354986703847151, + 0.29927811470471144, + 0.2383154494807866, + 0.5141196156926306, + 0.22852843826134475, + 0.3029632197461207, + 0.31816176388311806, + 0.34564495422999225, + 0.23122703430902647, + 0.3632080062076713, + 0.24864061849249355, + 0.38576724488116865, + 0.7549566371555385, + 0.38226483598986194, + 0.3089835438581367, + 0.3196984509042274, + 0.31093250092024294, + 0.20295548672822805, + 0.2987286196547741, + 0.38227810855050715, + 0.7372302038454784, + 0.5161641538138401, + 0.3920946178660357, + 0.31809910297796223, + 0.3062113632136757, + 0.3438778154239721, + 0.22419134349628717, + 0.38512030096603955, + 0.4697149708150653, + 0.27218658017712183, + 0.3589383518480768, + 0.3362069171865404, + 0.4206689302732104, + 0.3214301114628828, + 0.49929191539355444, + 0.33422135385945384, + 0.2610275526452105, + 0.3867226114995485, + 0.39034489503656933, + 0.5108746434829016, + 0.49872579016724516, + 0.22179942458119206, + 0.7035717425376569, + 0.6980336301241746, + 0.1916108455830147, + 0.25367249721191565, + 0.3311832270681193, + 0.4135708288810399, + 0.22798587423728073, + 0.3285520421282381, + 0.4328250515030768, + 0.45112573083593704, + 0.41773688636909917, + 0.3329464829493507, + 0.31492485758302996, + 0.31438123708250854, + 0.29169952404292343, + 0.4074972259084758, + 0.3491172698124794, + 0.3461771908168455, + 0.466634143157023, + 0.37134269395209907, + 0.5082503280798579, + 0.4125552558546188, + 0.5875251977085542, + 0.42135369358822405, + 0.3801626917359982, + 0.46337177794542805, + 0.42234890193345975, + 0.3049610793562382, + 0.5964281848319549, + 0.5794150857526061, + 0.407845524621732, + 0.3418023820245287, + 0.34168787764619457, + 0.4936615710828198, + 0.30884183151327416, + 0.38418651840443513, + 0.3601456974905017, + 0.4458598267365012, + 0.4039214764938503, + 0.3563127420975594, + 0.324287569516386, + 0.49614353306036096, + 0.4599310379160618, + 0.43709150262644675, + 0.453289925905289, + 0.34576802832208253, + 0.22131703448165793, + 0.20155518066404646, + 0.27377108282576623, + 0.48644437344409924, + 0.43257690438517515, + 0.21755455623267386, + 0.4090095438852967, + 0.3207747956825913, + 0.3249166169582282, + 0.3105550028938589, + 0.3431095038110433, + 0.30683403875447557, + 0.4173137788304406, + 0.6003730634285868, + 0.28246199582264586, + 0.39582191327633415, + 0.5121717120275247, + 0.32004798447516924, + 0.4010594807977009, + 0.5592586866268263, + 0.43024157942523616, + 0.42426137782062145, + 0.38207004276404444, + 0.3603609316615388, + 0.22506260888275148, + 0.3441266510441669, + 0.46055853385660295, + 0.46985930296799033, + 0.44806184522818865, + 0.2815892421057636, + 0.4258337528358749, + 0.4147468630371171, + 0.47836780193336237, + 0.38603052456028414, + 0.44782221544783946, + 0.3729985756908485, + 0.43362736970274274, + 0.42232045267984697, + 0.38568111831541235, + 0.2558297066937217, + 0.4280177757692885, + 0.4435611010666606, + 0.22993355802146706, + 0.32953630814701984, + 0.42461340008061627, + 0.36250462626431645, + 0.33005274200006024, + 0.46116707000445034, + 0.4759399876754337, + 0.3995862608988245, + 0.4179737407755563, + 0.30614288960771435, + 0.4233010250180882, + 0.3440237510487404, + 0.3363385689861757, + 0.3607489263339518, + 0.42168855659265103, + 0.3585137757211721, + 0.392471585651364, + 0.37632133083969804, + 0.7095262679812792, + 0.5269041334400133, + 0.48034402130151377, + 0.5200206267568744, + 0.45055163153620914, + 0.40154383982709696, + 0.5215583062167762, + 0.33738156855630097, + 0.3670012152357932, + 0.43765323787796867, + 0.3065614320362491, + 0.42143243673558334, + 0.479739346571168, + 0.4227299843147274, + 0.4321112726222309, + 0.43612153206547366, + 0.302180895190347, + 0.4969798167261069, + 0.47328798717477283, + 0.35182851234990026, + 0.4005928215481716, + 0.3579901889319222, + 0.3243619466375347, + 0.4173588519007031, + 0.4151442392686257, + 0.3945845987365226, + 0.43243147663286235, + 0.36421747079703864, + 0.45299562493519163, + 0.35579374828988003, + 0.38089139062877986, + 0.41477246013203795, + 0.4135357934997427, + 0.39488405289877876, + 0.44253163212628954, + 0.38200355195292035, + 0.3266502028620964, + 0.3318822463142453, + 0.4083072884740192, + 0.3919314614222489, + 0.42806450751958075, + 0.44714295224579986, + 0.40465881575499096, + 0.3959144224104844, + 0.3268405388954985, + 0.3854481239410261, + 0.5015091598947823, + 0.3657171396574473, + 0.19431573563991467, + 0.3233697974915148, + 0.39948662577899574, + 0.2861373336587472, + 0.28582011069490537, + 0.43204998733266275, + 0.4330561179875522, + 0.2563026552661348, + 0.3004774081304705, + 0.32747920353327487, + 0.31097900989241417, + 0.47749918500004246, + 0.25942527183096276, + 0.2601774504547419, + 0.2794792577721431, + 0.28352546543993723, + 0.20554794244405097, + 0.2685657537311299, + 0.24451535967343618, + 0.270242350209283, + 0.23467691646230837, + 0.3682237050770455, + 0.2614024119866476, + 0.3029514835272576, + 0.3127869308736746, + 0.37118462999968654, + 0.2687363192590794, + 0.3604254322276746, + 0.31628375685786253, + 0.3154818258234939, + 0.3176887148120039, + 0.3713420871194723, + 0.34063277132778863, + 0.5593464665719259, + 0.28659886778423743, + 0.35342881462486325, + 0.49541700285624884, + 0.4518619765045207, + 0.3558329527307281, + 0.4300042169298156, + 0.3885163975320872, + 0.4261656025250284, + 0.45010961116504494, + 0.34490251014978607, + 0.5592970385776868, + 0.4048183365774128, + 0.3928070342128559, + 0.33641157752459955, + 0.4054055938623086, + 0.35561019821042644, + 0.42511212827938427, + 0.4310603198829505, + 0.4207075745772124, + 0.4464867000980037, + 0.5567795771313682, + 0.41032452812885195, + 0.3765152793377639, + 0.4486926457260649, + 0.5165300754168325, + 0.4968257018494812, + 0.4266156667197031, + 0.5165580753630974, + 0.5134531619614175, + 0.3607482342363841, + 0.3578152251207684, + 0.4073427608718092, + 0.32567855243694016, + 0.5853369892601908, + 0.4716683498638905, + 0.3926445459420857, + 0.33308754188279954, + 0.4458228011461651, + 0.3757064731104621, + 0.48370798428140027, + 0.42084744481479675, + 0.3682739339277421, + 0.4871745230788159, + 0.26883027469983567, + 0.379278657079271, + 0.4896704614453145, + 0.3037588187091062, + 0.3856997434360264, + 0.29112377130464534, + 0.32369522912583, + 0.3503029626864545, + 0.3564883837139079, + 0.41842795528719057, + 0.3052025378107931, + 0.2948442970401768, + 0.3933167034876522, + 0.3570637928198856, + 0.34738607544446, + 0.4086970211957626, + 0.31698110467133944, + 0.3916649962494112, + 0.4811478103849921, + 0.543934525631077, + 0.42451486329015053, + 0.3231190742530227, + 0.441360718420222, + 0.4095047991683591, + 0.358426896940831, + 0.38467767078025245, + 0.2887465665321346, + 0.29596220547856156, + 0.2658699007887943, + 0.33780998113005, + 0.26092621422825624, + 0.6363231976333572, + 0.40643366987005297, + 0.37992620240304387, + 0.28751165725220634, + 0.28778668191914814, + 0.3065729470131412, + 0.3102193219199993, + 0.2800926347246414, + 0.28748137754135494, + 0.2790457686452915, + 0.3509336782796812, + 0.3460283575663375, + 0.4135631450589589, + 0.36955382745520593, + 0.33270089764964245, + 0.4729507966544752, + 0.4672247616296352, + 0.34225963956166583, + 0.38497443735605497, + 0.38838277728294235, + 0.3545178888329746, + 0.3240508332755915, + 0.4904983232090369, + 0.41992473005732195, + 0.5594677603134068, + 0.4169243804593508, + 0.40952148201117866, + 0.42179528417966244, + 0.425479916070913, + 0.3071977380770707, + 0.5399310232718298, + 0.38206988351329735, + 0.21971523073631422, + 0.42579422852886434, + 0.524386352199893, + 0.33996686776748297, + 0.40574361860511937, + 0.7275860394622785, + 0.4079022117697162, + 0.4711490664972132, + 0.2595337315284135, + 0.4388435776187657, + 0.24320003810280227, + 0.25420536714439057, + 0.27411744460849713, + 0.23043561788254444, + 0.41709684281926424, + 0.1652362528588712, + 0.27271649770262035, + 0.4111564722318734, + 0.3726466817336758, + 0.3712301871959365, + 0.37557385656016135, + 0.28874034375116103, + 0.3490233232897563, + 0.4858846319641563, + 0.5015992179201086, + 0.3191036289077358, + 0.4121627995385939, + 0.36920645813494285, + 0.7007406225655728, + 0.3452708582044739, + 0.535580358226832, + 0.7246188648650008, + 0.6988050110380895, + 0.40490502246001514, + 0.33373194965472125, + 0.27402314911302267, + 0.4126108020579496, + 0.45322785132844895, + 0.4646721617352249, + 0.5298179274201531, + 0.39162487397678813, + 0.5270658465571443, + 0.3772995534613739, + 0.3473644125397413, + 0.67569064273028, + 0.6651026970155488, + 0.4174792521475937, + 0.37912613375856324, + 0.42447013881810414, + 0.3787992733415952, + 0.41505326522911556, + 0.6260476043277399, + 0.6367284807441717, + 0.38489889195532395, + 0.39724210855460235, + 0.4020987147884175, + 0.32448056348579796, + 0.4621078050178996, + 0.29270707278461183, + 0.43944674522820454, + 0.43472152970489697, + 0.32092630420908486, + 0.4003613959479947, + 0.33902770783271396, + 0.29128327928950876, + 0.4676988723067487, + 0.39234686053245493, + 0.3522432561743395, + 0.33360087694967483, + 0.43453932387489264, + 0.373389549008439, + 0.2975562634250749, + 0.31231712063641914, + 0.2154974766344055, + 0.20758438014778524, + 0.2321179788566728, + 0.37992415700946996, + 0.411056259495774, + 0.2820994732876925, + 0.29917912176333605, + 0.28220460130394803, + 0.3794786028224427, + 0.37348714440854264, + 0.4387043175727541, + 0.29728839234007937, + 0.36552800808397334, + 0.4542532344415001, + 0.31073793179319936, + 0.47447528621330287, + 0.41465069855015724, + 0.3708524993341064, + 0.3653652736385736, + 0.30870071106439484, + 0.2008579206201649, + 0.25243810082370755, + 0.2467957859217543, + 0.4734658751982466, + 0.37657662986219737, + 0.3910688943041089, + 0.41204844972137095, + 0.4364768774628571, + 0.2889767668419769, + 0.35459853596847596, + 0.31656830716851897, + 0.635246894285114, + 0.39583765578780394, + 0.4267890088673366, + 0.3284258173326015, + 0.4100891759945411, + 0.4417975178617615, + 0.38662892642362134, + 0.3608951807422695, + 0.43129369846110593, + 0.40197214730364034, + 0.4963635593331362, + 0.3709544745862163, + 0.43200018408921925, + 0.3973825049758421, + 0.37575993418055537, + 0.35315429376289165, + 0.42667337731235394, + 0.3715207956500344, + 0.43753248380922793, + 0.30074151729470033, + 0.28072957088430667, + 0.28184568829892936, + 0.3486057440153385, + 0.3969412082671406, + 0.3178150904804135, + 0.36636641335247067, + 0.41023036657634165, + 0.38127944787647716, + 0.4082184760837829, + 0.3860869461723073, + 0.24277738775845176, + 0.37377442874316574, + 0.21555832053428597, + 0.3757703826571437, + 0.19091659767684438, + 0.6318510657280338, + 0.2926383589629861, + 0.3926122964721411, + 0.5009682485301208, + 0.41808158252584543, + 0.23148691304480368, + 0.36716776723167693, + 0.2947276353105842, + 0.7567566631298865, + 0.2513693746224998, + 0.45459262324827804, + 0.45055874011006836, + 0.46468867294239996, + 0.6164504095238068, + 0.3649123097898038, + 0.49784500196529224, + 0.4079746834315382, + 0.3258812771284439, + 0.2683105961297632, + 0.410399343385222, + 0.2664936462021966, + 0.6921362326731336 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.38062145313428886, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.38277768518901817, + 0.7383123473036689, + 0.7227498516765581, + 0.8047867038115637, + 0.6952944376012447, + 0.6129879147036991, + 0.5732034436473118, + 0.6305606609904256, + 0.3868424610431396, + 0.6699219619456633, + 0.2735781661853463, + 0.43243771199906245, + 0.28836373183172964, + 0.28589610418525935, + 0.262554329066267, + 0.25470960162320977, + 0.30333481264216305, + 0.48497701057014614, + 0.3378289105001015, + 0.35157214348808663, + 0.29822468928274415, + 0.6493608228955551, + 0.2635904840291214, + 0.5718120975140737, + 0.334527176336681, + 0.15061601662544924, + 0.3508267438838379, + 0.39284593116831146, + 0.4285070370798539, + 0.44628916619950126, + 0.2531055286004513, + 0.35653895138207137, + 0.3004725685948783, + 0.3155459018492246, + 0.3879066336350692, + 0.3137804952646289, + 0.2714404174199559, + 0.46328707333492125, + 0.4790116177446223, + 0.6143998773916577, + 0.7732316292631207, + 0.7659265571777883, + 0.7560822129957935, + 0.6217623992937279, + 0.35312072774533415, + 0.3074084052401024, + 0.7623537186546097, + 0.28656812630443607, + 0.28915817428880464, + 0.41886557101822286, + 0.3068871133144371, + 0.7615221129921207, + 0.7796309483225456, + 0.35341093181618316, + 0.2805189672003573, + 0.31403887764109883, + 0.29760284718317137, + 0.34941103804681534, + 0.3323548089208285, + 0.45203143018890923, + 0.38989389469710656, + 0.39098660383307077, + 0.3518563753256458, + 0.4537461143459537, + 0.3782688192409683, + 0.32600860931522335, + 0.3195482062678825, + 0.460437706214228, + 0.31873627666175147, + 0.39113496332956693, + 0.42954187886119816, + 0.37789233281955315, + 0.3596173630870755, + 0.3388308377777712, + 0.36387267403364354, + 0.34400529461228824, + 0.2452678945242375, + 0.23786371722482294, + 0.34799290184773274, + 0.2968963395621898, + 0.3894096578772871, + 0.27056516433118283, + 0.31095844089615604, + 0.2608670006917269, + 0.3392942050080105, + 0.44013038434733825, + 0.3602713518335138, + 0.24440783264829546, + 0.319581432618472, + 0.2889478875305673, + 0.26857710143200997, + 0.3007688106914995, + 0.4348426569002086, + 0.4607022853766301, + 0.2242986871114644, + 0.500786318902332, + 0.3048406669563183, + 0.468268615198575, + 0.2546448981409954, + 0.3991726014057868, + 0.38478240838823285, + 0.4585608882184244, + 0.20052082699098034, + 0.35145644831654715, + 0.2611619199049013, + 0.2883805656925027, + 0.41058001232240376, + 0.2822346867056807, + 0.2931052593586367, + 0.7951853658618325, + 0.7859783717336001, + 0.7928045250268011, + 0.8060387411784589, + 0.5679543341392074, + 0.6708753094458735, + 0.6925814999454467, + 0.741893464596952, + 0.6797237247211361, + 0.5731219888823851, + 0.6424743923246734, + 0.5619451212042328, + 0.7559653437780861, + 0.6391104711520731, + 0.39613789925258963, + 0.3009802981852702, + 0.24551945732450536, + 0.29282048146423917, + 0.4931435218765165, + 0.266538632658287, + 0.2528224734077721, + 0.31019133645533675, + 0.35145278812601877, + 0.2518562263259775, + 0.33431762740379617, + 0.3632139202636042, + 0.37911747097487736, + 0.3350733938501305, + 0.2806285147739148, + 0.31890876435648335, + 0.3479908306551814, + 0.27754047351262995, + 0.2819427439892274, + 0.23420131567841182, + 0.574446638011509, + 0.27806069432185754, + 0.507108217670802, + 0.3311529633474242, + 0.4040974962986664, + 0.21605346569305398, + 0.5384664235112088, + 0.26682191608748895, + 0.45882446613947275, + 0.28571862396917197, + 0.21004061821582345, + 0.305221809992839, + 0.2815415928168176, + 0.2373118478849639, + 0.1474234392174915, + 0.2714757663519809, + 0.2959199418657975, + 0.43605861365245685, + 0.39637759175697845, + 0.4493131252020485, + 0.43010861686249136, + 0.30751600049190436, + 0.40629086759291877, + 0.3477316634572254, + 0.380820416396225, + 0.371918030729332, + 0.36831253667545644, + 0.22624841500390835, + 0.1460543278374311, + 0.34377888835411713, + 0.2632436233341698, + 0.21678577537341295, + 0.22054672326280506, + 0.36948973016292735, + 0.4298150448481976, + 0.5387893758450805, + 0.39432287805569405, + 0.2109849421828965, + 0.42757196460298663, + 0.42265707410724873, + 0.4503848295445601, + 0.2552763617349392, + 0.46417706400604664, + 0.3740690526283342, + 0.44981261956558405, + 0.21572876923269937, + 0.3356490220186734, + 0.3060789380209896, + 0.22764528607805587, + 0.44533930086548706, + 0.37697387422192485, + 0.4814499505583775, + 0.2712586490746757, + 0.5788925546459242, + 0.26449991050611094, + 0.18528905617527128, + 0.3679831803898591, + 0.2552095770041859, + 0.24477358207063646, + 0.18698878867449917, + 0.3078616460617325, + 0.5794878172071521, + 0.21413188641678416, + 0.21776823028756118, + 0.383654955723191, + 0.30864412920217965, + 0.35367259857592886, + 0.4112619360844888, + 0.2881522402734121, + 0.22864508446339427, + 0.3167521832458587, + 0.2966047464267043, + 0.4757074691147521, + 0.48045057354926457, + 0.5184789065428143, + 0.4975870044909387, + 0.24167348207834582, + 0.3037879110229021, + 0.3985739453600221, + 0.23682408368038593, + 0.30104411789793595, + 0.2641236362708188, + 0.13543787319734119, + 0.4171321795557611, + 0.2924501161317165, + 0.21279668798361065, + 0.4797852082130023, + 0.21621758726700555, + 0.1715266493497545, + 0.35671503697886764, + 0.3536112586427656, + 0.25752025781611365, + 0.5578720828050654, + 0.29120901677425093, + 0.270793000192692, + 0.20826329290198034, + 0.3919650506276538, + 0.26681878103580203, + 0.42186372495567104, + 0.2480773303901405, + 0.36008327621917113, + 0.3361937279597879, + 0.33984796879210133, + 0.3004701356957079, + 0.29437529578304567, + 0.3032304101722694, + 0.301467919451016, + 0.4138602400869572, + 0.2593818844776639, + 0.3486869035841265, + 0.27295148629359, + 0.3000995884341089, + 0.3424279449286404, + 0.3492717405576785, + 0.31859271378928467, + 0.3878484302950001, + 0.25359646356445265, + 0.277199568721007, + 0.4615730379853412, + 0.23724230159695098, + 0.29740788480252633, + 0.276734845337718, + 0.2196129962435564, + 0.2803574605632946, + 0.2481003928290346, + 0.24199937761931745, + 0.2621535487500338, + 0.23919509677702355, + 0.28765480098496937, + 0.28813222210996037, + 0.33263598407292977, + 0.3372359543564996, + 0.41989762950786136, + 0.2089065307121764, + 0.36268648334644393, + 0.4356824193928576, + 0.2343076831644483, + 0.23766202070650708, + 0.5319143420058111, + 0.21177806021798423, + 0.21896131685946946, + 0.3609504623860509, + 0.3598427769691459, + 0.39275748192219545, + 0.319246688016825, + 0.4454373467622596, + 0.38651949594953433, + 0.38521030072211043, + 0.3099713791775547, + 0.38419628937219524, + 0.36525937169783046, + 0.3386253284525353, + 0.36251746188342, + 0.4335745943103276, + 0.39195919794898726, + 0.6398221688700048, + 0.2574202750277491, + 0.5318405426161381, + 0.6171301586697027, + 0.6239279801753421, + 0.6208619926818498, + 0.44227493489154307, + 0.42498991758845767, + 0.358212869620045, + 0.6446184180692857, + 0.3927141329687346, + 0.44260915849154864, + 0.2806939852209371, + 0.4210577238752621, + 0.31948072335876887, + 0.32489414101575004, + 0.36265397758760026, + 0.28154224715310844, + 0.39573667586329575, + 0.24616187830281328, + 0.28927357879194565, + 0.30868487513827336, + 0.3181641681013408, + 0.3666221388095019, + 0.2332172671870747, + 0.2993710430084499, + 0.3213009230753429, + 0.33205614215805146, + 0.284696107450392, + 0.32781013155221317, + 0.5025891047354298, + 0.3887325930297226, + 0.3381118329242977, + 0.40276428620761495, + 0.3881921267789556, + 0.2591728352440967, + 0.2841784758572911, + 0.32420044440516627, + 0.35686524941096764, + 0.2855821165597168, + 0.38432748398120936, + 0.47870455013720603, + 0.4723853930080032, + 0.3372960909555249, + 0.27029734202885225, + 0.41986605473965777, + 0.23819274776832738, + 0.2893135435054192, + 0.3957370097829982, + 0.28322730719674083, + 0.8058135123779017, + 0.46632698298991687, + 0.3679807120071657, + 0.3435047807070705, + 0.40468926279550305, + 0.41119414554128975, + 0.34144917194026064, + 0.4503576489212716, + 0.30442516476793774, + 0.4405353356980738, + 0.34424595735775493, + 0.42058649754513916, + 0.4100496356928235, + 0.39697335676836276, + 0.3753745935775227, + 0.29826960783209483, + 0.30994056025433125, + 0.4559143819280476, + 0.47952721425861805, + 0.36261088084316295, + 0.5426114184555808, + 0.37333090586681345, + 0.5060845517528529, + 0.34198175916142887, + 0.3764580087300169, + 0.39133687304443066, + 0.5070277655472718, + 0.49971904071085566, + 0.44527418713144906, + 0.47400481074039763, + 0.4522279389480468, + 0.45806216101137576, + 0.3427593883540523, + 0.3366582082736947, + 0.2903835528586978, + 0.23502184698619127, + 0.318686945362258, + 0.29268582205186305, + 0.2688960624609496, + 0.29249252018831345, + 0.39219897392699415, + 0.29255933297785364, + 0.26692193634284106, + 0.29177969632545153, + 0.4869611159106022, + 0.5323612549961271, + 0.4635043776349078, + 0.5018759108446822, + 0.3743955862547707, + 0.5140931422265897, + 0.3123797551367341, + 0.5842929568558218, + 0.39463191738458675, + 0.490870786516735, + 0.3898161544392806, + 0.48639031131691274, + 0.483272029554726, + 0.43500345483905534, + 0.40179297614486675, + 0.4420993593027978, + 0.44717460260256925, + 0.4699224720044452, + 0.2657629108344253, + 0.31509101234574105, + 0.26879637895392294, + 0.3116513438868304, + 0.4413614375685583, + 0.5028201551501167, + 0.25850550100565384, + 0.5112418651132484, + 0.42200920316104434, + 0.39769922429196203, + 0.32446739461045665, + 0.3969904392580307, + 0.4022218518586607, + 0.3104066579100642, + 0.32650002991282173, + 0.8307270292441725, + 0.37959772133683384, + 0.39064538052979847, + 0.42491598343653797, + 0.44076762853303675, + 0.4026528762573461, + 0.35912557323414157, + 0.5349182902506936, + 0.4087260395450465, + 0.31702875904933203, + 0.3720767600661888, + 0.3795323011917138, + 0.3910406370958445, + 0.36794274248753506, + 0.37824064002552477, + 0.4167145812190318, + 0.40899398004479215, + 0.2565420076897637, + 0.40238349269887097, + 0.25895715886549603, + 0.34761262203202237, + 0.19367089134029988, + 0.3988730625758214, + 0.32049719646090186, + 0.2455235297417123, + 0.41593091678247907, + 0.17858748992439746, + 0.2339069029026787, + 0.5073083545797208, + 0.44659240660046723, + 0.4080216835953354, + 0.4089622430861252, + 0.3588139094129051, + 0.3726781390860911, + 0.5465213139328127, + 0.30095781438411906, + 0.5246557489635676, + 0.30729427806810156, + 0.2392093456985137, + 0.18685282616840918, + 0.35016494874498166, + 0.42996422127801304, + 0.4715989540184759, + 0.5018319181584061, + 0.39108441285772383, + 0.44601314125534286, + 0.3143334933111205, + 0.42250587297507874, + 0.3202881070970239, + 0.256233417963868, + 0.30214666361771253, + 0.31679774253013987, + 0.38377464917385073, + 0.2967583503967893, + 0.32854155108936484, + 0.34711080878662964, + 0.3569402145050964, + 0.4169816897661028, + 0.3348311870433244, + 0.3112740503527762, + 0.43856411150316416, + 0.3950021014916406, + 0.4812080518077293, + 0.4407263357159544, + 0.5116835950672265, + 0.5211028536973829, + 0.6460803385036326, + 0.6303311680535492, + 0.45235902947970585, + 0.49935073709375033, + 0.38178727631449527, + 0.49975383159263187, + 0.3690224119968606, + 0.41396866475278593, + 0.34711782334534913, + 0.5504572664086528, + 0.3420022493730091, + 0.5090158802112767, + 0.6264821427584398, + 0.597055436235276, + 0.5379435418190818, + 0.6167779153875783, + 0.5148251129741959, + 0.43504840837755854, + 0.3668670713116375, + 0.3498771032121479, + 0.2442088070210227, + 0.3104239335708019, + 0.40873290580288546, + 0.48167180760794914, + 0.347960457093729, + 0.26981089157182275, + 0.19146652130879868, + 0.3963158028799727, + 0.38392699642846073, + 0.3741032675559854, + 0.26563307534278807, + 0.48010270801344296, + 0.35841648254264047, + 0.40276199218331815, + 0.30087304654771896, + 0.40096706702702933, + 0.4337530869101456, + 0.2903849977893462, + 0.3190359299154472, + 0.3825760097757355, + 0.3198166629089056, + 0.31748932008087466, + 0.30983728579490066, + 0.3200175084779342, + 0.2925583145820844, + 0.34402536319212756, + 0.3520500068720041, + 0.3558679414213038, + 0.2306361113543332, + 0.5103207476723765, + 0.35359735807427967, + 0.4433331979949387, + 0.45634759629859867, + 0.31715640942494305, + 0.4187543799118782, + 0.41867459879585117, + 0.42377631981106284, + 0.5261112658132674, + 0.35464382568074104, + 0.3870851520371497, + 0.4256284599361047, + 0.3152412662760584, + 0.4085062047928878, + 0.3683460491741825, + 0.28024192278660076, + 0.39010177488699194, + 0.4085937471178078, + 0.3670390344248122, + 0.2483278720809766, + 0.3586466515621874, + 0.21552968222343993, + 0.3286526034637896, + 0.5639345339199435, + 0.4832651048674984, + 0.4987962209299417, + 0.49395391652658166, + 0.5742911209931592, + 0.48815096720870854, + 0.5348403934957214, + 0.4915918578856301, + 0.5386808437624777, + 0.3455009052265127, + 0.3349351728851175, + 0.3557152405560218, + 0.2817708269703925, + 0.24516590238440572, + 0.3189667595254457, + 0.3098323780761088, + 0.2813931082110841, + 0.2315782836089431, + 0.3351018461784414, + 0.24668790949813993, + 0.3982787550192843, + 0.2808849203632005, + 0.48352822169716864, + 0.5198059452074387, + 0.4887741470492439, + 0.24839541273938476, + 0.33318252965330236, + 0.5441331462983672, + 0.26517591930404194, + 0.21448525052357412, + 0.2284532885226245, + 0.80439316478224, + 0.4726438450345937, + 0.37321491187189876, + 0.5025442081449982, + 0.4541747675680784, + 0.4396525798575059, + 0.3695042084245042, + 0.4213304822360518, + 0.4598260194669979, + 0.4003112522761184, + 0.5081484216817548, + 0.46639012935408025, + 0.2070317007479661, + 0.320348541680093, + 0.2217288242797547, + 0.3045734358937419, + 0.24570576779064862, + 0.44613371964886805, + 0.3613005407124993, + 0.28613417467341595, + 0.4018775026367703, + 0.45358399631370444, + 0.2890078392054864, + 0.4072964629560577, + 0.23543611773991843, + 0.38968970558561455, + 0.4475441264833111, + 0.29047135403857316, + 0.29049103010997224, + 0.2300583312733052, + 0.4935692715223212, + 0.22886613849358947, + 0.29296194045559315, + 0.46626536128172535, + 0.3560872962658742, + 0.38385882702062585, + 0.2789049296102409, + 0.29789453345532146, + 0.35930824177613513, + 0.33278023766901255, + 0.40614525425898446, + 0.28855500484224356, + 0.4411379070712847, + 0.2215720760988344, + 0.44391441188859526, + 0.377522961199841, + 0.34387401138292084, + 0.4333149483076629, + 0.34616660400047705, + 0.4457300541244299, + 0.2793862899734889, + 0.4499833556263793, + 0.33915425565277335, + 0.2652604517203428, + 0.392960510392662, + 0.27221643929377065, + 0.45685273389041814, + 0.19153818722138596, + 0.6546291787543782, + 0.6621209505181389, + 0.5525517149172515, + 0.5527725451220755, + 0.5882444202982038, + 0.5910008454518892, + 0.24447726002434358, + 0.3208878624808559, + 0.29176428687558853, + 0.34254775798803916, + 0.2925144727080149, + 0.310326192351838, + 0.32367553806104277, + 0.35225201984729804, + 0.37951467876449796, + 0.28912509835998207, + 0.32620023549896476, + 0.26478726865387464, + 0.44534746066849884, + 0.23429506401967312, + 0.3744625239496419, + 0.3848421330651462, + 0.32021721296691275, + 0.4419080580776849, + 0.3497111740204099, + 0.46404094927254713, + 0.22072436942846158, + 0.42621870909561627, + 0.332154306967944, + 0.3188629695105358, + 0.28184796236381765, + 0.333091684317699, + 0.30537833734511405, + 0.32843485222613106, + 0.47504626485215545, + 0.4291604394908193, + 0.5116531978672713, + 0.315796879621501, + 0.3718962113523937, + 0.3479798467270865, + 0.33406030634494044, + 0.42401509012772687, + 0.3264627027694357, + 0.2045577551028023, + 0.3805815575178028, + 0.33704500418185934, + 0.33146479727972283, + 0.37592530097852145, + 0.30270557092245076, + 0.24437166853165912, + 0.2270638680145124, + 0.29907997757490956, + 0.31827874649602317, + 0.26657582560435883, + 0.19115651796548339, + 0.28709082579156325, + 0.4640348207862493, + 0.5614021126866814, + 0.5016025920921038, + 0.4660900762070622, + 0.501554343617298, + 0.433919614285563, + 0.5004830407258996, + 0.45260446552477884, + 0.44361045275032235, + 0.316879772583754, + 0.35549883724239445, + 0.4295444152684038, + 0.5673445201936925, + 0.15096681681722324, + 0.27212012549340275, + 0.26784381876953195, + 0.2413005501621898, + 0.2560311643689326, + 0.3640046141905738, + 0.274974878675321, + 0.3381624176842221, + 0.7202642565147136, + 0.5958992254330048, + 0.7388447441956073, + 0.7543062771581538, + 0.6272957573582845, + 0.6976185824034932, + 0.7169783657501119, + 0.31768796873714433, + 0.3177467879178898, + 0.33537358750471696, + 0.45848562349721317, + 0.2968661392173812, + 0.32718721305863946, + 0.33060929059255184, + 0.3310986231182535, + 0.3450108600124105, + 0.32747009287120243, + 0.23499694980595454, + 0.40826714782020385, + 0.34369053492750395, + 0.3271043410739325, + 0.3832739610019684, + 0.2816407272310122, + 0.2922609036140584, + 0.2990001328933849, + 0.2893798417160451, + 0.23074140586874825, + 0.31386151471422796, + 0.23793422212228, + 0.48431766190636455, + 0.44595770229858545, + 0.31912717504504334, + 0.25533319495458245, + 0.19327823357986504, + 0.4785883731918192, + 0.5475647579335485, + 0.6246313866640939, + 0.2811439462807673, + 0.3354498458994981, + 0.22526767362487427, + 0.41646159014563316, + 0.3035610849520932, + 0.26611419856072466, + 0.4460992457417301, + 0.4583641321062377, + 0.42917492312331396, + 0.2913830618036705, + 0.39910792160712966, + 0.3911057831958461, + 0.44354468078219517, + 0.4064770061788413, + 0.4473736216345564, + 0.326867544950999, + 0.38544929095545993, + 0.39164603844726015, + 0.4062397758861392, + 0.23775210872718608, + 0.3163130893528681, + 0.36109252135303, + 0.3737603014533842, + 0.524744820969914, + 0.16861472908438194, + 0.5263079942838395, + 0.6875210135908356, + 0.4992414402207908, + 0.35206004981795724, + 0.279834299249431, + 0.30061667853516827, + 0.23828961333889342, + 0.3552191264938222, + 0.4361852379272727, + 0.4100838308586526, + 0.31530488136774837, + 0.23725096390246844, + 0.4302275356440695, + 0.45701281762247753, + 0.5109516967072625, + 0.5776991325878397, + 0.4151474712023482, + 0.4922329939830944, + 0.3171045958979121, + 0.42275045799468636, + 0.24139995147275672, + 0.2365799643692581, + 0.35354353483169626, + 0.330131952570298, + 0.47069127432995445, + 0.44540971967172277, + 0.5874026611512366, + 0.6375503934770186, + 0.7322861747840181, + 0.5719924808808149, + 0.7386740716725146, + 0.3677481044587056, + 0.3460465959428182, + 0.34805776703446517, + 0.41494662541961347, + 0.27832675566028325, + 0.44993777462231027, + 0.4042024478562346, + 0.45515862947777574, + 0.431652139828072, + 0.3699119024578262, + 0.37696590454411916, + 0.3843893327695838, + 0.3258523929146575, + 0.39001844849480144, + 0.3530542774143585, + 0.38085417300851204, + 0.4363527576929635, + 0.4583533944152064, + 0.44693518416471284, + 0.3283848494893905, + 0.3702335452343843, + 0.422680862838037, + 0.29745069054916556, + 0.3891441025943954, + 0.36000372758244925, + 0.3739347675815045, + 0.3428942403145445, + 0.39865465194008637, + 0.31139353341911546, + 0.37824902895563095, + 0.38940162676387335, + 0.4214290231323237, + 0.2952911528564827, + 0.44202308996383344, + 0.400723823200759, + 0.34683280587061016, + 0.4786236578301215, + 0.4536283407644397, + 0.7171923356257299, + 0.20476282353910152, + 0.26128115553108017, + 0.3239760463181153, + 0.30917032544914375, + 0.2361752471285344, + 0.20320078209916903, + 0.2656930982858789, + 0.6553901315453873, + 0.5355700805982123, + 0.6225384239582709, + 0.3045395490823823, + 0.28203689697492135, + 0.24276161268323415, + 0.3242736462391092, + 0.3621944241044535, + 0.19862246133577666, + 0.23057831540091744, + 0.3600060396977276, + 0.3263990049821135, + 0.3888299191909052, + 0.3354584703867746, + 0.38049138942442756, + 0.3270474473304096, + 0.27155121649692504, + 0.6782403664513924, + 0.4403918937204089, + 0.6788565505778388, + 0.5563827482150351, + 0.1993142022355951, + 0.45605467011142564, + 0.4291931083829032, + 0.45503119759416566, + 0.41791917103974846, + 0.34561341422267033, + 0.42035679691398764, + 0.42483135026748653, + 0.45866493955023807, + 0.4154857252128329, + 0.3011295337450407, + 0.36803916642820117, + 0.2282808573268554, + 0.2492931771983109, + 0.6214038583853407, + 0.4231201188833765, + 0.5532165340211691, + 0.4583655667198435, + 0.41207390113139364, + 0.3870426876213627, + 0.365285048194549, + 0.44383815985046166, + 0.364325421441148, + 0.21091891554603895, + 0.32748321412016806, + 0.4098964428678392, + 0.22463646141363094, + 0.7390053101714704, + 0.3853924990704681, + 0.5499609904622018, + 0.7048152543533532, + 0.6662691004438935, + 0.33800070933766135, + 0.7794210353923022, + 0.46088453883839386, + 0.6410567454315019, + 0.34685701914647704, + 0.4071191984749633, + 0.3173642971133757, + 0.30353143694072027, + 0.418263938554733, + 0.4601998565103825, + 0.38263234223651915, + 0.30810175327595657, + 0.327105962164762, + 0.33036340901973776, + 0.35548866111922284, + 0.46697762294632744, + 0.21754700938531155, + 0.43988389696222185, + 0.37255090693669596, + 0.35504799148888855, + 0.4369651431741668, + 0.5125530921537178, + 0.26831119449813373, + 0.35053270652504265, + 0.3765659077266461, + 0.48111388012916345, + 0.44210823808357286, + 0.36477867118734414, + 0.23922444305320587, + 0.45295927340227127, + 0.5139067582819757, + 0.4968944696469354, + 0.376435866010252, + 0.5513360262937158, + 0.5632268598102577, + 0.3813348483136314, + 0.4397395432237508, + 0.4826625097166579, + 0.4430993267195378, + 0.39979699118247763, + 0.29949745885281237, + 0.24203538699667201, + 0.2683447182785337, + 0.3196016188959741, + 0.26505148749552143, + 0.23944718186312747, + 0.2794592037799429, + 0.3022842444045453, + 0.3195037595765347, + 0.2674062990042041, + 0.28286334237538774, + 0.2696007012552013, + 0.2892899751120368, + 0.35225011219980323, + 0.2804818515387832, + 0.25657198050701935, + 0.3744598149558983, + 0.47467825231333316, + 0.6194055200338743, + 0.2559455202910168, + 0.4715007734524377, + 0.4781215076729345, + 0.4904092572389924, + 0.49008278034573594, + 0.3766404364786758, + 0.4076528477343691, + 0.33023329786218586, + 0.43342561925527223, + 0.35707045356291717, + 0.6079335528239457, + 0.40096080728201783, + 0.4374031800630286, + 0.2416906439691268, + 0.23455593773469063, + 0.3592362408078582, + 0.43796025796245447, + 0.3218094283923909, + 0.5403415714624789, + 0.2972107517190015, + 0.5924408783015279, + 0.4443479206156948, + 0.4553682375749723, + 0.46085452946970595, + 0.44487269110749567, + 0.373167554413181, + 0.4735041902013694, + 0.464085655400227, + 0.47175021209272666, + 0.46499942086896906, + 0.3892796442443284, + 0.3015429804221959, + 0.23523018844923932, + 0.3354444721460798, + 0.21145190539103723, + 0.3506802196395156, + 0.3804616245812455, + 0.3877351607310729, + 0.37274837334810795, + 0.40071280821536637, + 0.372485412491646, + 0.4546068857216249, + 0.41415739453200984, + 0.30434955845968664, + 0.4239422672962929, + 0.49170385776075826, + 0.31250270590933654, + 0.25406612834586034, + 0.30735462940753633, + 0.3464581306793453, + 0.4220801627329063, + 0.22420267758754608, + 0.3948896386636563, + 0.2240794226589314, + 0.36650307723103887, + 0.42892584029599573, + 0.4537063591701997, + 0.20016973359394202, + 0.2690143690841394, + 0.29163154155909926, + 0.19956639626569864, + 0.19243070149164995, + 0.29003919672944917, + 0.39335430548887207, + 0.24246151446822214, + 0.30721629798107813, + 0.4057340485608932, + 0.27727507347076, + 0.28869651601976526, + 0.38653844476673815, + 0.4538872075181521, + 0.1693100363785592, + 0.4201216415693239, + 0.3416477089873983, + 0.27635223082605376, + 0.2582361150938024, + 0.4081746581711962, + 0.36620917277025933, + 0.17831447631476705, + 0.43336412616298237, + 0.36804831333234345, + 0.3413447328193411, + 0.4501882521984986, + 0.3079571288247516, + 0.2571562601888342, + 0.3051057771305355, + 0.2174951174427832, + 0.32465990925067434, + 0.304125513270593, + 0.326344382631654, + 0.34964723088124583, + 0.35629523660362566, + 0.4433745630193095, + 0.3694050383102284, + 0.28769079927672514, + 0.33089456282197516, + 0.378295314687992, + 0.3642451931192692, + 0.3241443649767429, + 0.4274092816312925, + 0.3661245660946746, + 0.4672823540337449, + 0.5065033393282474, + 0.4775382052930729, + 0.4092411837725477, + 0.3029888177995054, + 0.21234198949911484, + 0.33916686846311367, + 0.4060691042158167, + 0.35480092043422046, + 0.40271082061667385, + 0.3967976494872176, + 0.27542963482529886, + 0.40700947562703954, + 0.2730237822878391, + 0.23630620843109265, + 0.35339519868628344, + 0.37513068274321604, + 0.40342880038555284, + 0.4081463100031466, + 0.2551214960348946, + 0.5259490010678433, + 0.42826370839016265, + 0.4366741635262415, + 0.46846779968006513, + 0.5425514620023197, + 0.3715300075159725, + 0.2765517794462855, + 0.3493075316619594, + 0.3561839502538916, + 0.3489867654361516, + 0.3286269839508024, + 0.2830469681133667, + 0.3018602357859774, + 0.3589708312526351, + 0.25770894248344617 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.18292135669208592, + "valid_tracks": 4, + "per_track_pearson": [ + 0.2270586253935691, + 0.16029173848931738, + 0.20350820110059617, + 0.14082686178486103 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.22073779191524612, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2509707182023045, + 0.22289894948482844, + 0.20970129696287892, + 0.222390563998171, + 0.23894684601182248, + 0.23642605291290086, + 0.19058060850556124, + 0.20881000334427746, + 0.23294321344204724, + 0.2323442657717988, + 0.2378382962942914, + 0.2423553162592303, + 0.24310640356439098, + 0.23974393945216635, + 0.23685128487696755, + 0.23689716201263694, + 0.2511126557091296, + 0.2390256989965337, + 0.23737461677733057, + 0.2367987070103284, + 0.24402271731089034, + 0.2338903816134166, + 0.22828524223448923, + 0.1681530625745891, + 0.23059708671231266, + 0.2284834524722823, + 0.2297339586921173, + 0.24988344677125568, + 0.24422075458170525, + 0.23500846862410388, + 0.22592096976963194, + 0.21170450545521657, + 0.18177879761602156, + 0.23143608163112914, + 0.142565980486252, + 0.21145719338218047, + 0.1992551165604515, + 0.24121540542352277, + 0.2196400330392345, + 0.25949174007624254, + 0.24568935345556622, + 0.19706131831861817, + 0.16936008607416925, + 0.19616041651229751, + 0.2208260618680496, + 0.20824044937035877, + 0.18494810649709642, + 0.23275588224847116, + 0.18691647429751165, + 0.2081199830815166, + 0.20262658800490913, + 0.18535226254580106, + 0.20947771321624176, + 0.2077909977013178, + 0.19668521483152998, + 0.22017707455982366, + 0.22583562130638354, + 0.2522866916184606, + 0.2399875578894227, + 0.2281647029846099, + 0.22681084356975773, + 0.23647183523154938, + 0.24691324873436316, + 0.23108101306063128, + 0.24854847922855694, + 0.22376907608301813, + 0.2484755210351396, + 0.21033149767070422, + 0.2266362430133286, + 0.2311083327636272, + 0.2382377076308603, + 0.23386378204434577, + 0.2426918086140308, + 0.2399277964145478, + 0.24377738381802516, + 0.24266082240709988, + 0.24194599373859368, + 0.2256719568293018, + 0.24378377109204313, + 0.2469192577109029, + 0.23840360218235118, + 0.2357397905845348, + 0.24197779155326787, + 0.23026010668616076, + 0.2335460492765027, + 0.25426905431582786, + 0.24396451994317817, + 0.23850469343690878, + 0.238157838770536, + 0.25059829160547226, + 0.24497175510573727, + 0.24249534933179787, + 0.24595351137219845, + 0.2373523921908909, + 0.2287920819727172, + 0.24618120930482598, + 0.241583327691182, + 0.22881952182645338, + 0.23580169564856626, + 0.23697147783527958, + 0.2354895069200453, + 0.22347780409990808, + 0.245987389605055, + 0.23936565342430013, + 0.24090343732424327, + 0.22341105212661552, + 0.1783162925884967, + 0.25072028160718046, + 0.24617427539971806, + 0.24190968542462754, + 0.23892588509693685, + 0.24456363921889276, + 0.24898210006439767, + 0.2439174750043756, + 0.2343916804316871, + 0.2266244194183631, + 0.2420947913272068, + 0.2276424508113529, + 0.2394152082250488, + 0.22650365067955117, + 0.24147323771891055, + 0.2302967733752622, + 0.23550130051169366, + 0.23523388551313748, + 0.21860231408560854, + null, + 0.24712271090847168, + 0.2245336571735924, + 0.22478753418913225, + 0.23772396132216886, + 0.1991105204717383, + 0.2480649146769519, + 0.23739378323564317, + 0.22870105033602897, + 0.23938296697870975, + 0.2411421672564894, + null, + 0.229606672354541, + 0.23638405604939122, + 0.20712893347039457, + 0.21006081779359426, + 0.22820425028581023, + 0.22136148397765923, + 0.23639998869954576, + 0.22191083181546764, + 0.22876301906045843, + 0.20135916873419998, + 0.22276380817172448, + 0.21908577705329219, + 0.22674377632049467, + 0.2173906278646434, + 0.23645376883708247, + 0.2434424937213089, + 0.21949815360871686, + 0.1993165729395403, + 0.20342275125072296, + 0.20931016766334198, + 0.21083140287930227, + 0.23827886041988283, + 0.22896860737354602, + 0.23092135214143644, + 0.2331970228905734, + 0.2484337764479008, + 0.2478961039218244, + 0.21609102672435768, + 0.22588729120302167, + 0.22231382528419547, + 0.229809846327576, + 0.21918107838311174, + 0.18572045035605217, + 0.2290327676023046, + 0.2387435712201847, + 0.23577616069986151, + 0.23691370699976033, + 0.23858323378935653, + 0.09102796652481858, + 0.20906134196602913, + 0.20644025288087903, + 0.233520666374147, + 0.22671556059195555, + 0.23577865542617218, + 0.2304079383384791, + 0.1328474941260248, + 0.22898720860554433, + 0.23495675935971175, + 0.1779984831807569, + 0.2539377314305057, + 0.24942314241867086, + 0.24502632801405871, + 0.23794161922626997, + 0.23811911137590488, + 0.21385884971836355, + 0.23422818824901614, + 0.2339869954363044, + 0.24798199728071532, + 0.24540897407845097, + 0.23392459166552668, + 0.23147873219065168, + 0.2453410638624898, + 0.22940414898301678, + null, + 0.23379219124330758, + 0.23127681237698297, + 0.2225394556194301, + 0.24082432014043548, + 0.22032044148345004, + 0.2288859490417846, + 0.22353701818457963, + null, + null, + 0.22687954526221307, + 0.2177739804256424, + null, + 0.24126083710615343, + 0.20509853641510684, + 0.22156404802721363, + 0.22271451909018425, + 0.24487104027939222, + 0.23311229078356663, + 0.22515311907288735, + null, + 0.23054290572903902, + 0.23681382114808933, + null, + 0.2148906924475225, + null, + 0.22772116371851506, + 0.22579003986501964, + 0.22847388585529335, + 0.2209764049254384, + null, + 0.22200665414618995, + 0.2184509127274081, + 0.231722713989812, + null, + 0.22972469741496665, + null, + 0.23408178345997283, + null, + 0.23495097290505562, + 0.21230844655823286, + null, + 0.22201485512901484, + 0.24179662722163223, + null, + null, + 0.22423534760510683, + 0.23538760135078884, + 0.2308143621082948, + 0.2291624920217641, + null, + 0.22196427100395258, + null, + 0.22037765184797292, + 0.2317322520193649, + null, + 0.1980206332030084, + 0.23658308597397312, + 0.23340812613703313, + null, + 0.22929415157833694, + 0.2239354427495364, + null, + 0.2291015830709335, + 0.2250434669828119, + 0.2265765780364261, + null, + null, + null, + null, + 0.2266455124211697, + null, + null, + 0.2349988625684981, + 0.21968051712393277, + 0.22658877921467938, + null, + 0.22524481068738955, + 0.23254308125064344, + null, + null, + 0.21472610960106053, + 0.23338080966316394, + 0.22023276581136977, + 0.2227985160302664, + 0.21459617483797935, + 0.23680236531428248, + 0.22821795557694693, + 0.24149893471446363, + 0.21503394479329413, + 0.23498735103507895, + 0.24269454292358425, + 0.23448674183256465, + null, + 0.22247821336577883, + 0.22405509626441014, + 0.21625162249949592, + 0.22429539846689658, + 0.20978065621129968, + 0.22718491320306441, + 0.21792325377542263, + 0.21706245104960678, + 0.22079690216381948, + 0.2258482805017207, + 0.22148688900948688, + 0.23525673671310965, + null, + 0.2202790616424649, + 0.24558809748532914, + 0.2282320262121938, + null, + 0.21993069839229878, + 0.20010138082298667, + null, + 0.2344311407775856, + null, + 0.23427000227761122, + 0.22308738757333385, + 0.22669061703295784, + 0.2262338160205259, + 0.21526997323615046, + 0.23415700786872443, + 0.22023411634641166, + null, + null, + 0.22129176264452974, + 0.23555500536822402, + null, + 0.23029663763670097, + null, + 0.22137708515166185, + 0.2282144080463014, + 0.23119821057525386, + 0.22806297994804828, + null, + null, + null, + null, + 0.2264408918240745, + null, + 0.2330049606144067, + 0.2291721237610867, + null, + null, + 0.23723267054803265, + 0.23936352835191513, + 0.23728376360470313, + null, + 0.23157485366899952, + 0.2305709324284839, + 0.22960203657832562, + null, + 0.23842680307446093, + null, + null, + 0.22671827645206483, + null, + 0.22647485443879242, + null, + null, + 0.22176436489427281, + 0.23864031307493874, + 0.23221275409170575, + 0.23790736064617019, + 0.2401524269075636, + null, + 0.23392729330544335, + 0.24406660455413132, + 0.20752261282527118, + 0.18894204829846503, + 0.1897784019335995, + 0.2423575050639799, + 0.216900036147479, + 0.2447437652989505, + 0.20821142281768756, + 0.22179135655652818, + 0.21415404524000475, + 0.2185073168248911, + 0.24920594255843162, + 0.23081524144914814, + 0.24589090414128337, + 0.18647692231317708, + 0.19545987094384978, + 0.2185556434014518, + 0.21571958033184097, + 0.24558430302901432, + 0.24124993174733506, + 0.23418528037955835, + 0.21585807167599091, + 0.19953922594939155, + 0.1550677835218486, + 0.2342159587027419, + 0.22966009339589577, + 0.21465417312124904, + 0.2393040690664549, + 0.17856895545861098, + 0.18259923638627812, + 0.2140683911071795, + 0.18401822416286803, + 0.16516688066808774, + 0.21747447883752266, + 0.1491777113858928, + 0.2052779763828955, + 0.14642392375292354, + 0.23799887346048484, + 0.22137211665563944, + 0.21636540516412123, + 0.21101032411246595, + 0.18358073486953777, + 0.18494188438251938, + 0.19170810171075423, + 0.21944612625885146, + 0.16611348549936095, + 0.17189846209137746, + 0.1762295223193314, + 0.19082371867176065, + 0.18281022720424175, + 0.17631333976360106, + 0.17277258088798259, + 0.18051700715343905, + 0.18321705808841873, + 0.1739360653861894, + 0.19305426987841467, + 0.20558220297434762, + 0.214812109013723, + 0.2155600642184204, + 0.22000527292784613, + 0.1931761753553957, + 0.23978528570133772, + 0.23919683549122306, + 0.21313088686654305, + 0.21259189838931322, + 0.22394845986680562, + 0.22610970432035823, + 0.21461321923822252, + 0.2315717386938766, + 0.18188604554296864, + 0.23047565230986464, + 0.22470388720885504, + 0.20907783216496292, + 0.21896035806090788, + 0.21274261981187023, + 0.2076250048550029, + 0.22615615197516512, + 0.23395311920032916, + 0.21773131351785804, + 0.22642710371130495, + 0.23084306575663643, + 0.2117529954314838, + 0.215224790667402, + 0.2214383993303998, + 0.22214532656947727, + 0.231131635145631, + 0.2145737376293834, + 0.2092798933695214, + 0.20414744712287727, + 0.21551259852996446, + 0.21533566956075786, + 0.22479894812165913, + 0.21515691623120015, + 0.21431192642852384, + 0.21020407819677625, + 0.2228462547405148, + 0.20817695426820693, + 0.23245681682175898, + 0.25287689955231735, + 0.23951673718339236, + 0.20461009694519763, + 0.22110652130651196, + 0.22516002895112702, + 0.22725846162762753, + 0.21957989490698637, + 0.21189850519864806, + 0.1762164647081997, + 0.2049816353566484, + 0.2033380517167101, + 0.22048273969843962, + 0.19036889171328736, + 0.20347192267833503, + 0.20826468645968665, + 0.1934415250745787, + 0.20295894134536135, + 0.2075381966221074, + 0.2158076419223579, + 0.19547400471041554, + 0.19758537712229762, + 0.20080266691849094, + 0.21821821317629156, + 0.21284840860861132, + 0.22029024149820675, + 0.23114485209084248, + 0.19738466522281728, + null, + 0.23743310198964326, + 0.20834642299868814, + 0.22573378524249488, + 0.21777697949908723, + 0.1942073069711394, + 0.20832678348853031, + 0.21838075076631097, + 0.22295397328519792, + 0.22588246338320433, + 0.19791737570596282, + null, + 0.21170181733672527, + 0.20073546557940714, + 0.2079219403908521, + 0.20927154425033895, + 0.20261885843187763, + 0.19851858907610415, + 0.19393483271986806, + 0.2061730081339031, + 0.23190860282716863, + 0.18524413504033185, + 0.2145307156275635, + 0.21408952036048756, + 0.20424197534085767, + 0.19956970127869075, + 0.1979366041322006, + 0.22344187898222642, + 0.19801013780051085, + 0.17650718926201414, + 0.19414242443002017, + 0.1999838707433898, + 0.22897784270402147, + 0.222774540682172, + 0.2080979170063034, + 0.21670548866873968, + 0.19887374403682792, + 0.23099422904666345, + 0.21760500962145216, + 0.20151415970756437, + 0.20852865106541116, + 0.2134604324135096, + 0.20253649835395535, + 0.17349481653581655, + 0.19840299073134418, + 0.20572704984210421, + 0.21015485295152053, + 0.20869441199669922, + 0.20123524546683572, + 0.19882605590691912, + 0.18468433798616776, + 0.2185770922485557, + 0.22562176995243052, + 0.20981269925701426, + 0.19237674388351478, + 0.20753584640081582, + 0.18792535096537905, + 0.17619262859789953, + 0.22268492716670216, + 0.21636795990503344, + 0.1880159012614866, + 0.23070013930858468, + 0.19342006010921728, + 0.22499697241995556, + 0.20370655064638518, + 0.21774675552815384, + 0.2310562379738841, + 0.22540499523480628, + 0.22260047260074736, + 0.21021641746949118, + 0.19587390089256293, + 0.2227452052020962, + 0.19931479957763876, + 0.23221994814073327, + 0.2513260091142453, + null, + 0.24749869518822007, + 0.23036773878008676, + 0.19911628392041836, + 0.2340637865060932, + 0.2306695782795583, + 0.2309195323286157, + 0.1916632651749317, + null, + null, + 0.24563212580709454, + 0.2155356589072931, + null, + 0.2251829677212033, + 0.1923239189144152, + 0.2005260119801527, + 0.20346244082462275, + 0.2320420800721853, + 0.22202823423886264, + 0.2350371158313793, + null, + 0.24839312073381414, + 0.24239930793013, + null, + 0.2069056678090727, + null, + 0.20453673180423262, + 0.24139933151753037, + 0.22986697328730266, + 0.20912864831647027, + null, + 0.2123479872391677, + 0.21694584942961995, + 0.2171681329112157, + null, + 0.1726029963554142, + null, + 0.2240493204991936, + null, + 0.21293000928551123, + 0.21078824712112307, + null, + 0.1897067984518134, + 0.24503306521361973, + null, + null, + 0.2315494658577455, + 0.21334217671747632, + 0.21955881372764105, + 0.21686164071354286, + null, + 0.20187285939896243, + null, + 0.20737662492415532, + 0.22571745585548572, + null, + 0.15627138486766382, + 0.2102260774264516, + 0.19694775685455007, + null, + 0.20843719466573898, + 0.23256864233224897, + null, + 0.239734181712724, + 0.23476691506508754, + 0.26005591445504034, + null, + null, + null, + null, + 0.2483170625947995, + null, + null, + 0.2182034021462658, + 0.2344413387249525, + 0.24405856357228986, + null, + 0.23784697253645576, + 0.23902998300354597, + null, + null, + 0.23166992663977484, + 0.23600404011254061, + 0.222060293206732, + 0.21286672118552954, + 0.21619383091420805, + 0.21744644880358804, + 0.20997118847789614, + 0.2091525579204129, + 0.20298042899478097, + 0.2149033035705548, + 0.22616556490787285, + 0.20366177934327734, + null, + 0.20409857670195003, + 0.19959768509861087, + 0.24122332024609333, + 0.21993398736714093, + 0.20201464801152036, + 0.2251665043694006, + 0.22867236627597265, + 0.23851753087548133, + 0.2490504797851745, + 0.2152260016729486, + 0.21687095832124928, + 0.19352046076153337, + null, + 0.2019392704853171, + 0.241528614122995, + 0.24977141704108838, + null, + 0.2246054217600176, + 0.18993546029415134, + null, + 0.21388550796642525, + null, + 0.24762901617971161, + 0.21077745301020545, + 0.22337237711933916, + 0.19400804740758965, + 0.23644053656590583, + 0.2060846872260812, + 0.1907735424509287, + null, + null, + 0.2328262085497406, + 0.23543865916013587, + null, + 0.20956758405862388, + null, + 0.20896652839293287, + 0.22339995321068043, + 0.20564458966374194, + 0.2506969168339787, + null, + null, + null, + null, + 0.21663060007778998, + null, + 0.213555159982003, + 0.24759850062200567, + null, + null, + 0.2350371723429493, + 0.2150811810744874, + 0.19350009999949283, + null, + 0.22763127866180205, + 0.22273062804841576, + 0.2026131746660195, + null, + 0.2340814312954415, + null, + null, + 0.21500117518755382, + null, + 0.20875634280066271, + null, + null, + 0.19175828345606247, + 0.2557539627590196, + 0.2475560556799951, + 0.2463092705930107, + 0.2523477604139763, + null, + 0.24470995843778454 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012, + 0.721932804832012 + ] + }, + "junctions": { + "mean_pearson": 0.179636230076655, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2844583807555112, + 0.27018118870562946, + 0.3783209692067856, + 0.26282879691422334, + 0.3258132201479353, + 0.23667132384520911, + 0.005763235507467452, + 0.11744121872456274, + 0.3210454514428416, + 0.19067836583958817, + 0.16164670075455584, + 0.10811166380843984, + 0.06717129541793285, + 0.07846809259994411, + 0.21004761654850324, + 0.22298207317137456, + 0.20694400173196778, + 0.0534562541764343, + 0.22078240693177442, + 0.20875494519858362, + 0.23403241172358022, + 0.20333169490766945, + 0.2479764068544412, + 0.16003854091647052, + 0.18382351681123715, + 0.24841976823882236, + 0.1702277705946532, + 0.2558947038954382, + 0.20576839013876463, + 0.17950763238217396, + 0.2766525046381171, + 0.3479110518194329, + 0.2292287322569531, + 0.31808511754581886, + 0.2865063567517287, + 0.37628599086532294, + 0.28048889182776693, + 0.18276530134457117, + 0.223005036843419, + 0.09862593473506606, + 0.10303332909269641, + 0.21876295239898175, + 0.27539135141387094, + 0.21436576130561003, + 0.18408328793331916, + 0.19562260393260977, + 0.28556616237541865, + 0.21526338813046592, + 0.21620463350195768, + 0.1839684089669717, + 0.22762056512796427, + 0.2691521567711689, + 0.1684397376134178, + 0.2618655749504888, + 0.13655622450818589, + 0.2532054305246176, + 0.1704949832585697, + 0.11020806648023079, + 0.1447087132426486, + 0.1321652378406192, + 0.06283907450992654, + 0.1574668250855348, + 0.19581196453434768, + 0.10978129980528434, + 0.06543186457255559, + 0.2464102560313125, + 0.18875840369453198, + 0.29008595774422313, + 0.11568228096665015, + 0.1944964980868987, + 0.16536306369444037, + 0.1788218877627163, + 0.19140863678984327, + 0.15475809122256265, + 0.39927136685538195, + 0.3335459281103966, + 0.2413864652466673, + 0.2177643933337331, + 0.23109414181185903, + 0.3441399957577581, + 0.28374564777746025, + 0.2679695803231174, + 0.1774316875682275, + 0.12363613248683039, + 0.23153084792297754, + 0.26690686420722426, + 0.28798772215937934, + 0.2891919614227104, + 0.2600814413285165, + 0.2492644588534774, + 0.31125162224862707, + 0.21574224417006393, + 0.15261804037615007, + 0.20729981068735992, + 0.20256636236455264, + 0.33639053063736346, + 0.26424972424529586, + 0.178159014639859, + 0.21425766928177942, + 0.18094226203227265, + 0.16717557161279517, + 0.22167864053655228, + 0.3724137718293376, + 0.20378853607098696, + 0.19736145371568292, + 0.24176968801259965, + 0.0753900423857118, + 0.2516934014622092, + 0.3034365727754454, + 0.32356748860488094, + 0.1587378820474928, + 0.33144555968016315, + 0.32777653821279035, + 0.3572632937638584, + 0.3007402411564745, + 0.1117258065407878, + 0.06914485483990371, + 0.1675980706867601, + 0.12336019433811633, + 0.25871512906934674, + 0.21736388713650326, + 0.13037385596651235, + 0.1446252888234776, + 0.2576154021110002, + 0.08209687197577814, + null, + 0.3472608735088374, + 0.17121895007572413, + 0.0717649572346232, + 0.16546774461559452, + 0.2204307863313354, + 0.23868424575881309, + 0.27114857454943625, + 0.2268474735125111, + 0.3166929834354794, + 0.12495176863242487, + null, + 0.3058879717908424, + 0.2942459813375849, + 0.2541488110991825, + 0.30494893046380533, + 0.16911356082083678, + 0.2323857475142608, + 0.20599250914444703, + 0.1981709395214919, + 0.15042082261658793, + 0.22313315581063767, + 0.203178389250608, + 0.17970726885890692, + 0.14633363555599185, + 0.13364656004454115, + 0.19613894854674488, + 0.14973110039826598, + 0.32707803176192196, + 0.40691293576188337, + 0.34100485857677376, + 0.37683505053013955, + 0.20394645598743188, + 0.15594816410675605, + 0.24727162243996634, + 0.16718482554118497, + 0.19475923878309448, + 0.24091329348457344, + 0.21706696748532783, + 0.08580024571901192, + 0.28895093169371733, + 0.23344685223060624, + 0.2531240634184285, + 0.08339219972984549, + 0.14276934555754872, + 0.19195364090072123, + 0.2404858262152004, + 0.2772719399576012, + 0.19039137530410977, + 0.09104629409645497, + 0.24025193569488218, + 0.1531222183098613, + 0.1716916430019571, + 0.17419652757910975, + 0.22746448505133865, + 0.24585121062922421, + 0.2015092274395377, + 0.04743736636516253, + 0.17213398620070539, + 0.23633632420379466, + 0.22784009367661726, + 0.2968127804465923, + 0.22056611691430789, + 0.2548449868549782, + 0.1826069900033219, + 0.28476528574491455, + 0.3587902558454471, + 0.18896895218465032, + 0.08804944213443043, + 0.1204536061354575, + 0.07131870043039747, + 0.06837700612188506, + 0.2003368298273845, + 0.26584129146084784, + 0.3699616077944031, + null, + 0.17589836762115507, + 0.19690996725498316, + 0.2351322947832965, + 0.2660327986707886, + 0.10862634702344, + 0.12328769846022118, + 0.08344099729152482, + null, + null, + 0.19781710629886126, + 0.22364040820156147, + null, + 0.14737723814655565, + 0.11878721658016131, + 0.08779034635018816, + 0.16657373474105713, + 0.09173767619245132, + 0.11327824337626625, + 0.07770242708522633, + null, + 0.269423084530916, + 0.1330073621530268, + null, + 0.060835026463275065, + null, + 0.23251819014536002, + 0.22385045744635174, + 0.14508921892530452, + 0.16332515424451327, + null, + 0.050257360341309766, + 0.04878503192860548, + 0.16929394116256688, + null, + 0.1140196035853508, + null, + 0.1069493586052358, + null, + 0.12780600722353888, + 0.10991164749144332, + null, + 0.18257127529875608, + 0.17598948891790586, + null, + null, + 0.23471889548458771, + 0.22819034041197372, + 0.03827568640022146, + 0.08015511078192614, + null, + 0.24760963974352473, + null, + 0.2107635299905633, + 0.14942012759869328, + null, + 0.16324066838041285, + 0.18497400101835587, + 0.19684670483109026, + null, + 0.14883564553488754, + 0.20130599806809477, + null, + 0.3375877974454653, + 0.2335866697788087, + 0.27927796425198037, + null, + null, + null, + null, + 0.2056935293304291, + null, + null, + 0.15201974775886457, + 0.2396049400285959, + 0.22969315533437448, + null, + 0.16383721741289545, + 0.29848518429144494, + null, + null, + 0.12661008427960171, + 0.08132170655462088, + 0.17908451851924373, + 0.1059583210701765, + 0.09668877472863988, + 0.07924696432047597, + 0.14980474766665391, + 0.1446533714850929, + 0.04159152202496807, + 0.06319195777308194, + 0.31853398451440845, + 0.11866072507030197, + null, + 0.1701484691190843, + 0.1788832572751104, + 0.05148305162554285, + 0.12678770542799525, + 0.10388005260928984, + 0.05527084582419611, + 0.121622708599299, + 0.14451633003970815, + 0.21437945435650146, + 0.14401795605216966, + 0.2020111755706708, + 0.17954331697050807, + null, + 0.16384257698932284, + 0.20124831848117053, + 0.21608269437183367, + null, + 0.15434409637698435, + 0.17020775680726025, + null, + 0.1326994690013508, + null, + 0.14364471930443873, + 0.09584010464626229, + 0.24458733223018686, + 0.122507549794763, + 0.15729299657081744, + 0.11384348333215538, + 0.10746008471990975, + null, + null, + 0.05810723168134404, + 0.1296240591544761, + null, + 0.17474683133961724, + null, + 0.18628799466773938, + 0.10267253738948846, + 0.0632879412236327, + 0.15539815219170383, + null, + null, + null, + null, + 0.022129238400500562, + null, + 0.07899291365088572, + 0.20850922314755987, + null, + null, + 0.20164590159891074, + 0.23970289714497317, + 0.11487704299612742, + null, + 0.2232585442698065, + 0.184885197870766, + 0.09149817232149875, + null, + 0.20538944976690415, + null, + null, + 0.2323999393650732, + null, + 0.06943098415210013, + null, + null, + 0.17097299889162798, + 0.1993643200647327, + 0.1861155335873945, + 0.17686715296665503, + 0.15362151233821714, + null, + 0.10945901364446657, + 0.07058716855391355, + 0.09503942242413814, + 0.3997154700605015, + 0.06759610945704612, + 0.24891761679173602, + 0.15789354196871633, + 0.07551059785507039, + 0.17134617073179545, + 0.0783451811770116, + 0.11954973518047957, + 0.11975923260787873, + 0.23632265488514614, + 0.2107653395329473, + 0.0720993149623513, + 0.1832276652286604, + 0.231798363435072, + 0.10252781706959221, + 0.08995451746416366, + 0.10273175548271415, + 0.2990698129887157, + 0.12478844151212132, + 0.16804319894910338, + 0.1973403071571537, + 0.30149474760352835, + 0.24414970096636807, + 0.20150512727994127, + 0.17214057342690484, + 0.1555283009446697, + 0.1317988200701541, + 0.09535823610653718, + 0.15555061708634368, + 0.22210584290048846, + 0.05222908418707518, + 0.25751629524724456, + 0.07371971558605231, + 0.26744237400103205, + 0.05261108553848055, + 0.11773429718084635, + 0.10880307531276183, + 0.28257593082107846, + 0.16801050411324825, + 0.053435670261300146, + 0.11568803847930687, + 0.06633604228696453, + 0.0851110196698628, + 0.09947062790103962, + 0.15368540778309045, + 0.205218721428822, + 0.20809931407991805, + 0.15298698959789397, + 0.08366990988155394, + 0.10094963383712352, + 0.14755211268456325, + 0.060450360984846616, + 0.07968196114345008, + 0.10828193703058286, + 0.1910783653235858, + 0.18399971275830104, + 0.21183546212301238, + 0.21096822696436052, + 0.2567177101259353, + 0.09676934328741063, + 0.11404485802405545, + 0.23244039372800912, + 0.07570504812741832, + 0.190971332894775, + 0.1484069260765962, + 0.1511807100006888, + 0.1392887544796339, + 0.0901767162201882, + 0.15295455417968318, + 0.22257299364581626, + 0.20571318361079124, + 0.0864673124011681, + 0.050931620715129064, + 0.09178821971841698, + 0.09877963742196641, + 0.19556752164720964, + 0.12057317495584026, + 0.12664937761479136, + 0.12487433818303673, + 0.25014958060593834, + 0.18120794322001785, + 0.0930236866158317, + 0.15698603937533795, + 0.1728131477658125, + 0.22847915366199023, + 0.11507925905130036, + 0.07311997602406045, + 0.05718347488899818, + 0.1559541790296936, + 0.09298513177994902, + 0.07263462310600989, + 0.0526147464760816, + 0.10411079609388785, + 0.11969923217720241, + 0.07917709145841732, + 0.1266396148314745, + 0.2123503301123792, + 0.20764135986847387, + 0.10321889318917102, + 0.26841521992042944, + 0.17256874518427567, + 0.19184051228189622, + 0.20306058971446864, + 0.13980861136409015, + 0.006763701429315128, + 0.207187048962326, + 0.07472792731031823, + 0.11458023640978468, + 0.05986090138545814, + 0.09673405987117538, + 0.09312282893259169, + 0.19257222658551135, + 0.06737929027617473, + 0.11168408312173476, + 0.2189586057416233, + 0.13058824076703937, + 0.2549125110111041, + 0.09185256654100693, + 0.08406582012659264, + 0.10408776666378715, + 0.06608307676015997, + 0.2636638912014108, + 0.217965645848028, + null, + 0.15307408455227656, + 0.18793200840691993, + 0.03030283566172638, + 0.2572024321621195, + 0.18424507357905842, + 0.06844240370197675, + 0.058729618115815, + 0.1669063500242318, + 0.162327516487172, + 0.17499777779905037, + null, + 0.15490891026578288, + 0.10725713431120372, + 0.043987516055784597, + 0.05846198410180613, + 0.18707020085287082, + 0.1675008735443163, + 0.17165039904284263, + 0.16651983740928983, + -0.017139248921525626, + 0.031561449947562735, + 0.10134880199228809, + 0.07928224287340456, + 0.13991305297309564, + 0.19108706025032277, + 0.2490550580354157, + 0.11242763738285867, + 0.3627708184276816, + 0.21562780074549162, + 0.3670240442749497, + 0.21467365068359295, + 0.10066049520202654, + 0.1301216715975092, + 0.10805249341636528, + 0.18473446274974828, + 0.13364001872840703, + 0.13069072328236445, + 0.2443960499128708, + 0.17447626206590713, + 0.16974129873795812, + 0.24881134156937096, + 0.19928136839923727, + 0.13213629825593923, + 0.1703324149068857, + 0.116532159150343, + 0.06414595634041521, + 0.17116060335389172, + 0.13896735959497125, + 0.16838354663630312, + 0.08463894722984122, + 0.2908755097686476, + 0.14936623721863757, + 0.20605905612199008, + 0.14131814453700467, + 0.09860967701241312, + 0.11292822347591794, + 0.2522806272193124, + 0.1995878226455283, + 0.10073785378043854, + 0.2961465571549746, + 0.24978377253018103, + 0.12069385868674966, + 0.20852880066305604, + 0.1715435217715767, + 0.3226043914508298, + 0.4576936196222382, + 0.14776253665844602, + 0.34391351720478874, + 0.2886259080199578, + 0.1777894569629597, + 0.3074524353805101, + 0.2166663323244355, + 0.15926838907149174, + 0.11713759926383496, + null, + 0.2498157900298854, + 0.19104882946220336, + 0.17204917812992818, + 0.21838598421986855, + 0.21062343376912204, + 0.10647527279371201, + 0.1815347823705039, + null, + null, + 0.19113781635020805, + 0.18861979916217633, + null, + 0.0463697829770835, + 0.15799796594151794, + 0.2468981213693049, + 0.16742565440988896, + 0.07978434052034201, + 0.3048422188947135, + 0.16908212040145773, + null, + 0.20899159249384627, + 0.19395579657793072, + null, + 0.1272363146371825, + null, + 0.23586009480640233, + 0.1583889823554571, + 0.1732667848502657, + 0.077108540325929, + null, + 0.09759117275999116, + 0.14035691374773168, + 0.14094378012506614, + null, + 0.2062002667288128, + null, + 0.3722267938541417, + null, + 0.106967521097231, + 0.18327289006772876, + null, + 0.20510429766084834, + 0.2177370090873375, + null, + null, + 0.1492045681159689, + 0.11226159283122387, + 0.1746451291441535, + 0.20058853329407386, + null, + 0.14032923491111646, + null, + 0.022448300135060852, + 0.10109740263435393, + null, + 0.25854464807904975, + 0.18629665740780108, + 0.22771295088655358, + null, + 0.2564719684317938, + 0.23868733948885223, + null, + 0.24731461145773323, + 0.2616005896397913, + 0.21282654223248315, + null, + null, + null, + null, + 0.18578282983841043, + null, + null, + 0.2074555853551363, + 0.12061044935407801, + 0.18615685179887306, + null, + 0.14095760769874113, + 0.16774867947467278, + null, + null, + 0.19764697042199153, + 0.05588290814548569, + 0.30691698736934425, + 0.1710415360551636, + 0.12164383342409353, + 0.12316047913717423, + 0.46291431253287496, + 0.12258540088469905, + 0.40162304635510776, + 0.17855592257267652, + 0.16130570399856153, + 0.1516212018327705, + null, + 0.14583148780342514, + 0.19855192350158, + 0.2406245730725977, + 0.28767486282958904, + 0.06872645252619565, + 0.33042998765410747, + 0.11689710231564023, + 0.09241466074146969, + 0.2895885951417858, + 0.2193942275184029, + 0.17297418459269542, + 0.15929593461795125, + null, + 0.10220296556718919, + 0.18216078519152673, + 0.08441392353049906, + null, + 0.14094793022054505, + 0.17286370290935557, + null, + 0.3203748182538522, + null, + 0.30566488984579787, + 0.17140945282187664, + 0.1150909553877591, + 0.2613826649443251, + 0.20762037990502732, + 0.15315140851115316, + 0.21744167907976614, + null, + null, + 0.10585030149765592, + 0.27747614714339225, + null, + 0.08308627810498571, + null, + 0.25670224666503594, + 0.18485801384536393, + 0.27161093065523667, + 0.18710328438878282, + null, + null, + null, + null, + 0.07633607101781532, + null, + 0.24968303215744694, + 0.2068173876382213, + null, + null, + 0.13725963963234467, + 0.24864657529693743, + 0.23448649061246568, + null, + 0.1040934400341926, + 0.17433283707131042, + 0.19788430174129532, + null, + 0.20603490888288895, + null, + null, + 0.10991944585518588, + null, + 0.1583298790031008, + null, + null, + 0.1612156921250306, + 0.19512993418108465, + 0.2507754791292043, + 0.2471119457165488, + 0.21025207451262448, + null, + 0.12947600168421391 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9503311258278145, + 0.9539473684210527, + 0.9466192170818505, + 0.9367088607594937, + 0.9635761589403974, + 0.9525423728813559, + 0.9507042253521126, + 0.9479166666666666, + 0.9551282051282052, + 0.950354609929078, + 0.9461279461279462, + 0.9389830508474576, + 0.9415584415584416, + 0.946875, + 0.9466666666666667, + 0.9473684210526315, + 0.9405594405594405, + 0.9551282051282052, + 0.9490445859872612, + 0.9509803921568627, + 0.9532374100719424, + 0.9483870967741935, + 0.9415584415584416, + 0.9405204460966543, + 0.9466192170818505, + 0.9572368421052632, + 0.9615384615384616, + 0.9559322033898305, + 0.9459459459459459, + 0.9448275862068966, + 0.9486301369863014, + 0.9579288025889967, + 0.9402985074626866, + 0.9513677811550152, + 0.9413919413919414, + 0.9463722397476341, + 0.9312977099236641, + 0.9373040752351097, + 0.9370629370629371, + 0.9463087248322147, + 0.9556313993174061, + 0.9518900343642611, + 0.9465648854961832, + 0.9407407407407408, + 0.9475409836065574, + 0.9358108108108109, + 0.9520295202952029, + 0.931899641577061, + 0.9290780141843972, + 0.9489795918367347, + 0.9416058394160584, + 0.9448529411764706, + 0.9431818181818182, + 0.9421768707482994, + 0.9253731343283582, + 0.9387755102040817, + 0.9377162629757786, + 0.9503311258278145, + 0.9417808219178082, + 0.9366197183098591, + 0.9482758620689655, + 0.9573770491803278, + 0.956953642384106, + 0.9522184300341296, + 0.9328859060402684, + 0.9620253164556962, + 0.9358108108108109, + 0.9530201342281879, + 0.9453924914675768, + 0.954248366013072, + 0.9501779359430605, + 0.9451612903225807, + 0.9494949494949495, + 0.956953642384106, + 0.942652329749104, + 0.9559322033898305, + 0.9517684887459807, + 0.939799331103679, + 0.9482758620689655, + 0.9489795918367347, + 0.9470198675496688, + 0.9620253164556962, + 0.934931506849315, + 0.9449838187702265, + 0.9543859649122807, + 0.9504643962848297, + 0.9417808219178082, + 0.954248366013072, + 0.95, + 0.9462025316455697, + 0.9477351916376306, + 0.9484536082474226, + 0.9578947368421052, + 0.9523809523809523, + 0.9493243243243243, + 0.9513888888888888, + 0.9389830508474576, + 0.9446366782006921, + 0.9503311258278145, + 0.9430379746835443, + 0.9433333333333334, + 0.9475409836065574, + 0.9430604982206405, + 0.9488817891373802, + 0.9522184300341296, + 0.9533333333333334, + 0.9213483146067416, + 0.9565217391304348, + 0.946843853820598, + 0.9493243243243243, + 0.9428571428571428, + 0.9463087248322147, + 0.9545454545454546, + 0.9525423728813559, + 0.9523809523809523, + 0.9425675675675675, + 0.9448275862068966, + 0.9493243243243243, + 0.9458483754512635, + 0.9556313993174061, + 0.9357142857142857, + 0.9491525423728814, + 0.9338235294117647, + 0.9572368421052632, + 0.9352226720647774, + 1.0, + 0.9515570934256056, + 0.9477351916376306, + 0.946969696969697, + 0.9514925373134329, + 0.9565217391304348, + 0.9482758620689655, + 0.9547038327526133, + 0.9360902255639098, + 0.9578947368421052, + 0.9431438127090301, + 1.0, + 0.9559322033898305, + 0.9514925373134329, + 0.9539473684210527, + 0.9635036496350365, + 0.9365671641791045, + 0.9632352941176471, + 0.9482758620689655, + 0.9403508771929825, + 0.9274809160305344, + 0.9488188976377953, + 0.9444444444444444, + 0.9338235294117647, + 0.9328621908127208, + 0.937037037037037, + 0.9468085106382979, + 0.9413919413919414, + 0.9675324675324676, + 0.9551724137931035, + 0.9484536082474226, + 0.9545454545454546, + 0.9602888086642599, + 0.9434628975265018, + 0.9452554744525548, + 0.9393939393939394, + 0.9444444444444444, + 0.9498432601880877, + 0.9372822299651568, + 0.9358490566037736, + 0.9568345323741008, + 0.9515570934256056, + 0.9735099337748344, + 0.939622641509434, + 0.9473684210526315, + 0.9487179487179487, + 0.9453924914675768, + 0.9586466165413534, + 0.9554794520547946, + 0.9461538461538461, + 0.953307392996109, + 0.9492753623188406, + 0.9442508710801394, + 0.9485294117647058, + 0.9440559440559441, + 0.9442508710801394, + 0.9477611940298507, + 0.955719557195572, + 0.9498207885304659, + 0.9507042253521126, + 0.9520295202952029, + 0.952054794520548, + 0.9393939393939394, + 0.931899641577061, + 0.9475409836065574, + 0.9409722222222222, + 0.9448529411764706, + 0.9357142857142857, + 0.9436619718309859, + 0.9372822299651568, + 0.9328621908127208, + 0.9338235294117647, + 0.9409722222222222, + 0.9477124183006536, + 0.9533333333333334, + 1.0, + 0.9539473684210527, + 0.9423728813559322, + 0.946843853820598, + 0.9459459459459459, + 0.954983922829582, + 0.9540636042402827, + 0.9375, + 1.0, + 0.9924242424242424, + 0.9603960396039604, + 0.9512987012987013, + 0.9923076923076923, + 0.9482758620689655, + 0.9377289377289377, + 0.9416058394160584, + 0.9419795221843004, + 0.9503311258278145, + 0.9413793103448276, + 0.9429530201342282, + 1.0, + 0.9472049689440993, + 0.9347079037800687, + 1.0, + 0.9296875, + 1.0, + 0.9418181818181818, + 0.9403508771929825, + 0.9401408450704225, + 0.9494949494949495, + 1.0, + 0.9384057971014492, + 0.9366197183098591, + 0.9333333333333333, + 1.0, + 0.936026936026936, + 0.9923076923076923, + 0.9383116883116883, + 1.0, + 0.95, + 0.9496644295302014, + 1.0, + 0.9256505576208178, + 0.940983606557377, + 1.0, + 1.0, + 0.9432624113475178, + 0.9509803921568627, + 0.9444444444444444, + 0.9393939393939394, + 1.0, + 0.9363957597173145, + 1.0, + 0.9383116883116883, + 0.9484536082474226, + 0.9930555555555556, + 0.9538461538461539, + 0.9389067524115756, + 0.9309090909090909, + 0.9927536231884058, + 0.9321428571428572, + 0.9464882943143813, + 1.0, + 0.9573770491803278, + 0.9333333333333333, + 0.9457627118644067, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9435215946843853, + 1.0, + 1.0, + 0.9489795918367347, + 0.9432624113475178, + 0.9455128205128205, + 0.9916666666666667, + 0.9419795221843004, + 0.9383561643835616, + 1.0, + 0.9923076923076923, + 0.9372822299651568, + 0.9420289855072463, + 0.9364548494983278, + 0.9425675675675675, + 0.9351535836177475, + 0.9319727891156463, + 0.9488054607508533, + 0.9411764705882353, + 0.9491525423728814, + 0.9252669039145908, + 0.945859872611465, + 0.9372693726937269, + 1.0, + 0.9438596491228071, + 0.9527027027027027, + 0.9464285714285714, + 0.9494949494949495, + 0.9523809523809523, + 0.9618055555555556, + 0.9438596491228071, + 0.9379310344827586, + 0.9377049180327869, + 0.9358108108108109, + 0.9420289855072463, + 0.9419795221843004, + 1.0, + 0.954983922829582, + 0.9425675675675675, + 0.9455128205128205, + 1.0, + 0.9488817891373802, + 0.9498207885304659, + 1.0, + 0.9312714776632303, + 1.0, + 0.9455128205128205, + 0.9479166666666666, + 0.9537953795379538, + 0.9347826086956522, + 0.9421768707482994, + 0.9417808219178082, + 0.9440559440559441, + 0.9929078014184397, + 1.0, + 0.9477124183006536, + 0.935374149659864, + 1.0, + 0.9386281588447654, + 1.0, + 0.9435215946843853, + 0.9429530201342282, + 0.9345454545454546, + 0.9368770764119602, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9486301369863014, + 1.0, + 0.9480968858131488, + 0.9587301587301588, + 1.0, + 1.0, + 0.9401408450704225, + 0.9435215946843853, + 0.9481481481481482, + 1.0, + 0.9486301369863014, + 0.9393939393939394, + 0.9326241134751773, + 0.9919354838709677, + 0.964968152866242, + 1.0, + 1.0, + 0.9488054607508533, + 1.0, + 0.9514563106796117, + 1.0, + 1.0, + 0.9420849420849421, + 0.948051948051948, + 0.9381443298969072, + 0.9381443298969072, + 0.9366197183098591, + 1.0, + 0.9403973509933775, + 0.9699248120300752, + 0.968421052631579, + 0.9818840579710145, + 0.9678571428571429, + 0.9841772151898734, + 0.9644128113879004, + 0.9686411149825784, + 0.9509433962264151, + 0.9726962457337884, + 0.9676258992805755, + 0.9584775086505191, + 0.9638989169675091, + 0.9726027397260274, + 0.976027397260274, + 0.9814126394052045, + 0.9633699633699634, + 0.9647887323943662, + 0.9629629629629629, + 0.975, + 0.9726962457337884, + 0.958041958041958, + 0.9681978798586572, + 0.9698996655518395, + 0.9736842105263158, + 0.9678571428571429, + 0.9745454545454545, + 0.9701986754966887, + 0.9790940766550522, + 0.967741935483871, + 0.9644268774703557, + 0.9614035087719298, + 0.9632352941176471, + 0.9637096774193549, + 0.975177304964539, + 0.978021978021978, + 0.975177304964539, + 0.9840637450199203, + 0.9574468085106383, + 0.9615384615384616, + 0.9640287769784173, + 0.9507042253521126, + 0.9781021897810219, + 0.9752650176678446, + 0.9686274509803922, + 0.9759450171821306, + 0.9640287769784173, + 0.9624060150375939, + 0.9732824427480916, + 0.9773584905660377, + 0.9730769230769231, + 0.9678714859437751, + 0.9814126394052045, + 0.9734848484848485, + 0.9786476868327402, + 0.9624060150375939, + 0.9739776951672863, + 0.9820143884892086, + 0.9713261648745519, + 0.9608540925266904, + 0.9565217391304348, + 0.9512195121951219, + 0.9662921348314607, + 0.9692832764505119, + 0.9743589743589743, + 0.9602888086642599, + 0.9705882352941176, + 0.9642857142857143, + 0.9652509652509652, + 0.9581749049429658, + 0.9605734767025089, + 0.9744525547445255, + 0.9528985507246377, + 0.9728813559322034, + 0.967741935483871, + 0.956989247311828, + 0.9710144927536232, + 0.9586206896551724, + 0.9646643109540636, + 0.9568345323741008, + 0.9811320754716981, + 0.9688581314878892, + 0.9691780821917808, + 0.9619771863117871, + 0.9740740740740741, + 0.9680851063829787, + 0.9742647058823529, + 0.9694656488549618, + 0.9652777777777778, + 0.9525547445255474, + 0.967032967032967, + 0.975609756097561, + 0.9644128113879004, + 0.9689655172413794, + 0.9553903345724907, + 0.968421052631579, + 0.974910394265233, + 0.9584775086505191, + 0.9516728624535316, + 0.9726027397260274, + 0.956989247311828, + 0.9501915708812261, + 0.9708029197080292, + 0.9782608695652174, + 0.975, + 0.9855072463768116, + 0.9756944444444444, + 0.969811320754717, + 0.9777777777777777, + 0.9719298245614035, + 0.9822064056939501, + 0.9683098591549296, + 0.9775280898876404, + 0.9701986754966887, + 0.9605734767025089, + 0.96415770609319, + 0.9730769230769231, + 0.9611307420494699, + 0.9538461538461539, + 0.9604743083003953, + 0.9704797047970479, + 0.9719298245614035, + 0.9713261648745519, + 0.9633699633699634, + 0.9823321554770318, + 0.9591078066914498, + 1.0, + 0.9680851063829787, + 0.9678571428571429, + 0.975177304964539, + 0.9644268774703557, + 0.9602888086642599, + 0.9790209790209791, + 0.9681978798586572, + 0.9736842105263158, + 0.975609756097561, + 0.9580152671755725, + 0.9906542056074766, + 0.9721254355400697, + 0.9669117647058824, + 0.9755244755244755, + 0.9721254355400697, + 0.9628252788104089, + 0.9609929078014184, + 0.9615384615384616, + 0.9609929078014184, + 0.9553903345724907, + 0.9589552238805971, + 0.9700374531835206, + 0.9675090252707581, + 0.967032967032967, + 0.9615384615384616, + 0.9630996309963099, + 0.974910394265233, + 0.977859778597786, + 0.969811320754717, + 0.9724137931034482, + 0.967032967032967, + 0.9661654135338346, + 0.9662162162162162, + 0.9814814814814815, + 0.967391304347826, + 0.9597069597069597, + 0.9741100323624595, + 0.958041958041958, + 0.9625468164794008, + 0.9763779527559056, + 0.976027397260274, + 0.974910394265233, + 0.9714285714285714, + 0.9739776951672863, + 0.9782608695652174, + 0.9592592592592593, + 0.9728682170542635, + 0.9752650176678446, + 0.955719557195572, + 0.9715447154471545, + 0.9645390070921985, + 0.975, + 0.972027972027972, + 0.9635036496350365, + 0.975609756097561, + 0.9628252788104089, + 0.9595588235294118, + 0.9597069597069597, + 0.9612676056338029, + 0.9547325102880658, + 0.966542750929368, + 0.9748201438848921, + 0.9577464788732394, + 0.9678571428571429, + 0.9656357388316151, + 0.977859778597786, + 0.9752650176678446, + 0.955719557195572, + 0.953405017921147, + 0.960431654676259, + 0.954225352112676, + 0.9615384615384616, + 0.9806451612903225, + 0.9726962457337884, + 0.9658119658119658, + 0.9787234042553191, + 0.9595959595959596, + 0.9626865671641791, + 0.9597069597069597, + 0.9794520547945206, + 0.9727891156462585, + 0.9575971731448764, + 0.9761904761904762, + 0.9833333333333333, + 0.9780564263322884, + 0.9804560260586319, + 0.984, + 0.968421052631579, + 0.9515570934256056, + 0.9692307692307692, + 0.9692832764505119, + 0.975609756097561, + 0.9608540925266904, + 0.9646643109540636, + 0.9921875, + 0.967741935483871, + 0.9740740740740741, + 0.9923076923076923, + 0.9701492537313433, + 0.976, + 0.9525423728813559, + 0.9679715302491103, + 0.9831081081081081, + 0.9719298245614035, + 0.9925925925925926, + 0.9560439560439561, + 0.9578947368421052, + 0.9558823529411765, + 1.0, + 0.9537366548042705, + 1.0, + 0.9681978798586572, + 1.0, + 0.9588014981273408, + 0.9781818181818182, + 0.9924812030075187, + 0.96, + 0.967741935483871, + 0.9925925925925926, + 1.0, + 0.9758620689655172, + 0.9575971731448764, + 0.9769736842105263, + 0.9663299663299664, + 0.9830508474576272, + 0.9649805447470817, + 1.0, + 0.9680851063829787, + 0.9626865671641791, + 1.0, + 0.959409594095941, + 0.9545454545454546, + 0.966542750929368, + 0.9922480620155039, + 0.9710144927536232, + 0.9577464788732394, + 0.9747899159663865, + 0.9658703071672355, + 0.9698996655518395, + 0.978494623655914, + 0.9926470588235294, + 0.9838709677419355, + 0.9682539682539683, + 0.990990990990991, + 0.972318339100346, + 0.9908256880733946, + 0.981651376146789, + 0.9716312056737588, + 0.9795918367346939, + 0.9695945945945946, + 0.9844961240310077, + 0.9755244755244755, + 0.9701986754966887, + 0.9915966386554622, + 0.976, + 0.9653979238754326, + 0.9725085910652921, + 0.9672727272727273, + 0.966996699669967, + 0.968421052631579, + 0.9671532846715328, + 0.9632352941176471, + 0.9795918367346939, + 0.969811320754717, + 0.9584905660377359, + 0.9661016949152542, + 0.9680851063829787, + 0.9834710743801653, + 0.9756944444444444, + 0.9605734767025089, + 0.9693877551020408, + 0.9649122807017544, + 0.9615384615384616, + 0.9685314685314685, + 0.9554794520547946, + 0.9743589743589743, + 0.9765100671140939, + 0.9611307420494699, + 0.9629629629629629, + 0.9636363636363636, + 1.0, + 0.9714285714285714, + 0.975, + 0.9733333333333334, + 0.983739837398374, + 0.9550173010380623, + 0.9765886287625418, + 1.0, + 0.9672131147540983, + 0.9836065573770492, + 0.9833887043189369, + 0.9786476868327402, + 0.9763513513513513, + 0.9568627450980393, + 0.9828767123287672, + 0.9440298507462687, + 0.9505703422053232, + 1.0, + 0.991869918699187, + 0.9683098591549296, + 0.956989247311828, + 0.9922480620155039, + 0.9715302491103203, + 1.0, + 0.9658703071672355, + 0.9619377162629758, + 0.958041958041958, + 0.9705882352941176, + 0.9927007299270073, + 0.9921259842519685, + 0.9745762711864406, + 0.9906542056074766, + 0.9683098591549296, + 0.991304347826087, + 0.95, + 0.9730639730639731, + 0.991869918699187, + 0.9923076923076923, + 0.9713261648745519, + 0.9754385964912281, + 0.9607142857142857, + 1.0, + 0.96415770609319, + 0.9790940766550522, + 0.9556313993174061, + 1.0, + 0.974025974025974, + 0.983739837398374, + 0.9752066115702479, + 0.9672727272727273, + 0.9833333333333333, + 0.9622641509433962, + 0.981651376146789, + 0.984, + 0.9630996309963099, + 0.9619377162629758, + 0.9638157894736842, + 0.9685314685314685, + 0.9627118644067797, + 0.9915966386554622, + 0.9728813559322034 + ] + }, + "contact_maps": { + "mean_pearson": 0.00024654172980162595, + "valid_tracks": 28, + "per_track_pearson": [ + 0.0, + -0.00041133908693874525, + 0.0, + 0.0, + 0.0, + -0.0006530761943649154, + -0.0008612295677638007, + 0.0, + -0.0011767106024870215, + -0.0003292888625238682, + 0.0003740389834899768, + 0.0, + -0.003925497840554737, + 0.0, + -0.0027454493693246495, + -0.0029537944948895612, + 0.0, + 0.0010026072594496254, + -0.0007797837612126436, + 0.0012394280535259934, + 0.0, + -0.00025948498682809833, + 0.0, + 0.0, + 0.017711661509012253, + 0.0, + -0.0009462038744215351, + 0.0016172912702772514 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.2919377648437051, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..d8d054a6142c2228f8b596eab26d7cc4e654ff07 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S011/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5481057858533107, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6386912422626709, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.09454684980460568, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7158244949683333, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2691925616679429, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.21506119550585595, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.21154402317862098, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.44553385438981186, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.189017859760019, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2085184509961343, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.37683453604744793, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.14523491139951616, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.1636743345537799, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S012/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S012/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..e90dabb04dcf96583e3c3ee395f51d52799c3869 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S012/training_provenance.json @@ -0,0 +1,66 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 1, + "offset": 0, + "global_step": 41020, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/local01", + "config": "/workspace/research/experiments/local01/config.yaml", + "resume": "/workspace/research/experiments/local01/step-040000.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 82040, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 100, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 6738.755934401153, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/README.md new file mode 100644 index 0000000000000000000000000000000000000000..05b7a7334a733eed10e7a15caa331e2a3deb573d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/README.md @@ -0,0 +1,95 @@ +# SeqFunc local01 after one complete training epoch with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,256,456 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 4 dilated residual depthwise blocks (701-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The compact model trained for 2,000 examples with balanced splice loss, then a zero-output local path was attached. Training continued through all 41,020 eligible intervals. Optimizer, sampler and RNG states were restored after infrastructure interruptions; unsaved updates were discarded. + +This checkpoint records **41,020 updates**, 1 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup training examples. The objective is `balanced_splice`. +Other family losses follow the reference. Recorded accumulated training-loop wall time is +112.3 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 701-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +## Measured validation results + +Family-mean Pearson: **0.327017** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.500901 | +| dnase | 0.614570 | +| procap | 0.377495 | +| cage | 0.209317 | +| rna_seq | 0.273101 | +| chip_tf | 0.435083 | +| chip_histone | 0.421491 | +| splice_sites | 0.194347 | +| splice_usage | 0.248031 | +| junctions | 0.189779 | +| contact_maps | 0.133076 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5794 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6302 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.1320 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7400 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2965 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2200 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.1936 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3211 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1677 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2263 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5832 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1547 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1918 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..430778c1cbb3c9662fdda1bb7e43279a2c3a56a6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/local01-epoch1/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..c2bd8520fcc5e01977b76a1ea1a3e6463a91e1b5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/package_manifest.json @@ -0,0 +1,33 @@ +{ + "parameters": 13256456, + "source_checkpoint_sha256": "1d2e8ee5cf27aad6d8a4b9f4617c9fe4725a3ac42e9d9c13c43e8b295d208dd5", + "files": { + "README.md": "dfd121ccc4548da52b1e6825eab5c4d95fafba791ab5479ce27f59cf2228f02c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "8d4ccf2f3ac91e272d2de59872447d8ee6d2c1245114634549b1eae9d6c140a6", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "2871dc8893879c3e14a526dea659701d72c8c4d0e2f835b0bfb857034a2c2e6d", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "a6653431e850d448c414f48cc65cbaaec8c17a430c08376b281441a6f6bce987", + "training_config.yaml": "deab298b32f0aef33e66ddf538541ec470eee0c05a6ebc983684f2575287d791", + "training_provenance.json": "20fcec7edebf4d173f1ad7454477334d597d2f53c2ac3c294d29baa40e12a814", + "decoding_policy.json": "fa72d3cf028143d91ea15a9317253624d1717f1f762a8d4572cacb14d0a764cf", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "validation_tracks.json": "d7c6d370dc187427ab33beac193bfcf8a780867da36ad972e9a2943f65efb91b", + "validation_variants.json": "e3481edc5d3ce0cf0d28e83f0ae7e92afef0afa5be70ec118f73a10767038252" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..316ae8b495a102a194b3b1c5088d815e5d66e74b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/model.py @@ -0,0 +1,426 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + + def __post_init__(self) -> None: + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path; 701-bp receptive field and aligned 128-bp pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, d) for d in [1, 4, 16, 64])) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..45bdf927f635baa77ab98cc8fcefbb60196aa0ff --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/train_single.py @@ -0,0 +1,155 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f9a3298d69b138a02fed00880182d4004b23327 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_config.yaml @@ -0,0 +1,28 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..e90dabb04dcf96583e3c3ee395f51d52799c3869 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/training_provenance.json @@ -0,0 +1,66 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32 + }, + "epoch": 1, + "offset": 0, + "global_step": 41020, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/local01", + "config": "/workspace/research/experiments/local01/config.yaml", + "resume": "/workspace/research/experiments/local01/step-040000.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 82040, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 100, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 6738.755934401153, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..e1739ab8929d6738347417b7a8c2da3004a8d29c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5009007147028535, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6185203140767616, + 0.6337654747904041, + 0.38362857080914775, + 0.5790602015579601, + 0.3785107106968191, + 0.37329687865475725, + 0.3922836515403589, + 0.47342841060425916, + 0.5116515651217635, + 0.5698325213158345, + 0.495687329983454, + 0.4595066391936, + 0.3225648356885936, + 0.42936757138601406, + 0.6157349616896458, + 0.40888137920540646, + 0.4670245230622458, + 0.428230918496098, + 0.4666147049336215, + 0.4496039339739506, + 0.4249727768129564, + 0.38549676377292624, + 0.44427677258478765, + 0.4386540143388153, + 0.4277492993481239, + 0.4124615871381051, + 0.3981701613222936, + 0.41942435737204864, + 0.4279068908105328, + 0.4796324193140268, + 0.411346563099859, + 0.42645559499536073, + 0.40242453004697676, + 0.326966249486066, + 0.2942623081433306, + 0.4903505021316574, + 0.43070977693060314, + 0.5309240332011876, + 0.4276847491462437, + 0.4131505194166142, + 0.4905711465669827, + 0.4919794023272441, + 0.4145882506083114, + 0.39409127076442646, + 0.48497824656892274, + 0.46418103671862776, + 0.35125187833181265, + 0.4023138829917393, + 0.39426607571083916, + 0.44600306748070667, + 0.4181072548575499, + 0.37939498338431515, + 0.46819987175936184, + 0.532134955782346, + 0.3826331293328174, + 0.6350435571068851, + 0.7330201428625471, + 0.46720364479736076, + 0.5384147632500974, + 0.5447819840159804, + 0.6681813686745998, + 0.4066859785510226, + 0.6406741167698511, + 0.5398465122772902, + 0.5712527795590573, + 0.42813850463932507, + 0.47781668408811295, + 0.5644341704178538, + 0.7131110342206542, + 0.6440275116750523, + 0.4601386182672312, + 0.4081709504926177, + 0.37133126765140767, + 0.45043298156075134, + 0.4229759658694227, + 0.41189877729208607, + 0.42658960565001686, + 0.46377678370564945, + 0.5029583317653792, + 0.4183053726538578, + 0.4885456514897415, + 0.5184229131096885, + 0.43669325970208245, + 0.4089183182950214, + 0.4348301102232859, + 0.4260046449413557, + 0.40795525549058165, + 0.40114672117066585, + 0.37610296742338434, + 0.43607189584017725, + 0.4255551519039607, + 0.42933635555843985, + 0.4230339070472779, + 0.43437514517960424, + 0.49838891513846634, + 0.3201671907844893, + 0.36246709660383303, + 0.3566069900123055, + 0.40204909386988735, + 0.4208088047463121, + 0.3856283430958672, + 0.3820846035084027, + 0.4188660703239773, + 0.3944329175062183, + 0.4595299181472506, + 0.4679884332020519, + 0.4570831751077103, + 0.43904814220594174, + 0.45845957100449347, + 0.4534386023201002, + 0.508500269425661, + 0.43796718275621954, + 0.4409712657404247, + 0.3644165134230115, + 0.38730149421799565, + 0.3661353366703198, + 0.3993008420109241, + 0.4129958516607055, + 0.41478582097321237, + 0.39779294571180024, + 0.598444663480199, + 0.591847505524127, + 0.6626617828282655, + 0.6630949654624323, + 0.6343434990067971, + 0.524506420497775, + 0.7986180101646914, + 0.6380347498693962, + 0.6935673625116536, + 0.6224642910038011, + 0.6701954766730832, + 0.5024040599806031, + 0.533999227149344, + 0.5070050343522893, + 0.6500439992931321, + 0.6433189489190148, + 0.6312343617189109, + 0.4690012144136973, + 0.37976597548190244, + 0.7477197485217772, + 0.7136538199823902, + 0.7302716539313079, + 0.7238609179887026, + 0.5350722100319478, + 0.8007507389426597, + 0.7093729558981811, + 0.6494785820817044, + 0.6561492817147565, + 0.7859640370710067, + 0.7409868630310668, + 0.5621229362634808, + 0.5598292182162703, + 0.648123128953559, + 0.6805314800972543, + 0.6840872516077652, + 0.5491471866379615, + 0.6690202903340311, + 0.5274594121134083, + 0.5458513397498495, + 0.6267710490505458, + 0.5090691519406209, + 0.7034986951702896, + 0.7187721762189506, + 0.568267619891077, + 0.5480431822415454, + 0.5171107999173885, + 0.6765535647097469 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.614569806897389, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5187156193118878, + 0.9111514248817341, + 0.7690954443855497, + 0.4763726579907475, + 0.4335121158195192, + 0.45533142053979325, + 0.9329700770724244, + 0.6975304846056711, + 0.38663896880964915, + 0.31118834595289446, + 0.8683904064593433, + 0.6541784323561416, + 0.4014585754566876, + 0.47229084563580276, + 0.45142164637215865, + 0.7923062542893331, + 0.7726466301550924, + 0.8435209296592983, + 0.8457112698294584, + 0.8795178987625905, + 0.6469955289875046, + 0.5825145767030818, + 0.8938363400675007, + 0.4678868269721175, + 0.48808509829017943, + 0.7681243521772354, + 0.4190973139046834, + 0.6308973335640276, + 0.46074388691043777, + 0.48305707942809034, + 0.5900614845579393, + 0.4207309735340692, + 0.8322723581867916, + 0.6143646137283473, + 0.8741942548994203, + 0.8802119231984766, + 0.4043452256364257, + 0.8172708900097914, + 0.8704015495211618, + 0.8517335685588674, + 0.8164221837004069, + 0.660499089733282, + 0.6965835385629184, + 0.6118993444981408, + 0.4451755520435361, + 0.6245482639267654, + 0.7228016667113297, + 0.4909236164011321, + 0.5627640171872191, + 0.3758365156732779, + 0.33371415135419763, + 0.5306279806951502, + 0.4599393634025604, + 0.82274234117761, + 0.3969795006577824, + 0.3772476404221636, + 0.40883400488988847, + 0.44785311457728627, + 0.4108299232370901, + 0.5031635473674161, + 0.390466421190698, + 0.593859408106014, + 0.4455379190228847, + 0.43230413397432405, + 0.45966546851504436, + 0.464277589554854, + 0.5154891155581193, + 0.4328041015498164, + 0.3436989938203522, + 0.38616893068559605, + 0.4908677262527932, + 0.6007274485104122, + 0.4446307322889331, + 0.5910461806404111, + 0.3083982103578502, + 0.8004274347351252, + 0.8884284211545559, + 0.6255582120871575, + 0.40534275236878664, + 0.41941248761048633, + 0.4615828424060205, + 0.455995793508925, + 0.4577691037767644, + 0.4338112275922422, + 0.5083525753791434, + 0.4684105770261409, + 0.560183339063397, + 0.4742238917498768, + 0.6124675193506727, + 0.41655512089952645, + 0.5983350277195284, + 0.38872248472109416, + 0.7089531306933472, + 0.43515911223721804, + 0.6939899151467941, + 0.47010876805403007, + 0.7165834871807815, + 0.6762550180264422, + 0.3152925078375149, + 0.32275479877330987, + 0.43557271210745563, + 0.34745308728422236, + 0.32623660082537853, + 0.6909463913629531, + 0.6601968143185306, + 0.3972140483056664, + 0.4195139013749943, + 0.5665254978709031, + 0.8361116337454944, + 0.5391918982435201, + 0.49449884763518953, + 0.5548052853560648, + 0.4631250294672824, + 0.4133194095477679, + 0.7627736307395326, + 0.5149485573881872, + 0.46534165048480775, + 0.5295986234153918, + 0.599760268589645, + 0.42572812252885944, + 0.4955149601149836, + 0.48354134451163633, + 0.5197027596773817, + 0.4932867746610949, + 0.5852090846053761, + 0.5169737226818627, + 0.5717439174070473, + 0.6636696034424452, + 0.6807633400512425, + 0.5763901618401573, + 0.4414929647431942, + 0.5780711427217595, + 0.6080627912150651, + 0.7806199176414185, + 0.41633528015173654, + 0.2945307076480459, + 0.5607524348452663, + 0.4511306505912628, + 0.6094994922632112, + 0.5800613453522523, + 0.5868479322731336, + 0.5386479455502732, + 0.43721813582543917, + 0.7656224250222331, + 0.5771643611269202, + 0.7503269067156437, + 0.3701608525049219, + 0.3544216350134514, + 0.3791732701872642, + 0.3175498410804037, + 0.8867805987819045, + 0.29968518076535905, + 0.8884605322769825, + 0.7028898477106382, + 0.4553099133470342, + 0.31338271087608005, + 0.3265890569990306, + 0.3482631587760273, + 0.6194266211307717, + 0.700342201702473, + 0.7054796123469913, + 0.633544518224074, + 0.5325608377684913, + 0.43649090061557777, + 0.6638703682589863, + 0.5751861639225231, + 0.4957720539639777, + 0.6069360684764501, + 0.43642352015120073, + 0.4549694609003162, + 0.36088772531617885, + 0.821617269245388, + 0.42405172167489574, + 0.4450435674847629, + 0.6316465398421235, + 0.6766231300467934, + 0.451881178325681, + 0.32387507362677226, + 0.870893157536887, + 0.3486669139081549, + 0.7475069651919928, + 0.41680866997736277, + 0.572927610366444, + 0.7043189495431468, + 0.44379736404765285, + 0.5383486883364572, + 0.5681780107582346, + 0.8094641582280551, + 0.40165449984573187, + 0.5863044573179143, + 0.522992307475356, + 0.6982802016870525, + 0.8913460007767485, + 0.38287807194040785, + 0.5615414307043888, + 0.4344430499465991, + 0.5951138352902602, + 0.6905542520707391, + 0.5517154837689919, + 0.41966652652833786, + 0.8627418805307229, + 0.8735075569007079, + 0.7478640585858062, + 0.828299183943691, + 0.8381389654228947, + 0.8269968185605356, + 0.8354894485275808, + 0.3379165804848848, + 0.350466616926393, + 0.4856712319368468, + 0.43571046463900565, + 0.41985284272876106, + 0.57034629098839, + 0.8552051327021506, + 0.841929275525015, + 0.64593456210546, + 0.8175787518144675, + 0.8153143189756238, + 0.8511697036206217, + 0.8988370260564559, + 0.45885982245669843, + 0.5151132811928583, + 0.6189006152504125, + 0.5323734895431684, + 0.8097262318849829, + 0.5299277786157714, + 0.9000789872941917, + 0.8738908174858419, + 0.8666108337862363, + 0.8946768510310983, + 0.8866484033684326, + 0.8623557013249368, + 0.821511348300383, + 0.8527569379820015, + 0.8985588046636002, + 0.3996521487247545, + 0.9167564259261582, + 0.8716727121423898, + 0.8131366698619117, + 0.5572324285768252, + 0.8657974483635115, + 0.4750335451946197, + 0.7446172298363837, + 0.8682326745903636, + 0.5361874505857839, + 0.5220372456690553, + 0.6555606358881001, + 0.4958726081636267, + 0.4909352104293351, + 0.7147541814627069, + 0.8750653132164388, + 0.6990757320506904, + 0.7983320380640762, + 0.5765986962832338, + 0.8564492995480095, + 0.7977562073960152, + 0.8288094759351249, + 0.8963840656499905, + 0.8262335317575816, + 0.8914178251790601, + 0.8625576502649779, + 0.5009435203321769, + 0.8707781247193529, + 0.5077680946933136, + 0.4861221426444926, + 0.8763149015369093, + 0.8986654303614249, + 0.8729721592568651, + 0.5323284929519712, + 0.5744471415532542, + 0.6375626714001863, + 0.7901520305610836, + 0.8812570744485712, + 0.6793518386632479, + 0.8976022340780608, + 0.9085362587402408, + 0.5738630052189109, + 0.2861395332800261, + 0.48334252061699734, + 0.8926306048494218, + 0.38646839605993244, + 0.5672335732960089, + 0.8726230524153102, + 0.8230508878893802, + 0.5718650103830293, + 0.8591967151339824, + 0.8877640149973224, + 0.8820735640299674, + 0.8701909863951219, + 0.8583217736931439, + 0.8889238689361423, + 0.8457288063393523, + 0.9200607981504676, + 0.7760704987828484, + 0.8250215534013173, + 0.9037014290613646, + 0.5093852110785356, + 0.5174402703374593, + 0.4948038821090652, + 0.48216873731638427, + 0.9016139727519286, + 0.7884299998092578, + 0.8415364613495613, + 0.7820383662496582, + 0.7526666655228288 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.37749507226017287, + "valid_tracks": 12, + "per_track_pearson": [ + 0.368043047241309, + 0.4067325418327992, + 0.31989370436046377, + 0.40774756591953737, + 0.4066971324248547, + 0.4061572041184215, + 0.3482098344321339, + 0.39549177656564805, + 0.2944942594720119, + 0.39099423103967995, + 0.3838901543935038, + 0.40158941532171166 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.2093173750006707, + "valid_tracks": 546, + "per_track_pearson": [ + 0.2153508707320226, + 0.24324370173538606, + 0.20607373356848524, + 0.13120968885674578, + 0.08015658335934678, + 0.21224413179845536, + 0.11193686528467649, + 0.14204712662714913, + 0.23137892301014049, + 0.20032242012063559, + 0.24791644746479036, + 0.13996492617582146, + 0.14578188893405883, + 0.12007729676376586, + 0.09961159723735286, + 0.1761987152832979, + 0.20025156162732177, + 0.24472874350787732, + 0.25088697987290876, + 0.2294257925324308, + 0.19873259545108518, + 0.1925404921561209, + 0.30324026311883334, + 0.17285486289596275, + 0.1698605823945869, + 0.22152787922545797, + 0.23492500515975512, + 0.23855894846459838, + 0.2569371361599748, + 0.24045601493913044, + 0.151262321677122, + 0.2533870122040615, + 0.18838474290714197, + 0.1323947133529073, + 0.2659326581568982, + 0.24730573466155964, + 0.23569983935385727, + 0.26405861397417535, + 0.22835373222338906, + 0.2927417200570085, + 0.3077881932018254, + 0.2636605282903816, + 0.23105924170643416, + 0.22849526236074333, + 0.12329932152601121, + 0.18419905174773502, + 0.20525150830710823, + 0.21245949759496496, + 0.12430447899962313, + 0.162726594265124, + 0.22957367694236447, + 0.25978058297077544, + 0.27234843961950655, + 0.18733699817904856, + 0.2841216423679478, + 0.3079023286284949, + 0.17071135053244607, + 0.2862184634617729, + 0.22185732628879615, + 0.25067877039200337, + 0.28775950850773213, + 0.1806363787017436, + 0.2519788813968875, + 0.19605085151274637, + 0.2690408789337124, + 0.2809446684614176, + 0.2983969949201408, + 0.17518804104167823, + 0.12434618239449809, + 0.11544991331431355, + 0.22993183116497085, + 0.26579305971845774, + 0.2257777329882584, + 0.23944678786599927, + 0.25645050117383583, + 0.21797989363657835, + 0.26392834964942447, + 0.23678641559014482, + 0.2627248919676388, + 0.2677853397982529, + 0.2231995445515429, + 0.13766639634357491, + 0.2398352106514738, + 0.27795816171386617, + 0.1371783412181595, + 0.25136191216077647, + 0.21091433161244666, + 0.28101135491317836, + 0.17588061181372766, + 0.2598733286484965, + 0.19054055018678273, + 0.2230947565596286, + 0.24208672919507696, + 0.11892924399388054, + 0.2569306129315669, + 0.22387399475742165, + 0.23730421931191703, + 0.24662442301447962, + 0.2292632669301823, + 0.14237525532381295, + 0.2206324208146421, + 0.2405702647205985, + 0.2443098518285668, + 0.3219710672124907, + 0.2796314841270067, + 0.28359262100707905, + 0.19083566266740565, + 0.24761526805351258, + 0.18770381228048677, + 0.20190021369549546, + 0.2164345801453986, + 0.21460288721011464, + 0.24058854747816857, + 0.236172354357439, + 0.19554317144599093, + 0.2393940581581662, + 0.2423622290637665, + 0.2498240027947806, + 0.2318186188523194, + 0.27156432404566266, + 0.27222708310351107, + 0.26771131077435123, + 0.1613618995625646, + 0.18680795318307275, + 0.20908970505360894, + 0.26018802381542383, + 0.23368520159126324, + 0.25469149248412803, + 0.3282466790852511, + 0.24779936168368377, + 0.2678085630324063, + 0.18838684884010368, + 0.2577926501308011, + 0.2685740742602281, + 0.30624329248305, + 0.22795621512981684, + 0.2769677037609238, + 0.1908594111129439, + 0.12112393853086378, + 0.24283338947630406, + 0.28998670674821203, + 0.2386397421043198, + 0.25473408779969114, + 0.20626814832895782, + 0.21811471087319162, + 0.22861390828468733, + 0.24498737663628536, + 0.2808808421178827, + 0.20867027160113832, + 0.2020037652837468, + 0.2339371518028019, + 0.1832769983644036, + 0.22691541670352558, + 0.24852351161845512, + 0.22333718931152374, + 0.16421176936995494, + 0.25394434311616665, + 0.2846507319047923, + 0.1907172659545968, + 0.16469307434262453, + 0.15425155049397027, + 0.21437780208007973, + 0.2068550163919851, + 0.23653027086658016, + 0.20476310479314583, + 0.16827251361368448, + 0.2291127200051695, + 0.16500010184603583, + 0.16897961642281384, + 0.2597932358074371, + 0.22887790740830255, + 0.18983957738661825, + 0.2101032653019255, + 0.3132860695291105, + 0.23260319711878785, + 0.16516428418112444, + 0.15252282089712604, + 0.1959749958396872, + 0.2451931761368282, + 0.2868077364019107, + 0.2723241438209452, + 0.27464180818667167, + 0.13469279790791425, + 0.2704809777892678, + 0.223286231384617, + 0.18747200533160124, + 0.1840117369486344, + 0.16030927926432492, + 0.2915355739712637, + 0.24381152914546933, + 0.19970880033124025, + 0.13338557915058372, + 0.12915951069801845, + 0.21223914662915483, + 0.21363928138418883, + 0.1940963320696552, + 0.22416377966683954, + 0.22122578153797579, + 0.17269614507624587, + 0.2131678820514882, + 0.12288974240922024, + 0.10145027246510988, + 0.12676056926802406, + 0.11952885721730679, + 0.1312196849142798, + 0.21797164715182915, + 0.21453925560554246, + 0.18376362295978008, + 0.12160429246508739, + 0.2059099097739504, + 0.2759619449940389, + 0.23396294575599497, + 0.24387954827466185, + 0.24154399250901462, + 0.22427606119211496, + 0.21501109158112058, + 0.26301672692517475, + 0.29397971073428575, + 0.2617639209679355, + 0.2548983359320512, + 0.23081897427396775, + 0.1714567300901476, + 0.2367703117986558, + 0.23329875458654595, + 0.246580249167965, + 0.30060974763662673, + 0.22659617918226052, + 0.1997108800092499, + 0.22072215116803834, + 0.23555259400845222, + 0.17414225520731777, + 0.169990329042481, + 0.28094699750369806, + 0.2180562823892302, + 0.19078820936596588, + 0.22342247395582063, + 0.29237539278092173, + 0.15869901258463148, + 0.23437115713258957, + 0.23388691118525393, + 0.12394930972794323, + 0.15499300870711782, + 0.15155767636673517, + 0.22975868695579255, + 0.17267647329563784, + 0.21187387005112296, + 0.09332131952882343, + 0.28151123589399063, + 0.2434708563006326, + 0.25217053265726386, + 0.25937684686442314, + 0.16323144903090483, + 0.29086705191586254, + 0.13880367269004773, + 0.26167019539443925, + 0.23447434619875185, + 0.33108074348638383, + 0.2473378976309854, + 0.25357619979908325, + 0.356357885250097, + 0.20919639422255432, + 0.23339784447549047, + 0.13606210826492143, + 0.24872238438333974, + 0.2365347449002395, + 0.1669983422881385, + 0.1411981818377776, + 0.144219699645479, + 0.16979194473721493, + 0.24185214614893302, + 0.29374553058822556, + 0.1615218160923637, + 0.13619349658137192, + 0.18228656215752803, + 0.2500417448782664, + 0.19569302024925095, + 0.13709083416373144, + 0.06708039271106166, + 0.19131639998398714, + 0.12882610958230148, + 0.1366444596234927, + 0.21150644909554872, + 0.18216979946888887, + 0.2187837876228678, + 0.13274823048432932, + 0.14501656326601622, + 0.1236416503187638, + 0.10731990181305971, + 0.1571715127998508, + 0.189025613088348, + 0.22223958156030427, + 0.23212722378568093, + 0.2098302999168057, + 0.18653072729307513, + 0.18335262626209414, + 0.2777361817911097, + 0.15766233191175774, + 0.15561799368471096, + 0.2172108211490102, + 0.20844705008696418, + 0.21375061376933455, + 0.23683841504973846, + 0.22331121647861518, + 0.13996479960713432, + 0.22316747138050544, + 0.17463455512003875, + 0.1205540659090938, + 0.2370460030689666, + 0.23302470056029545, + 0.23663153842191328, + 0.23787141677935217, + 0.22240360669109016, + 0.2679964319494205, + 0.27534163309540877, + 0.23290306668841718, + 0.2127857327455174, + 0.2117206351573226, + 0.1375738403495592, + 0.1751376532168934, + 0.19536520943713104, + 0.22142287893558507, + 0.11579773973926412, + 0.15697040642695542, + 0.19975247111525563, + 0.2429740035326666, + 0.24901510273298919, + 0.17663331573451768, + 0.25941214315164246, + 0.2748495710753791, + 0.1690115341267018, + 0.25688403585068886, + 0.19364175337860096, + 0.22931060152341654, + 0.25492967043737796, + 0.16826912193445387, + 0.23182662041735433, + 0.18099932559308252, + 0.2432965330514076, + 0.248371606590391, + 0.26194288563692353, + 0.15929382596174185, + 0.11639531349446024, + 0.1059224356065182, + 0.2059963959956915, + 0.24185645745804518, + 0.20952219490883164, + 0.2154107573363346, + 0.24186656555170996, + 0.20515946350380196, + 0.22363927021987787, + 0.2101424838264751, + 0.2329665661720229, + 0.2509454831678638, + 0.20532125792467737, + 0.12837523821106414, + 0.22548380129860457, + 0.24939051074614638, + 0.13285852633046563, + 0.2216317322474477, + 0.22460913307748695, + 0.26325025487021014, + 0.15462147379284105, + 0.2469885179121773, + 0.17894816375432285, + 0.19904562908960455, + 0.2239212317730446, + 0.11342461865520236, + 0.2272823985281185, + 0.19823701018895074, + 0.20189122968446033, + 0.2400942874203789, + 0.20602329838558064, + 0.1304478664454426, + 0.20325582249454027, + 0.22917766035589, + 0.21786782285639708, + 0.2917195906812895, + 0.24720028817589787, + 0.2660125431601925, + 0.17198501566039037, + 0.2212098271916704, + 0.1772364123460424, + 0.178077025618971, + 0.1946387524326607, + 0.1937191434641789, + 0.21851876850349874, + 0.22198637983597197, + 0.17995234928044682, + 0.2148247048984009, + 0.21414399725534491, + 0.22561285224615477, + 0.20606139913581778, + 0.2509693458356788, + 0.23808628980437468, + 0.2392387044259886, + 0.15021353592016853, + 0.1738087170114447, + 0.20462900725304764, + 0.2291305605930536, + 0.2075073453031028, + 0.23963677530430622, + 0.30451641600004054, + 0.2324550984573309, + 0.24333253184964973, + 0.17405856820258866, + 0.22110527135929461, + 0.22579366656427635, + 0.27310740338571043, + 0.21340832975629287, + 0.2580827696387414, + 0.18297919605955354, + 0.11257545085043162, + 0.22266646190867875, + 0.2538492535592365, + 0.20738023856228127, + 0.24767184479275708, + 0.18538960484022068, + 0.18260431961751936, + 0.19440658476242031, + 0.2156267734744363, + 0.24671405472411317, + 0.20408755271234955, + 0.18063091728255457, + 0.20764477119536798, + 0.1637083686617534, + 0.19801463349194023, + 0.208476400268076, + 0.19925668251203044, + 0.15171716858921983, + 0.22359134010293066, + 0.2537540442651802, + 0.1647452900265692, + 0.15192147933845035, + 0.13776452765803332, + 0.17934303926972398, + 0.18669136280438492, + 0.22608590625315217, + 0.1980146263492194, + 0.14867780054052276, + 0.20998728032078953, + 0.15648767053900753, + 0.15826080957430327, + 0.24891962114773403, + 0.2099489124496943, + 0.18131395025592706, + 0.1797614993133287, + 0.2929068313721421, + 0.21903204016483005, + 0.14525810144922213, + 0.14382760427412566, + 0.19970986329122067, + 0.19497952021937065, + 0.27665475337824447, + 0.25022916718188215, + 0.262768664482563, + 0.11889821525484685, + 0.24804022645049797, + 0.21657305478250957, + 0.1884923573537251, + 0.17319470398623765, + 0.15225500016313528, + 0.2503993959881756, + 0.21912023093897248, + 0.19229079148253753, + 0.14438808232242994, + 0.10257572419980558, + 0.15152949662161563, + 0.21454019703851032, + 0.1758924882929682, + 0.21900874796909273, + 0.20628587509002017, + 0.16316643134934863, + 0.18183878043609364, + 0.0950766436141179, + 0.07728239697269992, + 0.09780742568809266, + 0.10051465961963867, + 0.08653183002933235, + 0.20063303015644127, + 0.1625797197654678, + 0.14614925542116625, + 0.1131644244541903, + 0.19203342286590125, + 0.26876187933445134, + 0.24192155862030282, + 0.24382719516179635, + 0.24541766242697138, + 0.2280785440233486, + 0.21574545405591158, + 0.257234271311859, + 0.2821859490439964, + 0.2559496678760198, + 0.248902361480166, + 0.22937192324440211, + 0.15972342635520287, + 0.23554439543752442, + 0.20539995449847875, + 0.24785123194433198, + 0.28480152564934363, + 0.21121843262384724, + 0.1947582953243507, + 0.1952281345768265, + 0.21465536304084032, + 0.10988304873926091, + 0.10876989958130183, + 0.2633957469254327, + 0.20593160392438997, + 0.18208984430222852, + 0.20574512654634813, + 0.271299061354116, + 0.16081390409340562, + 0.2181082290806332, + 0.20801447102251416, + 0.11229400323858586, + 0.12283594134198321, + 0.1329619215486207, + 0.22321459361985233, + 0.14783373095932772, + 0.2039140145138559, + 0.08613487774028623, + 0.2688925260321863, + 0.23613753438650864, + 0.23471092219293396, + 0.24935448443813274, + 0.1735624344533932, + 0.26442738012624134, + 0.1251782070646701, + 0.22483518615734982, + 0.22472402597921332, + 0.30871493825849405, + 0.23993199172107696, + 0.2502789929681512, + 0.3357476081143604, + 0.20582108742734925, + 0.20966710753026688, + 0.12071675367308096, + 0.22124919103427493, + 0.22006860653462637, + 0.1683725249722256, + 0.12583832857598176, + 0.12583145704139514, + 0.1552918158573168, + 0.23576334918830033, + 0.27972735093594525, + 0.143907483801337, + 0.11158012288953485 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.27310141874746763, + "valid_tracks": 613, + "per_track_pearson": [ + 0.27933594314849836, + 0.28068104747497263, + 0.266132153977571, + 0.32542202702734296, + 0.2865527691348123, + 0.21241592482845606, + 0.2563791925212267, + 0.29290191336745613, + 0.2960289021789088, + 0.325057854169912, + 0.2323055077514305, + 0.26972501318437847, + 0.1967396462947911, + 0.33429034690549353, + 0.3135917871259616, + 0.32464429236108033, + 0.3242586660823164, + 0.2787037929686758, + 0.2738932526351682, + 0.27732740613301377, + 0.24726604833573054, + 0.28841671038251265, + 0.27725384218380866, + 0.26048733581879796, + 0.2085761089436977, + 0.2709140501762133, + 0.2248195245163052, + 0.29501376833345133, + 0.29087681796625176, + 0.30540003079155276, + 0.3141423477943922, + 0.267986336827743, + 0.3250926207409559, + 0.2546429619736808, + 0.3455273230618141, + 0.22002186440055957, + 0.3117425713927016, + 0.30987213613716735, + 0.2710990987670962, + 0.3118242154157619, + 0.32081490364391924, + 0.2621455419964126, + 0.2888579922611773, + 0.28919358209463997, + 0.2919591134515044, + 0.2157980870180971, + 0.26658173726215195, + 0.3370726670153321, + 0.2901445028514665, + 0.27007881394637584, + 0.28617718316356716, + 0.2963171684215778, + 0.25617524455564294, + 0.2873376483404127, + 0.2130922431019724, + 0.3036051192163354, + 0.2805919498358704, + 0.2974629519022369, + 0.3002303024874443, + 0.2842899370418132, + 0.26318661852912023, + 0.2788782184474174, + 0.2912639431324003, + 0.2318638262436198, + 0.29702050223574905, + 0.274119423545872, + 0.2973369303421721, + 0.29910237450196886, + 0.2854927954509163, + 0.29220663713976, + 0.2908942923920635, + 0.2942465315973003, + 0.3028348544626575, + 0.28493959195404167, + 0.2928306080123949, + 0.3080386939391952, + 0.3028591182965436, + 0.3050855153422108, + 0.321435886812077, + 0.30180626340204053, + 0.2770964190973576, + 0.2365776455570266, + 0.33776300880470445, + 0.3150721442074429, + 0.23505499311548422, + 0.26498844328216475, + 0.30100397708075693, + 0.2923127663313431, + 0.2942587929230562, + 0.294007646638181, + 0.24766177277109153, + 0.28368278467363905, + 0.28848785580275227, + 0.28888475027647686, + 0.2956515101736975, + 0.29652228972667605, + 0.29219320199673304, + 0.2955551872695898, + 0.27574081799804195, + 0.24446728297361495, + 0.23796054526001628, + 0.26933893381457363, + 0.2637422540182689, + 0.2837274181718942, + 0.2739292617145314, + 0.2986473326691886, + 0.2868822613196705, + 0.3320510498372217, + 0.3266874505606626, + 0.2909222731022294, + 0.2605932679900149, + 0.36440240166012555, + 0.3318429986109676, + 0.23174587448489, + 0.29062178510173753, + 0.31688874050687843, + 0.3568779065485947, + 0.29562707320191, + 0.29095948463797505, + 0.2931662151914178, + 0.25112240473449865, + 0.27971815027964925, + 0.26975573047809176, + 0.3018558931755098, + 0.33713333083424163, + 0.2658404193393366, + 0.28171194873972966, + 0.21722390423202784, + 0.2536759538188402, + 0.26069686219282057, + 0.36299173354331254, + 0.28855491112633447, + 0.26471611403104595, + 0.22110736998352848, + 0.2461145401878086, + 0.2999333994292572, + 0.32928393659672384, + 0.3531172916764393, + 0.313791072435205, + 0.3290772253364341, + 0.28002166946283547, + 0.2621128111704733, + 0.34636238459330826, + 0.3364797102909574, + 0.2432009865880928, + 0.2535220493486307, + 0.2315127547129781, + 0.2685967987263142, + 0.23166327981668144, + 0.2861490320383666, + 0.23789395938324406, + 0.2895942291828706, + 0.25979196287998946, + 0.23679325892520525, + 0.29601148198113886, + 0.24067262299369033, + 0.28219937769325193, + 0.30552204020435136, + 0.2668262130494939, + 0.3172772590080216, + 0.351598974318029, + 0.2961288233870978, + 0.3055023143674507, + 0.27960878725867444, + 0.27755492011492633, + 0.27472670133958826, + 0.28113994069252685, + 0.2513419872294393, + 0.28875009880861535, + 0.30220451677616983, + 0.24876310107769403, + 0.2581907630255938, + 0.2934248956868272, + 0.29464757486821125, + 0.25697004252906547, + 0.2541458253231618, + 0.2765687725892761, + 0.3146159614468334, + 0.3015161677004228, + 0.2316343019527999, + 0.28030025583216084, + 0.2666962476198101, + 0.2501234254722613, + 0.24387597684149218, + 0.23587341957545122, + 0.3100435176277411, + 0.2682081110639192, + 0.1731536549277742, + 0.23713602822470756, + 0.3514887079505023, + 0.4024773724264403, + 0.25114885362042333, + 0.22292179688020614, + 0.2886935182684332, + 0.23905760969038808, + 0.2538185191961858, + 0.2585396607563904, + 0.23977720542692593, + 0.2358338154443631, + 0.28229141880949693, + 0.3279930932159705, + 0.2309689889077477, + 0.41583454221800537, + 0.31003476870049607, + 0.25624055555410025, + 0.26531249040324895, + 0.26156009388518064, + 0.1897138465698638, + 0.269175967737516, + 0.23102128207639291, + 0.20264942817058804, + 0.20800044554361308, + 0.20634697850217695, + 0.22462704167391448, + 0.2390526458228599, + 0.2886139201186846, + 0.20558955444276616, + 0.23958039621621194, + 0.21949673772965134, + 0.24700943078802384, + 0.40576270406026493, + 0.2617943174130231, + 0.34820672589278845, + 0.3017747817218924, + 0.3027230269786071, + 0.3656880673246346, + 0.305245971566554, + 0.3532452540010199, + 0.23972796811855385, + 0.2506788960040265, + 0.2744308597098936, + 0.3891743007018443, + 0.2787350472978753, + 0.2849655127851939, + 0.2766754527193985, + 0.26613534681162365, + 0.1931315652380905, + 0.3001176716695451, + 0.30552511156179224, + 0.30362243393563226, + 0.31257432038107225, + 0.2584180966536149, + 0.2722980967914611, + 0.23477400804681614, + 0.20565215793481967, + 0.2917766028006814, + 0.3614832322203724, + 0.27873274596292313, + 0.27027067418671613, + 0.2572519411775637, + 0.25450355211370534, + 0.2750629423664214, + 0.23292891543856253, + 0.2510018042064556, + 0.28974977332198376, + 0.3087966240372904, + 0.24392134231189866, + 0.28037422667324524, + 0.2776769203151616, + 0.25440625819934265, + 0.3000864389786216, + 0.27870159503449277, + 0.30107245779756003, + 0.3287408506686811, + 0.28948326036448113, + 0.2569637432314988, + 0.270012096295431, + 0.23544646542676304, + 0.26902211090243333, + 0.28096610129985206, + 0.252440709103849, + 0.20699330504833882, + 0.28897284361374026, + 0.26584433867912416, + 0.3001222011664815, + 0.26865935255353635, + 0.17085388216018774, + 0.18918809704290268, + 0.30830866924012457, + 0.3055827718958495, + 0.29991678404446276, + 0.24716016407929783, + 0.23345061572458722, + 0.21271238362701628, + 0.26486878203611897, + 0.24171961725015922, + 0.24594343863092294, + 0.26711092936245456, + 0.26224203274171176, + 0.3043414096067379, + 0.23880518860590666, + 0.2096649565969395, + 0.2913631945882917, + 0.21849438634831644, + 0.1987001033374211, + 0.17157783122750206, + 0.2646271708137769, + 0.27457638702325815, + 0.3036134635019119, + 0.2596960378024739, + 0.2625365602384793, + 0.2616411459907672, + 0.2516902168113703, + 0.2073161287138005, + 0.2525823460183751, + 0.31552990517670704, + 0.23755930050661242, + 0.29291904208063757, + 0.2867534799905598, + 0.2705907055634679, + 0.2753030975704254, + 0.28431174769717243, + 0.2730943739232387, + 0.2907775144485131, + 0.2797049917001563, + 0.22961055715263742, + 0.19809138839868945, + 0.23819781403111787, + 0.2792429701555932, + 0.23789041895484422, + 0.2644523910954841, + 0.27332899756765455, + 0.26844027220431965, + 0.19501771410827834, + 0.23103660051061065, + 0.166270358659533, + 0.2614568162238258, + 0.24728718240436343, + 0.2775188719536394, + 0.2772242198064249, + 0.26099613353175166, + 0.26921679799318515, + 0.28926904867993686, + 0.2896973483090032, + 0.1915992265920916, + 0.29226923765982304, + 0.2888243734792566, + 0.30369063124886986, + 0.30827232148255335, + 0.2931283864524042, + 0.2970950888767802, + 0.27021990721986144, + 0.31263176264764164, + 0.2903555030717389, + 0.30179905516980216, + 0.2867965393493198, + 0.3019470334760678, + 0.2990425315827133, + 0.303476781097349, + 0.2563434048662313, + 0.2725926345216093, + 0.25311232695283115, + 0.220094911569721, + 0.31467353990788116, + 0.31584296082831653, + 0.24524594499917932, + 0.27154310426777, + 0.258134571245119, + 0.21995263073329813, + 0.25774863847974033, + 0.22994704984988476, + 0.219815582653535, + 0.2732368867195204, + 0.2675150118463206, + 0.2503984566040138, + 0.24656499674424104, + 0.27445244611552116, + 0.27756531187458805, + 0.2602189127344718, + 0.25270007005713946, + 0.22634470056009706, + 0.20079531104132592, + 0.20384052701835018, + 0.22300835290018206, + 0.298097551758236, + 0.2161312022171807, + 0.23913691131278167, + 0.27539031617926707, + 0.2728705949332212, + 0.3040955645443158, + 0.2442636753562248, + 0.24040878439707455, + 0.29541638603297893, + 0.25512407736904774, + 0.2341363000274448, + 0.29077407182389564, + 0.2590561887742359, + 0.2878966330623601, + 0.24218624089484161, + 0.24810651405911782, + 0.24725756669519122, + 0.25450735591679435, + 0.24219474452712816, + 0.2483928137294244, + 0.2733031881484303, + 0.2956014944727987, + 0.21563704069352232, + 0.21074157300214466, + 0.18490499826478826, + 0.1874856968139384, + 0.23613205017741204, + 0.32822824227254804, + 0.23770111128589477, + 0.22745575847493052, + 0.247354685847444, + 0.24046263266016155, + 0.30872100812254627, + 0.2671622197245215, + 0.3187303343064424, + 0.24758768872375306, + 0.27912306840713524, + 0.21660565133668389, + 0.23849993086380852, + 0.2955931718989435, + 0.27803463362969083, + 0.19727122217787643, + 0.21587318838803082, + 0.231456153086798, + 0.267350038766468, + 0.2615110524814113, + 0.23906836973196852, + 0.23988535137180958, + 0.25567359392255246, + 0.262238130696068, + 0.20306733555133927, + 0.22636278641813837, + 0.21896645776893278, + 0.27607995677653446, + 0.2199342204598329, + 0.27436626397756453, + 0.25927339029715185, + 0.294255015617117, + 0.24779347869897353, + 0.2661599077072027, + 0.1846139917911777, + 0.185194193424931, + 0.18618725765801947, + 0.1814484506162952, + 0.1800371717282415, + 0.2674826884969546, + 0.25570966972101045, + 0.2814629380087803, + 0.24532394382210135, + 0.32353892360589, + 0.26911664036872657, + 0.2228132797512717, + 0.21954429441503315, + 0.24936192413248964, + 0.3036942752381367, + 0.249942605540087, + 0.19788305497943695, + 0.24427242231266524, + 0.2971293408189174, + 0.18978750358251115, + 0.21937791868868742, + 0.1874011012565332, + 0.2775815547408455, + 0.25814160004161874, + 0.16578876745969856, + 0.21161279370347688, + 0.3113422463873142, + 0.34739508280377623, + 0.28771761834308585, + 0.21694336625726177, + 0.24950011752892345, + 0.2515463323925749, + 0.2536515757806659, + 0.246304788173044, + 0.2177466916675574, + 0.24063633178408117, + 0.27797493809529206, + 0.290905598578519, + 0.26666841414212444, + 0.3812661957266475, + 0.2797267475469601, + 0.2799955732916885, + 0.2550360177929621, + 0.25018502301626827, + 0.1470497824960223, + 0.31154264189996184, + 0.21442030553369223, + 0.2376432275722004, + 0.23656111707006885, + 0.2695939608891874, + 0.262653783314355, + 0.25952001045828227, + 0.3187123745202432, + 0.24932139907549722, + 0.24774321616252162, + 0.2694163631574395, + 0.2720252888421595, + 0.33942115378055177, + 0.2821896916478034, + 0.30088665308476453, + 0.2348254214044824, + 0.23639105894592707, + 0.30419445058407446, + 0.2166862797790531, + 0.3041288512403059, + 0.15901619698153405, + 0.2817397021597304, + 0.2125491885707608, + 0.3465602491898679, + 0.31345036817889427, + 0.2378659972705629, + 0.29230463549072405, + 0.22154118932815267, + 0.1535456063615721, + 0.3032383599108667, + 0.29384440475185286, + 0.2597940938160522, + 0.26825202650878344, + 0.2768973631071115, + 0.29373863596656846, + 0.276200862287254, + 0.21127142496940293, + 0.26365019469516937, + 0.3376170897757673, + 0.28445159692175, + 0.2491095629741576, + 0.22559048752014793, + 0.2419635570712924, + 0.2477965182742266, + 0.22887362706719871, + 0.2454804384368254, + 0.27077009386230083, + 0.26121189195807976, + 0.18233478480270093, + 0.2697323118804064, + 0.265848931890437, + 0.27614598957770686, + 0.23109214798993766, + 0.1840273391574269, + 0.26685795606577, + 0.29810921249634303, + 0.2656066657203458, + 0.2873308965583358, + 0.30050266383630975, + 0.17045151967479685, + 0.2993370455015246, + 0.27080112310523424, + 0.2138140035079732, + 0.25138228275031455, + 0.11780977809797692, + 0.3116293721550232, + 0.379804976191534, + 0.32038574315368135, + 0.32337125652513304, + 0.30510975675402935, + 0.309427216459232, + 0.3129962600133696, + 0.2989616571634309, + 0.243104394781862, + 0.2571221443878686, + 0.2613200255970744, + 0.25320659580766364, + 0.2519901745505945, + 0.2521500214253073, + 0.2639850793179393, + 0.24211271272268892, + 0.2405772711786647, + 0.2796555857807498, + 0.3109663777677253, + 0.31537631717065706, + 0.3108563590407128, + 0.31016731198147723, + 0.33677607506075957, + null, + 0.35608214210561645, + 0.3966265024048952, + null, + 0.16008264041172815, + 0.43695135054278306, + 0.16049785103547148, + 0.34981830138621645, + 0.3569953061704455, + 0.3887380828116521, + 0.4243885676286772, + 0.41748879171475833, + 0.42918138554522917, + 0.408387716692356, + 0.3683893913529269, + 0.34554346821783166, + 0.2778871740259708, + 0.4232507954820242, + 0.2509094268065796, + 0.3422051719082586, + 0.3131517140231441, + null, + 0.3322333148295513, + null, + null, + 0.3127057191660489, + null, + 0.2911926492364521, + 0.27412583055490586, + null, + 0.3619894356863606, + null, + null, + null, + null, + null, + null, + null, + 0.32771155575741995, + null, + null, + null, + null, + 0.2805909524217603, + null, + 0.32439030933824786, + null, + 0.2908488253071323, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3431379284180201, + 0.23046388068617335, + 0.3241710666682501, + 0.3332087315564517, + 0.2753512971583228, + 0.331798049835341, + null, + 0.3776947207097694, + null, + 0.30616144282147356, + null, + null, + 0.31842078282786385, + 0.35035042043923276, + 0.2937895871383883, + 0.2952430320727722, + null, + null, + 0.2900433320780422, + null, + null, + 0.3152102442871463, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.30872173493677646, + 0.2995118903149429, + 0.32194642204472296, + 0.2822777741863707, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4350833744422687, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.21882501745260038, + 0.34045515405381843, + 0.3038801399951839, + 0.4654102857949715, + 0.3958221294025091, + 0.2657809810587295, + 0.3769935532701229, + 0.4327297394815035, + 0.36280220014286846, + 0.3794822986066577, + 0.40092998009599073, + 0.3893964974010393, + 0.28156262506503343, + 0.43901594049286724, + 0.37548901959479075, + 0.37965774654047985, + 0.42062619239322296, + 0.35672234614817716, + 0.4791482916663711, + 0.3724717512541462, + 0.3778988778389443, + 0.34108122056313017, + 0.31514332996275846, + 0.4966213436207517, + 0.35444504737610805, + 0.3856731778933174, + 0.28871591609544567, + 0.3763007318816277, + 0.3909215211329431, + 0.3963705898194636, + 0.40073295420779814, + 0.4131758558422249, + 0.49653019232269413, + 0.28508017817370973, + 0.66567820195317, + 0.27922127481696646, + 0.2219544798826346, + 0.1682247466188644, + 0.33341514462458005, + 0.224849868731779, + 0.2727990749919327, + 0.4679000156338711, + 0.304709056976433, + 0.40537589951035247, + 0.2953189981580302, + 0.3964633472316879, + 0.3024850393117148, + 0.21194955951004046, + 0.47010958611069625, + 0.440501429866889, + 0.39197692019483504, + 0.36700610193233985, + 0.3798989414200085, + 0.3864874919908182, + 0.34142501058227775, + 0.39701021576499185, + 0.32359194524874363, + 0.4508265249186963, + 0.44031682231862207, + 0.4567781489596803, + 0.29502428680244386, + 0.20311460182358915, + 0.3993020969873822, + 0.42515699220867076, + 0.2770763117918935, + 0.34949736406355053, + 0.346161407227223, + 0.468701648474878, + 0.4937951617035247, + 0.42803306160486526, + 0.36481481106928065, + 0.7118616187303446, + 0.3603265015430608, + 0.37101546328787, + 0.3714048319602487, + 0.43406294940841733, + 0.2967407559012009, + 0.6439177058072848, + 0.41166714929389153, + 0.42313505458110395, + 0.3952474433448788, + 0.41858301221364, + 0.4885020961220343, + 0.44243869455531204, + 0.3462580788817472, + 0.30185354714569307, + 0.4580012952884049, + 0.4258393483530422, + 0.8056986593916337, + 0.47135695107414083, + 0.3824649964893254, + 0.3513018248439475, + 0.4258401306979222, + 0.3677244449728445, + 0.40510754887462364, + 0.5427066754085912, + 0.34399735365897327, + 0.3549678692146795, + 0.3740432124243017, + 0.3153095595025988, + 0.2573005646674989, + 0.3806030263466944, + 0.33543753495334694, + 0.3269606115539251, + 0.3730967283258543, + 0.2307010670784962, + 0.2858647840123512, + 0.30022055004955867, + 0.38018331381719833, + 0.3691951423713608, + 0.23969002548255366, + 0.3259345685865533, + 0.30580973835247005, + 0.35694468555097925, + 0.3646020556483133, + 0.44314000469923515, + 0.38452608058772947, + 0.3855348376367449, + 0.35410103078731353, + 0.40355466931134526, + 0.34905026386558585, + 0.4064036263762222, + 0.38225229208162054, + 0.2906539907580933, + 0.3658444375087719, + 0.3196648871439877, + 0.3331043440313669, + 0.3162104554143114, + 0.340583906521073, + 0.5349629836910553, + 0.3837574165331074, + 0.31173005939591714, + 0.40010663785653156, + 0.5987862211232853, + 0.3019777428606171, + 0.36846761426014857, + 0.35599907789106117, + 0.3039967694413197, + 0.3306933236576107, + 0.30909297782590267, + 0.32682756258793766, + 0.43805709951202704, + 0.5307197893816067, + 0.32586814337809067, + 0.39895221888295973, + 0.477819716924534, + 0.43045719541447314, + 0.2715725298751829, + 0.4154020507604638, + 0.3957919546280909, + 0.4011991463521433, + 0.33880136699700736, + 0.40230329626644346, + 0.46021259231166195, + 0.4055573832710553, + 0.4004479216143587, + 0.2877463540904672, + 0.3321378701777727, + 0.3316648894261559, + 0.41932202377773564, + 0.38253192668523783, + 0.35575993409702866, + 0.38612569796332885, + 0.3852579339226449, + 0.377115652995676, + 0.3669658184030549, + 0.3482996044800897, + 0.4612470539316162, + 0.2770524655911349, + 0.36936459659671106, + 0.35543742377753007, + 0.31646649883632294, + 0.48352027233646316, + 0.32841810495395024, + 0.2807574625693644, + 0.40161004232434067, + 0.2697541432936055, + 0.39578352965217484, + 0.38855871642376316, + 0.3170466871800914, + 0.37691893486001904, + 0.38273161066392875, + 0.31516901629528177, + 0.3867636783772689, + 0.32311790938432977, + 0.371570412555489, + 0.42291954782744, + 0.28465645576011633, + 0.31826451715204823, + 0.3913947604230905, + 0.29482448005499345, + 0.3419592863595597, + 0.427463224687754, + 0.42169966772447265, + 0.32133186636464284, + 0.3221263310100972, + 0.31445235329896076, + 0.3130119345949088, + 0.2670536029363718, + 0.4506416792352545, + 0.34384473104050134, + 0.3286752035916001, + 0.35525064380163296, + 0.31648770233163454, + 0.43945345798008617, + 0.3430549242536493, + 0.4105319803872983, + 0.3657699079237194, + 0.30357553933980214, + 0.48425178981542083, + 0.31900911130686277, + 0.4101504567264786, + 0.3285291244786193, + 0.3802124126270427, + 0.8575986114946371, + 0.41252249717281053, + 0.3907134223449578, + 0.37807794140847384, + 0.4506301066716995, + 0.5383208311811453, + 0.608000871963073, + 0.6232230451909749, + 0.4105832394469338, + 0.38754822350880813, + 0.4665003949058969, + 0.7300012779550539, + 0.32053607264822787, + 0.34826266849571513, + 0.5384182323992468, + 0.5151293564300746, + 0.5439594565086883, + 0.4390519026084874, + 0.31823062632715365, + 0.3824353961413342, + 0.5971746351850965, + 0.43246266585731863, + 0.3365450711166504, + 0.611484364923859, + 0.5448452926937742, + 0.44883004821194245, + 0.48293439829784135, + 0.7075044299692391, + 0.4642305899366904, + 0.5047398829483514, + 0.44662349727821193, + 0.7610181813104676, + 0.32070172886140985, + 0.6394854694936944, + 0.3912010215841065, + 0.5450045932125727, + 0.4274531015094065, + 0.3882501812988709, + 0.6994923110557746, + 0.5037458264523665, + 0.3079693492057037, + 0.37574961688739345, + 0.6425040294514819, + 0.3617620890575756, + 0.4464109180715619, + 0.4600611809364894, + 0.3615190682047879, + 0.46522506566129734, + 0.47438135715834784, + 0.41118505463564103, + 0.394506550319913, + 0.46719601661075605, + 0.3514901851119769, + 0.3224247033849972, + 0.6275214021712556, + 0.37058096035462657, + 0.4501036014573827, + 0.5648139874063477, + 0.3555704990808689, + 0.6574547754132931, + 0.34148266784081927, + 0.3421502674586858, + 0.5569227010380371, + 0.477722531486994, + 0.5821159186688933, + 0.7283391137659918, + 0.6635056232526813, + 0.40473583900661747, + 0.46758854577511405, + 0.5464496376677251, + 0.31487091417802676, + 0.33412198190290304, + 0.5628794877819827, + 0.5266118676349564, + 0.7260452171617042, + 0.46055347772655, + 0.3583122071020695, + 0.46072516580255796, + 0.37359243190967045, + 0.47390638579289396, + 0.7017555994832808, + 0.6239550582747272, + 0.4923614900540527, + 0.5265127701346155, + 0.4955688645780086, + 0.5138085389211591, + 0.7362126589999008, + 0.2996688247665839, + 0.43852750172564947, + 0.2929570927914701, + 0.5980180203278822, + 0.34462442390995957, + 0.7348318866560016, + 0.4114165456949577, + 0.4175300908115939, + 0.6917629616430607, + 0.5154502642095566, + 0.32533727387889516, + 0.5621891479032565, + 0.40152124803934197, + 0.5138738854188277, + 0.4997768554244373, + 0.3342003169423172, + 0.492773614561994, + 0.38948842189567, + 0.4592942874904829, + 0.37149634324817643, + 0.29692596226516, + 0.4011604550808942, + 0.5172943460587505, + 0.5770769793941732, + 0.657924680173607, + 0.39716909404328404, + 0.30071276666366276, + 0.4592103408329516, + 0.46179053602373266, + 0.4254696710532505, + 0.4854738980772949, + 0.4677319296987914, + 0.36900160557437467, + 0.3266822255228085, + 0.5020394222778738, + 0.765156333936315, + 0.5294292394885509, + 0.6009954958708683, + 0.5892752416930432, + 0.7689359619928438, + 0.5643458532767628, + 0.32663747520732533, + 0.5826875017344271, + 0.3780862258816573, + 0.5178647753032831, + 0.4529371926302071, + 0.3934690100550509, + 0.49240403838168934, + 0.7798196977730218, + 0.22699242669946243, + 0.4601810352978159, + 0.47097483714895416, + 0.6217087501210522, + 0.3784733123757858, + 0.2804360579973796, + 0.42071006236788355, + 0.7209735340442059, + 0.3841014645379864, + 0.4393962349761867, + 0.5060075699463837, + 0.44352876892871523, + 0.40500793607427066, + 0.4950191817884736, + 0.6112842006833437, + 0.4598339722589051, + 0.39523922489382407, + 0.5449487151259107, + 0.4560476080873796, + 0.5678301833032463, + 0.543025080432062, + 0.5295627108272563, + 0.5469293081929993, + 0.3084335638573366, + 0.3044297221138035, + 0.517408204195036, + 0.3070202482153527, + 0.6917763858890974, + 0.30284208679286084, + 0.5601383427799572, + 0.5621418073493439, + 0.683313594091123, + 0.3007466528320625, + 0.34440556706045805, + 0.3360096735349467, + 0.34627188228623373, + 0.29642331967804647, + 0.28646479704025907, + 0.6128733445456798, + 0.354300646518575, + 0.6803633718718394, + 0.577432835445594, + 0.4380157086025897, + 0.5684393567444554, + 0.382564935369384, + 0.5431991537927995, + 0.5325991785214576, + 0.4633975371985574, + 0.38647846872466074, + 0.4602473550268332, + 0.47783288872163243, + 0.5210841495922327, + 0.39180049971873426, + 0.3314237426896437, + 0.6179473569704912, + 0.4700613163848642, + 0.6562488261927685, + 0.6352844396777189, + 0.5060994510244526, + 0.31061617058206537, + 0.3377841432930379, + 0.3547550042806989, + 0.4529065735854834, + 0.7451778815680312, + 0.4682450606278535, + 0.3719724047107842, + 0.7403758280051096, + 0.4321715593876449, + 0.6231668148305651, + 0.5264260716343656, + 0.41104950520453437, + 0.4977325309651204, + 0.4546354090238499, + 0.4519012416783758, + 0.5759476789536704, + 0.5311087954409307, + 0.5214632316500889, + 0.4322452430150646, + 0.4575738739551438, + 0.4617217142221426, + 0.4806513131405995, + 0.7669851362125285, + 0.4858034926631415, + 0.468041822289462, + 0.4364639167767262, + 0.27867647969510906, + 0.6621242057798509, + 0.398849053231136, + 0.3221063048330882, + 0.511904401705139, + 0.42805613048155916, + 0.33098092691994163, + 0.47878009825506207, + 0.42632968571535784, + 0.7768858882206369, + 0.31288594736494785, + 0.3853285936970486, + 0.32639446550228446, + 0.42799879895770426, + 0.3921298212965602, + 0.2975281117575214, + 0.3969077807974527, + 0.27297385096633553, + 0.4137433577296076, + 0.7370494519797747, + 0.3057883888819916, + 0.4008868991902273, + 0.4919327969315543, + 0.4120105935187646, + 0.38211194380926905, + 0.47283474585079144, + 0.3531533290179036, + 0.5173459924026196, + 0.5193860257971349, + 0.21596878704534161, + 0.625924235513852, + 0.6835083317292152, + 0.49539237638123046, + 0.47246385680479547, + 0.5190162363789386, + 0.38295918602501533, + 0.36655342669886815, + 0.4471879711422015, + 0.777986041014157, + 0.36657408622508747, + 0.5992964161073224, + 0.3376568007842407, + 0.4517030787215752, + 0.4720504055833095, + 0.48480549823940505, + 0.38125973153214127, + 0.316173983154994, + 0.31511060988178563, + 0.2909174248588988, + 0.5007992753881109, + 0.4391426143761508, + 0.2885274772767965, + 0.3992156510257698, + 0.4469257260322794, + 0.2917139282481184, + 0.5510628746467068, + 0.4809650869793309, + 0.3898051005158727, + 0.449647686431258, + 0.298087961216038, + 0.41849564819482116, + 0.4904611273387277, + 0.5755976033300203, + 0.43231070361671053, + 0.7767426382404918, + 0.32158035564331033, + 0.5520678981181644, + 0.7340784253700177, + 0.6298063622505315, + 0.2566126553477872, + 0.4875864549183621, + 0.41571439238141306, + 0.33515312068628555, + 0.3558761411632804, + 0.3653952069302536, + 0.4594758895827061, + 0.49253466216461883, + 0.34286550132239396, + 0.6066699639736075, + 0.7633348518390164, + 0.43245763061768905, + 0.41565001578798066, + 0.43194771351223066, + 0.48621488385272044, + 0.5074207417060814, + 0.4907578038987877, + 0.4827510219442245, + 0.31753701558712244, + 0.6484453872029208, + 0.4734058660849892, + 0.5500614315928978, + 0.7086593022918196, + 0.30924874816953624, + 0.5888052240558447, + 0.3866660287004436, + 0.4476684457398404, + 0.34303421131407386, + 0.5104314404758531, + 0.4610756506114344, + 0.3204155334221209, + 0.415629927056388, + 0.4061451701600473, + 0.5365684860347644, + 0.5857464403220655, + 0.4408931683243928, + 0.45745049467293714, + 0.40728426470595425, + 0.7373562594167505, + 0.44496615051248906, + 0.5167540059471205, + 0.45610102873837144, + 0.464519434087228, + 0.6058135553202356, + 0.30206264623633444, + 0.5624688585570096, + 0.3728818535604606, + 0.5321979316873201, + 0.4138320633811382, + 0.3330201143423729, + 0.3338054502435634, + 0.46397748328984384, + 0.32283441428725856, + 0.5033965702910034, + 0.7299674611329645, + 0.7237897563803823, + 0.37942356794747745, + 0.3310561836398515, + 0.6992807346080309, + 0.4416288181208955, + 0.37606524371272326, + 0.5077922586863113, + 0.3526517997834547, + 0.5920289409294667, + 0.432875061225499, + 0.5599449539442038, + 0.5853895212659107, + 0.43112126726254757, + 0.5068614050882189, + 0.6529501765582103, + 0.5002230499642669, + 0.3512189592327401, + 0.4054870521609621, + 0.30000298851218266, + 0.5334569143216646, + 0.340237391910936, + 0.5114664478961164, + 0.49225827090636176, + 0.35523194578374667, + 0.7166796173965815, + 0.3171770991708881, + 0.7217354383327633, + 0.4310005591875364, + 0.5373071774887082, + 0.7389379781924887, + 0.4724444659054402, + 0.5460065797372405, + 0.3798259682766931, + 0.7010700464774292, + 0.40697338607054956, + 0.31506090335054965, + 0.5549229434835145, + 0.5123766424743443, + 0.6366127224624089, + 0.5440601778553957, + 0.384052944787046, + 0.4801848956758036, + 0.2777960887151633, + 0.44511897891194624, + 0.5610146691300212, + 0.5057395697897767, + 0.3445218170299694, + 0.534036962119084, + 0.5763699580229028, + 0.5632917880008728, + 0.4266781882224637, + 0.33875223377648417, + 0.5018716834986247, + 0.46110184531103515, + 0.5523657842488126, + 0.4829112549894874, + 0.3352298432702736, + 0.39916339159023223, + 0.49814876738986064, + 0.39585719219186083, + 0.7304428413948698, + 0.5058249454082608, + 0.4432549389647563, + 0.48889230078162643, + 0.3877449600449895, + 0.42546950010992324, + 0.5366026445252304, + 0.5451484482843056, + 0.7092335772144097, + 0.530843618423414, + 0.571769278410394, + 0.5220804966731324, + 0.47566168327993996, + 0.5698237410442001, + 0.37878283343114144, + 0.6281051600093435, + 0.5919798315669413, + 0.722424341641681, + 0.5103067276496848, + 0.5408636178836403, + 0.5371508632626093, + 0.40486562902057893, + 0.5321900598999216, + 0.3316898453889581, + 0.6740043794268322, + 0.4749212921699621, + 0.734916573751433, + 0.7565083001229731, + 0.46300437314024073, + 0.4420357682978271, + 0.4907788605354149, + 0.3433327276529187, + 0.49183507201702975, + 0.7574226017573998, + 0.48340528206313893, + 0.5014874114631406, + 0.5827714500859067, + 0.4952675438480739, + 0.3381895279795058, + 0.6053218393417913, + 0.422970660555822, + 0.3862902363985386, + 0.7321273356946161, + 0.403464673294177, + 0.4336319379658393, + 0.38862647034493736, + 0.47996425221878253, + 0.5578046513866654, + 0.6169078077595523, + 0.6781160314204909, + 0.6227955705851628, + 0.5857752930452825, + 0.45276860171775013, + 0.6011580023023019, + 0.5517808240040503, + 0.779588709076679, + 0.7395479534321738, + 0.3423636520770564, + 0.3659649273892656, + 0.3256196042135777, + 0.5141052785505024, + 0.4140049422945657, + 0.7869548062059839, + 0.37265443334063825, + 0.4169903037934814, + 0.5590010324387289, + 0.4802220148650324, + 0.558907858238449, + 0.6215995576729653, + 0.4671826039033969, + 0.6169460134085528, + 0.3763326057193979, + 0.681317580546525, + 0.7795109897667455, + 0.7622816404913033, + 0.5756741843063253, + 0.42662914442147215, + 0.5408124203258738, + 0.7400864108534758, + 0.688644556836712, + 0.6241137872109156, + 0.4809835213853613, + 0.43901465861128436, + 0.4903484582920972, + 0.7352639154139833, + 0.4152957239724208, + 0.42593998214825574, + 0.3070379465839928, + 0.59786084973276, + 0.2967169011415813, + 0.4048745037895954, + 0.6625373705472744, + 0.5379126574943407, + 0.6825690350640046, + 0.5426061627706354, + 0.5108369288922499, + 0.6671310689562719, + 0.633108130166752, + 0.5620566212629121, + 0.3811320895500894, + 0.6440981454818991, + 0.5191213262252881, + 0.40136290884884485, + 0.34599984389402416, + 0.47753083106380423, + 0.6371909955339137, + 0.5024177678101982, + 0.6107521285673682, + 0.5734667823253057, + 0.4182986776462233, + 0.40454484445187555, + 0.687065079355631, + 0.49160544551089824, + 0.5950997602447192, + 0.7238042648275541, + 0.4139336762036054, + 0.3630552839742534, + 0.3600640223675347, + 0.3648606309101556, + 0.7119359897622649, + 0.735678428461694, + 0.562598252637571, + 0.41915233365355337, + 0.45775995605658243, + 0.38645348587651757, + 0.780131449957534, + 0.3462885053344372, + 0.4808214577542309, + 0.33522472816013993, + 0.49980436204450357, + 0.366572631902444, + 0.5626939201502156, + 0.4795125197611242, + 0.3186779458096732, + 0.6962418380163103, + 0.5099007735786537, + 0.41788217525138815, + 0.33795834840515754, + 0.35639137228917983, + 0.4047548960603023, + 0.41984314808939194, + 0.3709493521784162, + 0.36538908769811707, + 0.4037050376201614, + 0.38133769816919844, + 0.4116958461560139, + 0.334202244835824, + 0.44365588621830626, + 0.2116572107493678, + 0.2727193331436961, + 0.32698974337547143, + 0.3887100957724761, + 0.4394897922035032, + 0.4238927785037702, + 0.3353027234019268, + 0.3358767156117897, + 0.33150370764008913, + 0.6685969784653482, + 0.32973589442207263, + 0.3106154948013664, + 0.7869423070693666, + 0.3633259212082645, + 0.4184738303774303, + 0.34816827609382156, + 0.4477518707481585, + 0.27665782282260676, + 0.4075026683901874, + 0.355168419959519, + 0.4827002399681957, + 0.2626453075837068, + 0.4138734906925895, + 0.3997780535215274, + 0.5057101102230066, + 0.38979617617540663, + 0.47199175607033556, + 0.30450091468731366, + 0.45533553587879166, + 0.408658741196903, + 0.3589347372302217, + 0.4706461525967899, + 0.37005081078293334, + 0.47258147585023014, + 0.38533731980605207, + 0.36904641479688755, + 0.8068581149646882, + 0.3060341994372348, + 0.43718743408089483, + 0.3645179096714337, + 0.40690424618091503, + 0.31779406910522795, + 0.37716357824557334, + 0.4559016975516327, + 0.5466077468998252, + 0.3419314515797742, + 0.381528457204375, + 0.2871743878680699, + 0.4105742300657838, + 0.42553709699734277, + 0.4305378069828252, + 0.4563526285239952, + 0.36497029233636885, + 0.35043068720858417, + 0.4647221779892469, + 0.40506624707450256, + 0.446889962435257, + 0.29836761883292656, + 0.3189996730128331, + 0.47755104437747226, + 0.454589423391815, + 0.4041045923891763, + 0.3659594655590652, + 0.36586928968496457, + 0.8364240066110371, + 0.32268186066510807, + 0.3334500749638459, + 0.3484933629082077, + 0.30896407305316265, + 0.3442966345270979, + 0.33949819574975976, + 0.29988668101219473, + 0.36287398067434035, + 0.35258055294423696, + 0.4335550418360907, + 0.3807921327165162, + 0.4347494531943518, + 0.47070416076553234, + 0.3946443311728597, + 0.3829292322940225, + 0.5562735503056593, + 0.39174258580501375, + 0.4147460918461371, + 0.4982398648912179, + 0.44843978097909704, + 0.35551811724005006, + 0.2999551636607511, + 0.36916290812293856, + 0.4893464905635565, + 0.4709553207560771, + 0.32240404005955287, + 0.36578544528087753, + 0.387522145043883, + 0.45727976961017486, + 0.33952709744893605, + 0.34951067282487097, + 0.49035091478408305, + 0.525858352138486, + 0.3097064604813054, + 0.4271808671723638, + 0.3179036983637453, + 0.33754139376075676, + 0.3392195934539049, + 0.7893704047003994, + 0.4205363067672308, + 0.44355500445285795, + 0.3120820344895036, + 0.3751685590232129, + 0.37277363459550483, + 0.7003843401286353, + 0.4206684044752588, + 0.37981231972391893, + 0.4459740608224705, + 0.41798807754428874, + 0.4121821054791619, + 0.4523296641270035, + 0.4876089930165707, + 0.2856405364958381, + 0.3474920028977734, + 0.4445148990171657, + 0.4220242844300343, + 0.41262104958257423, + 0.40965528641991106, + 0.37014925186239916, + 0.31450454296488617, + 0.35411369900355666, + 0.4050058742535218, + 0.37570328153715116, + 0.6105272370935607, + 0.35073420658807686, + 0.3725565509457066, + 0.3108968514749856, + 0.5592008145178645, + 0.3904313786018059, + 0.322334564907959, + 0.3877949715779759, + 0.4064685084276166, + 0.41757971699863045, + 0.3252665545635677, + 0.5447717542335674, + 0.4621755510800724, + 0.3182608172629553, + 0.41392029863434066, + 0.4436488525047196, + 0.4129769914855109, + 0.3774587625507417, + 0.4663904365833334, + 0.33621846096604013, + 0.654161106490802, + 0.3898340349308901, + 0.4572905097995524, + 0.459637671215076, + 0.39429864820883337, + 0.41219157179311783, + 0.3890909443903544, + 0.3016928204659661, + 0.3898473163546124, + 0.5937924016268229, + 0.37530996895904023, + 0.3256666741272832, + 0.5033097264171459, + 0.422097366883822, + 0.41764843054432627, + 0.6154490969473492, + 0.3863832279587604, + 0.44899068297842887, + 0.37033030936993117, + 0.3889300057407444, + 0.4767956800055221, + 0.3624494825356794, + 0.4697243785161032, + 0.41273442695914003, + 0.45052225428912224, + 0.31970038120204064, + 0.4282177534659728, + 0.3740891938706564, + 0.3258236689977596, + 0.3190798950535469, + 0.3383505558049615, + 0.4320238231469715, + 0.5334017421561198, + 0.4507054507294741, + 0.45202767678320166, + 0.4676275485041091, + 0.34667698026654636, + 0.3700103873994072, + 0.4786058551693059, + 0.36664327173313416, + 0.4000389627343568, + 0.4792639406364934, + 0.3844740305119771, + 0.42613167007816266, + 0.5012737024806081, + 0.4041517507590016, + 0.3706823215806667, + 0.5089688768086856, + 0.37900603569259916, + 0.43867950876305917, + 0.3189676606933628, + 0.4955716124406697, + 0.5980332461630286, + 0.7903737824021707, + 0.42517375123611234, + 0.38833396369541623, + 0.42792231754769267, + 0.36411705674752926, + 0.35990503339434105, + 0.3663465457082518, + 0.44065059719314986, + 0.3943679576856887, + 0.3656263024478576, + 0.448861521062164, + 0.429200507651767, + 0.4241991729449927, + 0.33445136534365044, + 0.3224448579709615, + 0.43437162272803304, + 0.41055797954464934, + 0.38328105615737695, + 0.39442305015998347, + 0.3950085746362374, + 0.36508723735064563, + 0.2765456720967182, + 0.42293347370419454, + 0.4394358232493337, + 0.36049222103737694, + 0.39860134361013044, + 0.3090473751466875, + 0.3889779615842237, + 0.26509288223066396, + 0.5213752148028321, + 0.3611977629500355, + 0.3284399649254961, + 0.4137429044676887, + 0.3865731364576429, + 0.3316006175077304, + 0.597019813218832, + 0.49773859522070335, + 0.3075483457694812, + 0.47186381187919346, + 0.41493328038843974, + 0.4194595896965543, + 0.386569374729113, + 0.2872328174125962, + 0.367699403653296, + 0.49646359291590797, + 0.3310761111004009, + 0.4307228958193857, + 0.31340496789191047, + 0.3533636854722245, + 0.3569966375177644, + 0.33581302161545357, + 0.4000833896493849, + 0.4277109631219022, + 0.37302936066819714, + 0.34405490625938984, + 0.4137683163952678, + 0.32533959573889715, + 0.39511935503869866, + 0.37663134298436274, + 0.4897803889200867, + 0.5270867772629223, + 0.4327883431776219, + 0.4119875791046678, + 0.4255298729458186, + 0.43896843770194327, + 0.43952806832106045, + 0.5283274799975659, + 0.4023707816137846, + 0.31572168587213, + 0.5006334815184639, + 0.39363810210507594, + 0.45047708060496766, + 0.44323708191454736, + 0.3874136592156799, + 0.492615888224944, + 0.32027979581913946, + 0.3672025158700878, + 0.31493797728708495, + 0.31104776379263993, + 0.35061578270554355, + 0.3976590644015309, + 0.3366425143115941, + 0.4101508285076556, + 0.5290374972638864, + 0.4176842430026134, + 0.4926309386321999, + 0.777532239857815, + 0.5547849430708497, + 0.3497416052078528, + 0.4164463377986912, + 0.45304847848852575, + 0.42125941503635855, + 0.3095188472764508, + 0.5765538253712, + 0.4699735262441638, + 0.3946054696042195, + 0.4582168643703926, + 0.3813766578724605, + 0.48895897918264486, + 0.5022119732416939, + 0.3950430065762332, + 0.5301300655505854, + 0.48668143618478227, + 0.3748291373846559, + 0.44870603805189724, + 0.5018324808367836, + 0.35611998267354483, + 0.38347547561000916, + 0.31076509219557286, + 0.5653780910914677, + 0.5089301660577351, + 0.8199667464149283, + 0.5281148674174035, + 0.4351561949591347, + 0.28120501143857, + 0.46863426577942213, + 0.43072816591742236, + 0.6026434359141286, + 0.343999222682848, + 0.5369182558606679, + 0.3285462155366472, + 0.35683334055637705, + 0.32631563066857866, + 0.31312160435063335, + 0.4160615137481524, + 0.3098709495560631, + 0.4249911061976278, + 0.4468981165013268, + 0.38946696840788325, + 0.32286604331109975, + 0.324629119806634, + 0.41364774318069236, + 0.310397441855161, + 0.5246894603487707, + 0.42565677903608284, + 0.49035826350527306, + 0.4190542333825673, + 0.3567976217078763, + 0.5537469927252255, + 0.46834013165168664, + 0.358092431521648, + 0.5892612906215228, + 0.3034053877206101, + 0.4437209896595325, + 0.3610858095737733, + 0.33291747502999186, + 0.32256523931293, + 0.3685525535824553, + 0.4739036166977365, + 0.3356412374497563, + 0.6894700931527497, + 0.44167688949928224, + 0.3276983427500806, + 0.38231201498404815, + 0.33517989818405813, + 0.5724692616788554, + 0.29544313883458184, + 0.364721800502486, + 0.378573694964701, + 0.4252120423543358, + 0.323319228072017, + 0.4022274965931488, + 0.30919726735502817, + 0.4316958760571449, + 0.7653350768255444, + 0.4441932243292318, + 0.3799847172074784, + 0.39458120723296236, + 0.389821159373106, + 0.2997783713203183, + 0.3712659409196611, + 0.44471714412554897, + 0.7490673845181163, + 0.5467660453488786, + 0.45009759764366564, + 0.38903194627974047, + 0.36330624590309196, + 0.40917425425684284, + 0.2982697692144968, + 0.4294474887957011, + 0.5072175596954681, + 0.3276354067822149, + 0.3829904870032535, + 0.3737996135498457, + 0.46188766148887406, + 0.33617321819869667, + 0.5398468041191455, + 0.4052114744139832, + 0.30052500831209866, + 0.4086853664550549, + 0.41858687093710106, + 0.5249082520298114, + 0.528426495098613, + 0.23372100503066162, + 0.7222597909354558, + 0.7099599916142694, + 0.2013622629477129, + 0.29777526137416693, + 0.38018120251013277, + 0.44222041481333557, + 0.25500194288766903, + 0.37337844465234277, + 0.4556961796881014, + 0.4681603584679939, + 0.4555530561642388, + 0.36509992313204936, + 0.3566048213554658, + 0.3446236326284365, + 0.31094197692909675, + 0.4257431815626578, + 0.3708212371993092, + 0.37249884534836497, + 0.4960399709450694, + 0.41090494051955806, + 0.526810643066163, + 0.43757036582246805, + 0.6034071450820743, + 0.43132394626585674, + 0.42239731692661836, + 0.48685614752087153, + 0.4456167156284043, + 0.32644287428666285, + 0.6240903720448107, + 0.6058134182152337, + 0.4346838116313217, + 0.360937342853482, + 0.37651108141694084, + 0.5254795664931299, + 0.35704140249803035, + 0.43071648248159916, + 0.4146560160095892, + 0.4691827884319823, + 0.43294158995863485, + 0.3778907596829936, + 0.37922356770594984, + 0.537698380789225, + 0.48216593281025355, + 0.45840495239981127, + 0.47752530676189614, + 0.3841400074423095, + 0.24466773141077966, + 0.2504234728032555, + 0.3172151664901587, + 0.5183061964867932, + 0.454925745082837, + 0.269900471375443, + 0.44125394809756296, + 0.34636929064472716, + 0.3348236133746585, + 0.36440841945221364, + 0.38823584882261947, + 0.3537789720003957, + 0.43606960132209266, + 0.6227122783142638, + 0.3300014009451023, + 0.42463269252824015, + 0.5285375011538397, + 0.34620546491091564, + 0.4358694638641434, + 0.5911176176544208, + 0.4509957262701245, + 0.44900194997196674, + 0.43775503502324664, + 0.4079484710447119, + 0.2611805992088679, + 0.35803289575449526, + 0.4830389577678975, + 0.49508185014838924, + 0.4707098685534747, + 0.3197524072249808, + 0.4579703046600466, + 0.4619481023605309, + 0.5033614515742952, + 0.4005638387423716, + 0.4643466337131322, + 0.383135102828924, + 0.4610032972470141, + 0.4407126397204955, + 0.41061949422416616, + 0.3072150086350808, + 0.44919436954145514, + 0.4606476288601804, + 0.2551743858679323, + 0.36336289937711275, + 0.4360599017198279, + 0.3957022047036544, + 0.3921135715816883, + 0.4848019762344166, + 0.49095631783694826, + 0.4395568635904647, + 0.47036792183131165, + 0.34520770472729145, + 0.45654219628080833, + 0.4073467542121324, + 0.38275692191112604, + 0.40241196625478703, + 0.4453126545343364, + 0.401716964239862, + 0.4199392003950241, + 0.4063767526975376, + 0.7214457099446382, + 0.5437735678078872, + 0.5029388671308436, + 0.5317871467892334, + 0.4784232579299674, + 0.43177815662466795, + 0.537912149324023, + 0.35525543042425295, + 0.39111429222781297, + 0.4665421631456268, + 0.3299983245271404, + 0.43923619573372247, + 0.49706250627036846, + 0.4395312604715886, + 0.44467006487102045, + 0.4580694683716705, + 0.32024744752448747, + 0.5322651176267604, + 0.4888060808024997, + 0.37494408021708786, + 0.42417913374095995, + 0.3787751052452975, + 0.3462682220285525, + 0.4494259710977374, + 0.4382543795603974, + 0.41535880016311005, + 0.45262215215458196, + 0.38380921983066846, + 0.4818210484858144, + 0.365403673854669, + 0.3910446529202685, + 0.4314965877122149, + 0.44411255163956076, + 0.4326881734238535, + 0.4679435175621267, + 0.4010159232816905, + 0.3428019377968395, + 0.35011886888155364, + 0.4321875785026007, + 0.4052057667975123, + 0.4424426159896599, + 0.4706626142687756, + 0.42384061114958627, + 0.4094596247544299, + 0.3501671916965005, + 0.40290517197427467, + 0.5239075550059936, + 0.39405524392601277, + 0.20387698921116285, + 0.3530298762795867, + 0.4147023998984974, + 0.3067780048262502, + 0.2982945474712165, + 0.4503439145561585, + 0.4505402937736413, + 0.2798310771114561, + 0.31863640549300615, + 0.3728783816934758, + 0.3637089166995233, + 0.5195414829115884, + 0.29688631796869774, + 0.29851357187680566, + 0.3180835652037805, + 0.32987176030905957, + 0.24921153941512708, + 0.3095045776631865, + 0.293243725234339, + 0.3192931002468817, + 0.29727497233169087, + 0.412644406551652, + 0.3209154723648104, + 0.36245340130705633, + 0.3661934239177502, + 0.41776279709016423, + 0.3185254531384367, + 0.39057214140800467, + 0.36873538496733027, + 0.3729506838100395, + 0.37468190703256443, + 0.39293365963113, + 0.37691904364088363, + 0.6056232517830333, + 0.3227160166488476, + 0.3813617263320277, + 0.545241864659999, + 0.48198608021554784, + 0.3706721786847428, + 0.4623593064884989, + 0.4149168755428333, + 0.45626311859429614, + 0.47120818814694376, + 0.3515180724386486, + 0.5795711082576717, + 0.4246647625280694, + 0.4254037953737775, + 0.3594708587460501, + 0.42977102251608645, + 0.3820913665422905, + 0.45553718005370303, + 0.4705514571509764, + 0.44737402770062573, + 0.46436293476160473, + 0.5721331004164099, + 0.4417262073318148, + 0.4031844224969779, + 0.4677680549232201, + 0.5318521117975442, + 0.5224300764586592, + 0.43982972307859497, + 0.5637385559619156, + 0.5543774288892344, + 0.38467423908986653, + 0.38197013920612494, + 0.4347243924311534, + 0.34678075714439655, + 0.6347759820231115, + 0.48397612000958756, + 0.4200334529566649, + 0.35182539840574706, + 0.46165384108013785, + 0.39981907287737367, + 0.5168041546395392, + 0.4352927705897089, + 0.41313855835727925, + 0.5071597754226035, + 0.3053948642352644, + 0.41253764372857116, + 0.5342426182685363, + 0.33535166375778264, + 0.4156692951035166, + 0.3201634931528913, + 0.3599150087216997, + 0.3816761672388277, + 0.3888924758285424, + 0.45985969415318145, + 0.3355800895810421, + 0.3260930172332048, + 0.4216412698778247, + 0.3898443918804967, + 0.3727164512852437, + 0.44436619063992566, + 0.37849962390672465, + 0.41591061352880465, + 0.502126739242483, + 0.558020245240192, + 0.45669262044550996, + 0.3538777621646539, + 0.4663167098891925, + 0.43730217639579544, + 0.38630406922977334, + 0.4207419372380236, + 0.31742969061118076, + 0.32403963801603064, + 0.29480995845684127, + 0.36536180769768484, + 0.30601239829412136, + 0.649456225521286, + 0.4289141414695595, + 0.41023011966525863, + 0.2957012482051813, + 0.3059377393147464, + 0.3171001103320664, + 0.3372219001841861, + 0.3042998287682645, + 0.3139377078001731, + 0.30174066848900005, + 0.37057274701415965, + 0.3840254892015406, + 0.4201824945594571, + 0.4036498581500002, + 0.36294456279171333, + 0.4988215198987738, + 0.49317714499921866, + 0.37163920071371465, + 0.41473026565678234, + 0.42499708332815034, + 0.3845382285415784, + 0.3528658658600842, + 0.5098050035274007, + 0.4429092960239457, + 0.5779870000258152, + 0.444628380268896, + 0.4381631163086848, + 0.44116365764418686, + 0.4478932445897959, + 0.3288954589405262, + 0.5922140626663632, + 0.4075434001139793, + 0.22822594098274646, + 0.4594862593905885, + 0.5716344229682069, + 0.36064917471015895, + 0.44297113721659487, + 0.7515835017460379, + 0.4356610414914727, + 0.5071102854861972, + 0.2851645887260279, + 0.4740261795850147, + 0.2823551917137836, + 0.2997798376483165, + 0.3174123786144578, + 0.270987925181081, + 0.444430488727255, + 0.19519762063517201, + 0.31721845928268766, + 0.44773712213941863, + 0.39337972557104256, + 0.3972106221066986, + 0.38846019550545774, + 0.3311630965292095, + 0.4069433355246303, + 0.5156961979345488, + 0.5346566632388132, + 0.3519576994596116, + 0.4447559050427418, + 0.4072476171657211, + 0.728018508089229, + 0.3917396089858594, + 0.5556453646637252, + 0.7344509897718359, + 0.7137371195679691, + 0.4405815263216349, + 0.37347152769203124, + 0.31044496930669646, + 0.44879886559741966, + 0.48931937893076954, + 0.4896562001667498, + 0.5691808158898688, + 0.42113023186475357, + 0.5483149094216003, + 0.4323026425417435, + 0.38240102845956675, + 0.6988311217974053, + 0.6800507894729338, + 0.46562620348203165, + 0.41340604159775335, + 0.46310447109064096, + 0.4159390541823896, + 0.44781437890394094, + 0.6402665382957075, + 0.6646732400433302, + 0.4202744651795779, + 0.429337393539547, + 0.43341491574281005, + 0.3629088848658456, + 0.48712851937090307, + 0.3283456592965269, + 0.48077550646760825, + 0.49620908274868253, + 0.3453501964645094, + 0.4385553721228294, + 0.3993363045537954, + 0.3385251230370624, + 0.4987008489642923, + 0.4427973488671541, + 0.3987541772105489, + 0.3700899654081674, + 0.4745488904070199, + 0.4191961626643286, + 0.3382850882981621, + 0.34349124987524127, + 0.24284009562091222, + 0.23999184909251933, + 0.2705024034923986, + 0.4396761216083052, + 0.46834725464228993, + 0.3350275683315895, + 0.31746703119595376, + 0.33685552024619764, + 0.420450670643129, + 0.4233920239133341, + 0.4907980618399266, + 0.3417477024869693, + 0.4099044778686753, + 0.5123300819767568, + 0.3682659664186755, + 0.5195292487630402, + 0.4663743033320804, + 0.4183740706204229, + 0.3963514229497726, + 0.35612929258699855, + 0.236931496497345, + 0.2815017636980112, + 0.2713929767056061, + 0.510047184858768, + 0.4269032012056549, + 0.4183106585695062, + 0.466949583380557, + 0.4906165673188584, + 0.3177631408017324, + 0.4121945152249392, + 0.3649718126450949, + 0.6603023723054177, + 0.4335222117460834, + 0.4547211455513849, + 0.35956040459984573, + 0.44694988092784965, + 0.4828196304956686, + 0.42870999353695394, + 0.39919444531722104, + 0.4783555813464258, + 0.44310708618014794, + 0.5501544717118563, + 0.4120383762392451, + 0.47782840409570215, + 0.43846029851544654, + 0.4173879690850045, + 0.3936982993966563, + 0.46389386165668983, + 0.41498688947853785, + 0.48077649888410223, + 0.3403139391046273, + 0.3186871850229628, + 0.318598423057907, + 0.3984836010800816, + 0.443174920566486, + 0.35969470420301247, + 0.42238515511541397, + 0.4653921282989605, + 0.41727094412925025, + 0.42928960404314775, + 0.4268124985436362, + 0.26516999105962297, + 0.41143532760558454, + 0.25010557531798205, + 0.422115272142134, + 0.21957992556527234, + 0.6471736891295945, + 0.31856154028007155, + 0.4379744751699853, + 0.5493057476370743, + 0.4669930987979117, + 0.2680892181556596, + 0.4148584760926813, + 0.32962840631077756, + 0.769167824211653, + 0.27369918578871355, + 0.5094868320249908, + 0.4975328367962922, + 0.5053877379643159, + 0.6268810764042007, + 0.4170159439753285, + 0.5508214171181466, + 0.45371991032173303, + 0.3795369213651114, + 0.32605721123235826, + 0.45083114240974603, + 0.3076969398060656, + 0.7107659767640241 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.4214908092061886, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.4607376899556623, + 0.7571870579677827, + 0.7343490088031139, + 0.8162683144726133, + 0.7356886605381673, + 0.649844973756041, + 0.6206643926463743, + 0.6580655630281119, + 0.39575455756863753, + 0.693719659748076, + 0.30811761793748427, + 0.4680753763389786, + 0.33189930187434363, + 0.32339510068087246, + 0.2953826977624352, + 0.28428263050874064, + 0.3260924866238675, + 0.530129586174908, + 0.404772646940383, + 0.4005016118927795, + 0.32224184369075004, + 0.6654645940596192, + 0.3285820214327129, + 0.584425621223712, + 0.35533634745878956, + 0.2995075574973439, + 0.3937073825495359, + 0.43412552017947725, + 0.48215856633644977, + 0.478113950641827, + 0.2738964428417341, + 0.3809697993230393, + 0.33535199044519365, + 0.3359088718206521, + 0.421137587034914, + 0.3383672798194744, + 0.3254031180186501, + 0.5078880173208881, + 0.5395837371535389, + 0.62577759694986, + 0.7865888443697898, + 0.7797553896433626, + 0.7698514337121138, + 0.6558309275461954, + 0.38129248668320803, + 0.3847311209254425, + 0.7783065690707793, + 0.3405289764930475, + 0.3306649674320893, + 0.47307158506228647, + 0.3780821169075704, + 0.7747488664366077, + 0.7931122530485031, + 0.3841517114928196, + 0.34190148662431413, + 0.3528582423227685, + 0.33051151895631775, + 0.3923863463274935, + 0.38703490226159776, + 0.5004258328167386, + 0.4440459859524003, + 0.42915143487049645, + 0.3890841614998636, + 0.49096456771787594, + 0.4547203757014913, + 0.36882620241135367, + 0.3302509442939859, + 0.5288314272692778, + 0.35930172067793614, + 0.4363518348847022, + 0.4677053954082826, + 0.4206713386049277, + 0.3959823148046011, + 0.3976115840407676, + 0.4053031068864427, + 0.4170406899607529, + 0.28071413419118163, + 0.27016912561057455, + 0.3633087071966174, + 0.31354129254341984, + 0.42462033805826815, + 0.34371612580350436, + 0.3328090449643089, + 0.27414778975365717, + 0.3960123849401402, + 0.47813737463193917, + 0.4010475822317232, + 0.29527701963000336, + 0.33728429996834003, + 0.3414899414659818, + 0.32113865787902285, + 0.32615669217798016, + 0.48357220462816924, + 0.49624948755948756, + 0.3015892449586325, + 0.5554082391600558, + 0.33732447246997066, + 0.5119017924489732, + 0.31627467468120285, + 0.447932793699676, + 0.4368095485621411, + 0.49637785521313277, + 0.23297557464207136, + 0.38670580041019953, + 0.2938375641141268, + 0.30869421136408093, + 0.44160433650329867, + 0.3454675669897479, + 0.3162964852150054, + 0.8080090603505286, + 0.8021069402257913, + 0.8072973150356054, + 0.8232545962305555, + 0.5815654050039757, + 0.6839180263841848, + 0.7054206026787188, + 0.7540589755438492, + 0.7040014411577306, + 0.5864331397534621, + 0.6521337020448983, + 0.574921689314925, + 0.7701732879369142, + 0.6550791808338098, + 0.41578798130367, + 0.3213292666639967, + 0.27866674604337116, + 0.3134781694189441, + 0.529997580380654, + 0.3167090439625908, + 0.27543559217890085, + 0.395570196303803, + 0.37483238299811517, + 0.31784978528140667, + 0.34625772866465043, + 0.3839477398905444, + 0.41805741138619584, + 0.37825395367175696, + 0.31588206542929764, + 0.34657558308469555, + 0.3736302860448755, + 0.32166641696505327, + 0.3034982133127542, + 0.2976047438315895, + 0.5894715280931178, + 0.29597929285109287, + 0.5619760768298893, + 0.3548726932912426, + 0.4293853438967985, + 0.29736850587614483, + 0.5530711917943488, + 0.28542758299164567, + 0.4996344312901744, + 0.3126484547768251, + 0.2645975518280737, + 0.34691820199995227, + 0.3141126946415245, + 0.25026024186301943, + 0.17353915252347338, + 0.36230393065884425, + 0.31091755159788004, + 0.47681068263205156, + 0.4443837477457747, + 0.4998586478370958, + 0.47028585137558643, + 0.31695877032232533, + 0.457559260957162, + 0.41267489844571575, + 0.4300604559967845, + 0.4092793992270254, + 0.4119654462548125, + 0.36618028181843265, + 0.2014067102654769, + 0.41178944803337425, + 0.3231663592218265, + 0.2427239015020924, + 0.23414234879803883, + 0.48247263886262803, + 0.4961966715639958, + 0.5635578210565269, + 0.45655258165964735, + 0.27192049434816307, + 0.48812051620100533, + 0.4746943437062003, + 0.5026200515667232, + 0.2869464042636326, + 0.5149787957487996, + 0.42628515696136005, + 0.5042813418590708, + 0.23976136802338022, + 0.38139313229970023, + 0.3332996520338084, + 0.2544067409237207, + 0.49767097629095836, + 0.4275795905344213, + 0.5372579025942769, + 0.3032450579529877, + 0.588137334671062, + 0.29445675530372983, + 0.267814383134643, + 0.38891938261309805, + 0.27977610081052395, + 0.2610945811762738, + 0.24757261055058608, + 0.3563020534709061, + 0.6223805612582484, + 0.25450935907413036, + 0.24208283679039053, + 0.412743873157886, + 0.3730529588404298, + 0.37553939124967106, + 0.4372587044840974, + 0.36444030939848904, + 0.27002397905391023, + 0.3371335533852905, + 0.3111294851454432, + 0.5136065835512955, + 0.5262670316284184, + 0.5590892882073147, + 0.5425720353116111, + 0.2889651426490698, + 0.3758573242273551, + 0.4224990497303258, + 0.2490153767974669, + 0.32227673716524147, + 0.31312226511848046, + 0.15929781320442196, + 0.4615344200961703, + 0.3090818514087709, + 0.28498327966456083, + 0.514084412513317, + 0.24229261956844028, + 0.2039567207915151, + 0.42802664494303805, + 0.38007884563593974, + 0.2713623599324806, + 0.5999492392252698, + 0.3601412348795373, + 0.2875364509904171, + 0.27983870376501074, + 0.41687128689690134, + 0.2775386214958387, + 0.44310074422146845, + 0.28461976900316605, + 0.39557818570028114, + 0.3915802761235116, + 0.37498717525780184, + 0.3275275468160074, + 0.31952071325829856, + 0.3622852425960186, + 0.33564237137430714, + 0.4284550188146806, + 0.2703150177509484, + 0.35989171493370437, + 0.2838325493439909, + 0.3333124074550009, + 0.3757211528711273, + 0.37935477253383626, + 0.34057612513010826, + 0.4144309967576572, + 0.3068417356326242, + 0.30428582757937056, + 0.49685575063897447, + 0.26454063005147627, + 0.3226794403825938, + 0.3116863935970049, + 0.24067604680093468, + 0.29651751751800637, + 0.3144096043826474, + 0.25881176912988973, + 0.28053314764026455, + 0.2528361976484986, + 0.3024507755181197, + 0.3037884137724175, + 0.34393226597302584, + 0.351379020004715, + 0.4500609012127736, + 0.23344248424666542, + 0.390523141600794, + 0.453063618802551, + 0.2555344335126314, + 0.2552462023835772, + 0.5786298244965754, + 0.27021326724577593, + 0.24227846821691235, + 0.44848611590080867, + 0.43218829754820487, + 0.4328621863982767, + 0.3709819499726344, + 0.4887014199162194, + 0.4325756373464088, + 0.44024939319842743, + 0.3775063021902455, + 0.44288665358448176, + 0.3994808473118943, + 0.40204123242212586, + 0.4205499448564356, + 0.46300428068683963, + 0.4179775738188466, + 0.651326088195613, + 0.3038746305352926, + 0.5702059024532463, + 0.6378565907548546, + 0.6561652688188653, + 0.6613680823707367, + 0.4903341093537144, + 0.5206593185826245, + 0.3981545732834393, + 0.6768980053045965, + 0.4157648666697945, + 0.4854012298230691, + 0.31140539952458957, + 0.4952172831336981, + 0.42051718906200025, + 0.38551916835921074, + 0.43688037319959605, + 0.292495347269902, + 0.4164280257596879, + 0.2677786885906859, + 0.3411540039089718, + 0.33843641386961787, + 0.3282194016147351, + 0.38584450384089114, + 0.2560687820054689, + 0.31832859910653627, + 0.33577010670836455, + 0.3414785234068033, + 0.2960684397742017, + 0.3411063330063656, + 0.5627172973177393, + 0.4602512048160331, + 0.3901353424675498, + 0.4578603940123076, + 0.45418992837846417, + 0.2724325117616396, + 0.29474541606434956, + 0.42642132214722367, + 0.39240044489238785, + 0.2995425888849451, + 0.4193805095557267, + 0.5215557271254889, + 0.5234716443422461, + 0.38361529628708446, + 0.287190671605405, + 0.4494677390027554, + 0.2726976205619193, + 0.3114117198114368, + 0.48547170528696254, + 0.3097527989033301, + 0.8208812739333232, + 0.5017981914275009, + 0.395560743460428, + 0.3811401923135269, + 0.4407663910209879, + 0.4374527638535369, + 0.3813114832705218, + 0.49275138526931345, + 0.3444381063077833, + 0.502734640503666, + 0.3834109095529074, + 0.4543463040635196, + 0.4557726414085644, + 0.431719720035093, + 0.4204398830346049, + 0.3289739859705186, + 0.33026879602170517, + 0.515953248731877, + 0.5136298790995661, + 0.39741698933950875, + 0.5755336112715191, + 0.43955962332818577, + 0.5411955356628916, + 0.38690191592347867, + 0.4293963689499226, + 0.4146490640404582, + 0.5485963877078656, + 0.553132818116159, + 0.5142904473932031, + 0.5230088043591576, + 0.4944178052926756, + 0.4932812312926219, + 0.4143638948519841, + 0.37927631049544336, + 0.36141840102540157, + 0.2766466388307292, + 0.3699617038281772, + 0.3212839812833789, + 0.3164110077649127, + 0.3473550681576529, + 0.46263840111400506, + 0.33649262911959266, + 0.2805800804973866, + 0.3421814905673017, + 0.5406394920432206, + 0.6002311955768002, + 0.5178833848227314, + 0.5595072702223907, + 0.39017770660297213, + 0.5551547165496521, + 0.36567074763896645, + 0.6248406043165918, + 0.46389909631919524, + 0.5496267713411717, + 0.42509475788981255, + 0.5342145164525935, + 0.5342386407891729, + 0.4718291945896968, + 0.4402183605043771, + 0.4924254557322506, + 0.500923798481009, + 0.512282721538709, + 0.31973546676472764, + 0.3640300241518223, + 0.30394777744787727, + 0.3376725487609485, + 0.4926767071420345, + 0.5601456788262649, + 0.30591409139833975, + 0.5812947525379953, + 0.48964670352838774, + 0.4626653346868524, + 0.3712656526278231, + 0.44064443722018215, + 0.46665801822092123, + 0.41993569435288736, + 0.38697107322148316, + 0.8483752790379564, + 0.43281278743805773, + 0.4673642599404266, + 0.49001264750423285, + 0.49713029655897084, + 0.4536762278256603, + 0.39403921528246166, + 0.580065413853819, + 0.4342938932474418, + 0.37337186303937153, + 0.4316130421941578, + 0.41594528600066927, + 0.42772833625587336, + 0.40332594401850275, + 0.4010289871735104, + 0.43221346957170464, + 0.42916245805619996, + 0.26710508487062146, + 0.430051455122077, + 0.27158945408674956, + 0.3682631899243408, + 0.2270231208276083, + 0.4300769296660506, + 0.355882292589726, + 0.26473144814154304, + 0.4808952665959647, + 0.24995178388057265, + 0.255082907231023, + 0.5402408545720111, + 0.46376779444990496, + 0.44981068437973837, + 0.45483947380858925, + 0.3968055495566341, + 0.3975753255195352, + 0.5926230307457917, + 0.3381540168416202, + 0.5411149808217948, + 0.35854615003140167, + 0.24932906762126233, + 0.20379001196836652, + 0.38635370408335623, + 0.476675516237798, + 0.5279991031303816, + 0.5557425182423045, + 0.43681505743039345, + 0.4908394544623606, + 0.3661475885706246, + 0.4762086305018319, + 0.3917000692947281, + 0.37840798205439546, + 0.383356178437281, + 0.37122157527331, + 0.4426013936093364, + 0.3538674253097956, + 0.3934381844650589, + 0.4357203264848026, + 0.41579901580288836, + 0.45560554454098057, + 0.38938073441444754, + 0.3724593325385232, + 0.4725422581225654, + 0.4397794932954294, + 0.5142907106706728, + 0.48754231168788875, + 0.548436549500624, + 0.574395525535786, + 0.6637213527529632, + 0.6477506615334773, + 0.47212625164196736, + 0.5174575371156128, + 0.4138251083556582, + 0.5152369748484824, + 0.3829543682042884, + 0.4538249457418696, + 0.38920511875983993, + 0.5727410125657014, + 0.37937949327265696, + 0.5431203910691655, + 0.6501341401471779, + 0.6207517366496659, + 0.5524761807474071, + 0.6381665607525416, + 0.5312093160510525, + 0.4437987710107404, + 0.3919388202349217, + 0.3901323789422442, + 0.2821676044341448, + 0.3615176439213815, + 0.4439330715228962, + 0.5282444165669242, + 0.4116101965490707, + 0.34470745322613483, + 0.26321724470414964, + 0.45337869376232837, + 0.44547529287904286, + 0.3931565734328697, + 0.3432507438739251, + 0.49762163487167027, + 0.37203007065151017, + 0.42372494222989726, + 0.3543365625518254, + 0.4460226787647034, + 0.4470703905731159, + 0.30831869737767986, + 0.3621167731025311, + 0.43755863431563, + 0.37771926703533165, + 0.36527533990115074, + 0.38095668681927114, + 0.36208038119858, + 0.3040515700973917, + 0.3876903617249273, + 0.41925974193143095, + 0.39416714450141493, + 0.2454197029022812, + 0.5617082516713351, + 0.3775365027403402, + 0.48841285861984324, + 0.5140500607165045, + 0.3479618626493642, + 0.4743088982090142, + 0.47776133939414417, + 0.47841002916391395, + 0.6298295434267184, + 0.4078348985232799, + 0.4442642028079746, + 0.47308298215124694, + 0.36769850682368954, + 0.4681754362140943, + 0.4538830156613117, + 0.3196002573701824, + 0.43122636887984683, + 0.45307684041653984, + 0.4043570035210256, + 0.2742237734451696, + 0.4036285184224983, + 0.31122639816495606, + 0.3926922462000999, + 0.6196159491204755, + 0.5306131414237395, + 0.5397505285762053, + 0.5506459915342907, + 0.6237965793598779, + 0.5339759412220203, + 0.5854666670694362, + 0.5553114038397781, + 0.5924479557949311, + 0.4279879579562777, + 0.4096545846837315, + 0.40389519050088557, + 0.3008223238424943, + 0.25652452117975105, + 0.3367339557299235, + 0.3390999907415038, + 0.2903949296199836, + 0.26734959399648767, + 0.4118146751990565, + 0.2770906027042809, + 0.43514163377966836, + 0.3097063066029728, + 0.5126194486214717, + 0.5628651605199089, + 0.5000179835738618, + 0.26788819212301107, + 0.3450788758306986, + 0.566839390333325, + 0.3055839166329456, + 0.3074502500990224, + 0.2723953027208555, + 0.8349814670625854, + 0.5139996525710554, + 0.406156310289801, + 0.5722525703362079, + 0.5090121882084708, + 0.47612829470151485, + 0.3963027039441489, + 0.46020468636007233, + 0.5072477635861847, + 0.42574839359171807, + 0.5617265998225324, + 0.5098462597070871, + 0.2645738979711098, + 0.3952367729642327, + 0.2846391359550045, + 0.34583585118049825, + 0.2737487097412227, + 0.4843798201787302, + 0.3902195206017856, + 0.32338635837214397, + 0.4504739447560461, + 0.5041990535634352, + 0.32061113019094795, + 0.4465709106623206, + 0.29101777712532984, + 0.42763775661802556, + 0.4836042875574497, + 0.33393300207812265, + 0.316849165405062, + 0.25631466519520113, + 0.5398393867101509, + 0.28020691120898944, + 0.34359201771250303, + 0.49906851341274483, + 0.38067518699413005, + 0.3994928205642488, + 0.2918191427690567, + 0.3205110245885776, + 0.38829615321067656, + 0.3780999602164013, + 0.42793475126604386, + 0.3079222191021166, + 0.4703391692870983, + 0.2579884786490128, + 0.48107933683728593, + 0.4016870687774366, + 0.3861617238848734, + 0.45061937656984574, + 0.3610055169153962, + 0.4600192072132856, + 0.3197571742427281, + 0.48199625441067995, + 0.3665382478416125, + 0.3184484043680943, + 0.4156769338131044, + 0.2858923938532957, + 0.4877742901778042, + 0.22707012489670328, + 0.6640761800203264, + 0.668488693058607, + 0.5625057032663577, + 0.5617725745269212, + 0.6110715292444749, + 0.6057724999828247, + 0.2551823293561968, + 0.3473774439462686, + 0.31025645363615123, + 0.3573597807347815, + 0.305569018683027, + 0.3281027142382424, + 0.3666384077127682, + 0.40109791062324923, + 0.40815266537270145, + 0.3156539909061712, + 0.3553329942743843, + 0.2987542798802582, + 0.45812043603426444, + 0.2467269773305026, + 0.4033785596512331, + 0.42202507142407675, + 0.3688233755125372, + 0.46741144750434804, + 0.3796130255472003, + 0.5025252011439754, + 0.2605874103814281, + 0.46779654685232047, + 0.3703698969786674, + 0.361033993636956, + 0.3189457859106905, + 0.3672964524263672, + 0.3494663899229204, + 0.386711350891126, + 0.4855119361505802, + 0.44836281724034593, + 0.5291866063689292, + 0.3364515953396842, + 0.3809939652803358, + 0.39083160461082933, + 0.40561260261615345, + 0.46413977114751565, + 0.3844158543601661, + 0.2574566573559703, + 0.44562595762437696, + 0.3878101134146379, + 0.37481478513302546, + 0.4622674417029904, + 0.34978805380010675, + 0.28523820825127477, + 0.26479413464589024, + 0.32992693022833863, + 0.3569527883478946, + 0.30951383477697086, + 0.26747023324629426, + 0.35018189169738395, + 0.5099313740418769, + 0.6002160671937156, + 0.5505560678082468, + 0.5119309727529833, + 0.5339376618099468, + 0.47711474796176534, + 0.5498547876932884, + 0.4922596801863613, + 0.49602572869521594, + 0.33606515595311376, + 0.3795513162977857, + 0.4851369940970654, + 0.5947226081926471, + 0.17322656628394711, + 0.3200862336183772, + 0.33164921357053234, + 0.2901214650124442, + 0.32430559933849606, + 0.4280169798115721, + 0.3078397218091741, + 0.4436287821335698, + 0.7414672812221904, + 0.6090607059377825, + 0.7741915513526548, + 0.7890615140690442, + 0.6414192529173716, + 0.7353445769510376, + 0.7519528385478559, + 0.357205553974943, + 0.3909287553322009, + 0.408934226161827, + 0.5023232205451812, + 0.325676763443145, + 0.42109395476307104, + 0.34492433466394223, + 0.37480786129510346, + 0.373283240928675, + 0.3507858437882309, + 0.28039469822710156, + 0.4282701812509512, + 0.38613638473699, + 0.3720311789588014, + 0.413335637699123, + 0.309237312008155, + 0.3146570192786815, + 0.31103728332583547, + 0.3039061491625667, + 0.2497560374829527, + 0.32817233199480045, + 0.2528972926196461, + 0.5095170473044298, + 0.48634084131412186, + 0.33628860255774506, + 0.2714086096805488, + 0.24041217303183793, + 0.5159627403365428, + 0.5786091636777129, + 0.6348538279532693, + 0.32801546908748824, + 0.3574153495282835, + 0.2512869798789732, + 0.4600892050237668, + 0.3170528837899106, + 0.30050702715469707, + 0.47053581025033303, + 0.49555028324768435, + 0.45639138502749105, + 0.3135804093053669, + 0.43188535334121686, + 0.4356556769173969, + 0.4798043851920331, + 0.43303172157054326, + 0.48561894784974147, + 0.39547054076934585, + 0.43912512140917265, + 0.43607001143758345, + 0.45551884923154506, + 0.28413341147762694, + 0.35182438737537036, + 0.4393565439199528, + 0.4071096307016257, + 0.5675388065259486, + 0.287050856218554, + 0.5479333695936889, + 0.7037232334830903, + 0.5551584769234528, + 0.36853381373958694, + 0.3075713432015666, + 0.323771532449442, + 0.2647108156786409, + 0.3999886187813398, + 0.48901348403410994, + 0.43686308396215073, + 0.34155743596440313, + 0.29433560998155917, + 0.4746729792648558, + 0.504131375492465, + 0.5305047017066828, + 0.588593883518473, + 0.4629005562741682, + 0.5126861311401378, + 0.36163960095042136, + 0.47930343000885256, + 0.27596100993324907, + 0.2941039837124292, + 0.40020513702308697, + 0.37280246286886026, + 0.5316841558298584, + 0.5028465485647162, + 0.6017451009634676, + 0.6533524574059182, + 0.7485838590113518, + 0.5930188553744322, + 0.7624408391546788, + 0.4307536043360757, + 0.3935579975861479, + 0.39383563795725784, + 0.47128624016365883, + 0.3272039928272017, + 0.4771112065753143, + 0.4523157739223405, + 0.4884411522475387, + 0.46101761981249, + 0.4088048482182609, + 0.4243563374350463, + 0.44877362147025995, + 0.38048767419320706, + 0.4477229925095582, + 0.417402941826922, + 0.437762970807484, + 0.49492422515826157, + 0.518836959780924, + 0.4932597498561642, + 0.3727613504755776, + 0.4392053725785006, + 0.4774199968248216, + 0.32814230634192726, + 0.4236351552672359, + 0.39396972969223576, + 0.41505187657247117, + 0.3802331273346801, + 0.4288087170164782, + 0.36029762312881614, + 0.42120882892875416, + 0.43514295354902877, + 0.45884821865771075, + 0.30790142315211744, + 0.46389648883339135, + 0.4571433673650975, + 0.3948065896784617, + 0.5228179801792998, + 0.5112990708033188, + 0.7433368231271813, + 0.2515601746116562, + 0.2972509358935474, + 0.371182903079071, + 0.3489824627372732, + 0.26045973536319744, + 0.2823079665057222, + 0.2848920180564941, + 0.6698445189880766, + 0.5639241636789277, + 0.6347619295165089, + 0.3229130306154097, + 0.357271039053248, + 0.28618613032622303, + 0.4059423836009494, + 0.4111064520038975, + 0.23068161610529828, + 0.2792715732834082, + 0.414666182490373, + 0.38490472949463217, + 0.4350673129227015, + 0.39305247709216273, + 0.41227472925391645, + 0.36381221769323235, + 0.28728493900917934, + 0.6845265851122249, + 0.4720064973201995, + 0.6868061399452279, + 0.5705739212485141, + 0.25692679001520863, + 0.5032235922159353, + 0.4796621488295817, + 0.49187166885071587, + 0.4567106666386404, + 0.37644160361786183, + 0.46460524514293666, + 0.4806560364809694, + 0.495832972687339, + 0.45155539232403197, + 0.3360251379838873, + 0.417375945732496, + 0.2806415685065429, + 0.3223638698481457, + 0.6317450294646109, + 0.47730387574798083, + 0.5804526985263686, + 0.47115959968523036, + 0.45107040653928004, + 0.42964230311491325, + 0.3915702586045954, + 0.4811976956389881, + 0.3969965790271112, + 0.24937658382893238, + 0.3813033148148342, + 0.4658011846309071, + 0.2509266784835637, + 0.7583271261257825, + 0.41134812927303765, + 0.5666917445860922, + 0.724908340639369, + 0.6801445144009041, + 0.3614744838371647, + 0.8068045998276017, + 0.48192864706542526, + 0.6576154634350707, + 0.37460650413072755, + 0.4721962165584623, + 0.3686993842349737, + 0.40769710867824555, + 0.49060317052958363, + 0.515101771603406, + 0.5122288451621076, + 0.35360240001591114, + 0.38011566367088273, + 0.39033584794668574, + 0.4014817189770194, + 0.5223495258252265, + 0.30400826627415733, + 0.5004095858692238, + 0.43511386498105475, + 0.43978108016902145, + 0.5103587579997327, + 0.5669851733424665, + 0.45403952084816673, + 0.3884685154998693, + 0.43026857890514053, + 0.5124035542523773, + 0.4907004307542813, + 0.39765891797026287, + 0.33337227607435244, + 0.4670216260742242, + 0.5292495697735545, + 0.5257697261379848, + 0.3975398801390049, + 0.5755564147701391, + 0.5845722398894293, + 0.41727554498412495, + 0.49512660888257976, + 0.5250260422024834, + 0.4836234259605491, + 0.4422760383687512, + 0.35224777889981224, + 0.29059306473428637, + 0.3378911319910138, + 0.3750200101536547, + 0.29700899491008875, + 0.3878875894922474, + 0.3267727889474307, + 0.3686343776743894, + 0.369595744710412, + 0.30029576832699983, + 0.3276903595440367, + 0.31005821446602205, + 0.3567036483851455, + 0.40667721530994433, + 0.3121077698615519, + 0.2724655089812442, + 0.3915188241253822, + 0.5025330539262765, + 0.6285537109640941, + 0.2681749502265599, + 0.5133168120921008, + 0.5334067027581155, + 0.5310571420134604, + 0.5411989270756816, + 0.40967539062755026, + 0.4612647488725156, + 0.3743523304649373, + 0.49483735666020884, + 0.4076563214410686, + 0.6155694000017707, + 0.46600243748307035, + 0.49521283090874374, + 0.2710119777190512, + 0.3061816812883213, + 0.3846163742557754, + 0.4645877570883892, + 0.36834589375651694, + 0.5533997144413428, + 0.31847609495434215, + 0.6049754633470072, + 0.4909406940607072, + 0.5062570599023418, + 0.5010121730480611, + 0.487362922261515, + 0.40921997098262625, + 0.52037607365372, + 0.517040676148106, + 0.5114133376479314, + 0.5070745297525411, + 0.42179463539943185, + 0.33366662642821604, + 0.3003418528718989, + 0.36752478343440964, + 0.2347260421212329, + 0.37935487189419925, + 0.426517073694783, + 0.4175480369902703, + 0.3948538706265402, + 0.43412635954080075, + 0.40665873665240937, + 0.48507970599340816, + 0.4705145812429389, + 0.3251157767880516, + 0.4579419055174925, + 0.5402861816169876, + 0.34792409792942997, + 0.2937541894218538, + 0.3642678598894266, + 0.39197562781462886, + 0.4741212957903794, + 0.3033874574738478, + 0.44871281527334156, + 0.2593055501107889, + 0.4422674052044993, + 0.4844035366641211, + 0.5000834372357602, + 0.27962553495947917, + 0.30798356394917364, + 0.3174928754886504, + 0.26251998572910296, + 0.24075690795139335, + 0.34895777653437887, + 0.44090586878367816, + 0.26383440092035815, + 0.3618826627345573, + 0.4355240118097028, + 0.3105216156656863, + 0.3592080996224111, + 0.4403979103655228, + 0.5073963404212254, + 0.25690211472790814, + 0.44640633409515057, + 0.39084396448693887, + 0.3177334218641402, + 0.31536042014801285, + 0.4566030859245305, + 0.40793081183978047, + 0.24770552671925247, + 0.4793787229424211, + 0.41238784212542523, + 0.3712514419961703, + 0.49142637412393536, + 0.38438152906808404, + 0.3177730110468976, + 0.32925410871708444, + 0.31251754574894214, + 0.38496211288696003, + 0.34244239455390324, + 0.37917113248013146, + 0.42319977146722976, + 0.40114572461673154, + 0.49401126239914234, + 0.43473834816186424, + 0.3396274454291061, + 0.3948482437891119, + 0.44218234644684234, + 0.41115622586871536, + 0.43511273921949345, + 0.5087412588779964, + 0.4549021061811127, + 0.5164537843135022, + 0.5461391552492452, + 0.5368994919989246, + 0.4439780082597055, + 0.3762137330001833, + 0.24180885741549416, + 0.40792639843790807, + 0.4722681110304899, + 0.40280238556583076, + 0.43628747268758744, + 0.4538296246948038, + 0.30531128910294114, + 0.45246419548095274, + 0.3568314335551039, + 0.2662032797401213, + 0.4049804881112013, + 0.4136320365903084, + 0.44648112369571363, + 0.4522352922517631, + 0.296470715409446, + 0.5654942386143329, + 0.4998035695170447, + 0.4750516441539291, + 0.5098315738993098, + 0.5949619887794356, + 0.4612683623384708, + 0.30024866410176665, + 0.38166160387021636, + 0.3808201010712228, + 0.36505855146728866, + 0.3688732748397051, + 0.3287542361711014, + 0.34745637351267916, + 0.398266242980794, + 0.28291117975156865 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.19434689748773426, + "valid_tracks": 4, + "per_track_pearson": [ + 0.2300303240829569, + 0.17566396001940104, + 0.2224608354277837, + 0.14923247042079532 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.24803095702897368, + "valid_tracks": 626, + "per_track_pearson": [ + 0.27231582089713474, + 0.2677241312367823, + 0.2731927995543778, + 0.2687137005737709, + 0.2788941297188383, + 0.27289980620350346, + 0.23162056170503662, + 0.23850084893750123, + 0.26992725750413515, + 0.2683520665759873, + 0.26196641913990465, + 0.2618845719138431, + 0.27195058421416324, + 0.2706813144145775, + 0.2699353876830177, + 0.27237522221923866, + 0.2647222977910305, + 0.2636268120025775, + 0.2708360961972534, + 0.2736829961186364, + 0.27274158077385846, + 0.2595613564227042, + 0.27454335924496975, + 0.24427243270074678, + 0.2728952205602529, + 0.25269783910040394, + 0.25604296063729703, + 0.2758027467268656, + 0.2669628752347692, + 0.2661803541357149, + 0.2643684706078457, + 0.2666493833527229, + 0.24535825207334794, + 0.2769629214603228, + 0.2336273958573493, + 0.27115358134222883, + 0.22867158553206987, + 0.26758327555289607, + 0.2626716293172045, + 0.26687969904461195, + 0.27146922667048046, + 0.2614993853554436, + 0.2587513116713796, + 0.26253108038663714, + 0.2701669008456923, + 0.24626774745581625, + 0.26984570775591643, + 0.2657554746376538, + 0.267106582083437, + 0.26320949435204216, + 0.27606005165235054, + 0.24575805440827814, + 0.2654165719794094, + 0.2693543077085871, + 0.2624393447256217, + 0.25490644755049074, + 0.27306666820147696, + 0.2660211029124402, + 0.2731679655714126, + 0.2758680209786722, + 0.24663515417597992, + 0.2594025404987114, + 0.26897567250299764, + 0.25790762382211985, + 0.2717738799928476, + 0.2614543000808143, + 0.27080299912270717, + 0.25563446732672657, + 0.26722964629465373, + 0.2552629252367952, + 0.2682487506682517, + 0.2621204437181784, + 0.2761223479064488, + 0.26858899705541434, + 0.2806726873406096, + 0.26632081239759137, + 0.26554760980499725, + 0.256694610651995, + 0.2622017440219624, + 0.26602181745998366, + 0.2821775979413019, + 0.27426310528051784, + 0.2660462675139382, + 0.2623062042860358, + 0.2664496508733562, + 0.2772020113631655, + 0.26758805935176444, + 0.2728326433408433, + 0.2672478391649356, + 0.2817903862622046, + 0.28134420338330957, + 0.27058373842920747, + 0.266933583521282, + 0.2679642455320979, + 0.26866473288755427, + 0.2774913130730737, + 0.2714539812963249, + 0.2583176680094311, + 0.25900123638985884, + 0.2483282315208907, + 0.26110511152045524, + 0.2679135860433211, + 0.26829679176744736, + 0.27588467026058666, + 0.2682439313917212, + 0.25750890260765885, + 0.2399068822809116, + 0.2630315062586942, + 0.2765644279406713, + 0.28410541267699946, + 0.26845311766543567, + 0.2786836514428336, + 0.27009568780957494, + 0.2752912023216306, + 0.26092853171644287, + 0.2687303002063703, + 0.2618454924194706, + 0.2561918581781553, + 0.2677180354849155, + 0.2702137730294907, + 0.2835756873988683, + 0.26837510705073675, + 0.26547962264301256, + 0.2646605975804859, + 0.2483834612946373, + null, + 0.27160383756529577, + 0.26051554641768115, + 0.2640310936660329, + 0.2725670523590367, + 0.262688751732554, + 0.2734687133219157, + 0.26612283475445003, + 0.26117675875985663, + 0.2683071723569826, + 0.254498269254992, + null, + 0.26249640956281134, + 0.27593448250740094, + 0.24064841843173732, + 0.25597351479492797, + 0.26085282882918803, + 0.2556811920757296, + 0.2620757190985959, + 0.25174391376288313, + 0.26075702069430184, + 0.23991294053658035, + 0.2605764733320037, + 0.2446126218379392, + 0.26322251707741084, + 0.2567317050707485, + 0.26782631257970746, + 0.26285509948235825, + 0.2730553847679277, + 0.2675445847069808, + 0.27523430556120576, + 0.26770853070495637, + 0.25121025135772773, + 0.27126332073404574, + 0.264140678526974, + 0.2513563209993476, + 0.2601224065525457, + 0.28007209350619644, + 0.264133192439663, + 0.25933698468398103, + 0.26504600920707844, + 0.2536930285681056, + 0.2576662419261184, + 0.24079694702665527, + 0.2591330764115121, + 0.27098699488577144, + 0.2665594363968351, + 0.26932967298163685, + 0.27115909310040665, + 0.2696720780929236, + 0.24322062715056314, + 0.26698439636638666, + 0.2629114006861767, + 0.24425879128766442, + 0.2575272464179388, + 0.2703170527514872, + 0.25572593723161, + 0.23609008676734222, + 0.2660308868623686, + 0.2624607236608932, + 0.22269200979126916, + 0.27413540023780203, + 0.26846300451201877, + 0.27033218308007234, + 0.26975506049305764, + 0.2721746420343795, + 0.24965781032709536, + 0.2681153756960065, + 0.26641522313666516, + 0.26517422295356097, + 0.2635229211620608, + 0.2567097950667177, + 0.25006392848793074, + 0.2794801468216295, + 0.2828395702550134, + null, + 0.2857906548024045, + 0.26959741386998637, + 0.25948815514326096, + 0.2766442911732272, + 0.28216434014440156, + 0.27958835732846304, + 0.2549067247619295, + null, + null, + 0.2717717709457068, + 0.2657523052487946, + null, + 0.2800071058191787, + 0.23961990310310144, + 0.2585818791484134, + 0.26127442661266154, + 0.28042447817925475, + 0.2640216713962691, + 0.2654062774318545, + null, + 0.28040123352790114, + 0.2751506371754241, + null, + 0.23504915814477348, + null, + 0.24379061972531724, + 0.26815298611817706, + 0.2765025178954167, + 0.2642599646407008, + null, + 0.26945038602970284, + 0.26564315357535756, + 0.25749487552226086, + null, + 0.2663248783158478, + null, + 0.2532400068198011, + null, + 0.265640657356554, + 0.2587465982783141, + null, + 0.2563679101790133, + 0.2775625824119648, + null, + null, + 0.27922227655743465, + 0.27333840234264895, + 0.271984133426985, + 0.2656951870277286, + null, + 0.2599155896547115, + null, + 0.2648121685993104, + 0.27569810664192435, + null, + 0.21204042996135633, + 0.27267740523668127, + 0.27409867854913383, + null, + 0.25893370425964496, + 0.2777081187153143, + null, + 0.2675812924190926, + 0.2605939013101764, + 0.2734897338245497, + null, + null, + null, + null, + 0.2686382355239438, + null, + null, + 0.2818484400827702, + 0.2689428676455233, + 0.26473604291937103, + null, + 0.27130132806679524, + 0.2639251126965731, + null, + null, + 0.27129819929117827, + 0.27490573760451376, + 0.26086547184186415, + 0.2668654619214417, + 0.26581069256013656, + 0.2626389483028915, + 0.2565449951444472, + 0.2674982844254968, + 0.2673147549349295, + 0.26469076080065024, + 0.2693665676209954, + 0.2653902033224658, + null, + 0.2764712342479328, + 0.27267155180407593, + 0.2699583119842284, + 0.2623328872258948, + 0.2660126123415301, + 0.2600751017711937, + 0.2687541045981967, + 0.26666447290579537, + 0.26555378252252276, + 0.2748785774251441, + 0.26189907867020623, + 0.26554046471448783, + null, + 0.2689861565629275, + 0.28056969837380835, + 0.27159138376305986, + null, + 0.2655429973690136, + 0.219755091401886, + null, + 0.2739275243301438, + null, + 0.2789619967819601, + 0.27126381640205766, + 0.27640489371933585, + 0.25242934081730906, + 0.2762213650819192, + 0.2621744888942028, + 0.2534934548336538, + null, + null, + 0.2596575209378706, + 0.2671767156994473, + null, + 0.26898812274490763, + null, + 0.2726935242820821, + 0.2615274653537652, + 0.2637765289466666, + 0.2686149342423546, + null, + null, + null, + null, + 0.26932940784876513, + null, + 0.2661625843418952, + 0.28052550844366386, + null, + null, + 0.2725142336392078, + 0.2792081375674253, + 0.2655981765488456, + null, + 0.2730417144312505, + 0.2693667210676331, + 0.2571697669615537, + null, + 0.2744626001048977, + null, + null, + 0.27012101339006006, + null, + 0.2652136412866987, + null, + null, + 0.2645990281366666, + 0.277673681436347, + 0.274892666498519, + 0.276057140559771, + 0.2715444211204672, + null, + 0.26417343415054206, + 0.2580536848656972, + 0.23217131163230986, + 0.2127858533259073, + 0.20976074046388424, + 0.25870066749810283, + 0.23042731473213252, + 0.22312031349190956, + 0.2227607492443092, + 0.24062374166442643, + 0.23331686316964698, + 0.23378954667057117, + 0.24613539929439282, + 0.22764709336449257, + 0.2347597003361768, + 0.21348668548477623, + 0.20580214856199366, + 0.23266651938279048, + 0.23758361161573255, + 0.257860255415697, + 0.25409442755936085, + 0.25040550543572965, + 0.23041193705197152, + 0.20803661380566157, + 0.1994823177686988, + 0.24707915988136775, + 0.25039083487967767, + 0.22467794372985944, + 0.26408462778775, + 0.20591174889300343, + 0.20918299134650484, + 0.21336264339172029, + 0.21647227205123512, + 0.20471678105810046, + 0.22702370057199633, + 0.20011866471580314, + 0.2311130176646745, + 0.19305845440864763, + 0.2498438921861204, + 0.2387618239383674, + 0.220045850035412, + 0.22474129952822292, + 0.20315886879086667, + 0.22525537014940222, + 0.2290138656639789, + 0.23640117421506562, + 0.20981963671321738, + 0.21982579454341836, + 0.2054588829685557, + 0.23086849243081536, + 0.2195056770050082, + 0.22152213661784692, + 0.19989002559321936, + 0.2308166202779578, + 0.224802530420933, + 0.2166568631113456, + 0.2235854569315519, + 0.24197645620753208, + 0.23712365600625462, + 0.23686814788434907, + 0.24436126375121967, + 0.23318971394251406, + 0.2547187182855254, + 0.2621186738744414, + 0.24289909920640818, + 0.2385858253329326, + 0.2537206413533318, + 0.2353303057230759, + 0.2245991255318705, + 0.250780024490332, + 0.20552822432047235, + 0.24768453141254756, + 0.23708124458256322, + 0.22890820110153415, + 0.23522153418933514, + 0.23505742962101375, + 0.23408027131999334, + 0.22925037306639834, + 0.24002050616613463, + 0.2412447253948486, + 0.237000838304538, + 0.23539119043329332, + 0.2434025938358484, + 0.23849339866727118, + 0.23013453666210526, + 0.23044673220101033, + 0.24390043574945397, + 0.22625550778530273, + 0.23304570798309523, + 0.23114441638970193, + 0.24097649547537003, + 0.23185228132107322, + 0.22406459258911252, + 0.24103353607994962, + 0.2334751074102112, + 0.22244873711545887, + 0.23628599702343533, + 0.22955791771751194, + 0.24299523670900833, + 0.24596135528164448, + 0.2581004563798466, + 0.2160223675012688, + 0.25508016951713663, + 0.2417447414206004, + 0.24864183890565614, + 0.24954965357244296, + 0.2428870875701345, + 0.20283664182868846, + 0.23807059465003155, + 0.22647638388613422, + 0.2270502545475303, + 0.2182614664527888, + 0.22942570528047465, + 0.22794592123930854, + 0.22699246730948222, + 0.2298790612006222, + 0.23632461570270274, + 0.21743435351529852, + 0.223011015464978, + 0.19970620663492236, + 0.22890658731709446, + 0.2311690476969602, + 0.23315942505191517, + 0.24003426622307125, + 0.2465448330093548, + 0.22867437407890082, + null, + 0.2537436545474858, + 0.2165969547004278, + 0.22907490044235548, + 0.22678545762920652, + 0.23216964956083033, + 0.24298867313501196, + 0.25365672217468443, + 0.24004930471886665, + 0.250405032028309, + 0.2413475933448209, + null, + 0.24154884310011507, + 0.2344692345195481, + 0.24186751335749104, + 0.24323878825320008, + 0.22754692027082546, + 0.234126994394225, + 0.23065607691566406, + 0.2315418470319574, + 0.22732675299927282, + 0.2300575537697295, + 0.2361223857675206, + 0.23415822972934378, + 0.22828197310487752, + 0.2198822283664136, + 0.24684270357094829, + 0.23955845799182232, + 0.22887914351261415, + 0.2009657336143419, + 0.21683416188428714, + 0.20883292787211682, + 0.24203182512114257, + 0.24707478257441454, + 0.24121768699191423, + 0.22140977825684854, + 0.23999652789906722, + 0.25238768065538225, + 0.2182096886973075, + 0.2248104159217256, + 0.23228282194291136, + 0.24928988882368544, + 0.23521685416149704, + 0.1995073338697476, + 0.2100463329997974, + 0.23943769537847487, + 0.23659582998012327, + 0.2452401107659716, + 0.22728144718724347, + 0.2212880115978032, + 0.2308805211700875, + 0.25333146033440185, + 0.25599752588068, + 0.23004123895856277, + 0.21256638935512914, + 0.2238367719128466, + 0.22145879214210631, + 0.21630708213917324, + 0.24978152961983213, + 0.24461876628975288, + 0.22940905196730177, + 0.24683479728808758, + 0.21018990976379154, + 0.23136712676167195, + 0.22844011892219823, + 0.23748352597045222, + 0.2559520863958947, + 0.24636795933566502, + 0.21236247648127024, + 0.21473067341944765, + 0.20640886790852073, + 0.21348529294189245, + 0.21142823052552182, + 0.2535076980217096, + 0.24606247555099006, + null, + 0.2593841273651373, + 0.2381264310647051, + 0.21216729468634357, + 0.2329098792306838, + 0.24899408949372667, + 0.2464365230666218, + 0.21838843988257628, + null, + null, + 0.2502534745848468, + 0.23665445054461934, + null, + 0.23678399119397808, + 0.1955453521699545, + 0.22778385499850523, + 0.20479853064995993, + 0.21549334000189177, + 0.23225317291010641, + 0.2518361527474143, + null, + 0.24691339082544106, + 0.23653033535682103, + null, + 0.20106625438268436, + null, + 0.22870616759607185, + 0.24608143058487467, + 0.23442161195090983, + 0.22315613203591028, + null, + 0.20704886543784254, + 0.21133325640041925, + 0.22885649975880012, + null, + 0.1958595201597788, + null, + 0.22555258527943392, + null, + 0.2230552892318148, + 0.2242346517139891, + null, + 0.2111040589737379, + 0.24926479207115812, + null, + null, + 0.2510293641751359, + 0.23747954413473443, + 0.22677682707404778, + 0.21327456101501138, + null, + 0.21335609584091408, + null, + 0.23026782048250088, + 0.22947870174768364, + null, + 0.17229941538859397, + 0.21600728983137824, + 0.20925058119577997, + null, + 0.22103794591790196, + 0.2268810959323616, + null, + 0.25450329352244744, + 0.2577174757156491, + 0.2584832750012574, + null, + null, + null, + null, + 0.26768949612715254, + null, + null, + 0.23217915040346704, + 0.22963588960369893, + 0.25545105712539706, + null, + 0.26710885195945105, + 0.2538862719723559, + null, + null, + 0.236477524410046, + 0.2443789124935541, + 0.25074046122437915, + 0.222131728980928, + 0.22532436181035387, + 0.2263717281296957, + 0.20028569846902355, + 0.229727856211307, + 0.20262910126734038, + 0.21114554504035293, + 0.23125513636073136, + 0.2062758008703434, + null, + 0.23433715678457775, + 0.21173723069790665, + 0.24353881216691725, + 0.23939680150698195, + 0.23046982351941786, + 0.24191273753319464, + 0.22908351593441287, + 0.23990881772499803, + 0.2452546073452355, + 0.21640180383252364, + 0.2131393842323906, + 0.20517305571884242, + null, + 0.21617006936178357, + 0.2603125973656437, + 0.2633219655154887, + null, + 0.22425733881976662, + 0.19184797087810898, + null, + 0.23198599848347368, + null, + 0.24407112190945912, + 0.22348414264664673, + 0.2464454896673821, + 0.2237173903962362, + 0.24138708013346669, + 0.20200853571324662, + 0.18485483952769816, + null, + null, + 0.25138722193243374, + 0.24134663916483795, + null, + 0.22521863312157178, + null, + 0.23512891778425377, + 0.22852791556754903, + 0.22377710662419717, + 0.24540502441280948, + null, + null, + null, + null, + 0.23490595170540343, + null, + 0.22996303410798768, + 0.2434218496079853, + null, + null, + 0.23505696598255227, + 0.21287695423511763, + 0.19661218035406786, + null, + 0.22018582858079075, + 0.22872510341068752, + 0.21750051263074552, + null, + 0.24664081933215373, + null, + null, + 0.2294001100130723, + null, + 0.2043108921340926, + null, + null, + 0.20894465939044243, + 0.2496851316983678, + 0.2541323167533626, + 0.2545311080644576, + 0.25125645820336434, + null, + 0.23230445288348903 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083, + 0.6893167232918083 + ] + }, + "junctions": { + "mean_pearson": 0.18977917292225174, + "valid_tracks": 626, + "per_track_pearson": [ + 0.3216333674965654, + 0.3136189659324995, + 0.3038355436386295, + 0.21204145063184254, + 0.37216997353786335, + 0.28140278145489767, + 0.0019329316348773362, + 0.20711668163436728, + 0.24793378023432683, + 0.25617014194804966, + 0.20194370830985728, + 0.17507417249536325, + 0.09621728146009603, + 0.09988550679277527, + 0.21307471636359476, + 0.23881033125421555, + 0.21368476848664686, + 0.08524073772240333, + 0.26821288186474374, + 0.2564609236164036, + 0.2435733779404811, + 0.27608899462817776, + 0.15919850120436768, + 0.2001824612890163, + 0.22319501733714292, + 0.3483341207774149, + 0.2402663360282219, + 0.32541353786886223, + 0.12052167592569672, + 0.1740222415010137, + 0.2918569765154138, + 0.30077285355096894, + 0.21485240121453905, + 0.2549820052372267, + 0.2393701535498509, + 0.27193911734473575, + 0.2722866505279498, + 0.2859620560489042, + 0.25350647073680554, + 0.15128938736191347, + 0.2196181262752483, + 0.1756106408435176, + 0.2114838409884864, + 0.18943575657648315, + 0.22422535421587642, + 0.17264611080424136, + 0.13195889540638223, + 0.19360746448079602, + 0.1727155342903691, + 0.1835693110729464, + 0.1502691952619319, + 0.2417479470683585, + 0.1354618312571428, + 0.26797628276232327, + 0.14981514138456245, + 0.2268467820104563, + 0.15396832561894394, + 0.15436939915081527, + 0.2118477309159868, + 0.1722582990769846, + 0.08766887557444947, + 0.1913539833438994, + 0.1791614411161464, + 0.13852222058824873, + 0.09469586890398177, + 0.3771287712502111, + 0.24365330139373242, + 0.24717612790638047, + 0.20288127435768136, + 0.16641220505090482, + 0.16954469731509403, + 0.23971415311358704, + 0.15654740407972786, + 0.09814539861963668, + 0.35312555783772553, + 0.4466367855211089, + 0.24624787641683799, + 0.20118539287079204, + 0.14446137942805393, + 0.34149163864645105, + 0.2558575869622052, + 0.2855044432719575, + 0.1893819483731988, + 0.1924647207710372, + 0.26236301641502585, + 0.27006383270746154, + 0.27964121447417, + 0.3173308793915824, + 0.3035673130489712, + 0.293649403651565, + 0.29620751791444877, + 0.2634739766949231, + 0.21790600095174867, + 0.32414185296632464, + 0.3100725486000222, + 0.30297544175443325, + 0.26840861890951717, + 0.21522882527277454, + 0.24106322575436834, + 0.20664838786448583, + 0.18862802716690194, + 0.2555495107385728, + 0.3555099486853753, + 0.22912347665342683, + 0.22374471843661173, + 0.1192179544935134, + 0.06439455886758857, + 0.3008954878135388, + 0.31943646961301936, + 0.3107465417901167, + 0.15153188873890938, + 0.3484558568944006, + 0.4104203140998163, + 0.4549623874413807, + 0.3398469484262059, + 0.13327454548316758, + 0.14767395633816724, + 0.19622634270304842, + 0.14093192678704392, + 0.2711249271087539, + 0.2729314230635147, + 0.19771431795853148, + 0.1821297653091258, + 0.2772007653362065, + 0.08067867040364618, + null, + 0.3657263718279279, + 0.2071567612741221, + 0.1452166585811016, + 0.18761738111686807, + 0.24034559119940424, + 0.27729153891359964, + 0.21375890101648043, + 0.28922111160131003, + 0.354696540604167, + 0.1456657524067388, + null, + 0.26237891173378675, + 0.2448570922863085, + 0.23584805108433216, + 0.26971607382149493, + 0.23887674773190024, + 0.2575661659118944, + 0.2606050720954229, + 0.22168073950167083, + 0.217167578775048, + 0.285096785374917, + 0.2681288183387926, + 0.18603605772100246, + 0.16517293881760797, + 0.158810337410604, + 0.22632654069146055, + 0.28732027335895033, + 0.2960977882460214, + 0.4312934318022197, + 0.29780897610808144, + 0.3865595651864297, + 0.25563712822385615, + 0.2066356559746874, + 0.2083837234393189, + 0.17335734240057932, + 0.2631691901716762, + 0.22448864975628643, + 0.25977530782115343, + 0.1049884628450304, + 0.29160967499982643, + 0.25720807321508316, + 0.29275858242161673, + 0.15126783337222285, + 0.18764724830365329, + 0.26112098875288653, + 0.21254896751029417, + 0.314722868800861, + 0.14605460646584398, + 0.1583557757299012, + 0.17890226628972497, + 0.23317536635270375, + 0.2365825468100717, + 0.19233147569211986, + 0.23659738755903786, + 0.2221844242171598, + 0.17465096917857173, + 0.06430000313587576, + 0.17133819773465297, + 0.34435523675072355, + 0.23734567426760889, + 0.28335779261979144, + 0.21570044127756688, + 0.27333150724409844, + 0.3221465334269636, + 0.34171255701683195, + 0.38591336529107695, + 0.18142613337953595, + 0.1809318855115126, + 0.1465208015464901, + 0.1642715725401279, + 0.14665932426850972, + 0.17596771475367357, + 0.2536268169344217, + 0.3746180672532256, + null, + 0.14236127802079981, + 0.2389028876079445, + 0.26941812613388844, + 0.35546302297385746, + 0.14847549684276493, + 0.14052020257137776, + 0.1287952312215947, + null, + null, + 0.22894742156518796, + 0.32642019289325014, + null, + 0.13679535059729472, + 0.1811234766739492, + 0.11187718588118795, + 0.14698471010482392, + 0.14110797674551612, + 0.23317839373496235, + 0.07466985577191125, + null, + 0.29572875556979383, + 0.21486443716541578, + null, + 0.0900959261944719, + null, + 0.2809640715450688, + 0.18481216031116343, + 0.18662567795072224, + 0.19064870269098516, + null, + 0.1316640675822669, + 0.11118438849552062, + 0.23950916104403794, + null, + 0.2000274821311093, + null, + 0.16757743194779026, + null, + 0.13300908880791712, + 0.17361934165179724, + null, + 0.1581305179924199, + 0.220684057075683, + null, + null, + 0.2737323351945348, + 0.2555676080543187, + 0.052769920034536635, + 0.11121411548261402, + null, + 0.2432209661466074, + null, + 0.23144721968533336, + 0.21899698527019787, + null, + 0.24013346694951726, + 0.29869389006196984, + 0.2861451877833789, + null, + 0.22660414360138342, + 0.2919384723941013, + null, + 0.34146346378403536, + 0.29687908071673286, + 0.3037818353607256, + null, + null, + null, + null, + 0.1852274669822888, + null, + null, + 0.1659750537844354, + 0.2337155104744866, + 0.2797704043882924, + null, + 0.15914672613896072, + 0.31343754364482346, + null, + null, + 0.23402738654655023, + 0.11648795886272284, + 0.26306290046052094, + 0.12728709640838914, + 0.16343026586141723, + 0.17922212571461243, + 0.14281741128568728, + 0.22627898950622546, + 0.04789505233373705, + 0.10406180463529946, + 0.3498621007146686, + 0.22233054861052728, + null, + 0.28187819591760566, + 0.23511936681582593, + 0.09581519679587482, + 0.15316470781898306, + 0.1152551338262349, + 0.08779375076230102, + 0.19565783016275468, + 0.16808738140792998, + 0.24589231769780145, + 0.1655303174714438, + 0.25380949887104665, + 0.20967717918931938, + null, + 0.12979924506342871, + 0.21469481222337713, + 0.2380669803762474, + null, + 0.2401745290987326, + 0.1942896280564362, + null, + 0.22720455914378812, + null, + 0.19129439686230995, + 0.1684419785821355, + 0.25409414100711186, + 0.10744312377844042, + 0.19594538399559466, + 0.24730402087560244, + 0.27758674950399703, + null, + null, + 0.11456375461008823, + 0.1620465104041164, + null, + 0.15392670041426354, + null, + 0.21852095822593, + 0.1450578697251766, + 0.12064445124902991, + 0.17443513948321615, + null, + null, + null, + null, + 0.11466928302993541, + null, + 0.11158859815827246, + 0.2453843288887007, + null, + null, + 0.2302118135241787, + 0.3127418970896703, + 0.28271806741197164, + null, + 0.3629747617328789, + 0.24485357337119365, + 0.11392313666927896, + null, + 0.24969092829598527, + null, + null, + 0.31840077019704455, + null, + 0.2596986121893002, + null, + null, + 0.20142179067795127, + 0.21322210917384207, + 0.2016040627156727, + 0.20006596892026796, + 0.16669771268228548, + null, + 0.1560315198112123, + 0.10874525672373807, + 0.09376866125960925, + 0.3385488732911665, + 0.06940048492946838, + 0.21121430939550517, + 0.13381300483760686, + 0.14968067388948425, + 0.10569540295937495, + 0.07337482551309364, + 0.10784822046062349, + 0.10359681437571028, + 0.1973046541705367, + 0.16044274994443958, + 0.14775811263579183, + 0.06992762777156111, + 0.2154848202942266, + 0.04992717389766735, + 0.1666111408571445, + 0.11286996075489238, + 0.21146707401023596, + 0.11331885252238427, + 0.21848615651522799, + 0.1724324543112217, + 0.321081789227494, + 0.17879129165241817, + 0.2117803941258728, + 0.11057269372474846, + 0.3077849096514202, + 0.123786968551335, + 0.033354359136806115, + 0.10889658858504304, + 0.20915305494398237, + 0.09397757163530726, + 0.1835667051489001, + 0.0867067583696633, + 0.2011176912400814, + 0.048934761428567695, + 0.15033827359228616, + 0.1621612162268884, + 0.21805807021949025, + 0.1656296271143588, + 0.09980239916630972, + 0.13049876134867122, + 0.07922420203747606, + 0.16347721340875956, + 0.1602372189352008, + 0.15423411768400544, + 0.2067237704675603, + 0.18118100487102612, + 0.15861422327459732, + 0.17488053116886182, + 0.11761740120122616, + 0.11668099656382096, + 0.10728951051536252, + 0.05116373224855433, + 0.19304902350116532, + 0.22423940723479457, + 0.08894694047122094, + 0.23203522353082334, + 0.21546605731702576, + 0.2514142804781679, + 0.01887595265749763, + 0.10357614845864523, + 0.13728625623378135, + 0.05842987927017165, + 0.048849845520419634, + 0.06788199680287749, + 0.08351764860665788, + 0.1291253121246669, + 0.06961919666823813, + 0.08593622417124028, + 0.15038346240150638, + 0.11499146106949583, + 0.0619941996870056, + 0.1129629306999203, + 0.1325557930356163, + 0.1091138921485364, + 0.07716693551904445, + 0.1373764340094438, + 0.0712025132494498, + 0.08471522502011576, + 0.18507063630550744, + 0.10379681379076136, + 0.0962395654704872, + 0.13669839014615984, + 0.1386913079646932, + 0.17655587945272927, + 0.0964150858074871, + 0.11495205735584037, + 0.06817403392278655, + 0.09446738626800662, + 0.05254851088683332, + 0.06802834504420373, + 0.0798484933550454, + 0.1559948645746764, + 0.11942265921259479, + 0.08983580357283545, + 0.11975303082908165, + 0.1730041542817819, + 0.21378175550434886, + 0.11079054204058308, + 0.1852087753552454, + 0.1138086632408482, + 0.22560584316449037, + 0.18588050362670697, + 0.08544377416628825, + 0.018911631627626654, + 0.1203755388046195, + 0.26068035645541443, + 0.16187109522894422, + 0.10489316821716542, + 0.15809654088287764, + 0.15830149742480432, + 0.18460194812589442, + 0.19061506526588926, + 0.042201753739913056, + 0.2492215196204155, + 0.08213771971857967, + 0.15819818136586644, + 0.0723991337791732, + 0.2310612907135368, + 0.13349219106403096, + 0.07794541849253014, + 0.13855464508623666, + 0.331724158614435, + null, + 0.27777539791077827, + 0.3583196742418101, + 0.09234456324499506, + 0.3605848547384348, + 0.21100842714814944, + 0.02768024518479059, + 0.09224710205517314, + 0.2698466940403014, + 0.07852981636419881, + 0.1295614357701043, + null, + 0.1308150281593029, + 0.10065918590270478, + 0.10534512650864808, + 0.09420868924323537, + 0.16378917400980658, + 0.1325100738474438, + 0.15527959699872237, + 0.35200015269634677, + 0.02928001257986559, + 0.14944445157725453, + 0.20384580644344166, + 0.15871290594370357, + 0.09951833040857176, + 0.23087146449407897, + 0.2466073285877475, + 0.11660116532905243, + 0.2558433764862375, + 0.307784910314055, + 0.3524575210017096, + 0.18659581835213812, + 0.1820576405281222, + 0.16277897478509604, + 0.062131955447802936, + 0.08109987374364183, + 0.15526701384934716, + 0.19059410710562774, + 0.18562895603832585, + 0.2857972695796209, + 0.18785834198841947, + 0.11467960642648624, + 0.09692461487143014, + 0.07828425371136878, + 0.17281731738491385, + 0.149474060832407, + 0.13344420112328279, + 0.15663170858137854, + 0.10888555178839576, + 0.20119741855073753, + 0.21606989260257495, + 0.35195039656897925, + 0.2198516262415072, + 0.10930323496094896, + 0.10641366831303174, + 0.0985734411939869, + 0.05274682985446133, + 0.2938528025370241, + 0.15547287015081077, + 0.11131540469463713, + 0.21216833819389871, + 0.1589139810970255, + 0.02421386568124322, + 0.13116582930465887, + 0.10235708680505828, + 0.21058829442533464, + 0.2654198130927191, + 0.0750648152144313, + 0.15840694191054924, + 0.1513528764272214, + 0.19079175495915474, + 0.15912999792543941, + 0.20250669165102447, + 0.24568292075159406, + 0.18676323059006802, + null, + 0.22604592722879188, + 0.16067183375975505, + 0.335586763852076, + 0.25496057792539273, + 0.2800728310637327, + 0.11290631004642487, + 0.23260491525519122, + null, + null, + 0.1786517101761674, + 0.3084060567500825, + null, + 0.14037640911602964, + 0.159735328829634, + 0.16151396449019317, + 0.11401085953030428, + 0.15873538412111154, + 0.20063938287076355, + 0.18579534183479232, + null, + 0.138905458995499, + 0.14655576820677485, + null, + 0.2551940048497472, + null, + 0.3221728590111676, + 0.09399688800509458, + 0.12240854518977622, + 0.03895766400311328, + null, + 0.17834490452792037, + 0.1999055645312408, + 0.13166016397507205, + null, + 0.25493574701735805, + null, + 0.2426362786137851, + null, + 0.23628085555272715, + 0.11503235022824816, + null, + 0.31734718335591094, + 0.20509325938367423, + null, + null, + 0.11712926724992383, + 0.11732015425151748, + 0.16612900943756137, + 0.18693686859847175, + null, + 0.1526344167020823, + null, + 0.08103452224696227, + 0.13893563996270597, + null, + 0.2184762607180414, + 0.22701451967503053, + 0.11624422311927424, + null, + 0.24264824400064658, + 0.2594776726605749, + null, + 0.1280178197307547, + 0.2074450269056223, + 0.11559801451802779, + null, + null, + null, + null, + 0.08394326022188285, + null, + null, + 0.23437425764569486, + 0.14243837932251743, + 0.17228218989414404, + null, + 0.0554282808805561, + 0.19965508945589844, + null, + null, + 0.18883795045919663, + 0.06846605193428935, + 0.20321647368496062, + 0.14620051421733124, + 0.05631972534222369, + 0.18005029941969022, + 0.04930345947963627, + 0.12166484258141233, + 0.1557714853721904, + 0.18087560625168247, + 0.21788602947776994, + 0.14668462743387545, + null, + 0.11296413286095575, + 0.16370274742143012, + 0.22876502478484323, + 0.20445519560718325, + 0.14238552074668748, + 0.30747116348371006, + 0.2065911805673315, + 0.11600077147940124, + 0.1974158022792838, + 0.21472206032077276, + 0.13672082837959307, + 0.06027448138427354, + null, + 0.0832752752689734, + 0.2547636196212411, + 0.16050978795559756, + null, + 0.14259566871808582, + 0.17059456221889743, + null, + 0.20361774975629365, + null, + 0.2587946912428467, + 0.2015942188058093, + 0.13489297597446412, + 0.3127796838224713, + 0.09967794611345525, + 0.30199060416059736, + 0.18708828258757723, + null, + null, + 0.14881252889162613, + 0.27849531152306345, + null, + 0.060706119237584226, + null, + 0.13888053613421142, + 0.26123378017159327, + 0.22857198770542775, + 0.1339623304273412, + null, + null, + null, + null, + 0.09608669356812757, + null, + 0.19986475128688336, + 0.18272432957013776, + null, + null, + 0.07528267078655337, + 0.14176679710348553, + 0.22003608091971935, + null, + 0.09590061600513425, + 0.15330736041103885, + 0.28063217440731997, + null, + 0.23072085294843167, + null, + null, + 0.07777113451027554, + null, + 0.14239650028852338, + null, + null, + 0.0665007895975951, + 0.13233800034922874, + 0.24381951739990465, + 0.2569330922499122, + 0.19279252762637797, + null, + 0.21004428607126818 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9437086092715232, + 0.9605263157894737, + 0.9608540925266904, + 0.9462025316455697, + 0.9602649006622517, + 0.9457627118644067, + 0.9471830985915493, + 0.9513888888888888, + 0.9583333333333334, + 0.950354609929078, + 0.9461279461279462, + 0.9389830508474576, + 0.9383116883116883, + 0.94375, + 0.95, + 0.9539473684210527, + 0.9405594405594405, + 0.9487179487179487, + 0.9522292993630573, + 0.9509803921568627, + 0.9388489208633094, + 0.9451612903225807, + 0.9383116883116883, + 0.9553903345724907, + 0.9466192170818505, + 0.9671052631578947, + 0.9647435897435898, + 0.9491525423728814, + 0.9493243243243243, + 0.9448275862068966, + 0.9452054794520548, + 0.9579288025889967, + 0.9552238805970149, + 0.9544072948328267, + 0.9633699633699634, + 0.9526813880126183, + 0.9541984732824428, + 0.9498432601880877, + 0.9265734265734266, + 0.9395973154362416, + 0.9419795221843004, + 0.9587628865979382, + 0.950381679389313, + 0.9592592592592593, + 0.9540983606557377, + 0.9527027027027027, + 0.959409594095941, + 0.942652329749104, + 0.9468085106382979, + 0.9659863945578231, + 0.9416058394160584, + 0.9595588235294118, + 0.9507575757575758, + 0.9591836734693877, + 0.9440298507462687, + 0.9591836734693877, + 0.9480968858131488, + 0.9602649006622517, + 0.9452054794520548, + 0.9471830985915493, + 0.9448275862068966, + 0.9508196721311475, + 0.956953642384106, + 0.9590443686006825, + 0.9563758389261745, + 0.9651898734177216, + 0.9425675675675675, + 0.959731543624161, + 0.9590443686006825, + 0.9509803921568627, + 0.9466192170818505, + 0.9516129032258065, + 0.9528619528619529, + 0.9635761589403974, + 0.946236559139785, + 0.9661016949152542, + 0.9517684887459807, + 0.9464882943143813, + 0.9517241379310345, + 0.9523809523809523, + 0.9470198675496688, + 0.9556962025316456, + 0.9452054794520548, + 0.9546925566343042, + 0.9508771929824561, + 0.9442724458204335, + 0.9486301369863014, + 0.9575163398692811, + 0.95, + 0.9493670886075949, + 0.9512195121951219, + 0.9450171821305842, + 0.9508771929824561, + 0.9523809523809523, + 0.9594594594594594, + 0.9583333333333334, + 0.9457627118644067, + 0.9377162629757786, + 0.9437086092715232, + 0.9493670886075949, + 0.9466666666666667, + 0.9508196721311475, + 0.9537366548042705, + 0.9488817891373802, + 0.9692832764505119, + 0.9533333333333334, + 0.9288389513108615, + 0.959866220735786, + 0.9534883720930233, + 0.9527027027027027, + 0.9464285714285714, + 0.9463087248322147, + 0.9642857142857143, + 0.9525423728813559, + 0.9591836734693877, + 0.9358108108108109, + 0.9448275862068966, + 0.9493243243243243, + 0.927797833935018, + 0.9522184300341296, + 0.9464285714285714, + 0.9525423728813559, + 0.9448529411764706, + 0.9605263157894737, + 0.9433198380566802, + 1.0, + 0.9619377162629758, + 0.9477351916376306, + 0.9545454545454546, + 0.9477611940298507, + 0.9601449275362319, + 0.9517241379310345, + 0.9651567944250871, + 0.9473684210526315, + 0.9508771929824561, + 0.9531772575250836, + 1.0, + 0.9627118644067797, + 0.9626865671641791, + 0.9671052631578947, + 0.9671532846715328, + 0.9440298507462687, + 0.9595588235294118, + 0.9482758620689655, + 0.9368421052631579, + 0.9427480916030534, + 0.952755905511811, + 0.9479166666666666, + 0.9411764705882353, + 0.950530035335689, + 0.9444444444444444, + 0.9432624113475178, + 0.945054945054945, + 0.9577922077922078, + 0.9620689655172414, + 0.9553264604810997, + 0.9583333333333334, + 0.9638989169675091, + 0.950530035335689, + 0.948905109489051, + 0.9356060606060606, + 0.9513888888888888, + 0.9467084639498433, + 0.9407665505226481, + 0.9433962264150944, + 0.9532374100719424, + 0.9480968858131488, + 0.9735099337748344, + 0.9509433962264151, + 0.9508771929824561, + 0.9560439560439561, + 0.9419795221843004, + 0.9624060150375939, + 0.9554794520547946, + 0.9538461538461539, + 0.9649805447470817, + 0.9420289855072463, + 0.9547038327526133, + 0.9522058823529411, + 0.9475524475524476, + 0.9442508710801394, + 0.9477611940298507, + 0.948339483394834, + 0.946236559139785, + 0.954225352112676, + 0.9630996309963099, + 0.9452054794520548, + 0.9393939393939394, + 0.931899641577061, + 0.9606557377049181, + 0.9409722222222222, + 0.9595588235294118, + 0.9428571428571428, + 0.9330985915492958, + 0.9442508710801394, + 0.9363957597173145, + 0.9227941176470589, + 0.9409722222222222, + 0.9411764705882353, + 0.9433333333333334, + 1.0, + 0.9506578947368421, + 0.9389830508474576, + 0.9534883720930233, + 0.9425675675675675, + 0.9614147909967846, + 0.9469964664310954, + 0.9375, + 1.0, + 1.0, + 0.9504950495049505, + 0.9512987012987013, + 1.0, + 0.9448275862068966, + 0.9304029304029304, + 0.948905109489051, + 0.9385665529010239, + 0.9470198675496688, + 0.9344827586206896, + 0.9429530201342282, + 1.0, + 0.9440993788819876, + 0.9381443298969072, + 1.0, + 0.94140625, + 1.0, + 0.9454545454545454, + 0.9403508771929825, + 0.9401408450704225, + 0.9393939393939394, + 1.0, + 0.9420289855072463, + 0.9436619718309859, + 0.9403508771929825, + 1.0, + 0.9427609427609428, + 1.0, + 0.9383116883116883, + 1.0, + 0.95, + 0.9429530201342282, + 1.0, + 0.9405204460966543, + 0.940983606557377, + 1.0, + 1.0, + 0.9432624113475178, + 0.9444444444444444, + 0.934640522875817, + 0.9393939393939394, + 1.0, + 0.9399293286219081, + 1.0, + 0.9383116883116883, + 0.9415807560137457, + 1.0, + 0.9538461538461539, + 0.9453376205787781, + 0.9309090909090909, + 1.0, + 0.9285714285714286, + 0.9431438127090301, + 1.0, + 0.9508196721311475, + 0.9368421052631579, + 0.9457627118644067, + 1.0, + 1.0, + 1.0, + 1.0, + 0.9401993355481728, + 1.0, + 0.9921259842519685, + 0.9489795918367347, + 0.9397163120567376, + 0.9487179487179487, + 1.0, + 0.9419795221843004, + 0.9383561643835616, + 1.0, + 1.0, + 0.9442508710801394, + 0.9347826086956522, + 0.9464882943143813, + 0.9358108108108109, + 0.9351535836177475, + 0.9421768707482994, + 0.9453924914675768, + 0.9342560553633218, + 0.9322033898305084, + 0.9288256227758007, + 0.945859872611465, + 0.9298892988929889, + 1.0, + 0.9508771929824561, + 0.9527027027027027, + 0.9464285714285714, + 0.9528619528619529, + 0.9591836734693877, + 0.9618055555555556, + 0.9508771929824561, + 0.9344827586206896, + 0.9442622950819672, + 0.9493243243243243, + 0.9492753623188406, + 0.9488054607508533, + 1.0, + 0.9485530546623794, + 0.9425675675675675, + 0.9551282051282052, + 1.0, + 0.939297124600639, + 0.953405017921147, + 1.0, + 0.9381443298969072, + 1.0, + 0.9455128205128205, + 0.9375, + 0.9471947194719472, + 0.9420289855072463, + 0.9455782312925171, + 0.9452054794520548, + 0.9440559440559441, + 1.0, + 1.0, + 0.9509803921568627, + 0.9421768707482994, + 1.0, + 0.924187725631769, + 1.0, + 0.9401993355481728, + 0.9362416107382551, + 0.9418181818181818, + 0.9335548172757475, + 1.0, + 1.0, + 1.0, + 0.9920634920634921, + 0.9452054794520548, + 1.0, + 0.9377162629757786, + 0.9523809523809523, + 1.0, + 1.0, + 0.9507042253521126, + 0.9435215946843853, + 0.9444444444444444, + 1.0, + 0.9554794520547946, + 0.9461279461279462, + 0.9326241134751773, + 0.9919354838709677, + 0.964968152866242, + 1.0, + 1.0, + 0.9522184300341296, + 1.0, + 0.9514563106796117, + 1.0, + 1.0, + 0.9305019305019305, + 0.9383116883116883, + 0.9381443298969072, + 0.9347079037800687, + 0.9366197183098591, + 1.0, + 0.9403973509933775, + 0.9360902255639098, + 0.9438596491228071, + 0.9202898550724637, + 0.925, + 0.9462025316455697, + 0.9181494661921709, + 0.9407665505226481, + 0.8981132075471698, + 0.931740614334471, + 0.9172661870503597, + 0.9342560553633218, + 0.9205776173285198, + 0.9315068493150684, + 0.934931506849315, + 0.9479553903345725, + 0.9157509157509157, + 0.9154929577464789, + 0.9148148148148149, + 0.9321428571428572, + 0.9249146757679181, + 0.9440559440559441, + 0.9257950530035336, + 0.9264214046822743, + 0.9172932330827067, + 0.8928571428571429, + 0.9381818181818182, + 0.9370860927152318, + 0.9442508710801394, + 0.9283154121863799, + 0.9130434782608695, + 0.9333333333333333, + 0.9154411764705882, + 0.9153225806451613, + 0.9326241134751773, + 0.9487179487179487, + 0.9326241134751773, + 0.9282868525896414, + 0.9148936170212766, + 0.9090909090909091, + 0.9172661870503597, + 0.9014084507042254, + 0.948905109489051, + 0.9293286219081273, + 0.9294117647058824, + 0.9209621993127147, + 0.9316546762589928, + 0.9172932330827067, + 0.916030534351145, + 0.9245283018867925, + 0.9153846153846154, + 0.9156626506024096, + 0.9368029739776952, + 0.9356060606060606, + 0.9395017793594306, + 0.9172932330827067, + 0.9256505576208178, + 0.9460431654676259, + 0.9247311827956989, + 0.9110320284697508, + 0.9057971014492754, + 0.9233449477351916, + 0.9250936329588015, + 0.9419795221843004, + 0.9267399267399268, + 0.9097472924187726, + 0.9338235294117647, + 0.9071428571428571, + 0.9073359073359073, + 0.9277566539923955, + 0.9175627240143369, + 0.9379562043795621, + 0.9094202898550725, + 0.9220338983050848, + 0.910394265232975, + 0.9283154121863799, + 0.9239130434782609, + 0.9206896551724137, + 0.8939929328621908, + 0.9244604316546763, + 0.9283018867924528, + 0.9204152249134948, + 0.9212328767123288, + 0.9125475285171103, + 0.9222222222222223, + 0.9361702127659575, + 0.9301470588235294, + 0.9351145038167938, + 0.9375, + 0.9124087591240876, + 0.9120879120879121, + 0.9163763066202091, + 0.9074733096085409, + 0.9275862068965517, + 0.9256505576208178, + 0.9298245614035088, + 0.9390681003584229, + 0.9238754325259516, + 0.9182156133828996, + 0.934931506849315, + 0.921146953405018, + 0.9080459770114943, + 0.9197080291970803, + 0.9384057971014492, + 0.9321428571428572, + 0.9311594202898551, + 0.9270833333333334, + 0.939622641509434, + 0.9333333333333333, + 0.9263157894736842, + 0.9395017793594306, + 0.9295774647887324, + 0.9438202247191011, + 0.9304635761589404, + 0.9354838709677419, + 0.9175627240143369, + 0.9230769230769231, + 0.9187279151943463, + 0.9230769230769231, + 0.8893280632411067, + 0.9040590405904059, + 0.9333333333333333, + 0.931899641577061, + 0.9010989010989011, + 0.9363957597173145, + 0.929368029739777, + 1.0, + 0.9361702127659575, + 0.9321428571428572, + 0.9290780141843972, + 0.924901185770751, + 0.927797833935018, + 0.9545454545454546, + 0.9257950530035336, + 0.9172932330827067, + 0.9198606271777003, + 0.9083969465648855, + 1.0, + 0.9372822299651568, + 0.9191176470588235, + 0.9370629370629371, + 0.9547038327526133, + 0.9330855018587361, + 0.925531914893617, + 0.9195804195804196, + 0.9219858156028369, + 0.9144981412639405, + 0.9216417910447762, + 0.9288389513108615, + 0.9097472924187726, + 0.9047619047619048, + 0.9153846153846154, + 0.9077490774907749, + 0.9283154121863799, + 0.9298892988929889, + 0.9320754716981132, + 0.9241379310344827, + 0.9304029304029304, + 0.9360902255639098, + 0.9155405405405406, + 0.9444444444444444, + 0.9166666666666666, + 0.9267399267399268, + 0.941747572815534, + 0.9125874125874126, + 0.9288389513108615, + 0.9173228346456693, + 0.9315068493150684, + 0.9498207885304659, + 0.9306122448979591, + 0.9405204460966543, + 0.9384057971014492, + 0.9222222222222223, + 0.9108527131782945, + 0.9434628975265018, + 0.933579335793358, + 0.9227642276422764, + 0.9290780141843972, + 0.9392857142857143, + 0.9405594405594405, + 0.916058394160584, + 0.9442508710801394, + 0.9330855018587361, + 0.9301470588235294, + 0.9230769230769231, + 0.9190140845070423, + 0.9218106995884774, + 0.9144981412639405, + 0.9316546762589928, + 0.9119718309859155, + 0.9285714285714286, + 0.9072164948453608, + 0.9298892988929889, + 0.9363957597173145, + 0.9114391143911439, + 0.931899641577061, + 0.9244604316546763, + 0.9190140845070423, + 0.9076923076923077, + 0.9451612903225807, + 0.9215017064846417, + 1.0, + 0.9290780141843972, + 0.9090909090909091, + 0.917910447761194, + 0.9194139194139194, + 0.9486301369863014, + 0.9319727891156463, + 0.9081272084805654, + 0.9920634920634921, + 0.9916666666666667, + 0.9529780564263323, + 0.9381107491856677, + 0.992, + 0.9403508771929825, + 0.9065743944636678, + 0.926923076923077, + 0.9215017064846417, + 0.9337979094076655, + 0.9110320284697508, + 0.911660777385159, + 0.984375, + 0.9247311827956989, + 0.9259259259259259, + 0.9923076923076923, + 0.914179104477612, + 0.984, + 0.8983050847457628, + 0.9323843416370107, + 0.9425675675675675, + 0.9228070175438596, + 1.0, + 0.9194139194139194, + 0.9298245614035088, + 0.9117647058823529, + 0.9922480620155039, + 0.9217081850533808, + 1.0, + 0.911660777385159, + 1.0, + 0.9138576779026217, + 0.9309090909090909, + 0.9924812030075187, + 0.9163636363636364, + 0.921146953405018, + 1.0, + 0.984, + 0.9344827586206896, + 0.911660777385159, + 0.930921052631579, + 0.9292929292929293, + 1.0, + 0.914396887159533, + 0.9920634920634921, + 0.9326241134751773, + 0.917910447761194, + 0.9920634920634921, + 0.933579335793358, + 0.9090909090909091, + 0.895910780669145, + 1.0, + 0.9202898550724637, + 0.9154929577464789, + 0.9915966386554622, + 0.9180887372013652, + 0.9331103678929766, + 0.921146953405018, + 0.9779411764705882, + 0.9919354838709677, + 1.0, + 0.9819819819819819, + 0.903114186851211, + 0.9908256880733946, + 0.9908256880733946, + 0.9148936170212766, + 0.9387755102040817, + 0.9222972972972973, + 0.9844961240310077, + 0.9300699300699301, + 0.9139072847682119, + 1.0, + 0.992, + 0.9307958477508651, + 0.9243986254295533, + 0.9054545454545454, + 0.9240924092409241, + 0.9298245614035088, + 0.9124087591240876, + 0.9117647058823529, + 0.935374149659864, + 0.9358490566037736, + 0.9283018867924528, + 0.9254237288135593, + 0.9113475177304965, + 0.9917355371900827, + 0.9340277777777778, + 0.921146953405018, + 0.9421768707482994, + 0.9192982456140351, + 0.9265734265734266, + 0.9230769230769231, + 0.9246575342465754, + 0.9084249084249084, + 0.9261744966442953, + 0.9293286219081273, + 0.9111111111111111, + 0.9163636363636364, + 1.0, + 0.9142857142857143, + 0.9321428571428572, + 0.93, + 1.0, + 0.9307958477508651, + 0.9264214046822743, + 0.9836065573770492, + 0.9180327868852459, + 1.0, + 0.946843853820598, + 0.9359430604982206, + 0.9425675675675675, + 0.9254901960784314, + 0.9178082191780822, + 0.9029850746268657, + 0.9201520912547528, + 0.9918032786885246, + 1.0, + 0.9366197183098591, + 0.8960573476702509, + 0.9922480620155039, + 0.9359430604982206, + 1.0, + 0.9283276450511946, + 0.9238754325259516, + 0.9090909090909091, + 0.9191176470588235, + 0.9854014598540146, + 0.9921259842519685, + 1.0, + 0.9906542056074766, + 0.9225352112676056, + 0.991304347826087, + 0.8961538461538462, + 0.9225589225589226, + 0.991869918699187, + 0.9923076923076923, + 0.9175627240143369, + 0.9403508771929825, + 0.9285714285714286, + 1.0, + 0.9283154121863799, + 0.9337979094076655, + 0.9215017064846417, + 1.0, + 0.9318181818181818, + 0.983739837398374, + 1.0, + 0.92, + 0.975, + 0.9358490566037736, + 1.0, + 0.992, + 0.915129151291513, + 0.9100346020761245, + 0.9276315789473685, + 0.9195804195804196, + 0.9254237288135593, + 1.0, + 0.9355932203389831 + ] + }, + "contact_maps": { + "mean_pearson": 0.1330763266198374, + "valid_tracks": 28, + "per_track_pearson": [ + 0.14631906711375872, + 0.13035636577691506, + 0.1329826124500324, + 0.04530476148224292, + 0.06226570453282049, + 0.07839910873837011, + 0.14701223701654556, + 0.11570035480424325, + 0.24782120431628366, + 0.1884545159473337, + 0.033475817828557436, + 0.11340451594297914, + 0.21535382293708902, + 0.02255805492074427, + 0.09597285221623485, + 0.21493325852548617, + 0.056536109143109727, + 0.1858925659281982, + 0.15972587595308982, + 0.1624511174051042, + 0.06838416857505245, + 0.08743543776640332, + 0.08328056391159362, + 0.1878109200727858, + 0.17889227797894336, + 0.19481089740549917, + 0.1342813791028456, + 0.236321577563185 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.32701744775598257, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..f7e31897291ac5a5832911766580d88bcc3f634c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S013/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5793698733965188, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.630227676656248, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.13204937683811374, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7400365468274857, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.29645498237375845, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.22002863111830218, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.19356315829006082, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.32105533178580553, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.16770820134417186, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22627638142454026, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5831934548667065, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.154692680776655, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.19178224857656814, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/README.md new file mode 100644 index 0000000000000000000000000000000000000000..7b03eb95bd651c86d63145c283e48851c6c51a9f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/README.md @@ -0,0 +1,82 @@ +# SeqFunc context09 at step 44,000 with transcript-direction junction decoding + +One model with **13,265,544 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The compact balanced-splice model gained a zero-output local path at step 2,000. At step 40,000, two identity-initialized residual blocks extended its local receptive field from 701 to 10,941 bp. All old parameters and optimizer moments were preserved, with optimizer IDs remapped by parameter name. The next 4,000 examples and learning rates match local01 through step 44,000. + +This checkpoint records **44,000 updates**, 1 complete training permutations and +2,980 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 3 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup training examples. The objective is `balanced_splice`. +Other family losses follow the reference. Recorded accumulated training-loop wall time is +127.2 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +## Measured validation results + +Family-mean Pearson: **0.330357** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.516952 | +| dnase | 0.622119 | +| procap | 0.404459 | +| cage | 0.216718 | +| rna_seq | 0.328743 | +| chip_tf | 0.435627 | +| chip_histone | 0.428129 | +| splice_sites | 0.041820 | +| splice_usage | 0.280484 | +| junctions | 0.224054 | +| contact_maps | 0.134829 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5851 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6859 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.1654 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7936 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3060 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2176 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.3017 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.1661 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1676 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2355 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.7073 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1344 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1816 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..b303311b59a8500d1f68011fe9d5cd8ace69018e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/context09/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..de7046dd95a983eb85cbe6bd5d4ba7b0246d2184 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/package_manifest.json @@ -0,0 +1,32 @@ +{ + "parameters": 13265544, + "source_checkpoint_sha256": "f8223af7bb526621769ac713df53b54cfe8d9d0cf0a34e2b1d18b9080d58cf8c", + "files": { + "README.md": "1282bbcd9d9c479d0bc084f5be0ca92e384eb44a47cf138b0634d74085085b9a", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "3554a6c79b220b41087f34adaa35dfe72d58f96f8060c78ea6c6786ed9324620", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "bcd41eba9c2f952250f76aac8812a9dd995ee206791fbe640c751fa073428b8b", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "a71ed8c14bb3dee86bdba5967e4152192bd35dad78d32934a1aaab17a4719835", + "training_provenance.json": "77969daf23c2d04f9d20de14fcb24651a911019f03619a5bd91ca59c712df4b0", + "decoding_policy.json": "56749f91c8b0d85201820dc20c59a1f685a8955cf467576fff326f67fe27f76f", + "validation_tracks.json": "d8488d7d6dd5676041f62ce55957c3a0429671b16fd52f92f238d89ebdcf4965", + "validation_variants.json": "ca151feb3ac50117a9bfc496623818acca991c5f9ce3552b715c0a392426a23d" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..dc0f0b20833fc99253b7dcf6023fd5e7a61b1128 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/predict.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3c2effcf45b2aa0eec7e31aa3f668107d7ff90 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/model.py @@ -0,0 +1,430 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e5a5cee0af43ebd5a92833bed8d5dcc8b264a8d3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_config.yaml @@ -0,0 +1,29 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..4e116d0456eb51c5525e13f6b75b10e78c9bc33a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/training_provenance.json @@ -0,0 +1,91 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6 + }, + "epoch": 1, + "offset": 2980, + "global_step": 44000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/context09", + "config": "/workspace/research/experiments/context09/config.yaml", + "resume": "/workspace/research/experiments/context09/start.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 44000, + "workers": 3, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 100, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 7631.602510094148, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + }, + "interventions": [ + { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..4ba663d4a105abd5615beddd1f56b83395eb39b6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5169519140758255, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6367338643690121, + 0.652573804100543, + 0.4011221061540954, + 0.5857791520362067, + 0.4017022557669607, + 0.39691120912728584, + 0.4165044503755358, + 0.4971073298189792, + 0.5223520181863803, + 0.5801151307804805, + 0.5120668002493944, + 0.48258714281149084, + 0.33870196434217137, + 0.451703796910298, + 0.6259839576174296, + 0.43027462143559886, + 0.48772101354663877, + 0.44911078419311745, + 0.4872650412467827, + 0.46878557497277074, + 0.4428987682112829, + 0.4050903818804531, + 0.4682818192545706, + 0.456966768318424, + 0.45016196895572463, + 0.43237186384837584, + 0.41702254318473697, + 0.436811952449506, + 0.44623985432481117, + 0.49920436897446313, + 0.42747744185430114, + 0.4499095684395622, + 0.41848369653890777, + 0.34574689076794934, + 0.31035708864130873, + 0.5118799269959629, + 0.452116833545577, + 0.5545645532173538, + 0.4468788501027642, + 0.43456572132407506, + 0.5117540671666333, + 0.5111914453383254, + 0.433890208149942, + 0.4129975261118156, + 0.49983739330722604, + 0.48800190349046507, + 0.37024563693015994, + 0.4215680160622504, + 0.41508459835151296, + 0.4627151771633768, + 0.4363941159846054, + 0.39702083766773355, + 0.487028836594565, + 0.5468630226855682, + 0.39814801144352036, + 0.6504203136967613, + 0.7400991503780373, + 0.47935737211700424, + 0.5499233175055978, + 0.5591853558358229, + 0.6788831344688152, + 0.41827953280464963, + 0.656346458497917, + 0.5631565104326005, + 0.5802301014117561, + 0.44596265690728293, + 0.4911146197003142, + 0.5755197442329281, + 0.726950136665914, + 0.6543051055807861, + 0.4740055588401607, + 0.42229486097853014, + 0.38863318357803334, + 0.47101376061671757, + 0.44151774017549117, + 0.4268451662009685, + 0.4424649545324114, + 0.4801377381397096, + 0.5190868992400702, + 0.43494647452255686, + 0.5072139496537842, + 0.5353283800558073, + 0.4536180908596607, + 0.4270048867808962, + 0.4539901793333471, + 0.44248564877266594, + 0.42648557820980043, + 0.41799891645300963, + 0.39357831318580544, + 0.45080000977980916, + 0.4438851787574938, + 0.44597058826472546, + 0.4380149034719638, + 0.4501152296339671, + 0.5126327935627687, + 0.3352311386641222, + 0.3808158640142165, + 0.3734655523338132, + 0.422643131324414, + 0.442537958077659, + 0.40013077934143815, + 0.3997698257404477, + 0.43844224672772664, + 0.4101279141886887, + 0.4791123269225001, + 0.4837706424144118, + 0.4746558669597684, + 0.45572934045260727, + 0.47744633048659146, + 0.4750852207130331, + 0.5306410907638325, + 0.45788381911196235, + 0.45656420925370816, + 0.38423022979387306, + 0.40590927137320565, + 0.38259714918192556, + 0.41668465591755, + 0.42869570603760243, + 0.4344927611380433, + 0.4153819041040363, + 0.6076649412827045, + 0.6032137867481219, + 0.6732535852752731, + 0.6730195864000755, + 0.6503792569452262, + 0.5331066980793979, + 0.8090480497838373, + 0.6575682101093205, + 0.7075242426725219, + 0.6342826988187508, + 0.680496957364782, + 0.5154409800463169, + 0.545420380043882, + 0.5178773309000131, + 0.663794281857404, + 0.6646872407126496, + 0.6462578398083674, + 0.48302470182501495, + 0.38997506272102633, + 0.758768870391049, + 0.7255997809873036, + 0.7425306208557043, + 0.7353695028418513, + 0.54420376648331, + 0.8092750377234093, + 0.7193696850696701, + 0.6677048751632753, + 0.6693856261100241, + 0.7931898511871772, + 0.7534906951480692, + 0.5711694887714259, + 0.5746790504706732, + 0.6583929975795046, + 0.6920682965364277, + 0.6997052822752481, + 0.5588523041190452, + 0.6838074332018426, + 0.5421398633618496, + 0.5558740178638902, + 0.6371272094445634, + 0.527197046182043, + 0.7146276354235583, + 0.7275914588438707, + 0.5835631862706739, + 0.5562402392152701, + 0.5286203006092831, + 0.6876685929456713 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.6221192617501552, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5289569097854437, + 0.9136216597353731, + 0.7744204072681982, + 0.4829222615619367, + 0.45440587010742145, + 0.468245639343208, + 0.935078699603243, + 0.7024662034553801, + 0.3948933991384384, + 0.3169254148037031, + 0.871699820590681, + 0.668960955541089, + 0.4056906479049398, + 0.4780199922273666, + 0.454920055251251, + 0.7947580556633095, + 0.7757826542833787, + 0.8448077515627624, + 0.8494559264908442, + 0.882424469444019, + 0.6601280419933987, + 0.5857509552849621, + 0.8956492004531903, + 0.4793049429502299, + 0.4980572339390654, + 0.7712359873065064, + 0.4295663831304184, + 0.6382143059856855, + 0.47333272395404563, + 0.4964070297624827, + 0.5947587036362088, + 0.42439019811525225, + 0.8346856498490669, + 0.6178080752839449, + 0.8757280655266343, + 0.8822935845454732, + 0.4147989232971965, + 0.8195206051811601, + 0.8714335659193031, + 0.853646132315553, + 0.8190797304234889, + 0.6620689101185859, + 0.6988350819978223, + 0.6156070322474595, + 0.46725865863085897, + 0.6282904630765517, + 0.7255567257544312, + 0.4934265444105113, + 0.5701004101695434, + 0.38268237703092717, + 0.34557826349170323, + 0.5376801522704717, + 0.46598872907211214, + 0.8240675121137458, + 0.40171589130055285, + 0.3834040814030519, + 0.42015870039445624, + 0.4575482143557857, + 0.4232261961308124, + 0.5080662924136998, + 0.40185244491435085, + 0.5973683217328789, + 0.4554004647602665, + 0.44176519471003367, + 0.4691712714286817, + 0.4738968167108436, + 0.5232433846048198, + 0.4418799273399047, + 0.3525450158910681, + 0.39484715803329257, + 0.4953363190964755, + 0.6100432139349335, + 0.4574303470558998, + 0.6004202782756266, + 0.3169308395835691, + 0.8032010952943468, + 0.890045727035608, + 0.6352743826114592, + 0.4111897884998479, + 0.42646411000903484, + 0.4741391834944844, + 0.4664378265929594, + 0.46068650543595635, + 0.45180711180414496, + 0.5207486356804719, + 0.47518175036301197, + 0.5701698974469238, + 0.4797598724822678, + 0.6176439723191436, + 0.4258066742507169, + 0.6048324735803841, + 0.3946392535171926, + 0.7122541239129481, + 0.44471750318938674, + 0.6998971158533647, + 0.47990025723802837, + 0.7190374478406389, + 0.6827877336603911, + 0.3261449953244557, + 0.33267276768727133, + 0.4409972706064492, + 0.35246139701838414, + 0.3310724172042645, + 0.7002075620726064, + 0.6624104557155696, + 0.39979839283883856, + 0.4320395977999298, + 0.5761301247101657, + 0.8393527054775677, + 0.538328692744639, + 0.5229829331662482, + 0.5583204059755366, + 0.4750834956289175, + 0.42827681168795106, + 0.7675236371330115, + 0.527951214996832, + 0.4739677240055193, + 0.534811105840829, + 0.6073703372087985, + 0.4144226641009887, + 0.5144950298292015, + 0.4957570731609815, + 0.5310842199120637, + 0.5093940127987275, + 0.5862917423389057, + 0.5341805740738351, + 0.5947463421850462, + 0.6723084397163342, + 0.6853829924630233, + 0.5963706317155957, + 0.4491630857202775, + 0.5892716260310421, + 0.6103531758066836, + 0.7818382932029204, + 0.4310839714434318, + 0.3020186425959827, + 0.5693987168241806, + 0.45946801448781116, + 0.6238897126845785, + 0.5892126483569936, + 0.5989808932095531, + 0.5460056023373139, + 0.458609623475162, + 0.7691265720670478, + 0.5858378007210024, + 0.7535173884113824, + 0.3720529888022095, + 0.36351224225822176, + 0.38523205799204513, + 0.3277112088739813, + 0.8879800759911984, + 0.3059173078632669, + 0.8912067305163245, + 0.718911300208406, + 0.45800543464148863, + 0.3223357619418515, + 0.3383296674918975, + 0.3550676801376659, + 0.6278380017560581, + 0.716799983601962, + 0.718797924527262, + 0.644452157692366, + 0.5488430033533034, + 0.44492756583940457, + 0.6698438388309591, + 0.5939941409249981, + 0.49543278423878884, + 0.6197859475285816, + 0.4427258395790773, + 0.4611515168935845, + 0.37307216256164805, + 0.8239019702209871, + 0.43538206166277565, + 0.453124175383279, + 0.6411321765207357, + 0.6788024387804833, + 0.465941802617538, + 0.33388193237443575, + 0.872498978446465, + 0.3581635881422971, + 0.7522670107922172, + 0.4339833820423245, + 0.5800169628255759, + 0.7079631461978387, + 0.4466356131105542, + 0.5496463809030283, + 0.5783532323063946, + 0.8123406608878571, + 0.40788552599540867, + 0.5940297812312122, + 0.5341092579210241, + 0.7005276658683963, + 0.8919110656366573, + 0.39169469467728696, + 0.5665252773104315, + 0.44575190239941975, + 0.6065790826842686, + 0.6908504121906084, + 0.5535200964730638, + 0.4293162258661189, + 0.8646156615648585, + 0.8761177399425063, + 0.7490416621427016, + 0.8301543765848138, + 0.8401777248470881, + 0.8290873937357023, + 0.8388469943073608, + 0.34433963629141023, + 0.3581620717177054, + 0.49173069795745733, + 0.44173148256348566, + 0.4278608643612276, + 0.5806229851859551, + 0.8573900633907607, + 0.8436712266142085, + 0.6578877564776912, + 0.8202623668325815, + 0.8196123880211491, + 0.8529197821117951, + 0.9008324420820415, + 0.47730693261515394, + 0.5401361316864963, + 0.6371249622656677, + 0.5601971480411133, + 0.8142394490803077, + 0.5483796194410848, + 0.9028921372282043, + 0.8760960193126819, + 0.8683259774431756, + 0.8965055697079255, + 0.8882877688593843, + 0.8657126379175616, + 0.8243406074341715, + 0.8543080713290537, + 0.9008725993627411, + 0.4163988033501991, + 0.9189655346867569, + 0.8735005517384891, + 0.815186193523814, + 0.5726263223468825, + 0.8672164473233501, + 0.4917412633571957, + 0.7462728900182481, + 0.8699578969816838, + 0.5538379283063054, + 0.5416971978154566, + 0.6705898928090168, + 0.5060259625939639, + 0.5072446606743921, + 0.7203428136530858, + 0.877165424506577, + 0.7151054746853457, + 0.8026087912312673, + 0.591261359381148, + 0.8586348250422243, + 0.8000096620575436, + 0.8307965893333832, + 0.8979522409267727, + 0.8273083914256977, + 0.8932768154074584, + 0.8648387445595235, + 0.5116590901697293, + 0.8725625793963703, + 0.534304131015614, + 0.5037120643257761, + 0.8781132515712108, + 0.899931552973062, + 0.8752067048424826, + 0.5523446931672954, + 0.5932906097922506, + 0.6478385891873385, + 0.7924589356911044, + 0.8842094770678337, + 0.6894476553254972, + 0.9003760838781516, + 0.910812397969864, + 0.5970062727655179, + 0.30090230992139844, + 0.49759671680208817, + 0.8939610355065162, + 0.40533869859761873, + 0.5856142578315172, + 0.8754281149127547, + 0.8247972006141591, + 0.5918270225968363, + 0.8606270299836738, + 0.889395610535412, + 0.883744485889877, + 0.872296930145089, + 0.8581295035990597, + 0.8880330511576036, + 0.8487030665402958, + 0.922208555846408, + 0.7777779274711822, + 0.8268161049023214, + 0.9056427270270733, + 0.5262712829466438, + 0.534393354527222, + 0.5153386368724466, + 0.5004423541344205, + 0.9032586593760489, + 0.7922409684186099, + 0.8435402155025687, + 0.7863830947823642, + 0.7570003328112224 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.4044585706229842, + "valid_tracks": 12, + "per_track_pearson": [ + 0.39107177315275327, + 0.4347724933304747, + 0.34902994839527485, + 0.43051866100332814, + 0.4283045824922067, + 0.4329144873706136, + 0.3725674246991704, + 0.4233261735647866, + 0.3354446417201755, + 0.4182055620688495, + 0.4058570369173547, + 0.4314900627608228 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.21671763887966514, + "valid_tracks": 546, + "per_track_pearson": [ + 0.20562340708420346, + 0.24512741414922554, + 0.20943097989496734, + 0.13134727325952922, + 0.07846448255018405, + 0.21402484128248597, + 0.11269562190180266, + 0.1429382768858765, + 0.2287595869940448, + 0.1937057013948044, + 0.24577786484833436, + 0.13757484154032704, + 0.1465508821278036, + 0.1237361586585224, + 0.09875329763605216, + 0.17639697750006803, + 0.19842755350464678, + 0.24389536437857756, + 0.2592327690847243, + 0.23090318837207588, + 0.19925830324105318, + 0.19475708238698852, + 0.3049010776638421, + 0.17270466700682166, + 0.16632785413164988, + 0.22430936211595778, + 0.23144980421957084, + 0.2342522794645249, + 0.25405141451983226, + 0.23649089280561847, + 0.1630915005068072, + 0.2562387367404046, + 0.1913766865405479, + 0.1369215241045783, + 0.2621948777128117, + 0.2435944744848156, + 0.24955637629252828, + 0.26944833267472273, + 0.241306469374647, + 0.2889720145153497, + 0.30114765859332415, + 0.2557849860921968, + 0.22609869260578627, + 0.2259678608649159, + 0.1394951668535957, + 0.18794860758436951, + 0.20400090008422564, + 0.21585205954114858, + 0.1295468815551105, + 0.16878851134806414, + 0.23160484932046413, + 0.24846244854177704, + 0.2648011533203724, + 0.18408012870063667, + 0.28724110518828433, + 0.30391792651532934, + 0.1671685511273831, + 0.2887857473459047, + 0.21443227563836892, + 0.26014379543112376, + 0.28198101923709307, + 0.18123376342544792, + 0.25136992850614603, + 0.1869067327630714, + 0.2613075279637714, + 0.2790487958225544, + 0.2912344498337312, + 0.18071092628198987, + 0.12016620534937435, + 0.11392244820247106, + 0.222624582087443, + 0.2617500065579066, + 0.21982342212825268, + 0.2330909801356625, + 0.2558062528599801, + 0.21276702281855006, + 0.25443801273347166, + 0.22927334282425796, + 0.2586078499828407, + 0.26615375130892677, + 0.21960473490751103, + 0.13595178938743546, + 0.24118764542736398, + 0.28669501617325943, + 0.13395796937363522, + 0.2439002125077165, + 0.2038605227381635, + 0.28201304290192797, + 0.172574836907248, + 0.25722630540231345, + 0.18497790940492417, + 0.22048662337251296, + 0.23701759795029007, + 0.11533844603105338, + 0.2523879338335284, + 0.21732621027253105, + 0.23133464690766467, + 0.2420735782474931, + 0.22794539596769295, + 0.1396665939179402, + 0.21724849255968295, + 0.23302614373397337, + 0.2430060332105477, + 0.3164449196492101, + 0.2767722814788953, + 0.2807902464440551, + 0.189447482189608, + 0.24138760335769444, + 0.18671213311627538, + 0.20389311286692324, + 0.21427504089779198, + 0.21215440380234457, + 0.23465400990265814, + 0.23188810451394415, + 0.19586186672173198, + 0.2359208040398986, + 0.2360005344057293, + 0.24267798688093728, + 0.22970633254133602, + 0.265893965968281, + 0.26327020803906326, + 0.2593806336083349, + 0.15894634814547032, + 0.17653552759666802, + 0.2083532897905059, + 0.25104210535948956, + 0.2260671570361928, + 0.24850433978572634, + 0.3265758412528506, + 0.2528813613905412, + 0.2613408410477281, + 0.18000683123730632, + 0.25359815407747843, + 0.26940215722473265, + 0.3101271774811303, + 0.22356408874260514, + 0.2763750735951846, + 0.2012908171745541, + 0.12496989358182105, + 0.24596662855515666, + 0.28453130497478996, + 0.23408410705524568, + 0.2534175350246954, + 0.20485038952868323, + 0.21131286668106594, + 0.22527006330603672, + 0.24942677043035053, + 0.2852700956394983, + 0.20712310804107248, + 0.19939320283481254, + 0.24524655309913465, + 0.18629011075988697, + 0.2329681904407709, + 0.2440819064511838, + 0.22220865862801545, + 0.17404086696063686, + 0.26563344228321517, + 0.2998014357397481, + 0.18650838161984742, + 0.1676539685643202, + 0.15000396536266628, + 0.20954280347914186, + 0.20857855836088884, + 0.24364921341409493, + 0.21710088746225248, + 0.17460214597507479, + 0.2293572477703677, + 0.16248403829615019, + 0.16979565797925109, + 0.267959898097536, + 0.2322342440802662, + 0.181320836296417, + 0.2195754820252675, + 0.32836317876293425, + 0.235177086816596, + 0.16471684704235331, + 0.15442896218293203, + 0.20536527474183644, + 0.25559216384046973, + 0.29316373407906987, + 0.2834258816250266, + 0.28435734262367895, + 0.14763229019952587, + 0.2711428857527873, + 0.22997885627708456, + 0.18158148906169727, + 0.20214781495913728, + 0.1650761117236785, + 0.3075066293881585, + 0.24198450499324498, + 0.196535101113058, + 0.1374797568378802, + 0.13597696806480963, + 0.2135020693470916, + 0.22344695582062518, + 0.1988689607760446, + 0.22836122856599542, + 0.2258146799730418, + 0.16765410543621287, + 0.2235211548145866, + 0.1260630751010572, + 0.11326920471911546, + 0.12759284750676458, + 0.12010899079231725, + 0.13410369461457378, + 0.2166560274521715, + 0.21665717475998353, + 0.18258811420995832, + 0.12040249867394029, + 0.20306743317092613, + 0.28448049750912335, + 0.24465542941948218, + 0.2629966285959256, + 0.2597366486531863, + 0.23615660814010228, + 0.2208938082484105, + 0.2829463551345749, + 0.3034371797712867, + 0.2617124970426645, + 0.2717998113275685, + 0.2447330722375395, + 0.18303044118356457, + 0.25240244611740004, + 0.23194151198556706, + 0.26592686885797073, + 0.31258133520516107, + 0.24796586279716593, + 0.21639180213647105, + 0.2325867821555055, + 0.25257910402483535, + 0.16799808244512404, + 0.167332521147609, + 0.28797179319158805, + 0.22153335344321806, + 0.19155264792419335, + 0.22379348198152393, + 0.30107298869634314, + 0.1665413270236291, + 0.2372857380167345, + 0.23172506867510106, + 0.12366258423302573, + 0.16608472014567416, + 0.15755775583053266, + 0.24792823676131306, + 0.18630123578188576, + 0.23004527973283664, + 0.09344840624024597, + 0.2874131262590886, + 0.24459592420426327, + 0.2592416871189403, + 0.2621818732876677, + 0.16545681307662818, + 0.3043923877672953, + 0.1391640419310085, + 0.264027906557273, + 0.2298746695559135, + 0.3382252344380831, + 0.2702168167066978, + 0.2760096613223597, + 0.3624970944151179, + 0.2184041599638742, + 0.24305607685328548, + 0.13680082647608321, + 0.24305395597551835, + 0.2440803534053086, + 0.16921840816346398, + 0.14352599055389031, + 0.14544438000483498, + 0.1694118607033486, + 0.24543478375350625, + 0.303669027986253, + 0.16167034359558874, + 0.13663421628044578, + 0.18713202629565429, + 0.2633271584818659, + 0.21202910009587334, + 0.1450928865098509, + 0.06707435513917978, + 0.21779697415145902, + 0.12747646118798764, + 0.15443623542969084, + 0.23114253170920523, + 0.18597951129317503, + 0.23251030517328727, + 0.13424210948735502, + 0.1521973504387164, + 0.13574037137029954, + 0.1181802003604337, + 0.15842493575848882, + 0.19474447939649223, + 0.2332437678877951, + 0.25180583637241455, + 0.2211776152358802, + 0.19435095493355262, + 0.19123594875505356, + 0.28979850053273604, + 0.15794968624880626, + 0.1605767792239566, + 0.2375502206657854, + 0.21861854746230377, + 0.220866917045465, + 0.24256576620594736, + 0.23081750003773216, + 0.1558041410485235, + 0.23909420725685376, + 0.17640128485189735, + 0.13005992543822079, + 0.24835815324043017, + 0.24224580809864918, + 0.26181903109818533, + 0.2619829860208141, + 0.2570409421622975, + 0.27623221233478507, + 0.28276533729840236, + 0.23989129621058766, + 0.22329202066780546, + 0.22156718233908396, + 0.14731846811603452, + 0.19484864929196985, + 0.2156613980339949, + 0.2386465068332694, + 0.1280711801967045, + 0.17017208354585833, + 0.2149898132012654, + 0.24962554802963752, + 0.2542243398273322, + 0.1790305539466146, + 0.2737507153836283, + 0.28297417866564, + 0.1730061463651474, + 0.27581593064937315, + 0.1966168244679369, + 0.25494177831843834, + 0.2658923208979796, + 0.17296023003564864, + 0.24238853831755555, + 0.1898100473732581, + 0.2530220235403051, + 0.26018602215877434, + 0.27184919288347925, + 0.1709252337704604, + 0.11869719650684941, + 0.10558162444859438, + 0.21103453636491157, + 0.25183579531430267, + 0.21522270269278895, + 0.22197264886311613, + 0.25033084109390547, + 0.2091181511279228, + 0.2320678942126082, + 0.21469072871070996, + 0.24246081485875443, + 0.2591665747216854, + 0.21375728059396362, + 0.13261459973641493, + 0.2383022313052564, + 0.27686110385832086, + 0.13380108121428597, + 0.23294969382265665, + 0.22410884559132113, + 0.280631469567876, + 0.1633661417156773, + 0.2560628600295315, + 0.18276080676827716, + 0.2134790517178706, + 0.22890788193064526, + 0.11314958156695121, + 0.23768874879293744, + 0.20342371007898308, + 0.21522910753054986, + 0.25099493469827566, + 0.21371008718390133, + 0.132394278062326, + 0.20729618167074265, + 0.23708771619967564, + 0.23034523326484357, + 0.30057215667792664, + 0.25619515788407143, + 0.275291488644103, + 0.18102133945590382, + 0.22999061812192328, + 0.1743444899487507, + 0.18595933578335505, + 0.200098391796633, + 0.2020691478119711, + 0.22808309902266077, + 0.23126147880409828, + 0.1839504506811663, + 0.22877134316110365, + 0.2167307360046092, + 0.23001732383769102, + 0.21428140394896023, + 0.25893653044305015, + 0.24659404838306498, + 0.24134255104271554, + 0.15928652343811367, + 0.181726417679691, + 0.21188204198737287, + 0.2427478039491924, + 0.2139398378240866, + 0.2452240109437541, + 0.3148886443961387, + 0.25509852672714006, + 0.2528207162990399, + 0.17554676248439957, + 0.22937944056050574, + 0.24470801256603789, + 0.2905884769196023, + 0.2214005639180496, + 0.2678183714749465, + 0.20281736136397213, + 0.12806035732641569, + 0.24087440970001817, + 0.2617307167193831, + 0.21391216734456026, + 0.25771071338803814, + 0.18900596191403612, + 0.19929916743746545, + 0.20891925733337724, + 0.2296990004239579, + 0.26551850372670677, + 0.2184713866669479, + 0.19116218687787495, + 0.2357895826288171, + 0.176971233909454, + 0.21534869432576578, + 0.22682127193937038, + 0.20771990089983852, + 0.17885425613929273, + 0.24995040051591658, + 0.27951254018990235, + 0.17968360369965086, + 0.16884706052235504, + 0.1481524249023122, + 0.1880158249400205, + 0.19797850225253066, + 0.24803357413142155, + 0.225920516679859, + 0.16150844982739068, + 0.22041766083717906, + 0.1746801478681904, + 0.15799461570111678, + 0.270802648530324, + 0.22330346884012003, + 0.18743459445386976, + 0.19920000941353452, + 0.31937987500882775, + 0.22980231585488295, + 0.1484968757393058, + 0.15156274097444714, + 0.21365575825082822, + 0.21095861576288197, + 0.2992506407286335, + 0.27483711252899773, + 0.28521502069486665, + 0.12717498690864024, + 0.26477964322031944, + 0.23105721796500642, + 0.1965889504620931, + 0.2080528131917117, + 0.16627560820081874, + 0.27848508569752045, + 0.22498137093377854, + 0.19806803470733225, + 0.15772524899516643, + 0.10845340522890085, + 0.1579432652840982, + 0.23504971122802865, + 0.19761293769250987, + 0.2353269824469692, + 0.22204766645283142, + 0.1708862309483549, + 0.2010898604207335, + 0.1007782024314163, + 0.08048656157763366, + 0.10158695958226815, + 0.09994904443895201, + 0.08898442366524373, + 0.22043617490494447, + 0.17282587396367594, + 0.1622375536276928, + 0.11250356611352763, + 0.19769006620529006, + 0.2902368234699835, + 0.2822493778893957, + 0.28402899750108557, + 0.27676903244967754, + 0.2609784234727267, + 0.23181471359268088, + 0.2885027396940468, + 0.3058848596574141, + 0.2671168501644488, + 0.2800370057238896, + 0.26066055969433954, + 0.19830361001981384, + 0.2695327123718187, + 0.23563303637431676, + 0.29720756057005415, + 0.3098773683708142, + 0.24117233735313928, + 0.22423309805375444, + 0.22061508182095774, + 0.24176738218897056, + 0.1247134357279558, + 0.11208218946322526, + 0.29087866918474653, + 0.21707864817164327, + 0.19938535910007554, + 0.2169976650898009, + 0.2973859604622079, + 0.1821213804544916, + 0.2298079909848144, + 0.21784247484091002, + 0.11202323061240577, + 0.1364674354597148, + 0.1367742449567964, + 0.25924908592657697, + 0.16698826382495388, + 0.23577893369912517, + 0.09111528621050785, + 0.28895989289357377, + 0.2487592772747449, + 0.25713625138517743, + 0.2641862687430382, + 0.17962115252545025, + 0.29492086872960666, + 0.12776699925817678, + 0.24469539704082244, + 0.22435498948489124, + 0.3247219161352488, + 0.2748185819507237, + 0.2913632848272675, + 0.3518359747197956, + 0.22696154033502125, + 0.22510721382090743, + 0.12402128599796533, + 0.23216047645507779, + 0.24102665377580124, + 0.18878884170374052, + 0.13111521949503444, + 0.12865744931596235, + 0.1582156554481372, + 0.24995602096140143, + 0.301633232959051, + 0.14547745294190154, + 0.11419777694721422 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.32874268536651374, + "valid_tracks": 613, + "per_track_pearson": [ + 0.30696611730231654, + 0.35084677273324816, + 0.327232788074233, + 0.3730953529236726, + 0.3501075623434754, + 0.22348536467885752, + 0.26953593681819293, + 0.35539191163506906, + 0.34958987109223766, + 0.38458570320814284, + 0.3045626750407335, + 0.30706343439566897, + 0.24741475923491338, + 0.36615729566324223, + 0.3384763647727203, + 0.35641034926872756, + 0.39869763757255056, + 0.3263869952170655, + 0.33076985738144066, + 0.3128737808780542, + 0.2731183577470696, + 0.35016443765490796, + 0.29175322350316774, + 0.26867187393173386, + 0.2420548833540739, + 0.3273469493369429, + 0.2947709300807259, + 0.35062751743691173, + 0.3350047650218868, + 0.34994435524876366, + 0.33576069370823, + 0.3298307014087009, + 0.386451487602127, + 0.29578347925793386, + 0.38962338135395175, + 0.26969409163391506, + 0.36340673844674537, + 0.3576036912823517, + 0.3185032807308652, + 0.35821615683925534, + 0.35704784321840727, + 0.31514380952144483, + 0.34744545839154184, + 0.3397981955858789, + 0.30605831906202013, + 0.23753874239279416, + 0.27951050581324005, + 0.36093815152779546, + 0.3070134205434667, + 0.3336904767802463, + 0.34058499205241055, + 0.3717133683064915, + 0.2620478142146016, + 0.31142174240683873, + 0.23981773610678625, + 0.3281132442392351, + 0.3087365929200704, + 0.34775038315108564, + 0.3713546229126144, + 0.3459932622171929, + 0.33369494148364653, + 0.3387752252606182, + 0.3582681807168717, + 0.248938308301753, + 0.3511061610180033, + 0.3331491572535664, + 0.36014053477129643, + 0.36479302785691475, + 0.3407359677648597, + 0.3541155261713658, + 0.3459694601205711, + 0.357536487866388, + 0.34474648351765286, + 0.3408231388925931, + 0.3479157498130683, + 0.35214018914431633, + 0.3607485158067828, + 0.3621463462035936, + 0.34121993875188045, + 0.35540207358265274, + 0.3232600699757556, + 0.2714058720178794, + 0.3936376119245242, + 0.394985214836079, + 0.30659428350933465, + 0.3252162568541452, + 0.3484138281666806, + 0.30280228894548156, + 0.3293316275837548, + 0.3019479169018741, + 0.27837134609494174, + 0.3533335667253134, + 0.34157035573457045, + 0.33577145258145585, + 0.339858123791541, + 0.3483963747978507, + 0.35184274330806375, + 0.3406729687741904, + 0.31353919999682384, + 0.2761756484895333, + 0.2578176767119178, + 0.2814591287125638, + 0.2807971189545109, + 0.343376671683254, + 0.2937417595477802, + 0.31482361863337216, + 0.34694453368797395, + 0.34781599733731633, + 0.37722518638195857, + 0.31407309777027587, + 0.30653120088283486, + 0.3769870523085767, + 0.34432390877635527, + 0.2970624325488076, + 0.33510292494624433, + 0.3532022568463358, + 0.39164577958796704, + 0.3112879196856036, + 0.31105948554816737, + 0.31817773095112933, + 0.3026266637598956, + 0.31270232517296664, + 0.2965880726223141, + 0.3406607237511136, + 0.3935886623845376, + 0.2715663759040944, + 0.2876207727822271, + 0.23856112330222962, + 0.25693294121648935, + 0.29107047117480594, + 0.40118708766877076, + 0.30591436651495924, + 0.2933840801741794, + 0.2719892241454626, + 0.2923729234444703, + 0.35821777971063457, + 0.3549347621596624, + 0.39518580645904827, + 0.3408048480796986, + 0.3725642023421633, + 0.2976366534805562, + 0.30516565306470145, + 0.3921543008340036, + 0.3514404610035897, + 0.25495902342276555, + 0.27757927401771454, + 0.26288545761637466, + 0.29851384379111806, + 0.27411049839781454, + 0.310997051870551, + 0.2731633221797884, + 0.3077103912384202, + 0.31857933429239194, + 0.26133073815418384, + 0.32837152494365845, + 0.26255492044724976, + 0.32174848238078396, + 0.32748588576467896, + 0.32973925444294044, + 0.3462896491547796, + 0.39636755086704983, + 0.3209673115641582, + 0.3525697077772785, + 0.3121438944994095, + 0.3097070491570827, + 0.3069862477941333, + 0.3145934160753259, + 0.2685694509009606, + 0.31054375160505077, + 0.3187579452329431, + 0.33800603538963125, + 0.32592931905299, + 0.3857093858807917, + 0.31568935121491243, + 0.2728032381507845, + 0.2961134015204494, + 0.30369000912355376, + 0.40213059507083715, + 0.3068083754939965, + 0.28120726695971643, + 0.2877661279691439, + 0.34511978167349516, + 0.26077816032957285, + 0.32561830047557555, + 0.24457497045871165, + 0.32621270199478886, + 0.33883409855696234, + 0.2149515255159232, + 0.2943962294021577, + 0.36111488682651754, + 0.4125048325753477, + 0.3131624187614433, + 0.2907750731238851, + 0.32630037401552797, + 0.32754553098864125, + 0.31886731926528245, + 0.320429681341178, + 0.23539586722404834, + 0.3085370456442626, + 0.33439473563395633, + 0.3362326517390479, + 0.32817745029546525, + 0.43500908671939853, + 0.3656882830420097, + 0.3304766491086102, + 0.3338631042396288, + 0.3348717192630044, + 0.19951149475017838, + 0.3374621595844586, + 0.31507719073354223, + 0.2485585625040532, + 0.244568526488536, + 0.2805562137716745, + 0.2974610033099723, + 0.2938214801679909, + 0.3542593117334275, + 0.26761958733584723, + 0.26639670782797925, + 0.3085835705978093, + 0.32783812628381725, + 0.426112410455706, + 0.36400531205995723, + 0.36851859517022784, + 0.34747092974107174, + 0.34389826328841955, + 0.3849582252763724, + 0.34818562108119905, + 0.37713658297140673, + 0.26873155613249067, + 0.3461134353596477, + 0.32369829313670595, + 0.3984872188654817, + 0.35496667143531, + 0.2913415034878126, + 0.3552036267965172, + 0.26139341183834786, + 0.19793461219203898, + 0.3660337055823058, + 0.3663031049495775, + 0.3586237427255727, + 0.373728871015813, + 0.34010610180959283, + 0.3486169956444898, + 0.3244335031785885, + 0.22079503999594005, + 0.3434078135845678, + 0.36457135411371117, + 0.3370291911580442, + 0.32077567902349047, + 0.30232555116481474, + 0.279148245319658, + 0.32494168019757874, + 0.28780744561400273, + 0.3000658498699322, + 0.3438437117252543, + 0.36037969066940506, + 0.2790632792309794, + 0.3479156948914915, + 0.3127811895557584, + 0.3418025393938255, + 0.3133193369692182, + 0.31101737777615285, + 0.37366578807486117, + 0.39414658057745217, + 0.33930064750231853, + 0.3467156032933559, + 0.34742565052846835, + 0.27337669087012395, + 0.3276370134930735, + 0.34197597163960564, + 0.2624415359308474, + 0.2625676335277623, + 0.40159950198135963, + 0.3663985479314744, + 0.39028749973792254, + 0.3782263059364041, + 0.2063216735735701, + 0.22879914019055433, + 0.4031627122549396, + 0.3962827395814516, + 0.39943019647477696, + 0.36395761438562485, + 0.29839494745637823, + 0.29788250457997856, + 0.3301205637365877, + 0.2997176748659424, + 0.31010412515155406, + 0.37663501136887306, + 0.3421007952768476, + 0.4027266479860295, + 0.31453102876945765, + 0.2603092066152707, + 0.4018512599531567, + 0.2647987298750037, + 0.23336084524749623, + 0.23136869724153555, + 0.36588678269712444, + 0.3831164578416641, + 0.398123409748863, + 0.34411602144020315, + 0.3461958888963144, + 0.3238801627508371, + 0.3674508095140935, + 0.28811594638160637, + 0.3359163416111051, + 0.40147646156721073, + 0.3258258978437936, + 0.3845523860936001, + 0.377711934204759, + 0.35926676106825356, + 0.35569573621764966, + 0.36082208064684856, + 0.3521728369663545, + 0.3829219775686492, + 0.36027501656424094, + 0.2838851067314901, + 0.2603057818445204, + 0.2908329961709212, + 0.3525164211309864, + 0.28956288944310365, + 0.371571174059207, + 0.38086542178427185, + 0.38338308940017846, + 0.23069202686183907, + 0.2928677430790796, + 0.20351759045115708, + 0.3167868428196727, + 0.31931728618500055, + 0.37432694798288313, + 0.3916805419450574, + 0.36362606992787894, + 0.37701861435458867, + 0.3977790233470155, + 0.41357424159816186, + 0.23314159343650043, + 0.39245608642329194, + 0.39679051058275266, + 0.4158265914135474, + 0.4167928831507466, + 0.4044275354979057, + 0.40727719996533135, + 0.37135770252174866, + 0.4130270632536204, + 0.37209481714783904, + 0.41230955900086486, + 0.38794747478040353, + 0.3950845039748113, + 0.40611134373752816, + 0.41274674215256707, + 0.3024392915944494, + 0.3747631083162634, + 0.34665366218014937, + 0.2922323972829604, + 0.4155478230628413, + 0.43935445959977637, + 0.3483135865737206, + 0.3797822399415647, + 0.35283226606607226, + 0.2562721132503569, + 0.3339542752233795, + 0.28276661244637785, + 0.27622436413338186, + 0.3847642233466797, + 0.35899929618242243, + 0.3402801248640533, + 0.3295549912577913, + 0.3696710754887287, + 0.37856799559949217, + 0.3443552525318372, + 0.334249869446647, + 0.30568958051589595, + 0.24849462378484058, + 0.2458905710729487, + 0.2740611659234395, + 0.4094139504740072, + 0.2622824535303184, + 0.2881053522077608, + 0.38365588924440774, + 0.3234392108227871, + 0.4122438314646206, + 0.30341390424134085, + 0.33270124587965405, + 0.3676469684295367, + 0.29644445246224815, + 0.33845042983486306, + 0.3980455345117516, + 0.32510384684182086, + 0.37017430426207365, + 0.2968653824936589, + 0.3042820823698719, + 0.31264282854519787, + 0.36929786673381193, + 0.3227569342152951, + 0.31462696120440137, + 0.3596950115276188, + 0.3926622750448705, + 0.24728867490618914, + 0.25386387050955, + 0.2315929382995154, + 0.23133227777062482, + 0.3097942133388937, + 0.41581759328022455, + 0.28261258359929603, + 0.29230763006838295, + 0.3425703244553467, + 0.3147544942894892, + 0.4240510429157538, + 0.3517080982402922, + 0.4180476542288045, + 0.29138476650352146, + 0.35414220274931657, + 0.2608173531481665, + 0.3175598600865156, + 0.3830648478159405, + 0.3173037731273525, + 0.24047114933794808, + 0.27205401568239473, + 0.29952234026033103, + 0.3474773434958535, + 0.35976657011190355, + 0.2940498512804127, + 0.31963414260047956, + 0.30570652323662095, + 0.3534708374408317, + 0.2657729637996656, + 0.28640209736050437, + 0.2729812855771844, + 0.36265125934328984, + 0.26229377578649504, + 0.37515551686066223, + 0.3322001129357345, + 0.3809682189079865, + 0.2930480713497266, + 0.3293005369522164, + 0.21600345497559692, + 0.22578531831812904, + 0.2232478969226436, + 0.21928333070077935, + 0.21663223745986523, + 0.33288303750951814, + 0.3104334603351749, + 0.4216269981282676, + 0.35215465623493175, + 0.44396800235312195, + 0.3235710051923464, + 0.2837830397983762, + 0.3023070044794623, + 0.29684901124002294, + 0.438643133326621, + 0.26301767344738763, + 0.28639580815645327, + 0.260110591738535, + 0.39578311881920586, + 0.20929980337546986, + 0.3048916962570442, + 0.21443580317736205, + 0.3271817012278067, + 0.3750218596196392, + 0.23524527297406805, + 0.3075802470798016, + 0.3289089714033119, + 0.38801657220681135, + 0.383419447686536, + 0.3017428450459674, + 0.31599772038973184, + 0.3529199479362704, + 0.3580822860779641, + 0.358052294463845, + 0.21892789365018933, + 0.34766942265342865, + 0.3881863380147131, + 0.30896441967967897, + 0.3980288748515757, + 0.423085671165578, + 0.3808704158052319, + 0.38716839536319714, + 0.36064359190470696, + 0.34128108014491737, + 0.17363930084740775, + 0.4048972026349358, + 0.31262886488217395, + 0.36312722160212374, + 0.3580839043855094, + 0.41012524565735853, + 0.39393590104116516, + 0.3545163448862822, + 0.43267242106645765, + 0.38486156708177544, + 0.30744200991093157, + 0.4118292700778533, + 0.40032786348842686, + 0.3843765310635773, + 0.4115688748384218, + 0.34027674240461936, + 0.2961161481504973, + 0.29620104264044894, + 0.33420873905664256, + 0.2601614504107372, + 0.33701029415864486, + 0.1934481389593916, + 0.4070301719837959, + 0.2621960053381935, + 0.3775164078332953, + 0.4229960950983846, + 0.25060062176299613, + 0.40936261219607134, + 0.23202252992158887, + 0.18736320288217959, + 0.40715501090619804, + 0.38694957796268675, + 0.3503050812359454, + 0.3610834997750225, + 0.38791868581594885, + 0.41717896168722746, + 0.4208623675936824, + 0.23195648347914974, + 0.3541772113553549, + 0.3732382562453465, + 0.3852660996873995, + 0.3293864395845709, + 0.3154774019447497, + 0.30510573018273274, + 0.34650543172952025, + 0.3151345565036105, + 0.32296698393548545, + 0.38059680755886, + 0.34991378401934425, + 0.22935019072915055, + 0.38343675970503344, + 0.3446511065483062, + 0.40562112322423216, + 0.2531197591834153, + 0.22213999560765607, + 0.37251731728177806, + 0.39536744443379457, + 0.36863252354942005, + 0.40253713507137845, + 0.41059087368762315, + 0.21267019649394886, + 0.38733686241331616, + 0.3829598739957327, + 0.23970078417557297, + 0.2706858094326482, + 0.13484712492276882, + 0.3236128388962424, + 0.3844344438696174, + 0.3271431428980142, + 0.3299665721072799, + 0.3129714918962562, + 0.32302961363996924, + 0.3294374511324987, + 0.31144491009833875, + 0.2566953287192895, + 0.2675005844609246, + 0.2701012846142884, + 0.26158615867140667, + 0.2615611213884853, + 0.26378063653065703, + 0.2702675056884941, + 0.2517727886858191, + 0.24918979242139647, + 0.2994665539251164, + 0.32130900159383013, + 0.3257767805677455, + 0.3211254586216962, + 0.3243483797687716, + 0.3523219264363045, + null, + 0.36923759107692466, + 0.40990582277332555, + null, + 0.18272961618190153, + 0.45160735758227993, + 0.17151606907713673, + 0.3583251239888548, + 0.3668844738706299, + 0.40656961398379865, + 0.4335860414578041, + 0.4265379479561546, + 0.44086205882685553, + 0.41943664893834903, + 0.3723359092828259, + 0.35625620250133416, + 0.28746833323740645, + 0.43714786930734134, + 0.2566538805755324, + 0.3540710173175543, + 0.32484348595875173, + null, + 0.34442389591752165, + null, + null, + 0.3288217431854921, + null, + 0.308486007225775, + 0.28479050492155483, + null, + 0.38353405617397074, + null, + null, + null, + null, + null, + null, + null, + 0.3427346183490027, + null, + null, + null, + null, + 0.28833016891274704, + null, + 0.3341536116011353, + null, + 0.30560850060873496, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3587925455373887, + 0.23334116267042562, + 0.333547439627136, + 0.34828447981283933, + 0.28566355081511446, + 0.34781042361193615, + null, + 0.39218907875882425, + null, + 0.3194239318386959, + null, + null, + 0.3326656731220206, + 0.3624068821376428, + 0.30305290349155806, + 0.30598153714416243, + null, + null, + 0.3039444764196164, + null, + null, + 0.3308459500461124, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3256342859674891, + 0.3179974750036098, + 0.3382021088721751, + 0.29940545277985503, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.43562683434095123, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.22331003599063046, + 0.3395591401496472, + 0.30662200301024156, + 0.4678220526309985, + 0.3985164173954446, + 0.2694922657319905, + 0.377576618194986, + 0.4339913544596826, + 0.3639143364211745, + 0.38181049493833114, + 0.40345493265639565, + 0.3906350477206884, + 0.2846019426041724, + 0.44327556826241843, + 0.3806324178318917, + 0.3836401964473994, + 0.4204742981939138, + 0.3580455240825296, + 0.48398736056068137, + 0.3741342805210052, + 0.3793263734665369, + 0.3451293385096214, + 0.3191763625872856, + 0.5024452077435738, + 0.3535544851126049, + 0.3877998430176644, + 0.2922633213617658, + 0.37959841047519655, + 0.38857734026873664, + 0.40159403415837314, + 0.39872055784267274, + 0.41524039619820496, + 0.49709012859722057, + 0.2877307839503172, + 0.6665672624188185, + 0.2816596650307794, + 0.22572239389634274, + 0.17517265097122411, + 0.33625214912768275, + 0.22753687539585427, + 0.2765811801934939, + 0.4694168947013349, + 0.30776603966643945, + 0.4090263850460249, + 0.299058000188185, + 0.3993529773316539, + 0.30543245506968153, + 0.2175432630207564, + 0.4730201813209411, + 0.44295756817309495, + 0.39345692163092816, + 0.368284918426738, + 0.3825374765982172, + 0.38909504524619976, + 0.3468307470830901, + 0.3994655706492355, + 0.32342519850795276, + 0.44811597400573866, + 0.440359472949732, + 0.45434651171600254, + 0.29528079364917675, + 0.2084289368976982, + 0.40209687231795677, + 0.4266436594039815, + 0.2799766337021323, + 0.35284117315442187, + 0.34823404610409636, + 0.4747895511874648, + 0.5007393094495832, + 0.43312765984066026, + 0.3712073583160347, + 0.7133354111097895, + 0.3672998416218434, + 0.37587601701231443, + 0.37601590729705636, + 0.43697738356620913, + 0.3038534768112966, + 0.6461154793061739, + 0.4181142591009891, + 0.426666320227737, + 0.40204117493027386, + 0.4274411769003879, + 0.49449565215299157, + 0.4501191033108654, + 0.35666658052128525, + 0.3136560720644164, + 0.46265732054264325, + 0.43436872440814045, + 0.808088531392567, + 0.4785920323294524, + 0.38823458406490285, + 0.35772202787456897, + 0.4324978025668693, + 0.3737807557647357, + 0.40897303673527396, + 0.5448028395613659, + 0.3495606127353719, + 0.36261253848983777, + 0.37896580623063497, + 0.3238268285322201, + 0.25863286464534513, + 0.3826737892839805, + 0.3371195510674108, + 0.32831626014681137, + 0.3742931203824685, + 0.2316069203908419, + 0.2908563187858141, + 0.3035609948880219, + 0.3826036674859208, + 0.37035846908457093, + 0.24188529747462073, + 0.3259534351806529, + 0.30711616016021315, + 0.3578802837341215, + 0.37015777071912814, + 0.4443602267772798, + 0.3830395358325097, + 0.38464885880353095, + 0.35438589617408356, + 0.4029154857826849, + 0.349411016120545, + 0.4053728984779128, + 0.3828161749148153, + 0.294557513726767, + 0.36617566971373783, + 0.3189951531086841, + 0.33325316329677307, + 0.3176545347710547, + 0.3432737545137622, + 0.5359858204314082, + 0.38699204574837354, + 0.3138361906119387, + 0.3994834764217675, + 0.6063619200233723, + 0.30757472790090246, + 0.3718743922416576, + 0.35968144685838616, + 0.3076797080702031, + 0.3307452783571377, + 0.31103446521942946, + 0.33212933364751646, + 0.44008906105874535, + 0.5332797416844667, + 0.3301219892948613, + 0.398877332469951, + 0.47696894155848996, + 0.4293334863779831, + 0.27382286168202596, + 0.4156962741412979, + 0.3992349989504333, + 0.40139903070464367, + 0.3390465394732935, + 0.3985321493781538, + 0.4647544412527371, + 0.4046726869744818, + 0.4012468966832311, + 0.2941844385730999, + 0.3352616585039463, + 0.33866513846932933, + 0.41916872466813865, + 0.38426734029294457, + 0.35595490719649014, + 0.38602510938403134, + 0.3867362758112844, + 0.38324765296054736, + 0.36481118305213833, + 0.34901890563076376, + 0.45920013646205443, + 0.28167679643982946, + 0.3683740365272693, + 0.3606070461926246, + 0.32060498266604753, + 0.4850681536608023, + 0.33003470552676956, + 0.28605203242856136, + 0.4043552815906593, + 0.2719727540464525, + 0.40578935748199096, + 0.3911489916428471, + 0.31906922084050987, + 0.38285956285135153, + 0.38087825528820746, + 0.31662497346868623, + 0.3884789204517407, + 0.3227434875996237, + 0.37765051838056235, + 0.4197934310560855, + 0.28694931634425436, + 0.31864035573316013, + 0.3908231866426377, + 0.2981960827300851, + 0.3431323764248672, + 0.42497100753006467, + 0.428489821730936, + 0.3231646686813536, + 0.32496197074317573, + 0.3185777253484799, + 0.31806172106677727, + 0.2670441371165932, + 0.45131761323211006, + 0.34635595598017516, + 0.3328065545281244, + 0.3550701833344913, + 0.3180564275625172, + 0.4408014669109843, + 0.3470174644887682, + 0.4096697916612542, + 0.3647799900218105, + 0.30665630648526665, + 0.48662301306806793, + 0.3238065958507511, + 0.4094735818376575, + 0.3301138364261439, + 0.37865700020201165, + 0.8606886850844164, + 0.41737993885778424, + 0.38795334492190514, + 0.3891203933541704, + 0.4521710761284527, + 0.5398814673020922, + 0.6080637906216889, + 0.6228040576218232, + 0.41070004535067867, + 0.3866362392736954, + 0.466994034940327, + 0.7305907079650226, + 0.3205951711981192, + 0.3477871404283578, + 0.5389183327912149, + 0.515081935718093, + 0.5487115798804602, + 0.4401595103612621, + 0.31657768870815933, + 0.37980018011304395, + 0.5964561113829661, + 0.43308504135724846, + 0.3368230005272978, + 0.6123080851044957, + 0.5454248889624527, + 0.4494265001240942, + 0.4829792448282677, + 0.7077058751187456, + 0.4629052585399501, + 0.5047825641637934, + 0.4461284585994312, + 0.7621877051914087, + 0.3208642619767206, + 0.6396597586551006, + 0.39265340628390477, + 0.5464860804354068, + 0.42666056573716965, + 0.38842060604261075, + 0.6999707187499382, + 0.5044152296169185, + 0.30596904365540234, + 0.37499280818862957, + 0.6424457918277104, + 0.3600393048272968, + 0.4455257234095596, + 0.46413453161560236, + 0.35610416917222615, + 0.4607348378131936, + 0.47672848617906977, + 0.4121698067054692, + 0.3930433936722329, + 0.46701257651869993, + 0.35081001166744813, + 0.3222103957375461, + 0.6280633478677262, + 0.36834275534634364, + 0.4501423131491488, + 0.5645320452901245, + 0.35400329660143176, + 0.6590273609964626, + 0.3418211471762543, + 0.34208780519764115, + 0.5567889810655622, + 0.47832712145728745, + 0.5829819766739756, + 0.7294922833993741, + 0.662947593259076, + 0.40462789002812766, + 0.46601083754083067, + 0.5453923575890548, + 0.3113983657839716, + 0.33246267072663843, + 0.56241097979771, + 0.5254944258530075, + 0.7254752287314163, + 0.4598373491583347, + 0.35843072307575496, + 0.46010807730539727, + 0.3731417782746428, + 0.4739293560541001, + 0.7015395778222084, + 0.6234472184571562, + 0.4892524882491876, + 0.5257518081173328, + 0.4953283601239038, + 0.514500377883088, + 0.7370643380775542, + 0.2976166773434778, + 0.43815693821100554, + 0.29080021083014485, + 0.5972692238144117, + 0.34227216740508093, + 0.7360826396774605, + 0.41039707955001525, + 0.4179604082353492, + 0.6917316924955693, + 0.5162185602291558, + 0.32421797533641805, + 0.5619438118823875, + 0.40158615233655365, + 0.5150965286564594, + 0.5012396135671805, + 0.33245823218624204, + 0.4941315504228336, + 0.3896521174583142, + 0.45845760819295545, + 0.3696306887151691, + 0.29555886028621847, + 0.4020347808085674, + 0.5164752800597694, + 0.5770548124790313, + 0.6597681888787584, + 0.3988888635814351, + 0.2979340982459002, + 0.4589221532967686, + 0.4622458849498789, + 0.42464426032219355, + 0.4847511420080725, + 0.47080746806979706, + 0.3703562979604819, + 0.32558911353376974, + 0.5020563779169956, + 0.7657431467043804, + 0.5294925624537287, + 0.6005484787827983, + 0.5892448712630372, + 0.7702087361365726, + 0.5656542543571638, + 0.32417371852799004, + 0.58224584540292, + 0.37803103533653365, + 0.5168879905848571, + 0.45211910603670397, + 0.3937636282800192, + 0.4926789133284808, + 0.7809376735081234, + 0.22616389339527282, + 0.4595905189743278, + 0.47146073289833307, + 0.6219178997996748, + 0.37872626377972124, + 0.2778900928062071, + 0.4197605609179524, + 0.721373891224438, + 0.38398141045770195, + 0.4393524177238579, + 0.5057911932045563, + 0.4440705436652087, + 0.40453531989690084, + 0.4959503331246935, + 0.6121993248717236, + 0.46002310243520106, + 0.39457943661420736, + 0.5449401780572999, + 0.4545427835948169, + 0.567303020361047, + 0.5423520569161125, + 0.5290534106953266, + 0.547494160982632, + 0.30926184613728847, + 0.30237627682464074, + 0.5200817166208199, + 0.30575933771531544, + 0.6929033011113322, + 0.3019582504777206, + 0.5601449665692134, + 0.5642595727947821, + 0.6845000772601535, + 0.29783296031327433, + 0.3436460566737829, + 0.33555057416633316, + 0.34314683964645226, + 0.29467697726236647, + 0.2838012912101802, + 0.6127644576175492, + 0.35333684275660765, + 0.6810855187867659, + 0.5773576884888677, + 0.4391572445108092, + 0.5688243118902694, + 0.3814515830867345, + 0.5440818654233985, + 0.531600975976775, + 0.4637907541746405, + 0.3845331348436472, + 0.45988839222307376, + 0.4777856618477359, + 0.520375940625369, + 0.39141301889277896, + 0.32949114701287724, + 0.6172720236459315, + 0.47089482127145715, + 0.65625961524441, + 0.6331746435109219, + 0.5042957673870477, + 0.30881026750820495, + 0.3358378843935976, + 0.3520860993826466, + 0.44617230399263363, + 0.7461750626798472, + 0.4686502229615195, + 0.3721883405163216, + 0.7410496692738765, + 0.433649058515121, + 0.623691245583243, + 0.527531646935405, + 0.40940685661560555, + 0.497777079678254, + 0.452385256988061, + 0.45087210534260225, + 0.5771802555747595, + 0.531629847574302, + 0.5216273582193759, + 0.43251161709512204, + 0.458255417017982, + 0.4618640588821335, + 0.4817640824270942, + 0.7677667626765837, + 0.48509947441884, + 0.46687301178550505, + 0.43611867474074845, + 0.2781536035321045, + 0.663259702898497, + 0.3982184030225226, + 0.3212602668813447, + 0.5116067213272514, + 0.426147995487826, + 0.33108431263616, + 0.48069396801833747, + 0.4259838706608684, + 0.7777671455443211, + 0.3129152570049126, + 0.3845647734778447, + 0.32474341440815757, + 0.42663492512298223, + 0.39128210413708264, + 0.2966882419908027, + 0.3974107284802778, + 0.270177279794066, + 0.41437782050904215, + 0.7393163268955082, + 0.30417486846369907, + 0.4002135119444603, + 0.49164165894826806, + 0.4117751398260271, + 0.3809773815323981, + 0.47195101987988913, + 0.35135692788357176, + 0.518297665523295, + 0.519432903557322, + 0.21648083757329553, + 0.6261464220743058, + 0.6837662906322052, + 0.49519852002748765, + 0.47380784699494277, + 0.5194519089920393, + 0.38311385786231744, + 0.37230672753798744, + 0.45288132400082964, + 0.7791379839196328, + 0.36410999666745364, + 0.5988362087286039, + 0.3375557239181163, + 0.45248088482492305, + 0.47160035193726024, + 0.4836639375505423, + 0.3791749670913694, + 0.3172318062884495, + 0.3141978170336145, + 0.30913699770718395, + 0.5021298495658991, + 0.4380725919147088, + 0.28634266405192665, + 0.3992087728128481, + 0.4465456502289053, + 0.28999155828671613, + 0.5513515394773877, + 0.4800371742977837, + 0.3899823998868643, + 0.4489908617258422, + 0.2954373862187052, + 0.41908094194321344, + 0.48883137073674743, + 0.5749415635413496, + 0.4313395241807911, + 0.7773737286669932, + 0.32067753974924357, + 0.5524203104188596, + 0.7350508174529528, + 0.6297514321743168, + 0.2558507832386817, + 0.48861830336771783, + 0.41531620703012867, + 0.3337625813332702, + 0.35490634895918094, + 0.3606752953673764, + 0.4590260312456899, + 0.49389873361754855, + 0.3408314767568274, + 0.606465772922958, + 0.7644929010573585, + 0.4313854205004317, + 0.4184696623787333, + 0.43214131551003315, + 0.48638757505502883, + 0.5071747442970744, + 0.4903563212775756, + 0.4824135346253375, + 0.31556643130458883, + 0.6484104166059688, + 0.47209511076396143, + 0.55028474111184, + 0.7090273003324505, + 0.3075408087416877, + 0.5894412209028733, + 0.3846464219960449, + 0.44813588008244976, + 0.3430459013826973, + 0.5129023044123953, + 0.46016126817497377, + 0.31926636868423136, + 0.41604171197613243, + 0.40960035899829106, + 0.537831461662178, + 0.5856916284018143, + 0.4411785712246451, + 0.4565275656336522, + 0.40566427257460974, + 0.7378474491172615, + 0.44275247288828085, + 0.5135363193099706, + 0.4568468992828316, + 0.4628050659565664, + 0.6055117965155824, + 0.29967609950415897, + 0.5621181324049073, + 0.37247381053695644, + 0.5316458411794, + 0.41385117380371583, + 0.33138318370837866, + 0.33230960193293113, + 0.46337211748341733, + 0.3218372652260802, + 0.5026102226019089, + 0.7304711988052972, + 0.7246595681955201, + 0.3788939607967538, + 0.3293706009619925, + 0.6994218718368574, + 0.44202716400896436, + 0.3767137536668939, + 0.5065212111802881, + 0.35326335049981183, + 0.5923919732613423, + 0.43247466573775806, + 0.5600221051720106, + 0.5849816920425036, + 0.4317071490284017, + 0.5071275276239526, + 0.6536078853383165, + 0.49922691086794024, + 0.35200215355420883, + 0.4038987731902405, + 0.2993681832374027, + 0.5326968970378924, + 0.3393751051605424, + 0.5116730592844176, + 0.4905098343399648, + 0.3534429667669461, + 0.7173167970099692, + 0.31592666292045196, + 0.7222426903943765, + 0.43131822183108715, + 0.5365392205040014, + 0.7396448907056041, + 0.4736573575595518, + 0.5477666685036625, + 0.3784126424150969, + 0.7027799043495663, + 0.407787412724571, + 0.31423347435502347, + 0.5551335194006164, + 0.512454455886931, + 0.6365374740131904, + 0.5436877071324735, + 0.3849554798671106, + 0.48039755577693155, + 0.2752661036837499, + 0.4438698003429662, + 0.5610113730338104, + 0.5040748989957455, + 0.3455611269446086, + 0.5341896431221006, + 0.5768280539461635, + 0.5637697959944457, + 0.4280346852236538, + 0.33987959384629135, + 0.5019642456019072, + 0.4605286033597356, + 0.5527680219046573, + 0.4819400052982934, + 0.3327583297075912, + 0.40095124306470836, + 0.49883675283425055, + 0.39595070477955213, + 0.7309776356851023, + 0.5083833988982625, + 0.4420166351203321, + 0.48932748695721673, + 0.3865484320073143, + 0.4263192241787834, + 0.5367940151988635, + 0.5443065281923428, + 0.7094918495355687, + 0.5307781914539003, + 0.5718018439633406, + 0.5221129158286755, + 0.4748732363164554, + 0.5701186027942452, + 0.3777279647613379, + 0.6282325883210288, + 0.5931619074662708, + 0.72353773995448, + 0.511168443442904, + 0.5415485450801022, + 0.5375874251803516, + 0.40015365616581405, + 0.5326282096077909, + 0.33090300171857195, + 0.6752745538544154, + 0.4745362313532928, + 0.7355280583373852, + 0.7584999263974478, + 0.46454085869781936, + 0.44069000562027855, + 0.4921620110168551, + 0.34020713784959394, + 0.4921870312387604, + 0.7582619989987169, + 0.48290094819140417, + 0.500412363711624, + 0.5837626551929738, + 0.4935343318840699, + 0.34014001307144076, + 0.6057761149901337, + 0.4222044967695073, + 0.3859109542437424, + 0.733004097408161, + 0.40554611817609776, + 0.43305756328735956, + 0.39000997121965725, + 0.48111101201306156, + 0.5576135457940594, + 0.617040285826947, + 0.6784004403576035, + 0.6223689451016104, + 0.5860825174139697, + 0.45380068189079364, + 0.6016218754216995, + 0.5505976726589781, + 0.7809097227988496, + 0.7414049964809024, + 0.34245115579541624, + 0.3660448075416062, + 0.3223087099982162, + 0.5151396419001257, + 0.41369010969359954, + 0.7880157805407817, + 0.37338475596037185, + 0.416616513999295, + 0.5601413637990448, + 0.480298219808855, + 0.5596465117570438, + 0.6216547441694665, + 0.4674803983128785, + 0.6178782209719026, + 0.37749409259257494, + 0.6811876273621675, + 0.7806237637480062, + 0.763515742252995, + 0.576719881129052, + 0.4262428144340435, + 0.5449619909501885, + 0.741553401735772, + 0.6890408704510649, + 0.623489697479835, + 0.4812556107224522, + 0.43883002316048203, + 0.4903960437117419, + 0.7358679728693545, + 0.4141833859554066, + 0.42429557881387914, + 0.30697585782875275, + 0.5980891965133359, + 0.2962009181493972, + 0.4056265666027942, + 0.6629262435611264, + 0.538124155370153, + 0.6828770301324774, + 0.5425049912942057, + 0.5118498149058074, + 0.6671892098837823, + 0.6335258129197983, + 0.5625948412992087, + 0.38065482498433745, + 0.6447604184852752, + 0.5191243452075902, + 0.4010191784940827, + 0.3464774693098417, + 0.4784463378421309, + 0.6376772147383737, + 0.5030973202649857, + 0.6107509711032378, + 0.5739201732548184, + 0.4178089030665672, + 0.40438223678583557, + 0.6871924664685535, + 0.4927278578283523, + 0.5947099423822944, + 0.7244137343779138, + 0.41350277642825045, + 0.36186168778015493, + 0.35767424696865663, + 0.3657489134149264, + 0.713005710122731, + 0.73661401256134, + 0.5644599040932388, + 0.4203237081801654, + 0.4585494992582472, + 0.38723914092098094, + 0.7818244812529016, + 0.3458617505581737, + 0.48090014788419966, + 0.3344890574543124, + 0.5003960209361931, + 0.36583685488990014, + 0.5619569869678956, + 0.48038074785403645, + 0.3149932996745291, + 0.6963813811526661, + 0.5100775512271222, + 0.420391218483512, + 0.33989339780476946, + 0.3586518477075137, + 0.4064921000882136, + 0.4222993738581071, + 0.37283934407738667, + 0.3679174427488767, + 0.4051231655017693, + 0.3843045832605413, + 0.4129803654985184, + 0.335493795423157, + 0.44461532594201486, + 0.21188416982012653, + 0.2821212467145472, + 0.335049459573231, + 0.39201485205530295, + 0.4462811376086182, + 0.4314437217933223, + 0.33978645822046916, + 0.3428114516818839, + 0.3346393496132945, + 0.6721708574199874, + 0.3378175494282749, + 0.31781897180638014, + 0.7884232288538936, + 0.36958370832371007, + 0.4221909820942315, + 0.35629799028937686, + 0.45257092207334715, + 0.28914108842994185, + 0.4072344958591592, + 0.36092937540188647, + 0.48857195316535185, + 0.269286908538406, + 0.41708784817494, + 0.4026662233887934, + 0.509482557343343, + 0.3937625456881589, + 0.4744582290924871, + 0.3105581493512856, + 0.4592532291915785, + 0.41338217039343333, + 0.3610970008403252, + 0.4728552746256681, + 0.37290145356552534, + 0.4770738552850097, + 0.38781713601946827, + 0.37230581336494234, + 0.8080455636454401, + 0.31232359093346246, + 0.43890168844847266, + 0.3680250916659896, + 0.41297462907664545, + 0.32758511245545496, + 0.3809039148295364, + 0.46184326050648644, + 0.5528605785845256, + 0.34768054182017083, + 0.3858405239826337, + 0.295597384915351, + 0.4143728354428148, + 0.42883672282856455, + 0.43036641321299496, + 0.4598087563197314, + 0.37180750206776275, + 0.3502082291650177, + 0.4704535561044929, + 0.4067888037865963, + 0.45371065119005266, + 0.3043698053724708, + 0.3260543371564298, + 0.48406831183151455, + 0.45965169460663213, + 0.41174344359428244, + 0.3745809616394242, + 0.3754903168281677, + 0.8373765068838567, + 0.33310811406738544, + 0.3385530760169898, + 0.3593668711487589, + 0.3179701802264148, + 0.3524839054317967, + 0.3557424532332586, + 0.30688931373667944, + 0.37131258163657, + 0.3536627971373499, + 0.43224790980288996, + 0.3754891330339303, + 0.4328399862143765, + 0.47268326808371747, + 0.39712226026421416, + 0.3812323672843461, + 0.5563315429316921, + 0.3842046157445325, + 0.4117618843753924, + 0.4981489097151532, + 0.44876887366723006, + 0.3520115668696075, + 0.293592022827938, + 0.3679501478664099, + 0.4836095052354127, + 0.46923653468130977, + 0.31747911622752617, + 0.3604319436805096, + 0.38073823399422624, + 0.4536942696425863, + 0.3342160673759598, + 0.34541032900154284, + 0.48984450810751023, + 0.5275699678198359, + 0.304131300465127, + 0.42608505734812857, + 0.31908298009387004, + 0.33493436189189196, + 0.3352901763947207, + 0.7886447566025779, + 0.420735755863665, + 0.4399118252699436, + 0.30949417653990235, + 0.37397622344835335, + 0.3715030959683378, + 0.6997233343563912, + 0.4200090527317764, + 0.3852508090153669, + 0.44760838779362344, + 0.4173597517457224, + 0.41268892024741294, + 0.44852423897578125, + 0.4866724089321276, + 0.27843944235188833, + 0.3407686862021885, + 0.44272083163929404, + 0.4175062592653316, + 0.4110379643200318, + 0.4069525830604227, + 0.366351802743941, + 0.3131043373526154, + 0.3455808470052427, + 0.4041600685534896, + 0.3751405953555927, + 0.6072351083392735, + 0.3485963110241786, + 0.3672024095690023, + 0.30619324290395195, + 0.5536270614337955, + 0.38671572289899814, + 0.3190083406328866, + 0.3798314529878347, + 0.3999921229336574, + 0.41106318084207444, + 0.3245111279272731, + 0.5449352972061483, + 0.4580126504657411, + 0.3108506993913761, + 0.41248508053883154, + 0.44236610810488536, + 0.421935831444417, + 0.37215766755093843, + 0.4660342022021768, + 0.3327424034893091, + 0.6540934875296026, + 0.3874834412640951, + 0.457478165251527, + 0.4557917126225714, + 0.3907336760174051, + 0.41181163968331563, + 0.385345520285648, + 0.2980038268109727, + 0.3899905459590069, + 0.5929367793107205, + 0.36998425833979554, + 0.321556859556741, + 0.5025747274926531, + 0.4214071731631406, + 0.4150411318875663, + 0.6149388410389934, + 0.3811002693249397, + 0.4417934020816852, + 0.36998942426693626, + 0.38320987309435656, + 0.4775714707388495, + 0.35868369138756695, + 0.4670877773226906, + 0.4113367624154988, + 0.44555431710809873, + 0.31563746773644713, + 0.42875394588535076, + 0.37054925840784364, + 0.3218027435700553, + 0.31761174649814244, + 0.33797918620721895, + 0.4300181671793528, + 0.5320589465461495, + 0.44812944481387124, + 0.45301772077062546, + 0.463003014823761, + 0.3405915859416727, + 0.3666568438030427, + 0.47766683487675254, + 0.36444306010057614, + 0.39804157521140454, + 0.4774038087975074, + 0.38281577414694734, + 0.42827086253585206, + 0.5015687973089314, + 0.402841401088828, + 0.3653192292492837, + 0.5075519292173001, + 0.3745792594370496, + 0.43786218027575846, + 0.31641341472968737, + 0.4949304712580017, + 0.5996546909161979, + 0.7905287082672429, + 0.4245825907511626, + 0.3839891056479695, + 0.4259823482595118, + 0.36456076180154456, + 0.35523644950320205, + 0.36285965661634373, + 0.4356354107439823, + 0.4026056568644564, + 0.365011627342655, + 0.4461651509440224, + 0.4255095785788457, + 0.4216508211700602, + 0.328659656591265, + 0.3186545167435765, + 0.4293501485387886, + 0.4113623117372862, + 0.38199089325855257, + 0.39394818897547135, + 0.393927885364331, + 0.3621222944793437, + 0.27209422822428114, + 0.4239046745299145, + 0.4391052744742099, + 0.35212736065841016, + 0.3969072570560346, + 0.3042874675427157, + 0.38742522749812097, + 0.25772053426699343, + 0.5223646912113213, + 0.35513181774816543, + 0.32264289809184804, + 0.4215454107715246, + 0.38523089118957665, + 0.3297549611284216, + 0.5965470646890305, + 0.4943530021244993, + 0.3017258630987595, + 0.46706422439014245, + 0.4120504787860754, + 0.41153229626278837, + 0.3844817577386932, + 0.28301373462528157, + 0.3643418834818193, + 0.4965778950254113, + 0.3241140437622518, + 0.4257484081894881, + 0.3154364197530768, + 0.3594602743149624, + 0.36249459472413975, + 0.3334801254051402, + 0.41413469103019457, + 0.42438295196253645, + 0.36876776126690136, + 0.3432956751182862, + 0.4091087390474632, + 0.3188794545614073, + 0.39063817521328714, + 0.3840687417655546, + 0.4893370846826104, + 0.5262634578556646, + 0.42599076087416904, + 0.40880293803386175, + 0.4222282306688773, + 0.43522234939744425, + 0.4387726346863055, + 0.5269298901322131, + 0.4009588905380877, + 0.31010953396242347, + 0.4996513230536671, + 0.3874690981640041, + 0.4505263534629163, + 0.44362240877222076, + 0.38227229081791514, + 0.4907240890028803, + 0.3135284844423742, + 0.3679870161205268, + 0.3090134264777769, + 0.3078409827749269, + 0.35015271801681336, + 0.39333742974797203, + 0.3311963026570927, + 0.4091692259023697, + 0.5283021447836622, + 0.4159497603639251, + 0.49374859578660846, + 0.7780793309032497, + 0.5542314878910382, + 0.3480581498043327, + 0.4147310212315643, + 0.44911761395680383, + 0.42497403482364693, + 0.3048335851783062, + 0.5772720988017427, + 0.46890468230986276, + 0.39022495111392075, + 0.45773862110172575, + 0.3786119422115637, + 0.4838551428477536, + 0.5047147211986455, + 0.391992038818749, + 0.5300560965151687, + 0.48293710999281736, + 0.3708495096128626, + 0.449762232422241, + 0.4975992452018941, + 0.3574982761642107, + 0.38081568600301235, + 0.3043168335031824, + 0.5650456707948569, + 0.5077567137492859, + 0.8204409479800324, + 0.5279569608341016, + 0.43055135308325043, + 0.2795464224569301, + 0.4697573603052431, + 0.4310423995672183, + 0.6031613730178058, + 0.3412967581162659, + 0.5346524044628068, + 0.32740700589828337, + 0.3527600196184177, + 0.3247294777545514, + 0.3130253622268082, + 0.4171476996447865, + 0.3044331849873531, + 0.42314217141205257, + 0.4461530629868149, + 0.3853916616037636, + 0.32030151049796607, + 0.32028849192795933, + 0.41217560176283585, + 0.3065344931123071, + 0.5228419770768635, + 0.42451957534960616, + 0.4897304693458345, + 0.4172986685873052, + 0.3538651085259514, + 0.5548138901362114, + 0.4701935786500853, + 0.35581372480058865, + 0.5894730597740228, + 0.29958791728406625, + 0.4393561005152143, + 0.35587014151543767, + 0.33300958644672585, + 0.3168057124416852, + 0.3650453599625866, + 0.46905485195765856, + 0.33404335422329345, + 0.6894609086875456, + 0.44586888024968924, + 0.3211652586888742, + 0.37727172361483513, + 0.33226116211570395, + 0.5681499869184641, + 0.28932528379269223, + 0.3636262786807896, + 0.3785270444865995, + 0.4205014095385593, + 0.32151344503856993, + 0.4002187287420918, + 0.3073884993695716, + 0.43348350857480283, + 0.7660299622493089, + 0.44627704031148807, + 0.37494241353356955, + 0.3899937309164478, + 0.38490921231268627, + 0.2961106086364629, + 0.3655935012519307, + 0.443047659611577, + 0.7497284591209734, + 0.5458336426505522, + 0.4486090940945247, + 0.38496839093678353, + 0.36177096405587866, + 0.40472662372024715, + 0.29201663870467015, + 0.42859413786189343, + 0.5023640116483262, + 0.3262879922222776, + 0.38623301805926696, + 0.3770529325078513, + 0.46275812359957746, + 0.3410611651828903, + 0.5415844757880602, + 0.4186832016656557, + 0.30628893940836394, + 0.4145846755140551, + 0.42435290079722576, + 0.5274892401756682, + 0.5321926962890476, + 0.2395334559039733, + 0.7231317589291899, + 0.7117642378243322, + 0.20404862394759474, + 0.29647310995631265, + 0.37946245416302615, + 0.4437446561315064, + 0.25651654932990414, + 0.37195081575444827, + 0.4572361167441996, + 0.469556635601744, + 0.457845912788147, + 0.3630804401660541, + 0.35663636381954195, + 0.34779359288797307, + 0.31286000428548605, + 0.42889578271327855, + 0.37174775025368795, + 0.3744266116817794, + 0.4984010870725768, + 0.40847158307794407, + 0.5232034782568419, + 0.4405508779466523, + 0.6046265871134088, + 0.433682413801217, + 0.4230406206710458, + 0.4887225530166219, + 0.44779981057891405, + 0.32790728649711576, + 0.6245449939656648, + 0.6038703754354798, + 0.4347771047847512, + 0.36400295152366446, + 0.3762204239105887, + 0.5272611413738173, + 0.3591961345811855, + 0.431125452119383, + 0.4137263507810754, + 0.4716829196639153, + 0.4341815439697165, + 0.3794826194278713, + 0.3763043835767482, + 0.536906142146383, + 0.4850953119422774, + 0.4608512364546785, + 0.4804728715316805, + 0.38319928633886835, + 0.24591805837098274, + 0.25593783710496004, + 0.3188395152385936, + 0.5191953762034839, + 0.4570761908337088, + 0.2735758382878668, + 0.4420329652452995, + 0.3475768401975452, + 0.33551070573458, + 0.36158521539932476, + 0.38885889309789023, + 0.3502667524886737, + 0.4377619258480159, + 0.6240422262244347, + 0.3300507543810492, + 0.42599459981362614, + 0.5312477409692264, + 0.35181858769692287, + 0.44072648024038286, + 0.5925526142330901, + 0.45271017759690646, + 0.4478116166366312, + 0.4346599679303117, + 0.40675749783110965, + 0.2666884218974284, + 0.3594400849129541, + 0.48566163803374457, + 0.4945069705469527, + 0.4732008038875598, + 0.3165088537386794, + 0.4595610166461654, + 0.4609637883256069, + 0.505807393136704, + 0.4027630848243445, + 0.46556780865392194, + 0.3841239700152076, + 0.46140566719868054, + 0.4435320477699979, + 0.4119189634029288, + 0.30722832613714846, + 0.4523541935632387, + 0.46282305487518344, + 0.2549617667229963, + 0.36298204116087573, + 0.43713124738162656, + 0.3982793334824274, + 0.3905200187574037, + 0.4884736686201511, + 0.4924788679296448, + 0.44184924511426554, + 0.47079535170215514, + 0.349037812332234, + 0.4588548883652495, + 0.4059042974963782, + 0.3843657575913012, + 0.4029396376292509, + 0.4482622434474891, + 0.40190373470292107, + 0.42446969630924136, + 0.41163423501414187, + 0.7220730047330919, + 0.5449179649026471, + 0.5049466328550373, + 0.5340273190454351, + 0.48223532807551545, + 0.437019801103046, + 0.5391852805182439, + 0.35672077953810105, + 0.39494454155649905, + 0.4723566545068504, + 0.33602390698650103, + 0.4440452637265966, + 0.4988958307956294, + 0.4421355719002137, + 0.4473034651984113, + 0.46253699614917426, + 0.3261949552056901, + 0.5354386838146796, + 0.49229368639308735, + 0.37881025381744987, + 0.42989280374148325, + 0.3822193714001733, + 0.35111912240697396, + 0.45287744481513814, + 0.44258598826606144, + 0.42043330335597484, + 0.45823082209922195, + 0.3895203199757849, + 0.4850456305839577, + 0.3669446403926367, + 0.39175681007274826, + 0.4350701236924617, + 0.45101911517725335, + 0.4399423907586192, + 0.4719336555296152, + 0.4040402071790746, + 0.34703908915926596, + 0.3543813085641314, + 0.4365653712865259, + 0.4095528856431672, + 0.44522354973949396, + 0.4749983824051022, + 0.42651388266229684, + 0.41234049252028676, + 0.3556266828068283, + 0.4051325279472369, + 0.5288021913503691, + 0.3983410299229377, + 0.20399853243673224, + 0.35406116450197944, + 0.41615698353515335, + 0.3065595019261505, + 0.29588486473459097, + 0.4533908328062266, + 0.45277600825581316, + 0.2859793536442211, + 0.31842103442586145, + 0.36409356799603354, + 0.3524164703356609, + 0.5225158005984567, + 0.296217563972152, + 0.29824693784355805, + 0.31957525772176054, + 0.3290755010223471, + 0.24729786072004015, + 0.3094355353483059, + 0.2927620625293884, + 0.3188848935442596, + 0.28658987311049455, + 0.4083402839843128, + 0.306963987726699, + 0.3504420444724535, + 0.3551912539426524, + 0.4149164554727489, + 0.30583564529796076, + 0.38963440462265286, + 0.3571106270405874, + 0.3600484004769871, + 0.3613321018598468, + 0.39193860175938067, + 0.36927550564378137, + 0.606558045322974, + 0.32670520489077964, + 0.38541461666662685, + 0.5495532989213655, + 0.48170428962534156, + 0.37130307485448083, + 0.46328520593871897, + 0.41740309514040835, + 0.4572798120286937, + 0.4719632733971133, + 0.35206579088961854, + 0.5804767796714373, + 0.4255915574750814, + 0.4266615852451307, + 0.361469454836656, + 0.43219090083965694, + 0.38352533895260543, + 0.4567321139126951, + 0.47219969062113276, + 0.44972063407656526, + 0.4654777514133149, + 0.5732044964012428, + 0.4432764297608058, + 0.4046472404247929, + 0.46985818303034416, + 0.5344432800184621, + 0.5242382471608549, + 0.4408703115148106, + 0.5651954728623687, + 0.5559018134572292, + 0.38742557789115667, + 0.3847405749413783, + 0.4344696495347941, + 0.34749794227297265, + 0.6363015284337586, + 0.4852374535826876, + 0.42122960667521747, + 0.35266587789824705, + 0.4628192995111642, + 0.4020257296717494, + 0.5185819343531167, + 0.43693595807998115, + 0.41367129006419384, + 0.5100286156118369, + 0.3095615611164899, + 0.41519772661880877, + 0.5352113905171504, + 0.3380822087601393, + 0.41575796844075924, + 0.3225735850527809, + 0.36196993787503245, + 0.38395597669035714, + 0.3927391714777276, + 0.46225770349755685, + 0.3381463487268555, + 0.32803506425925083, + 0.42451861047618644, + 0.39339891445705016, + 0.3737879920233084, + 0.44708823718943186, + 0.38881183597667424, + 0.4181052705273802, + 0.5045724470482128, + 0.5592597906974823, + 0.4595363982350429, + 0.3579248993870039, + 0.46810862369211925, + 0.4394890931971599, + 0.38868515540849236, + 0.4224869522829167, + 0.3214401445723046, + 0.3268065232503334, + 0.29999317896239736, + 0.3672473114710661, + 0.3165510711083621, + 0.6501091527602427, + 0.4297335094171178, + 0.410956286057799, + 0.2951031527390751, + 0.3123161550957474, + 0.3189365612570286, + 0.3446623795303264, + 0.3113053521537854, + 0.3202201237555861, + 0.3073887083936712, + 0.374608819877006, + 0.3883973872733823, + 0.4213967735157418, + 0.40665264414191116, + 0.3664587833301349, + 0.500090799980345, + 0.49396206931887277, + 0.37460041450513193, + 0.41880402520101023, + 0.42778325626827335, + 0.3864739847828184, + 0.3558055101384052, + 0.5108446545148732, + 0.44413961840293237, + 0.5791534888479993, + 0.4471830474008387, + 0.4405105667755145, + 0.44271001956494593, + 0.44930590643310203, + 0.32613048679738454, + 0.5940188679146657, + 0.41036200253084615, + 0.23012598037264312, + 0.46343618803967374, + 0.5719614324473271, + 0.3635502213879553, + 0.4540533640783699, + 0.750926713531713, + 0.43804159941640963, + 0.5105937774064945, + 0.2860293865396497, + 0.47501344942506213, + 0.2881482813439918, + 0.298440586623002, + 0.317654573099703, + 0.2760338336199167, + 0.4458006429061341, + 0.19966603717492973, + 0.31418648523780973, + 0.45136520148416137, + 0.3967534087894324, + 0.40012711026016307, + 0.3899736709920027, + 0.33208545917904153, + 0.41047102044467093, + 0.5175209766182488, + 0.5358249835917058, + 0.35345726957626306, + 0.4473564064148375, + 0.40783544993102105, + 0.7294634291818332, + 0.3924479748338592, + 0.5572900963465093, + 0.7357330467733332, + 0.7144132959449743, + 0.44164701447948035, + 0.3760647428533516, + 0.31442261316569936, + 0.4494975988130142, + 0.4907331098033215, + 0.49187203513027694, + 0.5700197837526055, + 0.4225839365611666, + 0.5494058841452346, + 0.43242438500846897, + 0.3858867269535689, + 0.6998420463299633, + 0.6809578382606961, + 0.46414193391762804, + 0.4153933277660704, + 0.4651248626004922, + 0.41819275429664066, + 0.4496736312250112, + 0.6415882574717872, + 0.665166327505988, + 0.4226424367053953, + 0.43303189053936036, + 0.43447908783840367, + 0.36370077073337037, + 0.49010402169667927, + 0.33064140180253726, + 0.48399103771504387, + 0.4976390050935858, + 0.34606343683897256, + 0.43603589050037433, + 0.39685418551003576, + 0.33661843030306393, + 0.5024470025957684, + 0.44384056333540844, + 0.3973012829201027, + 0.3667442341024304, + 0.47351685649997666, + 0.4164068430551691, + 0.3365920264624978, + 0.3413147771814358, + 0.2525187354208252, + 0.23882523193616007, + 0.2707649153993319, + 0.43374719258137956, + 0.4638985960950276, + 0.3321173707049459, + 0.3183089399757626, + 0.33312438500204683, + 0.4227871132849279, + 0.4236998456788445, + 0.4933259603446841, + 0.34453790465282685, + 0.41489635948048065, + 0.5096105844308455, + 0.36486402525616807, + 0.5225666948242752, + 0.4679758049837025, + 0.4139821252558154, + 0.3995775319343562, + 0.35227634700048377, + 0.23504738393372823, + 0.28332291263608556, + 0.27303537453858084, + 0.5138246553039183, + 0.4221279703426534, + 0.4166271813570954, + 0.46790981807955573, + 0.49198462461151266, + 0.3159705717742335, + 0.4097835094201798, + 0.3648774017776375, + 0.6631662840072837, + 0.43573664638787374, + 0.45551490357113517, + 0.3619672921803755, + 0.4492009546420307, + 0.4850075360551974, + 0.43228745436359417, + 0.4007915859059247, + 0.4823018573406428, + 0.4458618491568093, + 0.5542342405736996, + 0.41389834744192605, + 0.48043140073409246, + 0.4398219820893064, + 0.42041822190504236, + 0.39596518490829824, + 0.46596311770493015, + 0.4193922856152143, + 0.4845549967074681, + 0.34128112072312355, + 0.3184338301915716, + 0.3201634259956255, + 0.3976299668639872, + 0.44132249740776436, + 0.3620167758036589, + 0.4174467686888569, + 0.46138332386768405, + 0.41433382012712827, + 0.4288859581377576, + 0.42481641733699227, + 0.26817586395421866, + 0.40606344846351194, + 0.2540423424455957, + 0.4191575013889185, + 0.21915015991434425, + 0.6489906548832393, + 0.3194479445105004, + 0.4401200014933439, + 0.5463463680081418, + 0.4634697176534929, + 0.2672269432649498, + 0.4119609688965787, + 0.33015448636529043, + 0.7704856806635255, + 0.27498462157365394, + 0.5052188022845894, + 0.5034696267441392, + 0.5029621406280453, + 0.6277489064485188, + 0.41918640078256997, + 0.5507436019488505, + 0.45174423814183756, + 0.3778712046513979, + 0.3222636670148533, + 0.4486248557835911, + 0.3067107081430887, + 0.7118098371133914 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.4281288798270353, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.46656019131687143, + 0.7583070910576802, + 0.7353928419185474, + 0.8176371821020098, + 0.7379750963990221, + 0.6499281934087131, + 0.6264230020081837, + 0.6602774069629418, + 0.4066219558450325, + 0.697561965104789, + 0.31691672721518677, + 0.4758239202486567, + 0.3364313838820642, + 0.32136009831782725, + 0.30461254244333696, + 0.28922035260032797, + 0.3389896270001883, + 0.5435365053335406, + 0.41058279923730445, + 0.40299616121995047, + 0.3273652822861569, + 0.6695001675786593, + 0.33315367276821756, + 0.5868784386888967, + 0.3639440790689118, + 0.29567611269388905, + 0.3883339132218831, + 0.4352889361296461, + 0.5123563845287294, + 0.4855637310010544, + 0.2807244697662876, + 0.3815161807136012, + 0.3412375091983968, + 0.341622137713671, + 0.4341396434695882, + 0.34844868599584056, + 0.331181455571083, + 0.5143569900546336, + 0.5419558327794293, + 0.6270953427415484, + 0.7886917304851968, + 0.7815610779881241, + 0.7716948161798403, + 0.6567388176782276, + 0.3876208022473146, + 0.39174322925218696, + 0.7796581249261619, + 0.3496470354118719, + 0.3505569922857527, + 0.48605687503405354, + 0.3905064000857546, + 0.7771606521954764, + 0.7948961908346379, + 0.39338850573471773, + 0.3484209676919248, + 0.3599383903740781, + 0.34914048206099035, + 0.40470677859234255, + 0.39686936652072535, + 0.5049082510647491, + 0.4491019550610964, + 0.43250197512595473, + 0.4051627195919818, + 0.4953995285933013, + 0.45442824769305257, + 0.3787460420893833, + 0.35467517257496656, + 0.5336079506896385, + 0.363540774302704, + 0.43611556745216784, + 0.46896770575495966, + 0.43216302455249456, + 0.41186885991472755, + 0.4003238158211731, + 0.4080564693836408, + 0.4185409050845883, + 0.2854755361221134, + 0.2769091121279372, + 0.37021269075130864, + 0.3246231758347565, + 0.4323848605388848, + 0.3464543942773022, + 0.34007274622559197, + 0.3131211063595384, + 0.4046883333003473, + 0.48226418901748425, + 0.4027461853900655, + 0.2979508016759302, + 0.3408962243762627, + 0.35856144789362876, + 0.3400336526234451, + 0.3358889206945912, + 0.49241164015192557, + 0.4988012732632286, + 0.303954340202047, + 0.5584010314262483, + 0.3516457981532169, + 0.5262431355213454, + 0.3202209649708427, + 0.4557979522488516, + 0.4402309649767155, + 0.5088776413349527, + 0.24346557857042783, + 0.39100002065863004, + 0.300997890813308, + 0.3227221809102014, + 0.4565141875670874, + 0.34954671306336565, + 0.32561974319932285, + 0.8099725777481214, + 0.8041297515861252, + 0.8088971443812631, + 0.8249290096727914, + 0.5821354855780281, + 0.6859838918787811, + 0.7067460610842611, + 0.755355426863009, + 0.7059026561003575, + 0.5881252199720551, + 0.6539699956792712, + 0.5767792145022473, + 0.7726408393355203, + 0.6574575733884737, + 0.42188054457031016, + 0.32541889586956824, + 0.28342979673983243, + 0.32361424326478816, + 0.5389727331849697, + 0.32313562264267043, + 0.28039551697813125, + 0.40058018240315213, + 0.37873600030951965, + 0.32965807710758016, + 0.35618444444795494, + 0.3975682494196181, + 0.41673511160387905, + 0.3866572687624166, + 0.32413718592693064, + 0.3609546768807032, + 0.3834483864928094, + 0.33246072842533325, + 0.309229350835868, + 0.30400477185076497, + 0.5923583963763623, + 0.30516777188646654, + 0.5643097775575278, + 0.36055325531907195, + 0.4333306617835748, + 0.3020135362155363, + 0.5559043729499249, + 0.30346324357132026, + 0.49905961352096306, + 0.31877562841565027, + 0.2685873485569941, + 0.35533968010656025, + 0.3186836923445144, + 0.2631393088689305, + 0.18072314355440663, + 0.36905583131500014, + 0.3300832998447794, + 0.47962845495063483, + 0.44494040330423223, + 0.5031270529569533, + 0.4713337818382649, + 0.3402051562287155, + 0.45605676226815584, + 0.4148903903030823, + 0.4356671454042672, + 0.41390322017254777, + 0.42458668959064944, + 0.370451237790057, + 0.2139537206595189, + 0.41820806453083903, + 0.32991145575771075, + 0.25083009676600254, + 0.2518739133122761, + 0.48138307687446485, + 0.4998178873301098, + 0.5642487694575065, + 0.4677225533615942, + 0.28145569438348245, + 0.49419293171457596, + 0.47993501774093544, + 0.5116782164136116, + 0.29572305673021176, + 0.5205974246131514, + 0.43202265539253004, + 0.5131761775006941, + 0.24573644147279908, + 0.38740002837480497, + 0.34197964733951025, + 0.26242640197886, + 0.5025203875069845, + 0.432123629414925, + 0.5462719748322423, + 0.3130768211114686, + 0.5904917765917941, + 0.29578615551960874, + 0.27451468025140097, + 0.3933236352768607, + 0.3009023042118658, + 0.2689045987948718, + 0.2540845470429569, + 0.35800852078561995, + 0.6286841765856283, + 0.26138258611122234, + 0.2500731475144523, + 0.41364709523504334, + 0.3799943937199061, + 0.38581085830450357, + 0.44850346890786796, + 0.3927683006666799, + 0.27387979210393243, + 0.35614236773323343, + 0.3337446544427348, + 0.5189374952275891, + 0.5473295953263118, + 0.5628295802102036, + 0.5495710222354202, + 0.2898290419900689, + 0.3806464140128351, + 0.42438479686935043, + 0.26302929456432816, + 0.3319160739120177, + 0.3343041094323742, + 0.1584601944801178, + 0.46472610732298003, + 0.31480312623482776, + 0.2939467059484362, + 0.5205183942870506, + 0.24686426269871514, + 0.2053993086674685, + 0.4223008795703671, + 0.3807309557518312, + 0.2885317024673626, + 0.605335241955018, + 0.365220146897691, + 0.29278145163745983, + 0.27669989284336616, + 0.4211314066585971, + 0.2985784292873111, + 0.44900455057584626, + 0.2866114197738733, + 0.4009933522455402, + 0.3930500660743039, + 0.37640158124223544, + 0.3337360324784115, + 0.3250590354985444, + 0.36523050086733866, + 0.3386515402604782, + 0.4299446837729385, + 0.29419040275790004, + 0.3754610794875578, + 0.30919947998377634, + 0.33828030455593655, + 0.38342643267703524, + 0.383966736942098, + 0.35249450062881554, + 0.42498740816347247, + 0.3144517200870946, + 0.31079183057557874, + 0.5093976072637617, + 0.2732962210621321, + 0.32213447229034975, + 0.31903650025871966, + 0.25958582313133016, + 0.31626050549985546, + 0.32316454673437767, + 0.26665381983719344, + 0.3037117859406053, + 0.28194885751909193, + 0.3211099046643772, + 0.32844477098289027, + 0.36615480511858267, + 0.3632918366462331, + 0.45976161965132406, + 0.244592493224313, + 0.38925785889505443, + 0.4565629922326056, + 0.2668411858477138, + 0.27120652871379386, + 0.5893608615494336, + 0.27853353022563765, + 0.25439215700950996, + 0.4524659403639548, + 0.43967269171663975, + 0.43932736706516606, + 0.36850145556941427, + 0.4888069418775122, + 0.44106465359662694, + 0.4548449476166496, + 0.37720920803767294, + 0.4474841182772107, + 0.4023087917971252, + 0.39876664867579736, + 0.4210794142316618, + 0.4620787641114399, + 0.43268474426180137, + 0.652891309455475, + 0.3135016338018826, + 0.5705464977710827, + 0.6400536679642316, + 0.6577138223044862, + 0.6611042684861598, + 0.49680737272824543, + 0.5195142867052344, + 0.39753554442729255, + 0.6759956313711911, + 0.42101508544285915, + 0.48566626517258205, + 0.3169543365941837, + 0.49344045380632856, + 0.43127296806142246, + 0.3827636236958485, + 0.440661686282662, + 0.31403304342609656, + 0.4194059089530477, + 0.27356633219090926, + 0.34434621008234007, + 0.3429824781578894, + 0.3370394899289199, + 0.3905184118707049, + 0.27130225188545876, + 0.32794796920376323, + 0.3519303271332479, + 0.3587724382248888, + 0.3232680457920206, + 0.35866628663355293, + 0.56360524186998, + 0.465488164158385, + 0.40504976430677986, + 0.47325329746135925, + 0.4584949396810214, + 0.29722594588980167, + 0.30217947335933776, + 0.43422511725002505, + 0.4023534990956249, + 0.31416876438208285, + 0.42453601304515737, + 0.5202418116742442, + 0.528703510676125, + 0.3861150001775634, + 0.2973681050916786, + 0.458562496485694, + 0.2737199069610605, + 0.31679721101451663, + 0.48508061631481214, + 0.3072390889742379, + 0.8233769455103535, + 0.5053316510021464, + 0.40093334035970274, + 0.38615549359399837, + 0.4424851406568341, + 0.4412535530145277, + 0.3854693639525693, + 0.4939602302299519, + 0.3482349083224647, + 0.5029037314212983, + 0.39262740008724195, + 0.45437373046539353, + 0.46150751640884813, + 0.4367357021381091, + 0.42655833173831137, + 0.34188559848427763, + 0.35998581606324437, + 0.5201333072838731, + 0.5157391978574518, + 0.4047787854685532, + 0.5777832509850428, + 0.4401254340572119, + 0.5420674819627362, + 0.3939415120581259, + 0.4349874948539422, + 0.4202371521712451, + 0.5538192030058263, + 0.5606074842038942, + 0.5274724866529016, + 0.5298053153098576, + 0.5075795436853919, + 0.5069844577964943, + 0.42362195593804486, + 0.3906946535770142, + 0.3634316467205458, + 0.2742267918251928, + 0.3691751673993669, + 0.3186996692566795, + 0.3195686447296471, + 0.352333766979662, + 0.4568179768951529, + 0.3342380652899826, + 0.29866973713291983, + 0.3442217381197905, + 0.54267430853712, + 0.6032196455930489, + 0.521406976907398, + 0.5617992843524288, + 0.3955604910927979, + 0.5628862156913219, + 0.3725670928546805, + 0.6247164339949575, + 0.4638307884435056, + 0.5485538115103832, + 0.4205528472353269, + 0.5328227430477088, + 0.5375693228008153, + 0.46885836606724474, + 0.4437645872016454, + 0.4957111785879889, + 0.4978829145498288, + 0.5077436027979518, + 0.3467548698206091, + 0.37247619481712746, + 0.3150644138191728, + 0.3571699914802288, + 0.4979375500482621, + 0.5652728413885436, + 0.31446341548635864, + 0.579428480583676, + 0.49891448071591865, + 0.47029104412712425, + 0.38477036587935876, + 0.4575492910511891, + 0.4745070461886513, + 0.42013848410697635, + 0.4046062633494409, + 0.8502650532701947, + 0.4300678870901562, + 0.4668549687262351, + 0.4852172604932328, + 0.5025511554124477, + 0.4684475969118222, + 0.4057062903693824, + 0.5868413298830226, + 0.4492811920400439, + 0.38759399818270074, + 0.44761105190369893, + 0.42986211810848235, + 0.4354060631860583, + 0.41210960145559833, + 0.41939260342499823, + 0.4320026442779013, + 0.43418188970956123, + 0.28905040272783994, + 0.4327622907519314, + 0.28999398183309155, + 0.3764557221105333, + 0.23587264562600338, + 0.43436176631065304, + 0.36106717862309384, + 0.2796507084852296, + 0.4945904696727182, + 0.2569406182467826, + 0.26251579216651727, + 0.5440210314769408, + 0.46818802569029006, + 0.4515333243032536, + 0.4627592318054179, + 0.40372778967002987, + 0.40457715101841846, + 0.6013269757691243, + 0.346162677983615, + 0.5444349185603007, + 0.36493897790011554, + 0.2690656058017508, + 0.21707740716610657, + 0.38918169326246155, + 0.4788763449089589, + 0.5341294123787632, + 0.5611811037620178, + 0.44728296059520556, + 0.5064940210803995, + 0.37091798747653737, + 0.48115469965120966, + 0.39906821640618506, + 0.3915085034885978, + 0.3945626724169801, + 0.37403073057862984, + 0.44858610525594583, + 0.3707376784595839, + 0.4082365018888236, + 0.4462982795665941, + 0.4288834967319819, + 0.4566100766250015, + 0.38930570418448485, + 0.38400540637208747, + 0.47951051921768695, + 0.4472448624178452, + 0.5216481755679963, + 0.5006051631549523, + 0.5527666659436384, + 0.5788476870105753, + 0.6653274281692027, + 0.6484990733056737, + 0.4755340620214814, + 0.517940695290087, + 0.4161479381795282, + 0.5181097798740345, + 0.3905652417772892, + 0.46959395251597547, + 0.388758107317548, + 0.5748956250682853, + 0.389127925274209, + 0.5466659955177696, + 0.6519777684955159, + 0.6223354741877364, + 0.5541319079577454, + 0.6391809305801837, + 0.5348622771316381, + 0.45256340196104866, + 0.39557681858450366, + 0.39862093242035024, + 0.29261042893237804, + 0.3597515707028373, + 0.4526559913277071, + 0.5331874583751849, + 0.4220880834262727, + 0.3529455311786388, + 0.2724944580202876, + 0.46110892147749083, + 0.44582721571565204, + 0.39327980924874417, + 0.3474350290501432, + 0.4986196158469982, + 0.37532681307089527, + 0.4266647719769361, + 0.3738114504752857, + 0.45617516232871513, + 0.4526563042382214, + 0.3231968922088813, + 0.37505924314532696, + 0.4421544050546616, + 0.3775104578906056, + 0.37404313807303996, + 0.3862749548096748, + 0.3700260915964491, + 0.33010953242294455, + 0.39305474430025333, + 0.42213764719535146, + 0.39300505507363637, + 0.2678404382506704, + 0.5666162547233874, + 0.3846370622818161, + 0.49829488414404727, + 0.5216974711759478, + 0.3607041840192924, + 0.48774457233653434, + 0.4892333183265376, + 0.48350565756927416, + 0.6272315356342775, + 0.4102124625691879, + 0.44577659956232957, + 0.4726299713296033, + 0.381772546489668, + 0.4775128746819263, + 0.45539545144681065, + 0.3205529145957849, + 0.4296073752053213, + 0.45864295206414013, + 0.4033934751681701, + 0.28191232046649406, + 0.41770356676806636, + 0.3145862705145477, + 0.39906412697052424, + 0.6229570618540062, + 0.5355421633839528, + 0.5447643310293047, + 0.5513610371797286, + 0.6275452114550524, + 0.5420579582997126, + 0.5919852949649997, + 0.5597914992173215, + 0.5974439602945915, + 0.41964826831886415, + 0.4112144816671145, + 0.4064974936254811, + 0.31929610705179845, + 0.274311779222423, + 0.341261813736924, + 0.34022152579205667, + 0.31228214506855906, + 0.28041840864354084, + 0.4197047035846993, + 0.2875095420756036, + 0.4469651524645012, + 0.3259864474366632, + 0.5159440114851612, + 0.570139667062017, + 0.5020996175837605, + 0.26971381872913985, + 0.3611172057926299, + 0.566796453464855, + 0.3106007590737668, + 0.3202423979119696, + 0.2781828513767859, + 0.8361870702145795, + 0.5180535508158206, + 0.41038382731420303, + 0.5696772120787911, + 0.5136298466217742, + 0.47916404244946426, + 0.40733211627196964, + 0.4722221161924357, + 0.5087034434532058, + 0.43021121823310393, + 0.5630262179317495, + 0.5155681346753749, + 0.2777507200183981, + 0.40439081737383603, + 0.2924574123029328, + 0.35731526429467564, + 0.2911450228249345, + 0.48795341217804783, + 0.40344628014903544, + 0.3283455647627484, + 0.45642402441850105, + 0.5096123311097932, + 0.3300003647336093, + 0.4604597268697535, + 0.2953572063578585, + 0.43139245623326256, + 0.4845408538493046, + 0.34166840884140126, + 0.3198824167682632, + 0.27056938565858096, + 0.5491475016699731, + 0.2894207526698432, + 0.3504753234308731, + 0.5005475698856063, + 0.3813562441871832, + 0.40601534512775705, + 0.30699001567958756, + 0.33252826529125573, + 0.39141020550215005, + 0.3805829587749968, + 0.4338423576241767, + 0.32332645075589583, + 0.4747548135954051, + 0.270670677001008, + 0.4891189535113995, + 0.40564081059305074, + 0.38829081325510223, + 0.454423506319354, + 0.3746033295289792, + 0.46407699166959976, + 0.3382605756764401, + 0.4876936648621028, + 0.3711551506157865, + 0.32355792764293057, + 0.4210995903060436, + 0.30943207240720166, + 0.4934620385259318, + 0.2470746585588792, + 0.6645633237519154, + 0.6696868681222485, + 0.564653341851222, + 0.5626722114017274, + 0.611403565922816, + 0.6066165041747049, + 0.2687374135538656, + 0.3511162511248527, + 0.31145543802301184, + 0.36268643119118726, + 0.31902924138294847, + 0.3340217574938314, + 0.36945139741007876, + 0.4044387713116344, + 0.4148928420153238, + 0.331914909958968, + 0.36389764199247954, + 0.30472837971297057, + 0.4621697189337692, + 0.25020909575903494, + 0.40743772573423886, + 0.42654213783340056, + 0.372099052398188, + 0.4714967138547415, + 0.3942769664143106, + 0.5087144815184601, + 0.2749723732126442, + 0.4722737291857277, + 0.37354878583998413, + 0.3601352020480595, + 0.3230622528911176, + 0.37247360912314326, + 0.3484459671400913, + 0.38990183724661415, + 0.48821223539429637, + 0.4489465232456644, + 0.5303424146962465, + 0.3433761005735552, + 0.38662836627541475, + 0.39856643606456643, + 0.4120673608987907, + 0.4712800637112631, + 0.409679810122564, + 0.2665324785207347, + 0.44986694661872845, + 0.399401093947047, + 0.38438076469882015, + 0.469358050857827, + 0.3592193486195493, + 0.286164345803389, + 0.2663346737629529, + 0.3342751270275796, + 0.3726051926684324, + 0.33724734379789795, + 0.26491796268334855, + 0.3482832666039315, + 0.5132071073850414, + 0.6002459932107538, + 0.5533835166983645, + 0.517867245349668, + 0.5382237310348856, + 0.4822108172449703, + 0.5550222167027296, + 0.5058868236592658, + 0.49652032174223343, + 0.34319137042920284, + 0.3809172131076034, + 0.48532720453901534, + 0.6025754701477378, + 0.18772844363215202, + 0.34362777372369807, + 0.3350488117926936, + 0.29708677794815835, + 0.3327814074315273, + 0.431239855664546, + 0.32864016739811946, + 0.4598914437362837, + 0.7431428785145225, + 0.6114312040691388, + 0.7761871333990688, + 0.7914283044947845, + 0.6436455973358902, + 0.7354224903436511, + 0.7539085756845806, + 0.36413578433601246, + 0.3938319052269927, + 0.410473343761152, + 0.5108731620294447, + 0.33809848970129835, + 0.4204054016477298, + 0.34953277577281794, + 0.37730157901558714, + 0.38110043112889425, + 0.3672117496642532, + 0.29963326209254054, + 0.4271513736716355, + 0.38672260843614487, + 0.3746055048557936, + 0.41367205501253707, + 0.31413040005526793, + 0.3194317042319957, + 0.31398766393167105, + 0.30935932078874845, + 0.2573282501691207, + 0.33855798931255043, + 0.26454937580561994, + 0.5079499516259259, + 0.48911862309829984, + 0.3414315257384625, + 0.29143947612464943, + 0.2472138445790216, + 0.5110664773154142, + 0.5805662403750903, + 0.6348636848915917, + 0.3309948399485635, + 0.36006705720710935, + 0.2568239603009025, + 0.46305342572655184, + 0.3176608953893725, + 0.30595830000904534, + 0.4737157125843942, + 0.5082536145321651, + 0.46120814831523005, + 0.31719237795560706, + 0.4328010085727809, + 0.440826438967815, + 0.4857761172067625, + 0.4440461448961979, + 0.4908663373544331, + 0.4003454889668285, + 0.44409503912858617, + 0.4408379178913001, + 0.4730321145999128, + 0.28997153750221755, + 0.3508802672883244, + 0.4474892011690265, + 0.40991407559190085, + 0.5808978733300814, + 0.30263458670243687, + 0.5488506746308388, + 0.7054258546614496, + 0.5617319359211025, + 0.3716265157683498, + 0.31680183513429855, + 0.3350045320935791, + 0.2789434881286474, + 0.40036602796030857, + 0.49231191997650736, + 0.4417761198703313, + 0.3612152056361586, + 0.31013228866341214, + 0.47661793931533686, + 0.5059690558967921, + 0.5324246928587754, + 0.589900654058221, + 0.47913719559718615, + 0.5160750149817382, + 0.36254563367854314, + 0.4859597919532638, + 0.2802538351547144, + 0.29955370142360344, + 0.40510099551162676, + 0.3777745850364396, + 0.5327896517120783, + 0.5192655997067556, + 0.6037498775961686, + 0.6548648833697226, + 0.7507017947156456, + 0.5970163434386663, + 0.7653761433598576, + 0.43494467444769447, + 0.397330614838183, + 0.3990803003602122, + 0.4879255289431995, + 0.3313818291590371, + 0.4787168972467664, + 0.4573464806730954, + 0.4927014521645197, + 0.4692152941135653, + 0.4170765855627058, + 0.4267509333837099, + 0.4501755624731793, + 0.38603534140869716, + 0.4497327753318244, + 0.41566788095060425, + 0.4479110272999121, + 0.49580267551057294, + 0.5315062204758314, + 0.4979433738219191, + 0.3737754028514789, + 0.44640192950388635, + 0.4796912814510799, + 0.3328461473376756, + 0.4244734264490384, + 0.39613772929836705, + 0.4208456213886983, + 0.38485240985432695, + 0.42792681245244824, + 0.36881331593439415, + 0.4255483581923108, + 0.4431439675326713, + 0.4644539901556357, + 0.3103256278416466, + 0.46852139418355376, + 0.46122660532029963, + 0.4005231668798521, + 0.5253271142857536, + 0.5293972679126211, + 0.7466596370254074, + 0.2619876496410798, + 0.29710704405964405, + 0.3739478622429713, + 0.35337024516118903, + 0.26851850044984504, + 0.307884956973838, + 0.29535428940217023, + 0.6722427789547024, + 0.5653506691747407, + 0.6384034485131544, + 0.3376363977250311, + 0.36958584431970265, + 0.28911594024123033, + 0.4119947528300249, + 0.4144306712292582, + 0.23960883406518374, + 0.29277911413006125, + 0.41594420368697677, + 0.38715923547925707, + 0.4373500919405174, + 0.4129999179259089, + 0.41506471472802076, + 0.3749285975725931, + 0.29172422496315015, + 0.6870565340022879, + 0.4748861029440787, + 0.6904004091659266, + 0.5737339941437233, + 0.27068755282343215, + 0.5061945311080407, + 0.48301801669794736, + 0.4974113149229941, + 0.4674369037041004, + 0.38197542194056566, + 0.4709503807337745, + 0.4846020446073411, + 0.5011750666641941, + 0.4636609685172284, + 0.342416507871148, + 0.4239950610415826, + 0.28452075145063094, + 0.3258995743511355, + 0.6336488457359492, + 0.4935115229280981, + 0.5818776920490314, + 0.4695743955499942, + 0.45660722614140525, + 0.4339881520280753, + 0.39713711103761856, + 0.4885317351623682, + 0.4038200759678553, + 0.25261810236012633, + 0.3866566554055523, + 0.4674698617290781, + 0.25704532687961484, + 0.7607609156127296, + 0.41297093175829713, + 0.5679311508636172, + 0.7278019333868527, + 0.6823577447788624, + 0.3690930383063303, + 0.8105793241415767, + 0.4848434970067931, + 0.6593348631195564, + 0.3843894759974573, + 0.4753984644626931, + 0.36947536375600964, + 0.41737748543211506, + 0.49388789467651706, + 0.5469487398939107, + 0.5292173322176955, + 0.35605698194466184, + 0.38153905045892306, + 0.39747208310701837, + 0.40297844080566586, + 0.5524827564447675, + 0.31376653021327866, + 0.5074662340948659, + 0.4274510270164295, + 0.4495461667205207, + 0.5172914559762352, + 0.582724039611789, + 0.4616421881726485, + 0.3927514073183455, + 0.4343664814572309, + 0.5160861723852401, + 0.49979930531520206, + 0.4021211001801499, + 0.3426564748631431, + 0.4686761133548331, + 0.5291689381220998, + 0.5277188025741791, + 0.40210095093610665, + 0.579361306127932, + 0.5889423180667571, + 0.42173819424952635, + 0.498698555462801, + 0.5288247166939073, + 0.4952225221545199, + 0.44882668337330345, + 0.35683148220874555, + 0.29261945299342473, + 0.3440886132276007, + 0.3781136162724426, + 0.3126857610821109, + 0.3991111595917788, + 0.32928485289642323, + 0.3748318129718603, + 0.37444966356488263, + 0.3163200515564319, + 0.33086696435286717, + 0.3083312537300921, + 0.36201265468265176, + 0.4095241850604592, + 0.32971300817859256, + 0.2832072459049772, + 0.39472891980397257, + 0.5060505543953424, + 0.6326031229035106, + 0.28888322276244494, + 0.5167981908325129, + 0.5367005778936488, + 0.537291341298935, + 0.5501362380507139, + 0.4156158192055572, + 0.46055713989797, + 0.37658080414434486, + 0.4984327470158686, + 0.41280320458430286, + 0.6182307537016769, + 0.47022530408723767, + 0.49916997770988025, + 0.28795887618408395, + 0.3177256164025259, + 0.3886958227281402, + 0.467142146487749, + 0.36909584631648945, + 0.555330346317468, + 0.33155295195445533, + 0.6067543221451627, + 0.4961420522640465, + 0.5109339890525645, + 0.5060667797049692, + 0.4963868723978021, + 0.4148631155489368, + 0.5265183693593587, + 0.5210232719631417, + 0.5169917855235359, + 0.5155173045138984, + 0.4290659127577296, + 0.3334585956153898, + 0.3037399835484943, + 0.36867105320443516, + 0.2428287019193123, + 0.38329794242591775, + 0.43281820548684025, + 0.42215723246532594, + 0.407280249332062, + 0.4400067079870614, + 0.4091490776249868, + 0.48659597311875197, + 0.4732587955230346, + 0.33221915928382784, + 0.46696932097562643, + 0.5468519868035493, + 0.3509555430844014, + 0.29688845867258645, + 0.3671480995310316, + 0.39966381609635754, + 0.49770569269329473, + 0.3260372465924245, + 0.45474046703651894, + 0.26341581959398624, + 0.44753081611609064, + 0.48672540338340053, + 0.5274736663289928, + 0.2970412810631696, + 0.311288930576936, + 0.3194648948889046, + 0.2696577437028447, + 0.24669565163607476, + 0.3530852315865635, + 0.4434481540698736, + 0.27169921583540085, + 0.37235758829142834, + 0.4348706302334144, + 0.31344755438142474, + 0.3640502498946778, + 0.44394647837111967, + 0.532072431752263, + 0.27632953689609874, + 0.44532382409588417, + 0.39525538995237414, + 0.3224664088130923, + 0.32177139824402584, + 0.45981401369479336, + 0.4266648789751903, + 0.26306068365925844, + 0.4816573908552263, + 0.4110226061390944, + 0.3743226454837113, + 0.5117644054831785, + 0.39725570644862124, + 0.32194594710377983, + 0.33365014820114464, + 0.3171108118719583, + 0.3880580021670711, + 0.3639857349556349, + 0.38362229292767247, + 0.4307733888234682, + 0.40452831022032215, + 0.5131539404159141, + 0.43947965123679156, + 0.33734956779226083, + 0.4064913168818565, + 0.44731490089794457, + 0.43334012403852534, + 0.45144392842463626, + 0.5145405075554629, + 0.4606574174350366, + 0.5198348257415959, + 0.5513638412245601, + 0.5755968249954595, + 0.45048039639634135, + 0.38120594458473794, + 0.24588040483913975, + 0.4127839217700007, + 0.4750536407726904, + 0.4264865825600861, + 0.4415167473392452, + 0.4578279872341326, + 0.31256030718147787, + 0.4656814782963988, + 0.3752281918411639, + 0.27240325540233373, + 0.40837001924430627, + 0.4195001377992321, + 0.450642725902569, + 0.45677102180255574, + 0.30355741800672364, + 0.565567015327873, + 0.5084256037122463, + 0.48335448556178334, + 0.5239758814884666, + 0.6027614859170725, + 0.46935103866775973, + 0.30535208235666206, + 0.3850255457483094, + 0.38572161006881894, + 0.3728856831229572, + 0.3698354308317803, + 0.33117160047480365, + 0.3564859285097526, + 0.4033002333823855, + 0.29825708265342643 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.04181975991319257, + "valid_tracks": 4, + "per_track_pearson": [ + 0.044601949882622144, + 0.03949363759243322, + 0.041444042628299274, + 0.04173940954941565 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.28048366883296943, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2927660765103106, + 0.2721139970780286, + 0.28133554212370365, + 0.2967011748388399, + 0.2845263315667872, + 0.29938824200217395, + 0.2721895621982371, + 0.24362795696492262, + 0.27938970103854843, + 0.2705581266887339, + 0.2882888319766783, + 0.26759489364041583, + 0.2608809506591594, + 0.25193240581356324, + 0.2785104117980056, + 0.2949787939306581, + 0.27218873943103394, + 0.2670915207278411, + 0.2870558464994719, + 0.2708091984255448, + 0.2898669491226738, + 0.28197823437553543, + 0.28312889699867605, + 0.27155412879028307, + 0.2755660241585208, + 0.28453493120778356, + 0.2681405173931981, + 0.2996894754219908, + 0.27637500278192606, + 0.27829117782814583, + 0.2713056594993157, + 0.2919361108750996, + 0.2756308509229557, + 0.292665338357637, + 0.24426094320228134, + 0.28643327066248236, + 0.27045043247053163, + 0.2703956615829585, + 0.28132705660727253, + 0.2655944237662728, + 0.26325097314786294, + 0.2711196238521038, + 0.26816983711399806, + 0.2720133396660558, + 0.27177223517452015, + 0.26531790330901384, + 0.28057439819814034, + 0.2801440461294562, + 0.2852456782902342, + 0.28845756211263557, + 0.2806063870417231, + 0.27100664793495116, + 0.2772233790278373, + 0.2766725633432416, + 0.29104424786695565, + 0.25581781414639515, + 0.2822882357802728, + 0.28502931856869523, + 0.27986255103281205, + 0.2743760404750547, + 0.2643910354889973, + 0.273509100934536, + 0.28417374890159813, + 0.2846238718101628, + 0.2882226915798268, + 0.2761662999545708, + 0.30066367666939153, + 0.263221892570772, + 0.2801153611779926, + 0.2854353597246337, + 0.2796958842728697, + 0.2916402043096955, + 0.2725974579433363, + 0.2919543218976326, + 0.28629535610668505, + 0.3026202413358845, + 0.26566851987931156, + 0.28472775403641565, + 0.2835208303278231, + 0.2849319634536048, + 0.3014291695634362, + 0.27075520521450513, + 0.28913235931400977, + 0.2724059171770404, + 0.2822191868902089, + 0.2795764426171132, + 0.2640163898106108, + 0.28096982228012823, + 0.2734317093207227, + 0.2966265592594698, + 0.2850742417980326, + 0.2876338164265247, + 0.28421096276961644, + 0.27042861504069343, + 0.2748769706609767, + 0.30180914331565645, + 0.27768890800769624, + 0.28413323999340134, + 0.2646382689272549, + 0.27997048826573034, + 0.26068997013369405, + 0.29881651728327985, + 0.29437807171081326, + 0.2949349318313774, + 0.289710996826879, + 0.2896382024184023, + 0.26274042211107257, + 0.28213434743470617, + 0.29695479286748844, + 0.28957246239541284, + 0.2822857364452035, + 0.2936194964935258, + 0.28941145507715305, + 0.2916901158255843, + 0.2937496710958549, + 0.29375017873350306, + 0.2549602992622316, + 0.28263845862711495, + 0.28105774011170115, + 0.27637752805533666, + 0.26997684696702623, + 0.2826598697047777, + 0.27095002326444806, + 0.2771805402696457, + 0.25234891750324207, + null, + 0.278579274452312, + 0.27186532693533133, + 0.26884041674212633, + 0.29208646518593573, + 0.26940214746127306, + 0.27297921508146766, + 0.2887345407388609, + 0.28665021451927325, + 0.29860148641081175, + 0.30063262952660136, + null, + 0.27975389740211376, + 0.2758780627265488, + 0.2558964782862872, + 0.2622834564623829, + 0.2752669666414298, + 0.29460391675756487, + 0.24611864080177248, + 0.26829720896291576, + 0.2581992302743398, + 0.2607735323564376, + 0.25908252408989596, + 0.24896009345046546, + 0.2837801702144034, + 0.27044458144190064, + 0.2978636932144483, + 0.2859689239317472, + 0.30268197392926294, + 0.2784261111227856, + 0.2859653122515765, + 0.2834094105524998, + 0.25630090545267137, + 0.29393554919985665, + 0.2579764925887698, + 0.26842398421876923, + 0.2797486258964035, + 0.2961362373603979, + 0.29200306680658167, + 0.2754187695343082, + 0.27791217839571436, + 0.2817670288720972, + 0.28375980200389944, + 0.2522202327663996, + 0.2639156894276222, + 0.26961965035864344, + 0.2817442639695712, + 0.28612224206643844, + 0.2755088046642691, + 0.28789604654696915, + 0.2633976598767171, + 0.29950325348218554, + 0.29978651754526964, + 0.2746946150587278, + 0.2585496325492187, + 0.2872007290085049, + 0.2813576723952542, + 0.24990407783184512, + 0.28330593585784425, + 0.27815371164570235, + 0.25746765370836766, + 0.2981052797609088, + 0.2845293869637757, + 0.30275454367438304, + 0.26304574620253673, + 0.29091329496583324, + 0.3029225790463113, + 0.29480822256440753, + 0.2920658259951796, + 0.2921979048091582, + 0.2876957040106413, + 0.2959081102311436, + 0.2633232683243, + 0.30120560557755405, + 0.26831310578471507, + null, + 0.2662693363849417, + 0.27638630133092407, + 0.26941472878742867, + 0.2719013392870307, + 0.2706061367766245, + 0.26390437721077675, + 0.27278786768544117, + null, + null, + 0.2555402169256119, + 0.27757076446702633, + null, + 0.28158309147395644, + 0.24876491512742474, + 0.2724977942886914, + 0.26145194354308204, + 0.3056149680325427, + 0.2909303196340696, + 0.2610096981119219, + null, + 0.29596843092500436, + 0.27857308222831917, + null, + 0.26173009264089137, + null, + 0.2643526127588152, + 0.2888011928014402, + 0.2834561151024767, + 0.2832073442000969, + null, + 0.2636143549288496, + 0.281734777789062, + 0.26666399934187573, + null, + 0.27105657450023674, + null, + 0.26237245313198765, + null, + 0.2673551267742644, + 0.2782258997912483, + null, + 0.2737045544088369, + 0.2724216457539176, + null, + null, + 0.2927250075144747, + 0.26293679703787204, + 0.26082948182250393, + 0.2643334950514555, + null, + 0.28676472586164975, + null, + 0.284346376260307, + 0.2818565487642087, + null, + 0.20230324451584175, + 0.2685516655590385, + 0.2609827939541797, + null, + 0.25866787274391423, + 0.2865193248008661, + null, + 0.28199793644101795, + 0.28105645449219174, + 0.2717557271684251, + null, + null, + null, + null, + 0.27664918444278713, + null, + null, + 0.2693309229767057, + 0.2737749482203894, + 0.2735295441299488, + null, + 0.2826567050251228, + 0.2643395517678382, + null, + null, + 0.2742707758562524, + 0.2688267702887117, + 0.2820833759548521, + 0.2915457498078725, + 0.2874282536123562, + 0.29161058563547365, + 0.26973179741654296, + 0.2897439203263095, + 0.28062304374284736, + 0.29023602996152376, + 0.2541831077302542, + 0.29701945173231425, + null, + 0.2805831618809108, + 0.286608304367465, + 0.2776952095793623, + 0.2850941091007752, + 0.2843917411964927, + 0.27378655325737034, + 0.27608419156777, + 0.27183759205159136, + 0.2748390669907365, + 0.28955050746859523, + 0.2838777615667222, + 0.28557305409250305, + null, + 0.28293212344565744, + 0.27200839357014867, + 0.2594858718819212, + null, + 0.2671231669825222, + 0.19564407867605624, + null, + 0.28202577674777174, + null, + 0.27470032025996005, + 0.2759160592689721, + 0.2760938137890464, + 0.2651007191610619, + 0.26557891736407196, + 0.2641902220572971, + 0.24518508070332365, + null, + null, + 0.2842276054006015, + 0.2574275541078257, + null, + 0.29121074984343137, + null, + 0.2851568024775044, + 0.2778998630251041, + 0.2811999390906755, + 0.2716042899110435, + null, + null, + null, + null, + 0.29337824050241385, + null, + 0.27707633768155104, + 0.27226768939951745, + null, + null, + 0.28273370939461184, + 0.30185565910376516, + 0.29576084280661574, + null, + 0.28796468591577273, + 0.28715540697981173, + 0.2873707949592094, + null, + 0.26815678269622467, + null, + null, + 0.2852374037802311, + null, + 0.2940992219887311, + null, + null, + 0.2870441370633895, + 0.2700466359913459, + 0.27170264644038045, + 0.2733426129616842, + 0.26832941151362166, + null, + 0.2843797683571403, + 0.31889359326892097, + 0.3037316022680583, + 0.2521265015009247, + 0.25316227096441907, + 0.30935970739299984, + 0.2835479779754656, + 0.2662451230223744, + 0.29336145136035197, + 0.2914724826257243, + 0.2871638041946804, + 0.2920165622283853, + 0.29255133796670296, + 0.2857844581005674, + 0.2858998285549299, + 0.2412477291939754, + 0.2590021704046879, + 0.28401347639104385, + 0.3033256696943613, + 0.31643097159492567, + 0.3227611656601144, + 0.2990843329844308, + 0.2747471505490925, + 0.25330931814336377, + 0.23952913629684827, + 0.30426837752046004, + 0.32008080130496563, + 0.3032687363407109, + 0.32056095398539586, + 0.2547050931239014, + 0.25508444528092517, + 0.24754334095649597, + 0.2567587718087018, + 0.24621400000949217, + 0.2846506005383263, + 0.2363722679502153, + 0.2897693122907627, + 0.22542250400936392, + 0.30258990088463344, + 0.28746225913540585, + 0.26664984617993315, + 0.2797851946384009, + 0.24965677306212028, + 0.25517862008011494, + 0.2704208260860645, + 0.2923241848502626, + 0.24903489021147868, + 0.26560699211181504, + 0.25299667868233655, + 0.27859395982827506, + 0.2542939518464288, + 0.26422064357529723, + 0.24437159920558837, + 0.2791312970532842, + 0.26473492566074497, + 0.2479638624151927, + 0.27768354391007244, + 0.2982825288781345, + 0.2855986782890262, + 0.3016067947260859, + 0.2901289660539915, + 0.2892986754041812, + 0.32252701091708386, + 0.31681080338429335, + 0.3142921331643363, + 0.31738976663595475, + 0.3209982616335759, + 0.2843070981852178, + 0.2989998998085606, + 0.30658853843131834, + 0.25400291055553514, + 0.33351324516242725, + 0.2993274012562851, + 0.28114162354404787, + 0.2765974420539357, + 0.2934173505690109, + 0.2766319842526141, + 0.27619801211658407, + 0.32166044878312317, + 0.30502670886327043, + 0.30542192970007864, + 0.30013796949266763, + 0.302585634319374, + 0.2755497986623584, + 0.30499164130610634, + 0.2905751826606442, + 0.2845981007318409, + 0.2773444270775944, + 0.278898491670663, + 0.29229416494023497, + 0.3059289160287305, + 0.2858060742193283, + 0.2823630924042662, + 0.29378070296568953, + 0.2993555732464695, + 0.2979925305062755, + 0.2950117393450166, + 0.2907948964575095, + 0.320100501330187, + 0.3106989167301298, + 0.3204402572414637, + 0.2759755365425189, + 0.32918983942169067, + 0.31091505329317, + 0.3169115819475352, + 0.3069579567277499, + 0.3134726765006046, + 0.2592264466232192, + 0.30386655730065815, + 0.2729236387081344, + 0.281360101398533, + 0.27970368660916317, + 0.28147978692373804, + 0.27948250046777623, + 0.28607322407218516, + 0.28005381410304453, + 0.3150517297364169, + 0.28995875467088283, + 0.27753893119826606, + 0.25333678940829163, + 0.2968551085069213, + 0.2893759267845135, + 0.3021071006841185, + 0.30369851986502866, + 0.31777425259995323, + 0.27717328982643363, + null, + 0.2997695285974585, + 0.2651939209078536, + 0.27036104219302626, + 0.2704384630762419, + 0.29790869751555826, + 0.3081855416622152, + 0.309239991290469, + 0.3166040778010386, + 0.3044915897042605, + 0.2799469907144045, + null, + 0.3027118998637682, + 0.3067295872715948, + 0.3210527178340594, + 0.3016828434902144, + 0.2904204003611861, + 0.3005177769342876, + 0.2781496800289075, + 0.28176696448317223, + 0.3030107268478759, + 0.28956549551799826, + 0.30461203363742545, + 0.28137976403480647, + 0.3183091124429488, + 0.2751249517042726, + 0.3018264780589262, + 0.30046356804272756, + 0.2790886777944952, + 0.2534896921689081, + 0.24763716991505286, + 0.2531022920988768, + 0.29971200352918664, + 0.2905278771492757, + 0.28758870549014315, + 0.2804559371843344, + 0.2813311550693416, + 0.3039816579555068, + 0.26555591316381144, + 0.30815320352144415, + 0.28860766413322386, + 0.30303273143274406, + 0.30445975666506697, + 0.2788095897063503, + 0.2644772085129725, + 0.2932454242139512, + 0.3134446208807867, + 0.31457363448462505, + 0.28223528860404506, + 0.2648803149352835, + 0.28556066044789397, + 0.3070703645156966, + 0.3120957548780266, + 0.28808508030682883, + 0.24822615554660268, + 0.2935105117507071, + 0.29898375402655286, + 0.25646585029069147, + 0.2981395472451542, + 0.3051729487362862, + 0.287808264954901, + 0.3089638690834549, + 0.2557109052929672, + 0.2814216196025212, + 0.29052102085453385, + 0.27608653001506817, + 0.2876403941648981, + 0.3136101822420438, + 0.2603581663327862, + 0.26059263697253765, + 0.24829456139434425, + 0.2686084203062753, + 0.29337765067262533, + 0.3161682141060825, + 0.2955934962446422, + null, + 0.29055583585585937, + 0.2856796369771542, + 0.25979479584638016, + 0.28188080272156035, + 0.2993235784505806, + 0.3113916780964195, + 0.26470699374448736, + null, + null, + 0.3012956612263253, + 0.2872664232790692, + null, + 0.2661263246839258, + 0.2419858895018158, + 0.2831424363809504, + 0.2461114369349561, + 0.26194671893097576, + 0.2579885228631007, + 0.2965453061768928, + null, + 0.29864003130690464, + 0.29473543593484847, + null, + 0.24650577842886384, + null, + 0.2689180386325467, + 0.296390940522619, + 0.28643545081740235, + 0.2615140202227406, + null, + 0.25143424189584884, + 0.26411377462556224, + 0.27673332420150243, + null, + 0.22712298316302187, + null, + 0.2648992138538525, + null, + 0.2544175549384939, + 0.2608892465383517, + null, + 0.25008017276007016, + 0.29677975947928825, + null, + null, + 0.29505610805349813, + 0.286217989218794, + 0.26662641316962643, + 0.24585202821654517, + null, + 0.25901087991251753, + null, + 0.26154901813866305, + 0.264667616919346, + null, + 0.2292270207595904, + 0.25082188188689575, + 0.2504886858961225, + null, + 0.2634234222228462, + 0.2588733857729165, + null, + 0.3143827954554551, + 0.2975843900909078, + 0.3134323737335338, + null, + null, + null, + null, + 0.31253065161627575, + null, + null, + 0.29216315639771895, + 0.2878343801861941, + 0.3035686781059759, + null, + 0.3174789034465343, + 0.3131871468446434, + null, + null, + 0.2919657682116867, + 0.27562337771189804, + 0.300913835362684, + 0.2617207500781729, + 0.2621950634656635, + 0.27397889166580414, + 0.24576828658202207, + 0.26546620621192224, + 0.2424852343098119, + 0.2618784750944824, + 0.2724163189992058, + 0.2561346747437039, + null, + 0.2681053067121895, + 0.2587238812503895, + 0.31100440164675786, + 0.2948548264860583, + 0.28713813982508535, + 0.29025606240853374, + 0.2816772961440689, + 0.28870703749332677, + 0.3028736500199391, + 0.26034490777643593, + 0.2591532411956809, + 0.2503115815329209, + null, + 0.26066912174925905, + 0.31708673258241116, + 0.31620115041130853, + null, + 0.2586533963398708, + 0.2780078269284293, + null, + 0.2686112053776524, + null, + 0.2947523064504993, + 0.27851145050714216, + 0.2881712494718725, + 0.27006062163319344, + 0.3026445365966732, + 0.24561600172902034, + 0.22865943150537885, + null, + null, + 0.28821473111222023, + 0.30078263483580253, + null, + 0.2708345483269708, + null, + 0.2790905625201898, + 0.2729818411653068, + 0.2613040640465481, + 0.3006682691052434, + null, + null, + null, + null, + 0.27504545374281814, + null, + 0.28074574962971194, + 0.30010866705688566, + null, + null, + 0.29996346390699086, + 0.2570622145912148, + 0.250033099332434, + null, + 0.25549129115932306, + 0.2571704951070108, + 0.26779165859135245, + null, + 0.3025490407698007, + null, + null, + 0.27749361963995256, + null, + 0.23345527810821018, + null, + null, + 0.2649384192438135, + 0.3028008982921232, + 0.3057254041747847, + 0.2974463582266634, + 0.30152702835465633, + null, + 0.2800731123061684 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861 + ] + }, + "junctions": { + "mean_pearson": 0.2240537907436797, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2824461642353725, + 0.2993167044328379, + 0.28911622966794254, + 0.18126692223312765, + 0.39512697755346865, + 0.30407413474367817, + 0.012471228258718344, + 0.25622526207505397, + 0.2591423074959086, + 0.2631032289817984, + 0.21809653235102566, + 0.20046210567741438, + 0.08555284084955293, + 0.17719345187380403, + 0.20948125006762308, + 0.21016541425097293, + 0.23811141520396273, + 0.17630643777483399, + 0.19047895552601815, + 0.21724238334253962, + 0.25911338251124827, + 0.28458082881699454, + 0.14489611061902016, + 0.22674988620228492, + 0.22538655613520236, + 0.241400558076924, + 0.20324707195460176, + 0.27968482834063774, + 0.12239870387217161, + 0.2879253243001583, + 0.27070815909637774, + 0.32840800840895934, + 0.23647672778563208, + 0.28887865001251195, + 0.256479404882463, + 0.2799987287057979, + 0.25527990379293125, + 0.22407382261777642, + 0.28405712839942365, + 0.12112964573403914, + 0.19689031842839014, + 0.20664627645546374, + 0.20477128935087743, + 0.18481106874070607, + 0.1849055899700621, + 0.1501777440474809, + 0.13191316482991125, + 0.20465876483738757, + 0.15873984174273145, + 0.170646151662555, + 0.1834149420875143, + 0.19901421277541811, + 0.17035435861848797, + 0.24969104949905302, + 0.2155592208999413, + 0.2241292802009203, + 0.16560069722060666, + 0.1679274855025525, + 0.22749426209130028, + 0.214261363978728, + 0.11956675855109027, + 0.20777889975472588, + 0.21126052720115987, + 0.05302146171964449, + 0.10496220053919081, + 0.33240559808871234, + 0.2661461349391503, + 0.2944575586977847, + 0.1381273621487804, + 0.18707626245342263, + 0.1312111904246842, + 0.27737749807568873, + 0.17612237581063792, + 0.12398620466363826, + 0.3426647344807325, + 0.3962112041758353, + 0.19529207102984739, + 0.1882204453827454, + 0.1512852359622233, + 0.35557541204773224, + 0.24302816999738822, + 0.2717509331850244, + 0.2132939622872358, + 0.3222235202161386, + 0.26332476662695825, + 0.2844662120773463, + 0.2832028208260864, + 0.30577870198414103, + 0.28988456905854837, + 0.2776066044918514, + 0.30618342784721037, + 0.2766828239080145, + 0.18161583773250212, + 0.26476368554346025, + 0.2786841886440985, + 0.2679268159720965, + 0.27442565705333827, + 0.27455119113996873, + 0.26686317571356466, + 0.22826750740094373, + 0.1924195606232738, + 0.2516605806900299, + 0.3245868093950266, + 0.15351175280242824, + 0.1819371055611128, + 0.12261573082424744, + 0.07322096247355983, + 0.2626941138796848, + 0.2872753439822564, + 0.25636355677445116, + 0.1933993037068354, + 0.27746184875315577, + 0.28688055002513974, + 0.38576428665645396, + 0.28592955805703085, + 0.1465771987070091, + 0.1854380729117799, + 0.2457401218704906, + 0.17340339169251, + 0.2380802366589161, + 0.22076155156282198, + 0.18930779579500326, + 0.22120702300763637, + 0.21132967384362553, + 0.11644662061390346, + null, + 0.2673005454857849, + 0.26840335255544656, + 0.16385768369783577, + 0.22858676944150827, + 0.27567742191743927, + 0.22750241321045525, + 0.1446349265562069, + 0.36977343142916114, + 0.3499065169340458, + 0.15926021113479807, + null, + 0.23368377944091434, + 0.28701429848188736, + 0.2040865262402263, + 0.2176526497180297, + 0.28493467896860936, + 0.2684773367027385, + 0.32176333512943694, + 0.24867765411531495, + 0.21501110047302502, + 0.25645747580705686, + 0.23520457850327176, + 0.1932798226128312, + 0.15699273617451853, + 0.23126978444257365, + 0.3587085338988465, + 0.20522799960043672, + 0.296678606240697, + 0.38290106694769493, + 0.3409295505453182, + 0.26482912101781997, + 0.30761801690672225, + 0.28000499353932917, + 0.3194132911055824, + 0.23944912856791928, + 0.33576796208872944, + 0.1952371476085414, + 0.3083039410650497, + 0.12410437513929674, + 0.3206555645060305, + 0.3195601327466587, + 0.23644992385148259, + 0.17901634998930693, + 0.1509041268020434, + 0.2467694590420642, + 0.26010394097165934, + 0.3080855274050613, + 0.17199090599346004, + 0.20008221365591097, + 0.2464634118250196, + 0.27410559760063424, + 0.19432204347170745, + 0.212464394525661, + 0.18092620191372225, + 0.1650659214180188, + 0.1353260548836617, + 0.06997275979257278, + 0.26487173514970264, + 0.27868832494074147, + 0.26448261491208164, + 0.2249690998506194, + 0.19131432723731276, + 0.2481626391800114, + 0.26334733615752953, + 0.2724201985607786, + 0.3847031339029097, + 0.15740804060797534, + 0.13300399782871325, + 0.16748110523177684, + 0.14463472624724372, + 0.16468482200646173, + 0.15911450456050738, + 0.23069195917828036, + 0.3784769248829497, + null, + 0.10266939752245842, + 0.21300398851295435, + 0.2643625563174798, + 0.30987605396345047, + 0.1254766627976798, + 0.08939504835629168, + 0.19279059929687384, + null, + null, + 0.20305675463689324, + 0.29903683575421813, + null, + 0.15008630251150754, + 0.24269052030486243, + 0.21816925974801965, + 0.18210004137222016, + 0.11497699075059266, + 0.15053080127875024, + 0.09561672427245733, + null, + 0.27885058361847487, + 0.25704457579315315, + null, + 0.16281202598360886, + null, + 0.31239035010564553, + 0.15565476996510952, + 0.19579712892725762, + 0.2544552725738029, + null, + 0.11607420640834928, + 0.13374632880833695, + 0.22812257657113097, + null, + 0.1839291410293589, + null, + 0.2711307118371949, + null, + 0.10903192000041002, + 0.20821905395514131, + null, + 0.17163998958362378, + 0.2427076689401821, + null, + null, + 0.22659206696134265, + 0.2602384730196844, + 0.09052529053724193, + 0.09637738185648201, + null, + 0.36834855220283064, + null, + 0.20295205908095262, + 0.15105232067389307, + null, + 0.2104296069637429, + 0.25798799310919435, + 0.35818616401686915, + null, + 0.18635663656547086, + 0.23050719778296247, + null, + 0.3798071003773277, + 0.33948899421014817, + 0.36965737049611397, + null, + null, + null, + null, + 0.2253399406617692, + null, + null, + 0.20123014324957605, + 0.26619830058036015, + 0.3666255212140457, + null, + 0.18574054978094007, + 0.3412996361383495, + null, + null, + 0.30135586122840025, + 0.1507668733120532, + 0.21786558472159895, + 0.13407899903396203, + 0.15956829209581183, + 0.10970699126061786, + 0.20845871487699758, + 0.17658872533178804, + 0.14550156686657612, + 0.09203139707375584, + 0.313893641957374, + 0.2577447621493191, + null, + 0.25625786510453247, + 0.29655559570543266, + 0.053524336514237596, + 0.14514757164315192, + 0.0897369604371139, + 0.0552569147462417, + 0.21766816706841224, + 0.15133118142426372, + 0.23127866463378402, + 0.17808035048664003, + 0.25710237416516346, + 0.21930562868685952, + null, + 0.18614676066573638, + 0.20263711249691663, + 0.2126145891236034, + null, + 0.14389857152671653, + 0.19997082953622036, + null, + 0.23599256390383241, + null, + 0.13316216716980686, + 0.16068475443367486, + 0.2605346334450377, + 0.09450667904888031, + 0.17923335698615372, + 0.11695379523939903, + 0.13395384927491077, + null, + null, + 0.18845779331418186, + 0.1636350325200194, + null, + 0.18637952135755695, + null, + 0.2775915961077477, + 0.1946055223233524, + 0.12286587715402725, + 0.21979248286699934, + null, + null, + null, + null, + 0.10188267763360997, + null, + 0.08085733467390453, + 0.30514118687147984, + null, + null, + 0.26649693784675565, + 0.21082084696986475, + 0.16470552360446533, + null, + 0.3771023708329736, + 0.3136888721772168, + 0.17355004256582923, + null, + 0.33645512615897916, + null, + null, + 0.2530964726691911, + null, + 0.18124418064301917, + null, + null, + 0.26028697967595005, + 0.2614222941195829, + 0.2110810251881899, + 0.22570716739735985, + 0.199444904390449, + null, + 0.21029093294003467, + 0.18877863760950858, + 0.1683525435045144, + 0.31929778150346017, + 0.10843063787370204, + 0.3216751619770746, + 0.2465966595267891, + 0.1571235406138378, + 0.16680223056635587, + 0.2861145568042369, + 0.17353570055247888, + 0.25345552583612646, + 0.2991999257086431, + 0.22052396756635517, + 0.18671616033094793, + 0.15324964710170424, + 0.2852871741178276, + 0.2570070868212009, + 0.34445047261701917, + 0.115588659219032, + 0.26681067606396874, + 0.19577211283242996, + 0.2045987529877483, + 0.07743355907704685, + 0.23627221783873606, + 0.33541454588526687, + 0.329500365228332, + 0.3350669787026491, + 0.3349225911303318, + 0.24892238901708255, + 0.19377691955247284, + 0.19435475027402943, + 0.35486081187807994, + 0.227862971024791, + 0.24390542251408898, + 0.0624613376335234, + 0.3270256308540748, + 0.032622358471708156, + 0.19812161455832408, + 0.15163498014520715, + 0.31131260062336713, + 0.2490188518758334, + 0.019811052804197513, + 0.09876732690296643, + 0.23667165545993793, + 0.15943009729512236, + 0.15596379920169265, + 0.16090427631454352, + 0.19132351051905552, + 0.23716925619147716, + 0.2045721979798251, + 0.20159989394374317, + 0.10629406230385433, + 0.15366674036464065, + 0.07413654317894464, + 0.043782452516218394, + 0.10845602795955944, + 0.25906698347876767, + 0.2708145909847692, + 0.1915214048660593, + 0.2388315578387086, + 0.33201446519648986, + 0.29777766531999994, + 0.15555164817403058, + 0.224471172249441, + 0.19059073770498952, + 0.1685087063945003, + 0.12656826808702412, + 0.22267485650376562, + 0.12968101912418398, + 0.2651991772787135, + 0.33618108760236043, + 0.15814501098836178, + 0.18532802065226814, + 0.2208914874306654, + 0.08884881679413534, + 0.08024764243550131, + 0.305733447689078, + 0.3053017535209414, + 0.3272837877495389, + 0.19671104685236798, + 0.22892319234289618, + 0.2923712127255235, + 0.31061733515482565, + 0.15207844051892766, + 0.25818252876644837, + 0.23520390118086582, + 0.1961109493271364, + 0.26977298128123617, + 0.17366749952566962, + 0.25321783966343364, + 0.30736646073199525, + 0.2022489449917708, + 0.18117201316780654, + 0.16818000526219626, + 0.11748619345766206, + 0.1564224645127089, + 0.23769008225244376, + 0.20066090581522322, + 0.32585646862475504, + 0.21051433520015142, + 0.14002013216455939, + 0.24279571165341368, + 0.16194477447118935, + 0.1922447142398607, + 0.18955883635100904, + 0.2627334323534956, + 0.049883346090775584, + 0.2250748454498676, + 0.2236037709637277, + 0.08699303927098352, + 0.06007222082622346, + 0.14443177526174877, + 0.154465893562206, + 0.17867759628701663, + 0.1608485590283016, + 0.08060865711064645, + 0.32492710565780253, + 0.15533729719607678, + 0.11750220233422444, + 0.12131551563334396, + 0.32373738309371647, + 0.17200834791857264, + 0.13678889834317529, + 0.16768381672226368, + 0.2782643576365804, + null, + 0.366532198358971, + 0.3205210024411202, + 0.24875945069534058, + 0.42732534285395174, + 0.18925260402254102, + 0.08968453536490412, + 0.1380994090156205, + 0.3859062516642586, + 0.1774177775092803, + 0.14011594252850393, + null, + 0.1426889345556619, + 0.12447610316229457, + 0.10219861091709827, + 0.0720834716500037, + 0.19160240230085945, + 0.14955899538748266, + 0.15904183030915714, + 0.2793242034989631, + 0.07600440672015636, + 0.22597389468033727, + 0.26160549748344414, + 0.13660741974350293, + 0.1965661915484332, + 0.18455025515240314, + 0.34369264550392054, + 0.31517252730684786, + 0.28522379313078183, + 0.37897094473849297, + 0.30334162552982413, + 0.22744474144593718, + 0.2070195843966492, + 0.13850963291103657, + 0.04450016836825654, + 0.13227129507272326, + 0.1478274711724512, + 0.21824936162522426, + 0.1912584754670576, + 0.2506682207814725, + 0.2156188134629105, + 0.12285092580089403, + 0.14660526748489894, + 0.30896497064387396, + 0.13708854344678012, + 0.2186396153900472, + 0.28028721314529625, + 0.2236461856333961, + 0.07916787127020396, + 0.267823877439179, + 0.17188072971506044, + 0.29841955059566044, + 0.1541429255708039, + 0.10442768162743318, + 0.18121349644460297, + 0.23087619025970413, + 0.13433177707476102, + 0.2057631318334384, + 0.20617711826777507, + 0.34669009076289886, + 0.1316415253105998, + 0.23932834491758012, + 0.11609998494831385, + 0.20497074630006198, + 0.23464942056391058, + 0.2418970164561091, + 0.20276415644985143, + 0.12118932229973407, + 0.20818825682620387, + 0.28485667596127634, + 0.31136637506974885, + 0.26374752526230966, + 0.19135732207703315, + 0.19640210113052617, + 0.32794914664099767, + null, + 0.4226286940882423, + 0.23240913655862047, + 0.43458705543872855, + 0.36356659936732094, + 0.3545994909585473, + 0.2126359666741415, + 0.2457180254580169, + null, + null, + 0.21298328566661748, + 0.2499271682975163, + null, + 0.13719102535271366, + 0.23474878668782098, + 0.19437820252150467, + 0.2640059338350188, + 0.20047665187380062, + 0.28002290543062835, + 0.18739528546340467, + null, + 0.3179696211618675, + 0.2732926633178716, + null, + 0.3449500181034301, + null, + 0.42628280444655375, + 0.2511210483841074, + 0.17310705032607093, + 0.18590515108999334, + null, + 0.24925774720956972, + 0.16599590140991774, + 0.2873643711287458, + null, + 0.32114416438444826, + null, + 0.23685033498832034, + null, + 0.2875666737033515, + 0.34591954037168354, + null, + 0.3078386899997385, + 0.34062422982790647, + null, + null, + 0.35466629775048014, + 0.24937677887861107, + 0.04056355709429299, + 0.18443324578331055, + null, + 0.30005071471751765, + null, + 0.1591319891679374, + 0.2962370588392288, + null, + 0.27658976094193816, + 0.2791478645367696, + 0.19328217300168216, + null, + 0.3216320984141282, + 0.4252088431462733, + null, + 0.14270633793636897, + 0.2299358272435639, + 0.24931801698720246, + null, + null, + null, + null, + 0.2194696553832783, + null, + null, + 0.3710294462205733, + 0.2772221742035162, + 0.20018017148613246, + null, + 0.08301447422391693, + 0.2572143857461947, + null, + null, + 0.263427725048995, + 0.02515145501468169, + 0.25100377727410317, + 0.14758419104904555, + 0.1564074721019983, + 0.2901882749977938, + 0.0015216706030614026, + 0.26313132298400993, + 0.09095009257449546, + 0.3983058905676905, + 0.35482272830441497, + 0.28949673945306814, + null, + 0.08033694698201876, + 0.2863885099050951, + 0.24899295862707008, + 0.26118204354030355, + 0.1406417117374512, + 0.2381703365634153, + 0.24074705645232417, + 0.23324477635506063, + 0.26758405308783245, + 0.299896267170186, + 0.24037457263748438, + 0.1530028864061584, + null, + 0.21070255451773318, + 0.3642603131161859, + 0.25273360548469764, + null, + 0.2919199076612892, + 0.32064115657831016, + null, + 0.2533471102359787, + null, + 0.35274241598958395, + 0.3892817931142173, + 0.08394478703729497, + 0.29676314218921035, + 0.30310655234483763, + 0.20607470357470967, + 0.3344302029560856, + null, + null, + 0.2850982551873031, + 0.3054572625743474, + null, + 0.2830406271329694, + null, + 0.3577842192404553, + 0.333558255128818, + 0.3190062392509337, + 0.2854510120246089, + null, + null, + null, + null, + 0.320767608517843, + null, + 0.3545582289882456, + 0.3568320351050793, + null, + null, + 0.26581801841170405, + 0.05252302153991619, + 0.2643915959499457, + null, + 0.14754046586966527, + 0.22953922166003618, + 0.2516835257905133, + null, + 0.2684339837594022, + null, + null, + 0.2804140004601978, + null, + 0.21811389434816175, + null, + null, + 0.2823092424793934, + 0.3018851135045513, + 0.401290223927975, + 0.35890667998167824, + 0.3135748493268343, + null, + 0.28509454166774867 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9006622516556292, + 0.9243421052631579, + 0.9074733096085409, + 0.8892405063291139, + 0.9205298013245033, + 0.8813559322033898, + 0.9014084507042254, + 0.8888888888888888, + 0.907051282051282, + 0.8971631205673759, + 0.8922558922558923, + 0.8915254237288136, + 0.8961038961038961, + 0.90625, + 0.92, + 0.8848684210526315, + 0.8636363636363636, + 0.8846153846153846, + 0.9044585987261147, + 0.9150326797385621, + 0.9064748201438849, + 0.9, + 0.8798701298701299, + 0.895910780669145, + 0.9074733096085409, + 0.9177631578947368, + 0.9230769230769231, + 0.8983050847457628, + 0.9121621621621622, + 0.906896551724138, + 0.9006849315068494, + 0.912621359223301, + 0.9029850746268657, + 0.8875379939209727, + 0.9010989010989011, + 0.8990536277602523, + 0.9007633587786259, + 0.890282131661442, + 0.8461538461538461, + 0.9026845637583892, + 0.9010238907849829, + 0.9243986254295533, + 0.9045801526717557, + 0.9222222222222223, + 0.8950819672131147, + 0.8885135135135135, + 0.9261992619926199, + 0.899641577060932, + 0.8829787234042553, + 0.9013605442176871, + 0.9051094890510949, + 0.9117647058823529, + 0.875, + 0.9081632653061225, + 0.8880597014925373, + 0.9081632653061225, + 0.889273356401384, + 0.9205298013245033, + 0.886986301369863, + 0.8767605633802817, + 0.9137931034482759, + 0.9245901639344263, + 0.9139072847682119, + 0.8907849829351536, + 0.889261744966443, + 0.9272151898734177, + 0.8851351351351351, + 0.9161073825503355, + 0.8907849829351536, + 0.9117647058823529, + 0.896797153024911, + 0.9096774193548387, + 0.898989898989899, + 0.9271523178807947, + 0.910394265232975, + 0.911864406779661, + 0.9003215434083601, + 0.9096989966555183, + 0.9172413793103448, + 0.9149659863945578, + 0.9006622516556292, + 0.9177215189873418, + 0.9041095890410958, + 0.889967637540453, + 0.8947368421052632, + 0.8978328173374613, + 0.9075342465753424, + 0.9084967320261438, + 0.9071428571428571, + 0.9082278481012658, + 0.9128919860627178, + 0.9106529209621993, + 0.9017543859649123, + 0.8945578231292517, + 0.8986486486486487, + 0.9166666666666666, + 0.9152542372881356, + 0.889273356401384, + 0.9072847682119205, + 0.9050632911392406, + 0.8733333333333333, + 0.8918032786885246, + 0.900355871886121, + 0.9041533546325878, + 0.9249146757679181, + 0.9333333333333333, + 0.9026217228464419, + 0.9130434782608695, + 0.9069767441860465, + 0.9256756756756757, + 0.8964285714285715, + 0.8993288590604027, + 0.9090909090909091, + 0.8983050847457628, + 0.9047619047619048, + 0.8952702702702703, + 0.896551724137931, + 0.8716216216216216, + 0.8700361010830325, + 0.9044368600682594, + 0.8964285714285715, + 0.9084745762711864, + 0.8897058823529411, + 0.9144736842105263, + 0.8785425101214575, + 1.0, + 0.9065743944636678, + 0.8780487804878049, + 0.9166666666666666, + 0.917910447761194, + 0.894927536231884, + 0.906896551724138, + 0.9024390243902439, + 0.8947368421052632, + 0.9017543859649123, + 0.9096989966555183, + 1.0, + 0.9288135593220339, + 0.9029850746268657, + 0.9177631578947368, + 0.9051094890510949, + 0.8880597014925373, + 0.9227941176470589, + 0.8689655172413793, + 0.887719298245614, + 0.8702290076335878, + 0.8976377952755905, + 0.8993055555555556, + 0.875, + 0.8798586572438163, + 0.8740740740740741, + 0.900709219858156, + 0.8937728937728938, + 0.9285714285714286, + 0.9172413793103448, + 0.9072164948453608, + 0.8863636363636364, + 0.9314079422382672, + 0.8869257950530035, + 0.916058394160584, + 0.8977272727272727, + 0.8888888888888888, + 0.9090909090909091, + 0.9024390243902439, + 0.8905660377358491, + 0.8848920863309353, + 0.903114186851211, + 0.9337748344370861, + 0.9018867924528302, + 0.887719298245614, + 0.8901098901098901, + 0.8737201365187713, + 0.9210526315789473, + 0.9143835616438356, + 0.8884615384615384, + 0.9260700389105059, + 0.894927536231884, + 0.89198606271777, + 0.875, + 0.8916083916083916, + 0.8954703832752613, + 0.8805970149253731, + 0.9040590405904059, + 0.9032258064516129, + 0.9049295774647887, + 0.9261992619926199, + 0.9143835616438356, + 0.8888888888888888, + 0.8781362007168458, + 0.9278688524590164, + 0.8993055555555556, + 0.9264705882352942, + 0.8928571428571429, + 0.8697183098591549, + 0.8850174216027874, + 0.8586572438162544, + 0.8529411764705882, + 0.8958333333333334, + 0.9150326797385621, + 0.9, + 0.9922480620155039, + 0.8914473684210527, + 0.8813559322033898, + 0.8870431893687708, + 0.8851351351351351, + 0.9131832797427653, + 0.901060070671378, + 0.8717105263157895, + 1.0, + 0.9924242424242424, + 0.9042904290429042, + 0.9090909090909091, + 0.9923076923076923, + 0.9, + 0.8754578754578755, + 0.8759124087591241, + 0.8805460750853242, + 0.8940397350993378, + 0.8689655172413793, + 0.889261744966443, + 1.0, + 0.8850931677018633, + 0.8762886597938144, + 1.0, + 0.87109375, + 1.0, + 0.8945454545454545, + 0.887719298245614, + 0.897887323943662, + 0.8888888888888888, + 0.9927536231884058, + 0.8913043478260869, + 0.8838028169014085, + 0.8701754385964913, + 1.0, + 0.8787878787878788, + 1.0, + 0.8636363636363636, + 1.0, + 0.88, + 0.8926174496644296, + 1.0, + 0.8698884758364313, + 0.8786885245901639, + 1.0, + 1.0, + 0.8794326241134752, + 0.8921568627450981, + 0.8725490196078431, + 0.8821548821548821, + 1.0, + 0.8657243816254417, + 0.9844961240310077, + 0.8701298701298701, + 0.8865979381443299, + 0.9861111111111112, + 0.8884615384615384, + 0.864951768488746, + 0.8727272727272727, + 0.9927536231884058, + 0.8571428571428571, + 0.8996655518394648, + 0.9917355371900827, + 0.9049180327868852, + 0.8666666666666667, + 0.8915254237288136, + 1.0, + 1.0, + 0.9922480620155039, + 0.9916666666666667, + 0.8870431893687708, + 1.0, + 0.9921259842519685, + 0.9047619047619048, + 0.900709219858156, + 0.8878205128205128, + 1.0, + 0.9010238907849829, + 0.9041095890410958, + 1.0, + 0.9846153846153847, + 0.8815331010452961, + 0.8840579710144928, + 0.8795986622073578, + 0.8851351351351351, + 0.863481228668942, + 0.8843537414965986, + 0.89419795221843, + 0.8719723183391004, + 0.8813559322033898, + 0.8718861209964412, + 0.89171974522293, + 0.8708487084870848, + 1.0, + 0.8912280701754386, + 0.8783783783783784, + 0.8785714285714286, + 0.8888888888888888, + 0.9047619047619048, + 0.9166666666666666, + 0.9017543859649123, + 0.8724137931034482, + 0.8786885245901639, + 0.8851351351351351, + 0.8913043478260869, + 0.8703071672354948, + 1.0, + 0.887459807073955, + 0.8918918918918919, + 0.907051282051282, + 1.0, + 0.8849840255591054, + 0.899641577060932, + 1.0, + 0.8969072164948454, + 0.9850746268656716, + 0.8974358974358975, + 0.8888888888888888, + 0.8877887788778878, + 0.8840579710144928, + 0.9115646258503401, + 0.8732876712328768, + 0.8636363636363636, + 0.9858156028368794, + 1.0, + 0.8627450980392157, + 0.8809523809523809, + 1.0, + 0.8664259927797834, + 1.0, + 0.8870431893687708, + 0.8791946308724832, + 0.8690909090909091, + 0.8803986710963455, + 1.0, + 1.0, + 1.0, + 0.9841269841269841, + 0.8561643835616438, + 1.0, + 0.8858131487889274, + 0.926984126984127, + 1.0, + 1.0, + 0.897887323943662, + 0.8903654485049833, + 0.8925925925925926, + 1.0, + 0.8972602739726028, + 0.8888888888888888, + 0.8581560283687943, + 0.9919354838709677, + 0.9267515923566879, + 1.0, + 0.9922480620155039, + 0.8839590443686007, + 1.0, + 0.8737864077669902, + 1.0, + 1.0, + 0.8841698841698842, + 0.9025974025974026, + 0.8900343642611683, + 0.8831615120274914, + 0.8626760563380281, + 1.0, + 0.8675496688741722, + 0.9097744360902256, + 0.9122807017543859, + 0.9094202898550725, + 0.9214285714285714, + 0.9272151898734177, + 0.896797153024911, + 0.9163763066202091, + 0.9018867924528302, + 0.9215017064846417, + 0.9244604316546763, + 0.9238754325259516, + 0.8953068592057761, + 0.9006849315068494, + 0.9143835616438356, + 0.8996282527881041, + 0.8717948717948718, + 0.9084507042253521, + 0.8962962962962963, + 0.9071428571428571, + 0.8976109215017065, + 0.9090909090909091, + 0.9045936395759717, + 0.9297658862876255, + 0.9210526315789473, + 0.9035714285714286, + 0.9309090909090909, + 0.9105960264900662, + 0.9094076655052264, + 0.9032258064516129, + 0.9011857707509882, + 0.9087719298245615, + 0.9007352941176471, + 0.9153225806451613, + 0.9184397163120568, + 0.9304029304029304, + 0.9326241134751773, + 0.9083665338645418, + 0.8971631205673759, + 0.9020979020979021, + 0.9028776978417267, + 0.8943661971830986, + 0.9343065693430657, + 0.911660777385159, + 0.9215686274509803, + 0.9278350515463918, + 0.9244604316546763, + 0.9210526315789473, + 0.9198473282442748, + 0.909433962264151, + 0.926923076923077, + 0.9236947791164659, + 0.9442379182156134, + 0.9053030303030303, + 0.9181494661921709, + 0.9135338345864662, + 0.9107806691449815, + 0.9460431654676259, + 0.9283154121863799, + 0.896797153024911, + 0.894927536231884, + 0.9024390243902439, + 0.9138576779026217, + 0.9419795221843004, + 0.9230769230769231, + 0.9169675090252708, + 0.9117647058823529, + 0.9071428571428571, + 0.9073359073359073, + 0.9239543726235742, + 0.910394265232975, + 0.8905109489051095, + 0.894927536231884, + 0.9220338983050848, + 0.910394265232975, + 0.899641577060932, + 0.9130434782608695, + 0.9172413793103448, + 0.9045936395759717, + 0.9028776978417267, + 0.9018867924528302, + 0.903114186851211, + 0.8972602739726028, + 0.8935361216730038, + 0.9074074074074074, + 0.925531914893617, + 0.9191176470588235, + 0.9122137404580153, + 0.9166666666666666, + 0.8905109489051095, + 0.9084249084249084, + 0.9059233449477352, + 0.8896797153024911, + 0.906896551724138, + 0.9182156133828996, + 0.9087719298245615, + 0.9139784946236559, + 0.9238754325259516, + 0.8996282527881041, + 0.9006849315068494, + 0.899641577060932, + 0.9003831417624522, + 0.8978102189781022, + 0.9166666666666666, + 0.925, + 0.9166666666666666, + 0.9236111111111112, + 0.9169811320754717, + 0.9037037037037037, + 0.9122807017543859, + 0.9039145907473309, + 0.9084507042253521, + 0.9176029962546817, + 0.9172185430463576, + 0.9068100358422939, + 0.8853046594982079, + 0.9076923076923077, + 0.9187279151943463, + 0.9153846153846154, + 0.8814229249011858, + 0.915129151291513, + 0.9122807017543859, + 0.9175627240143369, + 0.9010989010989011, + 0.9151943462897526, + 0.9070631970260223, + 1.0, + 0.9361702127659575, + 0.9142857142857143, + 0.9290780141843972, + 0.9090909090909091, + 0.927797833935018, + 0.9265734265734266, + 0.9151943462897526, + 0.9210526315789473, + 0.9024390243902439, + 0.9007633587786259, + 1.0, + 0.9337979094076655, + 0.8933823529411765, + 0.9440559440559441, + 0.9512195121951219, + 0.9107806691449815, + 0.9042553191489362, + 0.9090909090909091, + 0.9113475177304965, + 0.9182156133828996, + 0.914179104477612, + 0.9026217228464419, + 0.9205776173285198, + 0.9157509157509157, + 0.8884615384615384, + 0.9077490774907749, + 0.9283154121863799, + 0.9188191881918819, + 0.9320754716981132, + 0.9241379310344827, + 0.9267399267399268, + 0.9323308270676691, + 0.9256756756756757, + 0.9222222222222223, + 0.9239130434782609, + 0.8937728937728938, + 0.912621359223301, + 0.8986013986013986, + 0.9176029962546817, + 0.9173228346456693, + 0.8972602739726028, + 0.942652329749104, + 0.9183673469387755, + 0.9182156133828996, + 0.927536231884058, + 0.9111111111111111, + 0.8953488372093024, + 0.9222614840989399, + 0.9077490774907749, + 0.9471544715447154, + 0.900709219858156, + 0.9107142857142857, + 0.9265734265734266, + 0.9051094890510949, + 0.9233449477351916, + 0.9144981412639405, + 0.9264705882352942, + 0.9047619047619048, + 0.9154929577464789, + 0.9259259259259259, + 0.895910780669145, + 0.920863309352518, + 0.8873239436619719, + 0.9214285714285714, + 0.9037800687285223, + 0.9114391143911439, + 0.911660777385159, + 0.8929889298892989, + 0.8888888888888888, + 0.89568345323741, + 0.8873239436619719, + 0.9038461538461539, + 0.9193548387096774, + 0.9078498293515358, + 0.9829059829059829, + 0.8900709219858156, + 0.8888888888888888, + 0.8917910447761194, + 0.8901098901098901, + 0.9212328767123288, + 0.8877551020408163, + 0.8904593639575972, + 0.9920634920634921, + 0.9833333333333333, + 0.9247648902821317, + 0.9185667752442996, + 1.0, + 0.9087719298245615, + 0.889273356401384, + 0.8923076923076924, + 0.9044368600682594, + 0.8989547038327527, + 0.896797153024911, + 0.9151943462897526, + 0.9921875, + 0.910394265232975, + 0.8925925925925926, + 0.9923076923076923, + 0.9029850746268657, + 1.0, + 0.9016949152542373, + 0.8932384341637011, + 0.9087837837837838, + 0.8982456140350877, + 1.0, + 0.9084249084249084, + 0.9228070175438596, + 0.8823529411764706, + 0.9844961240310077, + 0.9145907473309609, + 1.0, + 0.8869257950530035, + 1.0, + 0.9138576779026217, + 0.8945454545454545, + 0.9924812030075187, + 0.9018181818181819, + 0.9175627240143369, + 1.0, + 0.968, + 0.9206896551724137, + 0.901060070671378, + 0.8881578947368421, + 0.9023569023569024, + 0.9915254237288136, + 0.8910505836575876, + 0.9841269841269841, + 0.8936170212765957, + 0.8843283582089553, + 0.9920634920634921, + 0.9040590405904059, + 0.9020979020979021, + 0.8847583643122676, + 0.9922480620155039, + 0.8985507246376812, + 0.9049295774647887, + 1.0, + 0.9010238907849829, + 0.903010033444816, + 0.910394265232975, + 0.9852941176470589, + 0.9919354838709677, + 0.9920634920634921, + 1.0, + 0.8996539792387543, + 0.9908256880733946, + 0.9908256880733946, + 0.900709219858156, + 0.8945578231292517, + 0.9054054054054054, + 0.9922480620155039, + 0.9090909090909091, + 0.890728476821192, + 1.0, + 1.0, + 0.9204152249134948, + 0.9140893470790378, + 0.8945454545454545, + 0.9075907590759076, + 0.8912280701754386, + 0.8868613138686131, + 0.8860294117647058, + 0.9081632653061225, + 0.8867924528301887, + 0.9018867924528302, + 0.9084745762711864, + 0.8971631205673759, + 1.0, + 0.9305555555555556, + 0.910394265232975, + 0.9115646258503401, + 0.8982456140350877, + 0.9265734265734266, + 0.9020979020979021, + 0.9006849315068494, + 0.9010989010989011, + 0.9093959731543624, + 0.9151943462897526, + 0.8962962962962963, + 0.9054545454545454, + 1.0, + 0.8964285714285715, + 0.9035714285714286, + 0.91, + 1.0, + 0.9204152249134948, + 0.9130434782608695, + 0.9672131147540983, + 0.9114754098360656, + 1.0, + 0.9169435215946844, + 0.9217081850533808, + 0.9222972972972973, + 0.9215686274509803, + 0.8938356164383562, + 0.8843283582089553, + 0.8821292775665399, + 0.9918032786885246, + 0.991869918699187, + 0.8943661971830986, + 0.8888888888888888, + 0.9922480620155039, + 0.896797153024911, + 1.0, + 0.8976109215017065, + 0.8961937716262975, + 0.9020979020979021, + 0.8933823529411765, + 0.9781021897810219, + 0.9921259842519685, + 0.9915254237288136, + 1.0, + 0.9049295774647887, + 0.991304347826087, + 0.8884615384615384, + 0.9158249158249159, + 0.967479674796748, + 0.9846153846153847, + 0.8924731182795699, + 0.9192982456140351, + 0.9035714285714286, + 1.0, + 0.9247311827956989, + 0.9163763066202091, + 0.8976109215017065, + 1.0, + 0.922077922077922, + 0.983739837398374, + 0.9834710743801653, + 0.9054545454545454, + 0.9833333333333333, + 0.9056603773584906, + 1.0, + 1.0, + 0.8819188191881919, + 0.9065743944636678, + 0.9144736842105263, + 0.8986013986013986, + 0.9016949152542373, + 0.9915966386554622, + 0.9016949152542373 + ] + }, + "contact_maps": { + "mean_pearson": 0.1348292448886617, + "valid_tracks": 28, + "per_track_pearson": [ + 0.1480469600540382, + 0.13587073463442906, + 0.12985699737883036, + 0.04583933639668546, + 0.08361201863275683, + 0.08139833866982622, + 0.14520614930163683, + 0.11516511269788769, + 0.2450126987941672, + 0.18836399039982132, + 0.06404968605242306, + 0.11806815763723906, + 0.21358056304898168, + 0.05160177684832484, + 0.09815504988245578, + 0.1856335589463887, + 0.0560958998832771, + 0.18275321159413482, + 0.1569047471584472, + 0.16237147790310713, + 0.07534072803883315, + 0.08617589286717828, + 0.07955839764264477, + 0.18655053635865737, + 0.17938588460079047, + 0.19393938089940574, + 0.13112025181539358, + 0.23556131874476585 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.3303574772037849, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..4239e624a85c4d9533aa273d3a17284d793088bd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S014/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5851419159369071, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6858629037200465, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.16535341782125368, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7935816845280306, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.30600316518825144, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.21756527537889664, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.30172000282005795, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.1661049538452349, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.16760028985755937, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.23552216690812636, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.7072691711793893, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.13437076063427897, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.1815651064372292, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/README.md new file mode 100644 index 0000000000000000000000000000000000000000..d149784e1bb5561ab954762495286464d92f486d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/README.md @@ -0,0 +1,95 @@ +# SeqFunc context09 at step 44,000 with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,265,544 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. The compact balanced-splice model gained a zero-output local path at step 2,000. At step 40,000, two identity-initialized residual blocks extended its local receptive field from 701 to 10,941 bp. All old parameters and optimizer moments were preserved, with optimizer IDs remapped by parameter name. The next 4,000 examples and learning rates match local01 through step 44,000. + +This checkpoint records **44,000 updates**, 1 complete training permutations and +2,980 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 3 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup training examples. The objective is `balanced_splice`. +Other family losses follow the reference. Recorded accumulated training-loop wall time is +127.2 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +## Measured validation results + +Family-mean Pearson: **0.345014** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.516952 | +| dnase | 0.622119 | +| procap | 0.404459 | +| cage | 0.216718 | +| rna_seq | 0.328743 | +| chip_tf | 0.435627 | +| chip_histone | 0.428129 | +| splice_sites | 0.203039 | +| splice_usage | 0.280484 | +| junctions | 0.224054 | +| contact_maps | 0.134829 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5851 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6859 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.1654 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7579 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3230 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2425 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.1970 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3510 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1676 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2355 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.7073 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1344 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.1816 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..b303311b59a8500d1f68011fe9d5cd8ace69018e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/context09/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..ceffa34d61d51a09ac977496f8f7d049ec0f5eec --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/package_manifest.json @@ -0,0 +1,34 @@ +{ + "parameters": 13265544, + "source_checkpoint_sha256": "f8223af7bb526621769ac713df53b54cfe8d9d0cf0a34e2b1d18b9080d58cf8c", + "files": { + "README.md": "2a0ab49c9958cc3a2fb9737d5c4753cf7865aafd67fdeded6616532fc8a0040c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "3554a6c79b220b41087f34adaa35dfe72d58f96f8060c78ea6c6786ed9324620", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "bcd41eba9c2f952250f76aac8812a9dd995ee206791fbe640c751fa073428b8b", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "a71ed8c14bb3dee86bdba5967e4152192bd35dad78d32934a1aaab17a4719835", + "training_provenance.json": "77969daf23c2d04f9d20de14fcb24651a911019f03619a5bd91ca59c712df4b0", + "decoding_policy.json": "56749f91c8b0d85201820dc20c59a1f685a8955cf467576fff326f67fe27f76f", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "validation_tracks.json": "193050ed7ae41739e09a168fe9e44f6df679bf81be67cc430b6f19dbcf01ed11", + "validation_variants.json": "d3f874af119642c5d164598f6d34bc0b5c0400334d0c631a0d8dcca9bd16a069" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3c2effcf45b2aa0eec7e31aa3f668107d7ff90 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/model.py @@ -0,0 +1,430 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e5a5cee0af43ebd5a92833bed8d5dcc8b264a8d3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_config.yaml @@ -0,0 +1,29 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..4e116d0456eb51c5525e13f6b75b10e78c9bc33a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/training_provenance.json @@ -0,0 +1,91 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6 + }, + "epoch": 1, + "offset": 2980, + "global_step": 44000, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/context09", + "config": "/workspace/research/experiments/context09/config.yaml", + "resume": "/workspace/research/experiments/context09/start.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 44000, + "workers": 3, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 100, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 7631.602510094148, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + }, + "interventions": [ + { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..d8c32db6c9230897c55282e8925daf15937cb57f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5169519140758255, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6367338643690121, + 0.652573804100543, + 0.4011221061540954, + 0.5857791520362067, + 0.4017022557669607, + 0.39691120912728584, + 0.4165044503755358, + 0.4971073298189792, + 0.5223520181863803, + 0.5801151307804805, + 0.5120668002493944, + 0.48258714281149084, + 0.33870196434217137, + 0.451703796910298, + 0.6259839576174296, + 0.43027462143559886, + 0.48772101354663877, + 0.44911078419311745, + 0.4872650412467827, + 0.46878557497277074, + 0.4428987682112829, + 0.4050903818804531, + 0.4682818192545706, + 0.456966768318424, + 0.45016196895572463, + 0.43237186384837584, + 0.41702254318473697, + 0.436811952449506, + 0.44623985432481117, + 0.49920436897446313, + 0.42747744185430114, + 0.4499095684395622, + 0.41848369653890777, + 0.34574689076794934, + 0.31035708864130873, + 0.5118799269959629, + 0.452116833545577, + 0.5545645532173538, + 0.4468788501027642, + 0.43456572132407506, + 0.5117540671666333, + 0.5111914453383254, + 0.433890208149942, + 0.4129975261118156, + 0.49983739330722604, + 0.48800190349046507, + 0.37024563693015994, + 0.4215680160622504, + 0.41508459835151296, + 0.4627151771633768, + 0.4363941159846054, + 0.39702083766773355, + 0.487028836594565, + 0.5468630226855682, + 0.39814801144352036, + 0.6504203136967613, + 0.7400991503780373, + 0.47935737211700424, + 0.5499233175055978, + 0.5591853558358229, + 0.6788831344688152, + 0.41827953280464963, + 0.656346458497917, + 0.5631565104326005, + 0.5802301014117561, + 0.44596265690728293, + 0.4911146197003142, + 0.5755197442329281, + 0.726950136665914, + 0.6543051055807861, + 0.4740055588401607, + 0.42229486097853014, + 0.38863318357803334, + 0.47101376061671757, + 0.44151774017549117, + 0.4268451662009685, + 0.4424649545324114, + 0.4801377381397096, + 0.5190868992400702, + 0.43494647452255686, + 0.5072139496537842, + 0.5353283800558073, + 0.4536180908596607, + 0.4270048867808962, + 0.4539901793333471, + 0.44248564877266594, + 0.42648557820980043, + 0.41799891645300963, + 0.39357831318580544, + 0.45080000977980916, + 0.4438851787574938, + 0.44597058826472546, + 0.4380149034719638, + 0.4501152296339671, + 0.5126327935627687, + 0.3352311386641222, + 0.3808158640142165, + 0.3734655523338132, + 0.422643131324414, + 0.442537958077659, + 0.40013077934143815, + 0.3997698257404477, + 0.43844224672772664, + 0.4101279141886887, + 0.4791123269225001, + 0.4837706424144118, + 0.4746558669597684, + 0.45572934045260727, + 0.47744633048659146, + 0.4750852207130331, + 0.5306410907638325, + 0.45788381911196235, + 0.45656420925370816, + 0.38423022979387306, + 0.40590927137320565, + 0.38259714918192556, + 0.41668465591755, + 0.42869570603760243, + 0.4344927611380433, + 0.4153819041040363, + 0.6076649412827045, + 0.6032137867481219, + 0.6732535852752731, + 0.6730195864000755, + 0.6503792569452262, + 0.5331066980793979, + 0.8090480497838373, + 0.6575682101093205, + 0.7075242426725219, + 0.6342826988187508, + 0.680496957364782, + 0.5154409800463169, + 0.545420380043882, + 0.5178773309000131, + 0.663794281857404, + 0.6646872407126496, + 0.6462578398083674, + 0.48302470182501495, + 0.38997506272102633, + 0.758768870391049, + 0.7255997809873036, + 0.7425306208557043, + 0.7353695028418513, + 0.54420376648331, + 0.8092750377234093, + 0.7193696850696701, + 0.6677048751632753, + 0.6693856261100241, + 0.7931898511871772, + 0.7534906951480692, + 0.5711694887714259, + 0.5746790504706732, + 0.6583929975795046, + 0.6920682965364277, + 0.6997052822752481, + 0.5588523041190452, + 0.6838074332018426, + 0.5421398633618496, + 0.5558740178638902, + 0.6371272094445634, + 0.527197046182043, + 0.7146276354235583, + 0.7275914588438707, + 0.5835631862706739, + 0.5562402392152701, + 0.5286203006092831, + 0.6876685929456713 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.6221192617501552, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5289569097854437, + 0.9136216597353731, + 0.7744204072681982, + 0.4829222615619367, + 0.45440587010742145, + 0.468245639343208, + 0.935078699603243, + 0.7024662034553801, + 0.3948933991384384, + 0.3169254148037031, + 0.871699820590681, + 0.668960955541089, + 0.4056906479049398, + 0.4780199922273666, + 0.454920055251251, + 0.7947580556633095, + 0.7757826542833787, + 0.8448077515627624, + 0.8494559264908442, + 0.882424469444019, + 0.6601280419933987, + 0.5857509552849621, + 0.8956492004531903, + 0.4793049429502299, + 0.4980572339390654, + 0.7712359873065064, + 0.4295663831304184, + 0.6382143059856855, + 0.47333272395404563, + 0.4964070297624827, + 0.5947587036362088, + 0.42439019811525225, + 0.8346856498490669, + 0.6178080752839449, + 0.8757280655266343, + 0.8822935845454732, + 0.4147989232971965, + 0.8195206051811601, + 0.8714335659193031, + 0.853646132315553, + 0.8190797304234889, + 0.6620689101185859, + 0.6988350819978223, + 0.6156070322474595, + 0.46725865863085897, + 0.6282904630765517, + 0.7255567257544312, + 0.4934265444105113, + 0.5701004101695434, + 0.38268237703092717, + 0.34557826349170323, + 0.5376801522704717, + 0.46598872907211214, + 0.8240675121137458, + 0.40171589130055285, + 0.3834040814030519, + 0.42015870039445624, + 0.4575482143557857, + 0.4232261961308124, + 0.5080662924136998, + 0.40185244491435085, + 0.5973683217328789, + 0.4554004647602665, + 0.44176519471003367, + 0.4691712714286817, + 0.4738968167108436, + 0.5232433846048198, + 0.4418799273399047, + 0.3525450158910681, + 0.39484715803329257, + 0.4953363190964755, + 0.6100432139349335, + 0.4574303470558998, + 0.6004202782756266, + 0.3169308395835691, + 0.8032010952943468, + 0.890045727035608, + 0.6352743826114592, + 0.4111897884998479, + 0.42646411000903484, + 0.4741391834944844, + 0.4664378265929594, + 0.46068650543595635, + 0.45180711180414496, + 0.5207486356804719, + 0.47518175036301197, + 0.5701698974469238, + 0.4797598724822678, + 0.6176439723191436, + 0.4258066742507169, + 0.6048324735803841, + 0.3946392535171926, + 0.7122541239129481, + 0.44471750318938674, + 0.6998971158533647, + 0.47990025723802837, + 0.7190374478406389, + 0.6827877336603911, + 0.3261449953244557, + 0.33267276768727133, + 0.4409972706064492, + 0.35246139701838414, + 0.3310724172042645, + 0.7002075620726064, + 0.6624104557155696, + 0.39979839283883856, + 0.4320395977999298, + 0.5761301247101657, + 0.8393527054775677, + 0.538328692744639, + 0.5229829331662482, + 0.5583204059755366, + 0.4750834956289175, + 0.42827681168795106, + 0.7675236371330115, + 0.527951214996832, + 0.4739677240055193, + 0.534811105840829, + 0.6073703372087985, + 0.4144226641009887, + 0.5144950298292015, + 0.4957570731609815, + 0.5310842199120637, + 0.5093940127987275, + 0.5862917423389057, + 0.5341805740738351, + 0.5947463421850462, + 0.6723084397163342, + 0.6853829924630233, + 0.5963706317155957, + 0.4491630857202775, + 0.5892716260310421, + 0.6103531758066836, + 0.7818382932029204, + 0.4310839714434318, + 0.3020186425959827, + 0.5693987168241806, + 0.45946801448781116, + 0.6238897126845785, + 0.5892126483569936, + 0.5989808932095531, + 0.5460056023373139, + 0.458609623475162, + 0.7691265720670478, + 0.5858378007210024, + 0.7535173884113824, + 0.3720529888022095, + 0.36351224225822176, + 0.38523205799204513, + 0.3277112088739813, + 0.8879800759911984, + 0.3059173078632669, + 0.8912067305163245, + 0.718911300208406, + 0.45800543464148863, + 0.3223357619418515, + 0.3383296674918975, + 0.3550676801376659, + 0.6278380017560581, + 0.716799983601962, + 0.718797924527262, + 0.644452157692366, + 0.5488430033533034, + 0.44492756583940457, + 0.6698438388309591, + 0.5939941409249981, + 0.49543278423878884, + 0.6197859475285816, + 0.4427258395790773, + 0.4611515168935845, + 0.37307216256164805, + 0.8239019702209871, + 0.43538206166277565, + 0.453124175383279, + 0.6411321765207357, + 0.6788024387804833, + 0.465941802617538, + 0.33388193237443575, + 0.872498978446465, + 0.3581635881422971, + 0.7522670107922172, + 0.4339833820423245, + 0.5800169628255759, + 0.7079631461978387, + 0.4466356131105542, + 0.5496463809030283, + 0.5783532323063946, + 0.8123406608878571, + 0.40788552599540867, + 0.5940297812312122, + 0.5341092579210241, + 0.7005276658683963, + 0.8919110656366573, + 0.39169469467728696, + 0.5665252773104315, + 0.44575190239941975, + 0.6065790826842686, + 0.6908504121906084, + 0.5535200964730638, + 0.4293162258661189, + 0.8646156615648585, + 0.8761177399425063, + 0.7490416621427016, + 0.8301543765848138, + 0.8401777248470881, + 0.8290873937357023, + 0.8388469943073608, + 0.34433963629141023, + 0.3581620717177054, + 0.49173069795745733, + 0.44173148256348566, + 0.4278608643612276, + 0.5806229851859551, + 0.8573900633907607, + 0.8436712266142085, + 0.6578877564776912, + 0.8202623668325815, + 0.8196123880211491, + 0.8529197821117951, + 0.9008324420820415, + 0.47730693261515394, + 0.5401361316864963, + 0.6371249622656677, + 0.5601971480411133, + 0.8142394490803077, + 0.5483796194410848, + 0.9028921372282043, + 0.8760960193126819, + 0.8683259774431756, + 0.8965055697079255, + 0.8882877688593843, + 0.8657126379175616, + 0.8243406074341715, + 0.8543080713290537, + 0.9008725993627411, + 0.4163988033501991, + 0.9189655346867569, + 0.8735005517384891, + 0.815186193523814, + 0.5726263223468825, + 0.8672164473233501, + 0.4917412633571957, + 0.7462728900182481, + 0.8699578969816838, + 0.5538379283063054, + 0.5416971978154566, + 0.6705898928090168, + 0.5060259625939639, + 0.5072446606743921, + 0.7203428136530858, + 0.877165424506577, + 0.7151054746853457, + 0.8026087912312673, + 0.591261359381148, + 0.8586348250422243, + 0.8000096620575436, + 0.8307965893333832, + 0.8979522409267727, + 0.8273083914256977, + 0.8932768154074584, + 0.8648387445595235, + 0.5116590901697293, + 0.8725625793963703, + 0.534304131015614, + 0.5037120643257761, + 0.8781132515712108, + 0.899931552973062, + 0.8752067048424826, + 0.5523446931672954, + 0.5932906097922506, + 0.6478385891873385, + 0.7924589356911044, + 0.8842094770678337, + 0.6894476553254972, + 0.9003760838781516, + 0.910812397969864, + 0.5970062727655179, + 0.30090230992139844, + 0.49759671680208817, + 0.8939610355065162, + 0.40533869859761873, + 0.5856142578315172, + 0.8754281149127547, + 0.8247972006141591, + 0.5918270225968363, + 0.8606270299836738, + 0.889395610535412, + 0.883744485889877, + 0.872296930145089, + 0.8581295035990597, + 0.8880330511576036, + 0.8487030665402958, + 0.922208555846408, + 0.7777779274711822, + 0.8268161049023214, + 0.9056427270270733, + 0.5262712829466438, + 0.534393354527222, + 0.5153386368724466, + 0.5004423541344205, + 0.9032586593760489, + 0.7922409684186099, + 0.8435402155025687, + 0.7863830947823642, + 0.7570003328112224 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.4044585706229842, + "valid_tracks": 12, + "per_track_pearson": [ + 0.39107177315275327, + 0.4347724933304747, + 0.34902994839527485, + 0.43051866100332814, + 0.4283045824922067, + 0.4329144873706136, + 0.3725674246991704, + 0.4233261735647866, + 0.3354446417201755, + 0.4182055620688495, + 0.4058570369173547, + 0.4314900627608228 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.21671763887966514, + "valid_tracks": 546, + "per_track_pearson": [ + 0.20562340708420346, + 0.24512741414922554, + 0.20943097989496734, + 0.13134727325952922, + 0.07846448255018405, + 0.21402484128248597, + 0.11269562190180266, + 0.1429382768858765, + 0.2287595869940448, + 0.1937057013948044, + 0.24577786484833436, + 0.13757484154032704, + 0.1465508821278036, + 0.1237361586585224, + 0.09875329763605216, + 0.17639697750006803, + 0.19842755350464678, + 0.24389536437857756, + 0.2592327690847243, + 0.23090318837207588, + 0.19925830324105318, + 0.19475708238698852, + 0.3049010776638421, + 0.17270466700682166, + 0.16632785413164988, + 0.22430936211595778, + 0.23144980421957084, + 0.2342522794645249, + 0.25405141451983226, + 0.23649089280561847, + 0.1630915005068072, + 0.2562387367404046, + 0.1913766865405479, + 0.1369215241045783, + 0.2621948777128117, + 0.2435944744848156, + 0.24955637629252828, + 0.26944833267472273, + 0.241306469374647, + 0.2889720145153497, + 0.30114765859332415, + 0.2557849860921968, + 0.22609869260578627, + 0.2259678608649159, + 0.1394951668535957, + 0.18794860758436951, + 0.20400090008422564, + 0.21585205954114858, + 0.1295468815551105, + 0.16878851134806414, + 0.23160484932046413, + 0.24846244854177704, + 0.2648011533203724, + 0.18408012870063667, + 0.28724110518828433, + 0.30391792651532934, + 0.1671685511273831, + 0.2887857473459047, + 0.21443227563836892, + 0.26014379543112376, + 0.28198101923709307, + 0.18123376342544792, + 0.25136992850614603, + 0.1869067327630714, + 0.2613075279637714, + 0.2790487958225544, + 0.2912344498337312, + 0.18071092628198987, + 0.12016620534937435, + 0.11392244820247106, + 0.222624582087443, + 0.2617500065579066, + 0.21982342212825268, + 0.2330909801356625, + 0.2558062528599801, + 0.21276702281855006, + 0.25443801273347166, + 0.22927334282425796, + 0.2586078499828407, + 0.26615375130892677, + 0.21960473490751103, + 0.13595178938743546, + 0.24118764542736398, + 0.28669501617325943, + 0.13395796937363522, + 0.2439002125077165, + 0.2038605227381635, + 0.28201304290192797, + 0.172574836907248, + 0.25722630540231345, + 0.18497790940492417, + 0.22048662337251296, + 0.23701759795029007, + 0.11533844603105338, + 0.2523879338335284, + 0.21732621027253105, + 0.23133464690766467, + 0.2420735782474931, + 0.22794539596769295, + 0.1396665939179402, + 0.21724849255968295, + 0.23302614373397337, + 0.2430060332105477, + 0.3164449196492101, + 0.2767722814788953, + 0.2807902464440551, + 0.189447482189608, + 0.24138760335769444, + 0.18671213311627538, + 0.20389311286692324, + 0.21427504089779198, + 0.21215440380234457, + 0.23465400990265814, + 0.23188810451394415, + 0.19586186672173198, + 0.2359208040398986, + 0.2360005344057293, + 0.24267798688093728, + 0.22970633254133602, + 0.265893965968281, + 0.26327020803906326, + 0.2593806336083349, + 0.15894634814547032, + 0.17653552759666802, + 0.2083532897905059, + 0.25104210535948956, + 0.2260671570361928, + 0.24850433978572634, + 0.3265758412528506, + 0.2528813613905412, + 0.2613408410477281, + 0.18000683123730632, + 0.25359815407747843, + 0.26940215722473265, + 0.3101271774811303, + 0.22356408874260514, + 0.2763750735951846, + 0.2012908171745541, + 0.12496989358182105, + 0.24596662855515666, + 0.28453130497478996, + 0.23408410705524568, + 0.2534175350246954, + 0.20485038952868323, + 0.21131286668106594, + 0.22527006330603672, + 0.24942677043035053, + 0.2852700956394983, + 0.20712310804107248, + 0.19939320283481254, + 0.24524655309913465, + 0.18629011075988697, + 0.2329681904407709, + 0.2440819064511838, + 0.22220865862801545, + 0.17404086696063686, + 0.26563344228321517, + 0.2998014357397481, + 0.18650838161984742, + 0.1676539685643202, + 0.15000396536266628, + 0.20954280347914186, + 0.20857855836088884, + 0.24364921341409493, + 0.21710088746225248, + 0.17460214597507479, + 0.2293572477703677, + 0.16248403829615019, + 0.16979565797925109, + 0.267959898097536, + 0.2322342440802662, + 0.181320836296417, + 0.2195754820252675, + 0.32836317876293425, + 0.235177086816596, + 0.16471684704235331, + 0.15442896218293203, + 0.20536527474183644, + 0.25559216384046973, + 0.29316373407906987, + 0.2834258816250266, + 0.28435734262367895, + 0.14763229019952587, + 0.2711428857527873, + 0.22997885627708456, + 0.18158148906169727, + 0.20214781495913728, + 0.1650761117236785, + 0.3075066293881585, + 0.24198450499324498, + 0.196535101113058, + 0.1374797568378802, + 0.13597696806480963, + 0.2135020693470916, + 0.22344695582062518, + 0.1988689607760446, + 0.22836122856599542, + 0.2258146799730418, + 0.16765410543621287, + 0.2235211548145866, + 0.1260630751010572, + 0.11326920471911546, + 0.12759284750676458, + 0.12010899079231725, + 0.13410369461457378, + 0.2166560274521715, + 0.21665717475998353, + 0.18258811420995832, + 0.12040249867394029, + 0.20306743317092613, + 0.28448049750912335, + 0.24465542941948218, + 0.2629966285959256, + 0.2597366486531863, + 0.23615660814010228, + 0.2208938082484105, + 0.2829463551345749, + 0.3034371797712867, + 0.2617124970426645, + 0.2717998113275685, + 0.2447330722375395, + 0.18303044118356457, + 0.25240244611740004, + 0.23194151198556706, + 0.26592686885797073, + 0.31258133520516107, + 0.24796586279716593, + 0.21639180213647105, + 0.2325867821555055, + 0.25257910402483535, + 0.16799808244512404, + 0.167332521147609, + 0.28797179319158805, + 0.22153335344321806, + 0.19155264792419335, + 0.22379348198152393, + 0.30107298869634314, + 0.1665413270236291, + 0.2372857380167345, + 0.23172506867510106, + 0.12366258423302573, + 0.16608472014567416, + 0.15755775583053266, + 0.24792823676131306, + 0.18630123578188576, + 0.23004527973283664, + 0.09344840624024597, + 0.2874131262590886, + 0.24459592420426327, + 0.2592416871189403, + 0.2621818732876677, + 0.16545681307662818, + 0.3043923877672953, + 0.1391640419310085, + 0.264027906557273, + 0.2298746695559135, + 0.3382252344380831, + 0.2702168167066978, + 0.2760096613223597, + 0.3624970944151179, + 0.2184041599638742, + 0.24305607685328548, + 0.13680082647608321, + 0.24305395597551835, + 0.2440803534053086, + 0.16921840816346398, + 0.14352599055389031, + 0.14544438000483498, + 0.1694118607033486, + 0.24543478375350625, + 0.303669027986253, + 0.16167034359558874, + 0.13663421628044578, + 0.18713202629565429, + 0.2633271584818659, + 0.21202910009587334, + 0.1450928865098509, + 0.06707435513917978, + 0.21779697415145902, + 0.12747646118798764, + 0.15443623542969084, + 0.23114253170920523, + 0.18597951129317503, + 0.23251030517328727, + 0.13424210948735502, + 0.1521973504387164, + 0.13574037137029954, + 0.1181802003604337, + 0.15842493575848882, + 0.19474447939649223, + 0.2332437678877951, + 0.25180583637241455, + 0.2211776152358802, + 0.19435095493355262, + 0.19123594875505356, + 0.28979850053273604, + 0.15794968624880626, + 0.1605767792239566, + 0.2375502206657854, + 0.21861854746230377, + 0.220866917045465, + 0.24256576620594736, + 0.23081750003773216, + 0.1558041410485235, + 0.23909420725685376, + 0.17640128485189735, + 0.13005992543822079, + 0.24835815324043017, + 0.24224580809864918, + 0.26181903109818533, + 0.2619829860208141, + 0.2570409421622975, + 0.27623221233478507, + 0.28276533729840236, + 0.23989129621058766, + 0.22329202066780546, + 0.22156718233908396, + 0.14731846811603452, + 0.19484864929196985, + 0.2156613980339949, + 0.2386465068332694, + 0.1280711801967045, + 0.17017208354585833, + 0.2149898132012654, + 0.24962554802963752, + 0.2542243398273322, + 0.1790305539466146, + 0.2737507153836283, + 0.28297417866564, + 0.1730061463651474, + 0.27581593064937315, + 0.1966168244679369, + 0.25494177831843834, + 0.2658923208979796, + 0.17296023003564864, + 0.24238853831755555, + 0.1898100473732581, + 0.2530220235403051, + 0.26018602215877434, + 0.27184919288347925, + 0.1709252337704604, + 0.11869719650684941, + 0.10558162444859438, + 0.21103453636491157, + 0.25183579531430267, + 0.21522270269278895, + 0.22197264886311613, + 0.25033084109390547, + 0.2091181511279228, + 0.2320678942126082, + 0.21469072871070996, + 0.24246081485875443, + 0.2591665747216854, + 0.21375728059396362, + 0.13261459973641493, + 0.2383022313052564, + 0.27686110385832086, + 0.13380108121428597, + 0.23294969382265665, + 0.22410884559132113, + 0.280631469567876, + 0.1633661417156773, + 0.2560628600295315, + 0.18276080676827716, + 0.2134790517178706, + 0.22890788193064526, + 0.11314958156695121, + 0.23768874879293744, + 0.20342371007898308, + 0.21522910753054986, + 0.25099493469827566, + 0.21371008718390133, + 0.132394278062326, + 0.20729618167074265, + 0.23708771619967564, + 0.23034523326484357, + 0.30057215667792664, + 0.25619515788407143, + 0.275291488644103, + 0.18102133945590382, + 0.22999061812192328, + 0.1743444899487507, + 0.18595933578335505, + 0.200098391796633, + 0.2020691478119711, + 0.22808309902266077, + 0.23126147880409828, + 0.1839504506811663, + 0.22877134316110365, + 0.2167307360046092, + 0.23001732383769102, + 0.21428140394896023, + 0.25893653044305015, + 0.24659404838306498, + 0.24134255104271554, + 0.15928652343811367, + 0.181726417679691, + 0.21188204198737287, + 0.2427478039491924, + 0.2139398378240866, + 0.2452240109437541, + 0.3148886443961387, + 0.25509852672714006, + 0.2528207162990399, + 0.17554676248439957, + 0.22937944056050574, + 0.24470801256603789, + 0.2905884769196023, + 0.2214005639180496, + 0.2678183714749465, + 0.20281736136397213, + 0.12806035732641569, + 0.24087440970001817, + 0.2617307167193831, + 0.21391216734456026, + 0.25771071338803814, + 0.18900596191403612, + 0.19929916743746545, + 0.20891925733337724, + 0.2296990004239579, + 0.26551850372670677, + 0.2184713866669479, + 0.19116218687787495, + 0.2357895826288171, + 0.176971233909454, + 0.21534869432576578, + 0.22682127193937038, + 0.20771990089983852, + 0.17885425613929273, + 0.24995040051591658, + 0.27951254018990235, + 0.17968360369965086, + 0.16884706052235504, + 0.1481524249023122, + 0.1880158249400205, + 0.19797850225253066, + 0.24803357413142155, + 0.225920516679859, + 0.16150844982739068, + 0.22041766083717906, + 0.1746801478681904, + 0.15799461570111678, + 0.270802648530324, + 0.22330346884012003, + 0.18743459445386976, + 0.19920000941353452, + 0.31937987500882775, + 0.22980231585488295, + 0.1484968757393058, + 0.15156274097444714, + 0.21365575825082822, + 0.21095861576288197, + 0.2992506407286335, + 0.27483711252899773, + 0.28521502069486665, + 0.12717498690864024, + 0.26477964322031944, + 0.23105721796500642, + 0.1965889504620931, + 0.2080528131917117, + 0.16627560820081874, + 0.27848508569752045, + 0.22498137093377854, + 0.19806803470733225, + 0.15772524899516643, + 0.10845340522890085, + 0.1579432652840982, + 0.23504971122802865, + 0.19761293769250987, + 0.2353269824469692, + 0.22204766645283142, + 0.1708862309483549, + 0.2010898604207335, + 0.1007782024314163, + 0.08048656157763366, + 0.10158695958226815, + 0.09994904443895201, + 0.08898442366524373, + 0.22043617490494447, + 0.17282587396367594, + 0.1622375536276928, + 0.11250356611352763, + 0.19769006620529006, + 0.2902368234699835, + 0.2822493778893957, + 0.28402899750108557, + 0.27676903244967754, + 0.2609784234727267, + 0.23181471359268088, + 0.2885027396940468, + 0.3058848596574141, + 0.2671168501644488, + 0.2800370057238896, + 0.26066055969433954, + 0.19830361001981384, + 0.2695327123718187, + 0.23563303637431676, + 0.29720756057005415, + 0.3098773683708142, + 0.24117233735313928, + 0.22423309805375444, + 0.22061508182095774, + 0.24176738218897056, + 0.1247134357279558, + 0.11208218946322526, + 0.29087866918474653, + 0.21707864817164327, + 0.19938535910007554, + 0.2169976650898009, + 0.2973859604622079, + 0.1821213804544916, + 0.2298079909848144, + 0.21784247484091002, + 0.11202323061240577, + 0.1364674354597148, + 0.1367742449567964, + 0.25924908592657697, + 0.16698826382495388, + 0.23577893369912517, + 0.09111528621050785, + 0.28895989289357377, + 0.2487592772747449, + 0.25713625138517743, + 0.2641862687430382, + 0.17962115252545025, + 0.29492086872960666, + 0.12776699925817678, + 0.24469539704082244, + 0.22435498948489124, + 0.3247219161352488, + 0.2748185819507237, + 0.2913632848272675, + 0.3518359747197956, + 0.22696154033502125, + 0.22510721382090743, + 0.12402128599796533, + 0.23216047645507779, + 0.24102665377580124, + 0.18878884170374052, + 0.13111521949503444, + 0.12865744931596235, + 0.1582156554481372, + 0.24995602096140143, + 0.301633232959051, + 0.14547745294190154, + 0.11419777694721422 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.32874268536651374, + "valid_tracks": 613, + "per_track_pearson": [ + 0.30696611730231654, + 0.35084677273324816, + 0.327232788074233, + 0.3730953529236726, + 0.3501075623434754, + 0.22348536467885752, + 0.26953593681819293, + 0.35539191163506906, + 0.34958987109223766, + 0.38458570320814284, + 0.3045626750407335, + 0.30706343439566897, + 0.24741475923491338, + 0.36615729566324223, + 0.3384763647727203, + 0.35641034926872756, + 0.39869763757255056, + 0.3263869952170655, + 0.33076985738144066, + 0.3128737808780542, + 0.2731183577470696, + 0.35016443765490796, + 0.29175322350316774, + 0.26867187393173386, + 0.2420548833540739, + 0.3273469493369429, + 0.2947709300807259, + 0.35062751743691173, + 0.3350047650218868, + 0.34994435524876366, + 0.33576069370823, + 0.3298307014087009, + 0.386451487602127, + 0.29578347925793386, + 0.38962338135395175, + 0.26969409163391506, + 0.36340673844674537, + 0.3576036912823517, + 0.3185032807308652, + 0.35821615683925534, + 0.35704784321840727, + 0.31514380952144483, + 0.34744545839154184, + 0.3397981955858789, + 0.30605831906202013, + 0.23753874239279416, + 0.27951050581324005, + 0.36093815152779546, + 0.3070134205434667, + 0.3336904767802463, + 0.34058499205241055, + 0.3717133683064915, + 0.2620478142146016, + 0.31142174240683873, + 0.23981773610678625, + 0.3281132442392351, + 0.3087365929200704, + 0.34775038315108564, + 0.3713546229126144, + 0.3459932622171929, + 0.33369494148364653, + 0.3387752252606182, + 0.3582681807168717, + 0.248938308301753, + 0.3511061610180033, + 0.3331491572535664, + 0.36014053477129643, + 0.36479302785691475, + 0.3407359677648597, + 0.3541155261713658, + 0.3459694601205711, + 0.357536487866388, + 0.34474648351765286, + 0.3408231388925931, + 0.3479157498130683, + 0.35214018914431633, + 0.3607485158067828, + 0.3621463462035936, + 0.34121993875188045, + 0.35540207358265274, + 0.3232600699757556, + 0.2714058720178794, + 0.3936376119245242, + 0.394985214836079, + 0.30659428350933465, + 0.3252162568541452, + 0.3484138281666806, + 0.30280228894548156, + 0.3293316275837548, + 0.3019479169018741, + 0.27837134609494174, + 0.3533335667253134, + 0.34157035573457045, + 0.33577145258145585, + 0.339858123791541, + 0.3483963747978507, + 0.35184274330806375, + 0.3406729687741904, + 0.31353919999682384, + 0.2761756484895333, + 0.2578176767119178, + 0.2814591287125638, + 0.2807971189545109, + 0.343376671683254, + 0.2937417595477802, + 0.31482361863337216, + 0.34694453368797395, + 0.34781599733731633, + 0.37722518638195857, + 0.31407309777027587, + 0.30653120088283486, + 0.3769870523085767, + 0.34432390877635527, + 0.2970624325488076, + 0.33510292494624433, + 0.3532022568463358, + 0.39164577958796704, + 0.3112879196856036, + 0.31105948554816737, + 0.31817773095112933, + 0.3026266637598956, + 0.31270232517296664, + 0.2965880726223141, + 0.3406607237511136, + 0.3935886623845376, + 0.2715663759040944, + 0.2876207727822271, + 0.23856112330222962, + 0.25693294121648935, + 0.29107047117480594, + 0.40118708766877076, + 0.30591436651495924, + 0.2933840801741794, + 0.2719892241454626, + 0.2923729234444703, + 0.35821777971063457, + 0.3549347621596624, + 0.39518580645904827, + 0.3408048480796986, + 0.3725642023421633, + 0.2976366534805562, + 0.30516565306470145, + 0.3921543008340036, + 0.3514404610035897, + 0.25495902342276555, + 0.27757927401771454, + 0.26288545761637466, + 0.29851384379111806, + 0.27411049839781454, + 0.310997051870551, + 0.2731633221797884, + 0.3077103912384202, + 0.31857933429239194, + 0.26133073815418384, + 0.32837152494365845, + 0.26255492044724976, + 0.32174848238078396, + 0.32748588576467896, + 0.32973925444294044, + 0.3462896491547796, + 0.39636755086704983, + 0.3209673115641582, + 0.3525697077772785, + 0.3121438944994095, + 0.3097070491570827, + 0.3069862477941333, + 0.3145934160753259, + 0.2685694509009606, + 0.31054375160505077, + 0.3187579452329431, + 0.33800603538963125, + 0.32592931905299, + 0.3857093858807917, + 0.31568935121491243, + 0.2728032381507845, + 0.2961134015204494, + 0.30369000912355376, + 0.40213059507083715, + 0.3068083754939965, + 0.28120726695971643, + 0.2877661279691439, + 0.34511978167349516, + 0.26077816032957285, + 0.32561830047557555, + 0.24457497045871165, + 0.32621270199478886, + 0.33883409855696234, + 0.2149515255159232, + 0.2943962294021577, + 0.36111488682651754, + 0.4125048325753477, + 0.3131624187614433, + 0.2907750731238851, + 0.32630037401552797, + 0.32754553098864125, + 0.31886731926528245, + 0.320429681341178, + 0.23539586722404834, + 0.3085370456442626, + 0.33439473563395633, + 0.3362326517390479, + 0.32817745029546525, + 0.43500908671939853, + 0.3656882830420097, + 0.3304766491086102, + 0.3338631042396288, + 0.3348717192630044, + 0.19951149475017838, + 0.3374621595844586, + 0.31507719073354223, + 0.2485585625040532, + 0.244568526488536, + 0.2805562137716745, + 0.2974610033099723, + 0.2938214801679909, + 0.3542593117334275, + 0.26761958733584723, + 0.26639670782797925, + 0.3085835705978093, + 0.32783812628381725, + 0.426112410455706, + 0.36400531205995723, + 0.36851859517022784, + 0.34747092974107174, + 0.34389826328841955, + 0.3849582252763724, + 0.34818562108119905, + 0.37713658297140673, + 0.26873155613249067, + 0.3461134353596477, + 0.32369829313670595, + 0.3984872188654817, + 0.35496667143531, + 0.2913415034878126, + 0.3552036267965172, + 0.26139341183834786, + 0.19793461219203898, + 0.3660337055823058, + 0.3663031049495775, + 0.3586237427255727, + 0.373728871015813, + 0.34010610180959283, + 0.3486169956444898, + 0.3244335031785885, + 0.22079503999594005, + 0.3434078135845678, + 0.36457135411371117, + 0.3370291911580442, + 0.32077567902349047, + 0.30232555116481474, + 0.279148245319658, + 0.32494168019757874, + 0.28780744561400273, + 0.3000658498699322, + 0.3438437117252543, + 0.36037969066940506, + 0.2790632792309794, + 0.3479156948914915, + 0.3127811895557584, + 0.3418025393938255, + 0.3133193369692182, + 0.31101737777615285, + 0.37366578807486117, + 0.39414658057745217, + 0.33930064750231853, + 0.3467156032933559, + 0.34742565052846835, + 0.27337669087012395, + 0.3276370134930735, + 0.34197597163960564, + 0.2624415359308474, + 0.2625676335277623, + 0.40159950198135963, + 0.3663985479314744, + 0.39028749973792254, + 0.3782263059364041, + 0.2063216735735701, + 0.22879914019055433, + 0.4031627122549396, + 0.3962827395814516, + 0.39943019647477696, + 0.36395761438562485, + 0.29839494745637823, + 0.29788250457997856, + 0.3301205637365877, + 0.2997176748659424, + 0.31010412515155406, + 0.37663501136887306, + 0.3421007952768476, + 0.4027266479860295, + 0.31453102876945765, + 0.2603092066152707, + 0.4018512599531567, + 0.2647987298750037, + 0.23336084524749623, + 0.23136869724153555, + 0.36588678269712444, + 0.3831164578416641, + 0.398123409748863, + 0.34411602144020315, + 0.3461958888963144, + 0.3238801627508371, + 0.3674508095140935, + 0.28811594638160637, + 0.3359163416111051, + 0.40147646156721073, + 0.3258258978437936, + 0.3845523860936001, + 0.377711934204759, + 0.35926676106825356, + 0.35569573621764966, + 0.36082208064684856, + 0.3521728369663545, + 0.3829219775686492, + 0.36027501656424094, + 0.2838851067314901, + 0.2603057818445204, + 0.2908329961709212, + 0.3525164211309864, + 0.28956288944310365, + 0.371571174059207, + 0.38086542178427185, + 0.38338308940017846, + 0.23069202686183907, + 0.2928677430790796, + 0.20351759045115708, + 0.3167868428196727, + 0.31931728618500055, + 0.37432694798288313, + 0.3916805419450574, + 0.36362606992787894, + 0.37701861435458867, + 0.3977790233470155, + 0.41357424159816186, + 0.23314159343650043, + 0.39245608642329194, + 0.39679051058275266, + 0.4158265914135474, + 0.4167928831507466, + 0.4044275354979057, + 0.40727719996533135, + 0.37135770252174866, + 0.4130270632536204, + 0.37209481714783904, + 0.41230955900086486, + 0.38794747478040353, + 0.3950845039748113, + 0.40611134373752816, + 0.41274674215256707, + 0.3024392915944494, + 0.3747631083162634, + 0.34665366218014937, + 0.2922323972829604, + 0.4155478230628413, + 0.43935445959977637, + 0.3483135865737206, + 0.3797822399415647, + 0.35283226606607226, + 0.2562721132503569, + 0.3339542752233795, + 0.28276661244637785, + 0.27622436413338186, + 0.3847642233466797, + 0.35899929618242243, + 0.3402801248640533, + 0.3295549912577913, + 0.3696710754887287, + 0.37856799559949217, + 0.3443552525318372, + 0.334249869446647, + 0.30568958051589595, + 0.24849462378484058, + 0.2458905710729487, + 0.2740611659234395, + 0.4094139504740072, + 0.2622824535303184, + 0.2881053522077608, + 0.38365588924440774, + 0.3234392108227871, + 0.4122438314646206, + 0.30341390424134085, + 0.33270124587965405, + 0.3676469684295367, + 0.29644445246224815, + 0.33845042983486306, + 0.3980455345117516, + 0.32510384684182086, + 0.37017430426207365, + 0.2968653824936589, + 0.3042820823698719, + 0.31264282854519787, + 0.36929786673381193, + 0.3227569342152951, + 0.31462696120440137, + 0.3596950115276188, + 0.3926622750448705, + 0.24728867490618914, + 0.25386387050955, + 0.2315929382995154, + 0.23133227777062482, + 0.3097942133388937, + 0.41581759328022455, + 0.28261258359929603, + 0.29230763006838295, + 0.3425703244553467, + 0.3147544942894892, + 0.4240510429157538, + 0.3517080982402922, + 0.4180476542288045, + 0.29138476650352146, + 0.35414220274931657, + 0.2608173531481665, + 0.3175598600865156, + 0.3830648478159405, + 0.3173037731273525, + 0.24047114933794808, + 0.27205401568239473, + 0.29952234026033103, + 0.3474773434958535, + 0.35976657011190355, + 0.2940498512804127, + 0.31963414260047956, + 0.30570652323662095, + 0.3534708374408317, + 0.2657729637996656, + 0.28640209736050437, + 0.2729812855771844, + 0.36265125934328984, + 0.26229377578649504, + 0.37515551686066223, + 0.3322001129357345, + 0.3809682189079865, + 0.2930480713497266, + 0.3293005369522164, + 0.21600345497559692, + 0.22578531831812904, + 0.2232478969226436, + 0.21928333070077935, + 0.21663223745986523, + 0.33288303750951814, + 0.3104334603351749, + 0.4216269981282676, + 0.35215465623493175, + 0.44396800235312195, + 0.3235710051923464, + 0.2837830397983762, + 0.3023070044794623, + 0.29684901124002294, + 0.438643133326621, + 0.26301767344738763, + 0.28639580815645327, + 0.260110591738535, + 0.39578311881920586, + 0.20929980337546986, + 0.3048916962570442, + 0.21443580317736205, + 0.3271817012278067, + 0.3750218596196392, + 0.23524527297406805, + 0.3075802470798016, + 0.3289089714033119, + 0.38801657220681135, + 0.383419447686536, + 0.3017428450459674, + 0.31599772038973184, + 0.3529199479362704, + 0.3580822860779641, + 0.358052294463845, + 0.21892789365018933, + 0.34766942265342865, + 0.3881863380147131, + 0.30896441967967897, + 0.3980288748515757, + 0.423085671165578, + 0.3808704158052319, + 0.38716839536319714, + 0.36064359190470696, + 0.34128108014491737, + 0.17363930084740775, + 0.4048972026349358, + 0.31262886488217395, + 0.36312722160212374, + 0.3580839043855094, + 0.41012524565735853, + 0.39393590104116516, + 0.3545163448862822, + 0.43267242106645765, + 0.38486156708177544, + 0.30744200991093157, + 0.4118292700778533, + 0.40032786348842686, + 0.3843765310635773, + 0.4115688748384218, + 0.34027674240461936, + 0.2961161481504973, + 0.29620104264044894, + 0.33420873905664256, + 0.2601614504107372, + 0.33701029415864486, + 0.1934481389593916, + 0.4070301719837959, + 0.2621960053381935, + 0.3775164078332953, + 0.4229960950983846, + 0.25060062176299613, + 0.40936261219607134, + 0.23202252992158887, + 0.18736320288217959, + 0.40715501090619804, + 0.38694957796268675, + 0.3503050812359454, + 0.3610834997750225, + 0.38791868581594885, + 0.41717896168722746, + 0.4208623675936824, + 0.23195648347914974, + 0.3541772113553549, + 0.3732382562453465, + 0.3852660996873995, + 0.3293864395845709, + 0.3154774019447497, + 0.30510573018273274, + 0.34650543172952025, + 0.3151345565036105, + 0.32296698393548545, + 0.38059680755886, + 0.34991378401934425, + 0.22935019072915055, + 0.38343675970503344, + 0.3446511065483062, + 0.40562112322423216, + 0.2531197591834153, + 0.22213999560765607, + 0.37251731728177806, + 0.39536744443379457, + 0.36863252354942005, + 0.40253713507137845, + 0.41059087368762315, + 0.21267019649394886, + 0.38733686241331616, + 0.3829598739957327, + 0.23970078417557297, + 0.2706858094326482, + 0.13484712492276882, + 0.3236128388962424, + 0.3844344438696174, + 0.3271431428980142, + 0.3299665721072799, + 0.3129714918962562, + 0.32302961363996924, + 0.3294374511324987, + 0.31144491009833875, + 0.2566953287192895, + 0.2675005844609246, + 0.2701012846142884, + 0.26158615867140667, + 0.2615611213884853, + 0.26378063653065703, + 0.2702675056884941, + 0.2517727886858191, + 0.24918979242139647, + 0.2994665539251164, + 0.32130900159383013, + 0.3257767805677455, + 0.3211254586216962, + 0.3243483797687716, + 0.3523219264363045, + null, + 0.36923759107692466, + 0.40990582277332555, + null, + 0.18272961618190153, + 0.45160735758227993, + 0.17151606907713673, + 0.3583251239888548, + 0.3668844738706299, + 0.40656961398379865, + 0.4335860414578041, + 0.4265379479561546, + 0.44086205882685553, + 0.41943664893834903, + 0.3723359092828259, + 0.35625620250133416, + 0.28746833323740645, + 0.43714786930734134, + 0.2566538805755324, + 0.3540710173175543, + 0.32484348595875173, + null, + 0.34442389591752165, + null, + null, + 0.3288217431854921, + null, + 0.308486007225775, + 0.28479050492155483, + null, + 0.38353405617397074, + null, + null, + null, + null, + null, + null, + null, + 0.3427346183490027, + null, + null, + null, + null, + 0.28833016891274704, + null, + 0.3341536116011353, + null, + 0.30560850060873496, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3587925455373887, + 0.23334116267042562, + 0.333547439627136, + 0.34828447981283933, + 0.28566355081511446, + 0.34781042361193615, + null, + 0.39218907875882425, + null, + 0.3194239318386959, + null, + null, + 0.3326656731220206, + 0.3624068821376428, + 0.30305290349155806, + 0.30598153714416243, + null, + null, + 0.3039444764196164, + null, + null, + 0.3308459500461124, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3256342859674891, + 0.3179974750036098, + 0.3382021088721751, + 0.29940545277985503, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.43562683434095123, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.22331003599063046, + 0.3395591401496472, + 0.30662200301024156, + 0.4678220526309985, + 0.3985164173954446, + 0.2694922657319905, + 0.377576618194986, + 0.4339913544596826, + 0.3639143364211745, + 0.38181049493833114, + 0.40345493265639565, + 0.3906350477206884, + 0.2846019426041724, + 0.44327556826241843, + 0.3806324178318917, + 0.3836401964473994, + 0.4204742981939138, + 0.3580455240825296, + 0.48398736056068137, + 0.3741342805210052, + 0.3793263734665369, + 0.3451293385096214, + 0.3191763625872856, + 0.5024452077435738, + 0.3535544851126049, + 0.3877998430176644, + 0.2922633213617658, + 0.37959841047519655, + 0.38857734026873664, + 0.40159403415837314, + 0.39872055784267274, + 0.41524039619820496, + 0.49709012859722057, + 0.2877307839503172, + 0.6665672624188185, + 0.2816596650307794, + 0.22572239389634274, + 0.17517265097122411, + 0.33625214912768275, + 0.22753687539585427, + 0.2765811801934939, + 0.4694168947013349, + 0.30776603966643945, + 0.4090263850460249, + 0.299058000188185, + 0.3993529773316539, + 0.30543245506968153, + 0.2175432630207564, + 0.4730201813209411, + 0.44295756817309495, + 0.39345692163092816, + 0.368284918426738, + 0.3825374765982172, + 0.38909504524619976, + 0.3468307470830901, + 0.3994655706492355, + 0.32342519850795276, + 0.44811597400573866, + 0.440359472949732, + 0.45434651171600254, + 0.29528079364917675, + 0.2084289368976982, + 0.40209687231795677, + 0.4266436594039815, + 0.2799766337021323, + 0.35284117315442187, + 0.34823404610409636, + 0.4747895511874648, + 0.5007393094495832, + 0.43312765984066026, + 0.3712073583160347, + 0.7133354111097895, + 0.3672998416218434, + 0.37587601701231443, + 0.37601590729705636, + 0.43697738356620913, + 0.3038534768112966, + 0.6461154793061739, + 0.4181142591009891, + 0.426666320227737, + 0.40204117493027386, + 0.4274411769003879, + 0.49449565215299157, + 0.4501191033108654, + 0.35666658052128525, + 0.3136560720644164, + 0.46265732054264325, + 0.43436872440814045, + 0.808088531392567, + 0.4785920323294524, + 0.38823458406490285, + 0.35772202787456897, + 0.4324978025668693, + 0.3737807557647357, + 0.40897303673527396, + 0.5448028395613659, + 0.3495606127353719, + 0.36261253848983777, + 0.37896580623063497, + 0.3238268285322201, + 0.25863286464534513, + 0.3826737892839805, + 0.3371195510674108, + 0.32831626014681137, + 0.3742931203824685, + 0.2316069203908419, + 0.2908563187858141, + 0.3035609948880219, + 0.3826036674859208, + 0.37035846908457093, + 0.24188529747462073, + 0.3259534351806529, + 0.30711616016021315, + 0.3578802837341215, + 0.37015777071912814, + 0.4443602267772798, + 0.3830395358325097, + 0.38464885880353095, + 0.35438589617408356, + 0.4029154857826849, + 0.349411016120545, + 0.4053728984779128, + 0.3828161749148153, + 0.294557513726767, + 0.36617566971373783, + 0.3189951531086841, + 0.33325316329677307, + 0.3176545347710547, + 0.3432737545137622, + 0.5359858204314082, + 0.38699204574837354, + 0.3138361906119387, + 0.3994834764217675, + 0.6063619200233723, + 0.30757472790090246, + 0.3718743922416576, + 0.35968144685838616, + 0.3076797080702031, + 0.3307452783571377, + 0.31103446521942946, + 0.33212933364751646, + 0.44008906105874535, + 0.5332797416844667, + 0.3301219892948613, + 0.398877332469951, + 0.47696894155848996, + 0.4293334863779831, + 0.27382286168202596, + 0.4156962741412979, + 0.3992349989504333, + 0.40139903070464367, + 0.3390465394732935, + 0.3985321493781538, + 0.4647544412527371, + 0.4046726869744818, + 0.4012468966832311, + 0.2941844385730999, + 0.3352616585039463, + 0.33866513846932933, + 0.41916872466813865, + 0.38426734029294457, + 0.35595490719649014, + 0.38602510938403134, + 0.3867362758112844, + 0.38324765296054736, + 0.36481118305213833, + 0.34901890563076376, + 0.45920013646205443, + 0.28167679643982946, + 0.3683740365272693, + 0.3606070461926246, + 0.32060498266604753, + 0.4850681536608023, + 0.33003470552676956, + 0.28605203242856136, + 0.4043552815906593, + 0.2719727540464525, + 0.40578935748199096, + 0.3911489916428471, + 0.31906922084050987, + 0.38285956285135153, + 0.38087825528820746, + 0.31662497346868623, + 0.3884789204517407, + 0.3227434875996237, + 0.37765051838056235, + 0.4197934310560855, + 0.28694931634425436, + 0.31864035573316013, + 0.3908231866426377, + 0.2981960827300851, + 0.3431323764248672, + 0.42497100753006467, + 0.428489821730936, + 0.3231646686813536, + 0.32496197074317573, + 0.3185777253484799, + 0.31806172106677727, + 0.2670441371165932, + 0.45131761323211006, + 0.34635595598017516, + 0.3328065545281244, + 0.3550701833344913, + 0.3180564275625172, + 0.4408014669109843, + 0.3470174644887682, + 0.4096697916612542, + 0.3647799900218105, + 0.30665630648526665, + 0.48662301306806793, + 0.3238065958507511, + 0.4094735818376575, + 0.3301138364261439, + 0.37865700020201165, + 0.8606886850844164, + 0.41737993885778424, + 0.38795334492190514, + 0.3891203933541704, + 0.4521710761284527, + 0.5398814673020922, + 0.6080637906216889, + 0.6228040576218232, + 0.41070004535067867, + 0.3866362392736954, + 0.466994034940327, + 0.7305907079650226, + 0.3205951711981192, + 0.3477871404283578, + 0.5389183327912149, + 0.515081935718093, + 0.5487115798804602, + 0.4401595103612621, + 0.31657768870815933, + 0.37980018011304395, + 0.5964561113829661, + 0.43308504135724846, + 0.3368230005272978, + 0.6123080851044957, + 0.5454248889624527, + 0.4494265001240942, + 0.4829792448282677, + 0.7077058751187456, + 0.4629052585399501, + 0.5047825641637934, + 0.4461284585994312, + 0.7621877051914087, + 0.3208642619767206, + 0.6396597586551006, + 0.39265340628390477, + 0.5464860804354068, + 0.42666056573716965, + 0.38842060604261075, + 0.6999707187499382, + 0.5044152296169185, + 0.30596904365540234, + 0.37499280818862957, + 0.6424457918277104, + 0.3600393048272968, + 0.4455257234095596, + 0.46413453161560236, + 0.35610416917222615, + 0.4607348378131936, + 0.47672848617906977, + 0.4121698067054692, + 0.3930433936722329, + 0.46701257651869993, + 0.35081001166744813, + 0.3222103957375461, + 0.6280633478677262, + 0.36834275534634364, + 0.4501423131491488, + 0.5645320452901245, + 0.35400329660143176, + 0.6590273609964626, + 0.3418211471762543, + 0.34208780519764115, + 0.5567889810655622, + 0.47832712145728745, + 0.5829819766739756, + 0.7294922833993741, + 0.662947593259076, + 0.40462789002812766, + 0.46601083754083067, + 0.5453923575890548, + 0.3113983657839716, + 0.33246267072663843, + 0.56241097979771, + 0.5254944258530075, + 0.7254752287314163, + 0.4598373491583347, + 0.35843072307575496, + 0.46010807730539727, + 0.3731417782746428, + 0.4739293560541001, + 0.7015395778222084, + 0.6234472184571562, + 0.4892524882491876, + 0.5257518081173328, + 0.4953283601239038, + 0.514500377883088, + 0.7370643380775542, + 0.2976166773434778, + 0.43815693821100554, + 0.29080021083014485, + 0.5972692238144117, + 0.34227216740508093, + 0.7360826396774605, + 0.41039707955001525, + 0.4179604082353492, + 0.6917316924955693, + 0.5162185602291558, + 0.32421797533641805, + 0.5619438118823875, + 0.40158615233655365, + 0.5150965286564594, + 0.5012396135671805, + 0.33245823218624204, + 0.4941315504228336, + 0.3896521174583142, + 0.45845760819295545, + 0.3696306887151691, + 0.29555886028621847, + 0.4020347808085674, + 0.5164752800597694, + 0.5770548124790313, + 0.6597681888787584, + 0.3988888635814351, + 0.2979340982459002, + 0.4589221532967686, + 0.4622458849498789, + 0.42464426032219355, + 0.4847511420080725, + 0.47080746806979706, + 0.3703562979604819, + 0.32558911353376974, + 0.5020563779169956, + 0.7657431467043804, + 0.5294925624537287, + 0.6005484787827983, + 0.5892448712630372, + 0.7702087361365726, + 0.5656542543571638, + 0.32417371852799004, + 0.58224584540292, + 0.37803103533653365, + 0.5168879905848571, + 0.45211910603670397, + 0.3937636282800192, + 0.4926789133284808, + 0.7809376735081234, + 0.22616389339527282, + 0.4595905189743278, + 0.47146073289833307, + 0.6219178997996748, + 0.37872626377972124, + 0.2778900928062071, + 0.4197605609179524, + 0.721373891224438, + 0.38398141045770195, + 0.4393524177238579, + 0.5057911932045563, + 0.4440705436652087, + 0.40453531989690084, + 0.4959503331246935, + 0.6121993248717236, + 0.46002310243520106, + 0.39457943661420736, + 0.5449401780572999, + 0.4545427835948169, + 0.567303020361047, + 0.5423520569161125, + 0.5290534106953266, + 0.547494160982632, + 0.30926184613728847, + 0.30237627682464074, + 0.5200817166208199, + 0.30575933771531544, + 0.6929033011113322, + 0.3019582504777206, + 0.5601449665692134, + 0.5642595727947821, + 0.6845000772601535, + 0.29783296031327433, + 0.3436460566737829, + 0.33555057416633316, + 0.34314683964645226, + 0.29467697726236647, + 0.2838012912101802, + 0.6127644576175492, + 0.35333684275660765, + 0.6810855187867659, + 0.5773576884888677, + 0.4391572445108092, + 0.5688243118902694, + 0.3814515830867345, + 0.5440818654233985, + 0.531600975976775, + 0.4637907541746405, + 0.3845331348436472, + 0.45988839222307376, + 0.4777856618477359, + 0.520375940625369, + 0.39141301889277896, + 0.32949114701287724, + 0.6172720236459315, + 0.47089482127145715, + 0.65625961524441, + 0.6331746435109219, + 0.5042957673870477, + 0.30881026750820495, + 0.3358378843935976, + 0.3520860993826466, + 0.44617230399263363, + 0.7461750626798472, + 0.4686502229615195, + 0.3721883405163216, + 0.7410496692738765, + 0.433649058515121, + 0.623691245583243, + 0.527531646935405, + 0.40940685661560555, + 0.497777079678254, + 0.452385256988061, + 0.45087210534260225, + 0.5771802555747595, + 0.531629847574302, + 0.5216273582193759, + 0.43251161709512204, + 0.458255417017982, + 0.4618640588821335, + 0.4817640824270942, + 0.7677667626765837, + 0.48509947441884, + 0.46687301178550505, + 0.43611867474074845, + 0.2781536035321045, + 0.663259702898497, + 0.3982184030225226, + 0.3212602668813447, + 0.5116067213272514, + 0.426147995487826, + 0.33108431263616, + 0.48069396801833747, + 0.4259838706608684, + 0.7777671455443211, + 0.3129152570049126, + 0.3845647734778447, + 0.32474341440815757, + 0.42663492512298223, + 0.39128210413708264, + 0.2966882419908027, + 0.3974107284802778, + 0.270177279794066, + 0.41437782050904215, + 0.7393163268955082, + 0.30417486846369907, + 0.4002135119444603, + 0.49164165894826806, + 0.4117751398260271, + 0.3809773815323981, + 0.47195101987988913, + 0.35135692788357176, + 0.518297665523295, + 0.519432903557322, + 0.21648083757329553, + 0.6261464220743058, + 0.6837662906322052, + 0.49519852002748765, + 0.47380784699494277, + 0.5194519089920393, + 0.38311385786231744, + 0.37230672753798744, + 0.45288132400082964, + 0.7791379839196328, + 0.36410999666745364, + 0.5988362087286039, + 0.3375557239181163, + 0.45248088482492305, + 0.47160035193726024, + 0.4836639375505423, + 0.3791749670913694, + 0.3172318062884495, + 0.3141978170336145, + 0.30913699770718395, + 0.5021298495658991, + 0.4380725919147088, + 0.28634266405192665, + 0.3992087728128481, + 0.4465456502289053, + 0.28999155828671613, + 0.5513515394773877, + 0.4800371742977837, + 0.3899823998868643, + 0.4489908617258422, + 0.2954373862187052, + 0.41908094194321344, + 0.48883137073674743, + 0.5749415635413496, + 0.4313395241807911, + 0.7773737286669932, + 0.32067753974924357, + 0.5524203104188596, + 0.7350508174529528, + 0.6297514321743168, + 0.2558507832386817, + 0.48861830336771783, + 0.41531620703012867, + 0.3337625813332702, + 0.35490634895918094, + 0.3606752953673764, + 0.4590260312456899, + 0.49389873361754855, + 0.3408314767568274, + 0.606465772922958, + 0.7644929010573585, + 0.4313854205004317, + 0.4184696623787333, + 0.43214131551003315, + 0.48638757505502883, + 0.5071747442970744, + 0.4903563212775756, + 0.4824135346253375, + 0.31556643130458883, + 0.6484104166059688, + 0.47209511076396143, + 0.55028474111184, + 0.7090273003324505, + 0.3075408087416877, + 0.5894412209028733, + 0.3846464219960449, + 0.44813588008244976, + 0.3430459013826973, + 0.5129023044123953, + 0.46016126817497377, + 0.31926636868423136, + 0.41604171197613243, + 0.40960035899829106, + 0.537831461662178, + 0.5856916284018143, + 0.4411785712246451, + 0.4565275656336522, + 0.40566427257460974, + 0.7378474491172615, + 0.44275247288828085, + 0.5135363193099706, + 0.4568468992828316, + 0.4628050659565664, + 0.6055117965155824, + 0.29967609950415897, + 0.5621181324049073, + 0.37247381053695644, + 0.5316458411794, + 0.41385117380371583, + 0.33138318370837866, + 0.33230960193293113, + 0.46337211748341733, + 0.3218372652260802, + 0.5026102226019089, + 0.7304711988052972, + 0.7246595681955201, + 0.3788939607967538, + 0.3293706009619925, + 0.6994218718368574, + 0.44202716400896436, + 0.3767137536668939, + 0.5065212111802881, + 0.35326335049981183, + 0.5923919732613423, + 0.43247466573775806, + 0.5600221051720106, + 0.5849816920425036, + 0.4317071490284017, + 0.5071275276239526, + 0.6536078853383165, + 0.49922691086794024, + 0.35200215355420883, + 0.4038987731902405, + 0.2993681832374027, + 0.5326968970378924, + 0.3393751051605424, + 0.5116730592844176, + 0.4905098343399648, + 0.3534429667669461, + 0.7173167970099692, + 0.31592666292045196, + 0.7222426903943765, + 0.43131822183108715, + 0.5365392205040014, + 0.7396448907056041, + 0.4736573575595518, + 0.5477666685036625, + 0.3784126424150969, + 0.7027799043495663, + 0.407787412724571, + 0.31423347435502347, + 0.5551335194006164, + 0.512454455886931, + 0.6365374740131904, + 0.5436877071324735, + 0.3849554798671106, + 0.48039755577693155, + 0.2752661036837499, + 0.4438698003429662, + 0.5610113730338104, + 0.5040748989957455, + 0.3455611269446086, + 0.5341896431221006, + 0.5768280539461635, + 0.5637697959944457, + 0.4280346852236538, + 0.33987959384629135, + 0.5019642456019072, + 0.4605286033597356, + 0.5527680219046573, + 0.4819400052982934, + 0.3327583297075912, + 0.40095124306470836, + 0.49883675283425055, + 0.39595070477955213, + 0.7309776356851023, + 0.5083833988982625, + 0.4420166351203321, + 0.48932748695721673, + 0.3865484320073143, + 0.4263192241787834, + 0.5367940151988635, + 0.5443065281923428, + 0.7094918495355687, + 0.5307781914539003, + 0.5718018439633406, + 0.5221129158286755, + 0.4748732363164554, + 0.5701186027942452, + 0.3777279647613379, + 0.6282325883210288, + 0.5931619074662708, + 0.72353773995448, + 0.511168443442904, + 0.5415485450801022, + 0.5375874251803516, + 0.40015365616581405, + 0.5326282096077909, + 0.33090300171857195, + 0.6752745538544154, + 0.4745362313532928, + 0.7355280583373852, + 0.7584999263974478, + 0.46454085869781936, + 0.44069000562027855, + 0.4921620110168551, + 0.34020713784959394, + 0.4921870312387604, + 0.7582619989987169, + 0.48290094819140417, + 0.500412363711624, + 0.5837626551929738, + 0.4935343318840699, + 0.34014001307144076, + 0.6057761149901337, + 0.4222044967695073, + 0.3859109542437424, + 0.733004097408161, + 0.40554611817609776, + 0.43305756328735956, + 0.39000997121965725, + 0.48111101201306156, + 0.5576135457940594, + 0.617040285826947, + 0.6784004403576035, + 0.6223689451016104, + 0.5860825174139697, + 0.45380068189079364, + 0.6016218754216995, + 0.5505976726589781, + 0.7809097227988496, + 0.7414049964809024, + 0.34245115579541624, + 0.3660448075416062, + 0.3223087099982162, + 0.5151396419001257, + 0.41369010969359954, + 0.7880157805407817, + 0.37338475596037185, + 0.416616513999295, + 0.5601413637990448, + 0.480298219808855, + 0.5596465117570438, + 0.6216547441694665, + 0.4674803983128785, + 0.6178782209719026, + 0.37749409259257494, + 0.6811876273621675, + 0.7806237637480062, + 0.763515742252995, + 0.576719881129052, + 0.4262428144340435, + 0.5449619909501885, + 0.741553401735772, + 0.6890408704510649, + 0.623489697479835, + 0.4812556107224522, + 0.43883002316048203, + 0.4903960437117419, + 0.7358679728693545, + 0.4141833859554066, + 0.42429557881387914, + 0.30697585782875275, + 0.5980891965133359, + 0.2962009181493972, + 0.4056265666027942, + 0.6629262435611264, + 0.538124155370153, + 0.6828770301324774, + 0.5425049912942057, + 0.5118498149058074, + 0.6671892098837823, + 0.6335258129197983, + 0.5625948412992087, + 0.38065482498433745, + 0.6447604184852752, + 0.5191243452075902, + 0.4010191784940827, + 0.3464774693098417, + 0.4784463378421309, + 0.6376772147383737, + 0.5030973202649857, + 0.6107509711032378, + 0.5739201732548184, + 0.4178089030665672, + 0.40438223678583557, + 0.6871924664685535, + 0.4927278578283523, + 0.5947099423822944, + 0.7244137343779138, + 0.41350277642825045, + 0.36186168778015493, + 0.35767424696865663, + 0.3657489134149264, + 0.713005710122731, + 0.73661401256134, + 0.5644599040932388, + 0.4203237081801654, + 0.4585494992582472, + 0.38723914092098094, + 0.7818244812529016, + 0.3458617505581737, + 0.48090014788419966, + 0.3344890574543124, + 0.5003960209361931, + 0.36583685488990014, + 0.5619569869678956, + 0.48038074785403645, + 0.3149932996745291, + 0.6963813811526661, + 0.5100775512271222, + 0.420391218483512, + 0.33989339780476946, + 0.3586518477075137, + 0.4064921000882136, + 0.4222993738581071, + 0.37283934407738667, + 0.3679174427488767, + 0.4051231655017693, + 0.3843045832605413, + 0.4129803654985184, + 0.335493795423157, + 0.44461532594201486, + 0.21188416982012653, + 0.2821212467145472, + 0.335049459573231, + 0.39201485205530295, + 0.4462811376086182, + 0.4314437217933223, + 0.33978645822046916, + 0.3428114516818839, + 0.3346393496132945, + 0.6721708574199874, + 0.3378175494282749, + 0.31781897180638014, + 0.7884232288538936, + 0.36958370832371007, + 0.4221909820942315, + 0.35629799028937686, + 0.45257092207334715, + 0.28914108842994185, + 0.4072344958591592, + 0.36092937540188647, + 0.48857195316535185, + 0.269286908538406, + 0.41708784817494, + 0.4026662233887934, + 0.509482557343343, + 0.3937625456881589, + 0.4744582290924871, + 0.3105581493512856, + 0.4592532291915785, + 0.41338217039343333, + 0.3610970008403252, + 0.4728552746256681, + 0.37290145356552534, + 0.4770738552850097, + 0.38781713601946827, + 0.37230581336494234, + 0.8080455636454401, + 0.31232359093346246, + 0.43890168844847266, + 0.3680250916659896, + 0.41297462907664545, + 0.32758511245545496, + 0.3809039148295364, + 0.46184326050648644, + 0.5528605785845256, + 0.34768054182017083, + 0.3858405239826337, + 0.295597384915351, + 0.4143728354428148, + 0.42883672282856455, + 0.43036641321299496, + 0.4598087563197314, + 0.37180750206776275, + 0.3502082291650177, + 0.4704535561044929, + 0.4067888037865963, + 0.45371065119005266, + 0.3043698053724708, + 0.3260543371564298, + 0.48406831183151455, + 0.45965169460663213, + 0.41174344359428244, + 0.3745809616394242, + 0.3754903168281677, + 0.8373765068838567, + 0.33310811406738544, + 0.3385530760169898, + 0.3593668711487589, + 0.3179701802264148, + 0.3524839054317967, + 0.3557424532332586, + 0.30688931373667944, + 0.37131258163657, + 0.3536627971373499, + 0.43224790980288996, + 0.3754891330339303, + 0.4328399862143765, + 0.47268326808371747, + 0.39712226026421416, + 0.3812323672843461, + 0.5563315429316921, + 0.3842046157445325, + 0.4117618843753924, + 0.4981489097151532, + 0.44876887366723006, + 0.3520115668696075, + 0.293592022827938, + 0.3679501478664099, + 0.4836095052354127, + 0.46923653468130977, + 0.31747911622752617, + 0.3604319436805096, + 0.38073823399422624, + 0.4536942696425863, + 0.3342160673759598, + 0.34541032900154284, + 0.48984450810751023, + 0.5275699678198359, + 0.304131300465127, + 0.42608505734812857, + 0.31908298009387004, + 0.33493436189189196, + 0.3352901763947207, + 0.7886447566025779, + 0.420735755863665, + 0.4399118252699436, + 0.30949417653990235, + 0.37397622344835335, + 0.3715030959683378, + 0.6997233343563912, + 0.4200090527317764, + 0.3852508090153669, + 0.44760838779362344, + 0.4173597517457224, + 0.41268892024741294, + 0.44852423897578125, + 0.4866724089321276, + 0.27843944235188833, + 0.3407686862021885, + 0.44272083163929404, + 0.4175062592653316, + 0.4110379643200318, + 0.4069525830604227, + 0.366351802743941, + 0.3131043373526154, + 0.3455808470052427, + 0.4041600685534896, + 0.3751405953555927, + 0.6072351083392735, + 0.3485963110241786, + 0.3672024095690023, + 0.30619324290395195, + 0.5536270614337955, + 0.38671572289899814, + 0.3190083406328866, + 0.3798314529878347, + 0.3999921229336574, + 0.41106318084207444, + 0.3245111279272731, + 0.5449352972061483, + 0.4580126504657411, + 0.3108506993913761, + 0.41248508053883154, + 0.44236610810488536, + 0.421935831444417, + 0.37215766755093843, + 0.4660342022021768, + 0.3327424034893091, + 0.6540934875296026, + 0.3874834412640951, + 0.457478165251527, + 0.4557917126225714, + 0.3907336760174051, + 0.41181163968331563, + 0.385345520285648, + 0.2980038268109727, + 0.3899905459590069, + 0.5929367793107205, + 0.36998425833979554, + 0.321556859556741, + 0.5025747274926531, + 0.4214071731631406, + 0.4150411318875663, + 0.6149388410389934, + 0.3811002693249397, + 0.4417934020816852, + 0.36998942426693626, + 0.38320987309435656, + 0.4775714707388495, + 0.35868369138756695, + 0.4670877773226906, + 0.4113367624154988, + 0.44555431710809873, + 0.31563746773644713, + 0.42875394588535076, + 0.37054925840784364, + 0.3218027435700553, + 0.31761174649814244, + 0.33797918620721895, + 0.4300181671793528, + 0.5320589465461495, + 0.44812944481387124, + 0.45301772077062546, + 0.463003014823761, + 0.3405915859416727, + 0.3666568438030427, + 0.47766683487675254, + 0.36444306010057614, + 0.39804157521140454, + 0.4774038087975074, + 0.38281577414694734, + 0.42827086253585206, + 0.5015687973089314, + 0.402841401088828, + 0.3653192292492837, + 0.5075519292173001, + 0.3745792594370496, + 0.43786218027575846, + 0.31641341472968737, + 0.4949304712580017, + 0.5996546909161979, + 0.7905287082672429, + 0.4245825907511626, + 0.3839891056479695, + 0.4259823482595118, + 0.36456076180154456, + 0.35523644950320205, + 0.36285965661634373, + 0.4356354107439823, + 0.4026056568644564, + 0.365011627342655, + 0.4461651509440224, + 0.4255095785788457, + 0.4216508211700602, + 0.328659656591265, + 0.3186545167435765, + 0.4293501485387886, + 0.4113623117372862, + 0.38199089325855257, + 0.39394818897547135, + 0.393927885364331, + 0.3621222944793437, + 0.27209422822428114, + 0.4239046745299145, + 0.4391052744742099, + 0.35212736065841016, + 0.3969072570560346, + 0.3042874675427157, + 0.38742522749812097, + 0.25772053426699343, + 0.5223646912113213, + 0.35513181774816543, + 0.32264289809184804, + 0.4215454107715246, + 0.38523089118957665, + 0.3297549611284216, + 0.5965470646890305, + 0.4943530021244993, + 0.3017258630987595, + 0.46706422439014245, + 0.4120504787860754, + 0.41153229626278837, + 0.3844817577386932, + 0.28301373462528157, + 0.3643418834818193, + 0.4965778950254113, + 0.3241140437622518, + 0.4257484081894881, + 0.3154364197530768, + 0.3594602743149624, + 0.36249459472413975, + 0.3334801254051402, + 0.41413469103019457, + 0.42438295196253645, + 0.36876776126690136, + 0.3432956751182862, + 0.4091087390474632, + 0.3188794545614073, + 0.39063817521328714, + 0.3840687417655546, + 0.4893370846826104, + 0.5262634578556646, + 0.42599076087416904, + 0.40880293803386175, + 0.4222282306688773, + 0.43522234939744425, + 0.4387726346863055, + 0.5269298901322131, + 0.4009588905380877, + 0.31010953396242347, + 0.4996513230536671, + 0.3874690981640041, + 0.4505263534629163, + 0.44362240877222076, + 0.38227229081791514, + 0.4907240890028803, + 0.3135284844423742, + 0.3679870161205268, + 0.3090134264777769, + 0.3078409827749269, + 0.35015271801681336, + 0.39333742974797203, + 0.3311963026570927, + 0.4091692259023697, + 0.5283021447836622, + 0.4159497603639251, + 0.49374859578660846, + 0.7780793309032497, + 0.5542314878910382, + 0.3480581498043327, + 0.4147310212315643, + 0.44911761395680383, + 0.42497403482364693, + 0.3048335851783062, + 0.5772720988017427, + 0.46890468230986276, + 0.39022495111392075, + 0.45773862110172575, + 0.3786119422115637, + 0.4838551428477536, + 0.5047147211986455, + 0.391992038818749, + 0.5300560965151687, + 0.48293710999281736, + 0.3708495096128626, + 0.449762232422241, + 0.4975992452018941, + 0.3574982761642107, + 0.38081568600301235, + 0.3043168335031824, + 0.5650456707948569, + 0.5077567137492859, + 0.8204409479800324, + 0.5279569608341016, + 0.43055135308325043, + 0.2795464224569301, + 0.4697573603052431, + 0.4310423995672183, + 0.6031613730178058, + 0.3412967581162659, + 0.5346524044628068, + 0.32740700589828337, + 0.3527600196184177, + 0.3247294777545514, + 0.3130253622268082, + 0.4171476996447865, + 0.3044331849873531, + 0.42314217141205257, + 0.4461530629868149, + 0.3853916616037636, + 0.32030151049796607, + 0.32028849192795933, + 0.41217560176283585, + 0.3065344931123071, + 0.5228419770768635, + 0.42451957534960616, + 0.4897304693458345, + 0.4172986685873052, + 0.3538651085259514, + 0.5548138901362114, + 0.4701935786500853, + 0.35581372480058865, + 0.5894730597740228, + 0.29958791728406625, + 0.4393561005152143, + 0.35587014151543767, + 0.33300958644672585, + 0.3168057124416852, + 0.3650453599625866, + 0.46905485195765856, + 0.33404335422329345, + 0.6894609086875456, + 0.44586888024968924, + 0.3211652586888742, + 0.37727172361483513, + 0.33226116211570395, + 0.5681499869184641, + 0.28932528379269223, + 0.3636262786807896, + 0.3785270444865995, + 0.4205014095385593, + 0.32151344503856993, + 0.4002187287420918, + 0.3073884993695716, + 0.43348350857480283, + 0.7660299622493089, + 0.44627704031148807, + 0.37494241353356955, + 0.3899937309164478, + 0.38490921231268627, + 0.2961106086364629, + 0.3655935012519307, + 0.443047659611577, + 0.7497284591209734, + 0.5458336426505522, + 0.4486090940945247, + 0.38496839093678353, + 0.36177096405587866, + 0.40472662372024715, + 0.29201663870467015, + 0.42859413786189343, + 0.5023640116483262, + 0.3262879922222776, + 0.38623301805926696, + 0.3770529325078513, + 0.46275812359957746, + 0.3410611651828903, + 0.5415844757880602, + 0.4186832016656557, + 0.30628893940836394, + 0.4145846755140551, + 0.42435290079722576, + 0.5274892401756682, + 0.5321926962890476, + 0.2395334559039733, + 0.7231317589291899, + 0.7117642378243322, + 0.20404862394759474, + 0.29647310995631265, + 0.37946245416302615, + 0.4437446561315064, + 0.25651654932990414, + 0.37195081575444827, + 0.4572361167441996, + 0.469556635601744, + 0.457845912788147, + 0.3630804401660541, + 0.35663636381954195, + 0.34779359288797307, + 0.31286000428548605, + 0.42889578271327855, + 0.37174775025368795, + 0.3744266116817794, + 0.4984010870725768, + 0.40847158307794407, + 0.5232034782568419, + 0.4405508779466523, + 0.6046265871134088, + 0.433682413801217, + 0.4230406206710458, + 0.4887225530166219, + 0.44779981057891405, + 0.32790728649711576, + 0.6245449939656648, + 0.6038703754354798, + 0.4347771047847512, + 0.36400295152366446, + 0.3762204239105887, + 0.5272611413738173, + 0.3591961345811855, + 0.431125452119383, + 0.4137263507810754, + 0.4716829196639153, + 0.4341815439697165, + 0.3794826194278713, + 0.3763043835767482, + 0.536906142146383, + 0.4850953119422774, + 0.4608512364546785, + 0.4804728715316805, + 0.38319928633886835, + 0.24591805837098274, + 0.25593783710496004, + 0.3188395152385936, + 0.5191953762034839, + 0.4570761908337088, + 0.2735758382878668, + 0.4420329652452995, + 0.3475768401975452, + 0.33551070573458, + 0.36158521539932476, + 0.38885889309789023, + 0.3502667524886737, + 0.4377619258480159, + 0.6240422262244347, + 0.3300507543810492, + 0.42599459981362614, + 0.5312477409692264, + 0.35181858769692287, + 0.44072648024038286, + 0.5925526142330901, + 0.45271017759690646, + 0.4478116166366312, + 0.4346599679303117, + 0.40675749783110965, + 0.2666884218974284, + 0.3594400849129541, + 0.48566163803374457, + 0.4945069705469527, + 0.4732008038875598, + 0.3165088537386794, + 0.4595610166461654, + 0.4609637883256069, + 0.505807393136704, + 0.4027630848243445, + 0.46556780865392194, + 0.3841239700152076, + 0.46140566719868054, + 0.4435320477699979, + 0.4119189634029288, + 0.30722832613714846, + 0.4523541935632387, + 0.46282305487518344, + 0.2549617667229963, + 0.36298204116087573, + 0.43713124738162656, + 0.3982793334824274, + 0.3905200187574037, + 0.4884736686201511, + 0.4924788679296448, + 0.44184924511426554, + 0.47079535170215514, + 0.349037812332234, + 0.4588548883652495, + 0.4059042974963782, + 0.3843657575913012, + 0.4029396376292509, + 0.4482622434474891, + 0.40190373470292107, + 0.42446969630924136, + 0.41163423501414187, + 0.7220730047330919, + 0.5449179649026471, + 0.5049466328550373, + 0.5340273190454351, + 0.48223532807551545, + 0.437019801103046, + 0.5391852805182439, + 0.35672077953810105, + 0.39494454155649905, + 0.4723566545068504, + 0.33602390698650103, + 0.4440452637265966, + 0.4988958307956294, + 0.4421355719002137, + 0.4473034651984113, + 0.46253699614917426, + 0.3261949552056901, + 0.5354386838146796, + 0.49229368639308735, + 0.37881025381744987, + 0.42989280374148325, + 0.3822193714001733, + 0.35111912240697396, + 0.45287744481513814, + 0.44258598826606144, + 0.42043330335597484, + 0.45823082209922195, + 0.3895203199757849, + 0.4850456305839577, + 0.3669446403926367, + 0.39175681007274826, + 0.4350701236924617, + 0.45101911517725335, + 0.4399423907586192, + 0.4719336555296152, + 0.4040402071790746, + 0.34703908915926596, + 0.3543813085641314, + 0.4365653712865259, + 0.4095528856431672, + 0.44522354973949396, + 0.4749983824051022, + 0.42651388266229684, + 0.41234049252028676, + 0.3556266828068283, + 0.4051325279472369, + 0.5288021913503691, + 0.3983410299229377, + 0.20399853243673224, + 0.35406116450197944, + 0.41615698353515335, + 0.3065595019261505, + 0.29588486473459097, + 0.4533908328062266, + 0.45277600825581316, + 0.2859793536442211, + 0.31842103442586145, + 0.36409356799603354, + 0.3524164703356609, + 0.5225158005984567, + 0.296217563972152, + 0.29824693784355805, + 0.31957525772176054, + 0.3290755010223471, + 0.24729786072004015, + 0.3094355353483059, + 0.2927620625293884, + 0.3188848935442596, + 0.28658987311049455, + 0.4083402839843128, + 0.306963987726699, + 0.3504420444724535, + 0.3551912539426524, + 0.4149164554727489, + 0.30583564529796076, + 0.38963440462265286, + 0.3571106270405874, + 0.3600484004769871, + 0.3613321018598468, + 0.39193860175938067, + 0.36927550564378137, + 0.606558045322974, + 0.32670520489077964, + 0.38541461666662685, + 0.5495532989213655, + 0.48170428962534156, + 0.37130307485448083, + 0.46328520593871897, + 0.41740309514040835, + 0.4572798120286937, + 0.4719632733971133, + 0.35206579088961854, + 0.5804767796714373, + 0.4255915574750814, + 0.4266615852451307, + 0.361469454836656, + 0.43219090083965694, + 0.38352533895260543, + 0.4567321139126951, + 0.47219969062113276, + 0.44972063407656526, + 0.4654777514133149, + 0.5732044964012428, + 0.4432764297608058, + 0.4046472404247929, + 0.46985818303034416, + 0.5344432800184621, + 0.5242382471608549, + 0.4408703115148106, + 0.5651954728623687, + 0.5559018134572292, + 0.38742557789115667, + 0.3847405749413783, + 0.4344696495347941, + 0.34749794227297265, + 0.6363015284337586, + 0.4852374535826876, + 0.42122960667521747, + 0.35266587789824705, + 0.4628192995111642, + 0.4020257296717494, + 0.5185819343531167, + 0.43693595807998115, + 0.41367129006419384, + 0.5100286156118369, + 0.3095615611164899, + 0.41519772661880877, + 0.5352113905171504, + 0.3380822087601393, + 0.41575796844075924, + 0.3225735850527809, + 0.36196993787503245, + 0.38395597669035714, + 0.3927391714777276, + 0.46225770349755685, + 0.3381463487268555, + 0.32803506425925083, + 0.42451861047618644, + 0.39339891445705016, + 0.3737879920233084, + 0.44708823718943186, + 0.38881183597667424, + 0.4181052705273802, + 0.5045724470482128, + 0.5592597906974823, + 0.4595363982350429, + 0.3579248993870039, + 0.46810862369211925, + 0.4394890931971599, + 0.38868515540849236, + 0.4224869522829167, + 0.3214401445723046, + 0.3268065232503334, + 0.29999317896239736, + 0.3672473114710661, + 0.3165510711083621, + 0.6501091527602427, + 0.4297335094171178, + 0.410956286057799, + 0.2951031527390751, + 0.3123161550957474, + 0.3189365612570286, + 0.3446623795303264, + 0.3113053521537854, + 0.3202201237555861, + 0.3073887083936712, + 0.374608819877006, + 0.3883973872733823, + 0.4213967735157418, + 0.40665264414191116, + 0.3664587833301349, + 0.500090799980345, + 0.49396206931887277, + 0.37460041450513193, + 0.41880402520101023, + 0.42778325626827335, + 0.3864739847828184, + 0.3558055101384052, + 0.5108446545148732, + 0.44413961840293237, + 0.5791534888479993, + 0.4471830474008387, + 0.4405105667755145, + 0.44271001956494593, + 0.44930590643310203, + 0.32613048679738454, + 0.5940188679146657, + 0.41036200253084615, + 0.23012598037264312, + 0.46343618803967374, + 0.5719614324473271, + 0.3635502213879553, + 0.4540533640783699, + 0.750926713531713, + 0.43804159941640963, + 0.5105937774064945, + 0.2860293865396497, + 0.47501344942506213, + 0.2881482813439918, + 0.298440586623002, + 0.317654573099703, + 0.2760338336199167, + 0.4458006429061341, + 0.19966603717492973, + 0.31418648523780973, + 0.45136520148416137, + 0.3967534087894324, + 0.40012711026016307, + 0.3899736709920027, + 0.33208545917904153, + 0.41047102044467093, + 0.5175209766182488, + 0.5358249835917058, + 0.35345726957626306, + 0.4473564064148375, + 0.40783544993102105, + 0.7294634291818332, + 0.3924479748338592, + 0.5572900963465093, + 0.7357330467733332, + 0.7144132959449743, + 0.44164701447948035, + 0.3760647428533516, + 0.31442261316569936, + 0.4494975988130142, + 0.4907331098033215, + 0.49187203513027694, + 0.5700197837526055, + 0.4225839365611666, + 0.5494058841452346, + 0.43242438500846897, + 0.3858867269535689, + 0.6998420463299633, + 0.6809578382606961, + 0.46414193391762804, + 0.4153933277660704, + 0.4651248626004922, + 0.41819275429664066, + 0.4496736312250112, + 0.6415882574717872, + 0.665166327505988, + 0.4226424367053953, + 0.43303189053936036, + 0.43447908783840367, + 0.36370077073337037, + 0.49010402169667927, + 0.33064140180253726, + 0.48399103771504387, + 0.4976390050935858, + 0.34606343683897256, + 0.43603589050037433, + 0.39685418551003576, + 0.33661843030306393, + 0.5024470025957684, + 0.44384056333540844, + 0.3973012829201027, + 0.3667442341024304, + 0.47351685649997666, + 0.4164068430551691, + 0.3365920264624978, + 0.3413147771814358, + 0.2525187354208252, + 0.23882523193616007, + 0.2707649153993319, + 0.43374719258137956, + 0.4638985960950276, + 0.3321173707049459, + 0.3183089399757626, + 0.33312438500204683, + 0.4227871132849279, + 0.4236998456788445, + 0.4933259603446841, + 0.34453790465282685, + 0.41489635948048065, + 0.5096105844308455, + 0.36486402525616807, + 0.5225666948242752, + 0.4679758049837025, + 0.4139821252558154, + 0.3995775319343562, + 0.35227634700048377, + 0.23504738393372823, + 0.28332291263608556, + 0.27303537453858084, + 0.5138246553039183, + 0.4221279703426534, + 0.4166271813570954, + 0.46790981807955573, + 0.49198462461151266, + 0.3159705717742335, + 0.4097835094201798, + 0.3648774017776375, + 0.6631662840072837, + 0.43573664638787374, + 0.45551490357113517, + 0.3619672921803755, + 0.4492009546420307, + 0.4850075360551974, + 0.43228745436359417, + 0.4007915859059247, + 0.4823018573406428, + 0.4458618491568093, + 0.5542342405736996, + 0.41389834744192605, + 0.48043140073409246, + 0.4398219820893064, + 0.42041822190504236, + 0.39596518490829824, + 0.46596311770493015, + 0.4193922856152143, + 0.4845549967074681, + 0.34128112072312355, + 0.3184338301915716, + 0.3201634259956255, + 0.3976299668639872, + 0.44132249740776436, + 0.3620167758036589, + 0.4174467686888569, + 0.46138332386768405, + 0.41433382012712827, + 0.4288859581377576, + 0.42481641733699227, + 0.26817586395421866, + 0.40606344846351194, + 0.2540423424455957, + 0.4191575013889185, + 0.21915015991434425, + 0.6489906548832393, + 0.3194479445105004, + 0.4401200014933439, + 0.5463463680081418, + 0.4634697176534929, + 0.2672269432649498, + 0.4119609688965787, + 0.33015448636529043, + 0.7704856806635255, + 0.27498462157365394, + 0.5052188022845894, + 0.5034696267441392, + 0.5029621406280453, + 0.6277489064485188, + 0.41918640078256997, + 0.5507436019488505, + 0.45174423814183756, + 0.3778712046513979, + 0.3222636670148533, + 0.4486248557835911, + 0.3067107081430887, + 0.7118098371133914 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.4281288798270353, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.46656019131687143, + 0.7583070910576802, + 0.7353928419185474, + 0.8176371821020098, + 0.7379750963990221, + 0.6499281934087131, + 0.6264230020081837, + 0.6602774069629418, + 0.4066219558450325, + 0.697561965104789, + 0.31691672721518677, + 0.4758239202486567, + 0.3364313838820642, + 0.32136009831782725, + 0.30461254244333696, + 0.28922035260032797, + 0.3389896270001883, + 0.5435365053335406, + 0.41058279923730445, + 0.40299616121995047, + 0.3273652822861569, + 0.6695001675786593, + 0.33315367276821756, + 0.5868784386888967, + 0.3639440790689118, + 0.29567611269388905, + 0.3883339132218831, + 0.4352889361296461, + 0.5123563845287294, + 0.4855637310010544, + 0.2807244697662876, + 0.3815161807136012, + 0.3412375091983968, + 0.341622137713671, + 0.4341396434695882, + 0.34844868599584056, + 0.331181455571083, + 0.5143569900546336, + 0.5419558327794293, + 0.6270953427415484, + 0.7886917304851968, + 0.7815610779881241, + 0.7716948161798403, + 0.6567388176782276, + 0.3876208022473146, + 0.39174322925218696, + 0.7796581249261619, + 0.3496470354118719, + 0.3505569922857527, + 0.48605687503405354, + 0.3905064000857546, + 0.7771606521954764, + 0.7948961908346379, + 0.39338850573471773, + 0.3484209676919248, + 0.3599383903740781, + 0.34914048206099035, + 0.40470677859234255, + 0.39686936652072535, + 0.5049082510647491, + 0.4491019550610964, + 0.43250197512595473, + 0.4051627195919818, + 0.4953995285933013, + 0.45442824769305257, + 0.3787460420893833, + 0.35467517257496656, + 0.5336079506896385, + 0.363540774302704, + 0.43611556745216784, + 0.46896770575495966, + 0.43216302455249456, + 0.41186885991472755, + 0.4003238158211731, + 0.4080564693836408, + 0.4185409050845883, + 0.2854755361221134, + 0.2769091121279372, + 0.37021269075130864, + 0.3246231758347565, + 0.4323848605388848, + 0.3464543942773022, + 0.34007274622559197, + 0.3131211063595384, + 0.4046883333003473, + 0.48226418901748425, + 0.4027461853900655, + 0.2979508016759302, + 0.3408962243762627, + 0.35856144789362876, + 0.3400336526234451, + 0.3358889206945912, + 0.49241164015192557, + 0.4988012732632286, + 0.303954340202047, + 0.5584010314262483, + 0.3516457981532169, + 0.5262431355213454, + 0.3202209649708427, + 0.4557979522488516, + 0.4402309649767155, + 0.5088776413349527, + 0.24346557857042783, + 0.39100002065863004, + 0.300997890813308, + 0.3227221809102014, + 0.4565141875670874, + 0.34954671306336565, + 0.32561974319932285, + 0.8099725777481214, + 0.8041297515861252, + 0.8088971443812631, + 0.8249290096727914, + 0.5821354855780281, + 0.6859838918787811, + 0.7067460610842611, + 0.755355426863009, + 0.7059026561003575, + 0.5881252199720551, + 0.6539699956792712, + 0.5767792145022473, + 0.7726408393355203, + 0.6574575733884737, + 0.42188054457031016, + 0.32541889586956824, + 0.28342979673983243, + 0.32361424326478816, + 0.5389727331849697, + 0.32313562264267043, + 0.28039551697813125, + 0.40058018240315213, + 0.37873600030951965, + 0.32965807710758016, + 0.35618444444795494, + 0.3975682494196181, + 0.41673511160387905, + 0.3866572687624166, + 0.32413718592693064, + 0.3609546768807032, + 0.3834483864928094, + 0.33246072842533325, + 0.309229350835868, + 0.30400477185076497, + 0.5923583963763623, + 0.30516777188646654, + 0.5643097775575278, + 0.36055325531907195, + 0.4333306617835748, + 0.3020135362155363, + 0.5559043729499249, + 0.30346324357132026, + 0.49905961352096306, + 0.31877562841565027, + 0.2685873485569941, + 0.35533968010656025, + 0.3186836923445144, + 0.2631393088689305, + 0.18072314355440663, + 0.36905583131500014, + 0.3300832998447794, + 0.47962845495063483, + 0.44494040330423223, + 0.5031270529569533, + 0.4713337818382649, + 0.3402051562287155, + 0.45605676226815584, + 0.4148903903030823, + 0.4356671454042672, + 0.41390322017254777, + 0.42458668959064944, + 0.370451237790057, + 0.2139537206595189, + 0.41820806453083903, + 0.32991145575771075, + 0.25083009676600254, + 0.2518739133122761, + 0.48138307687446485, + 0.4998178873301098, + 0.5642487694575065, + 0.4677225533615942, + 0.28145569438348245, + 0.49419293171457596, + 0.47993501774093544, + 0.5116782164136116, + 0.29572305673021176, + 0.5205974246131514, + 0.43202265539253004, + 0.5131761775006941, + 0.24573644147279908, + 0.38740002837480497, + 0.34197964733951025, + 0.26242640197886, + 0.5025203875069845, + 0.432123629414925, + 0.5462719748322423, + 0.3130768211114686, + 0.5904917765917941, + 0.29578615551960874, + 0.27451468025140097, + 0.3933236352768607, + 0.3009023042118658, + 0.2689045987948718, + 0.2540845470429569, + 0.35800852078561995, + 0.6286841765856283, + 0.26138258611122234, + 0.2500731475144523, + 0.41364709523504334, + 0.3799943937199061, + 0.38581085830450357, + 0.44850346890786796, + 0.3927683006666799, + 0.27387979210393243, + 0.35614236773323343, + 0.3337446544427348, + 0.5189374952275891, + 0.5473295953263118, + 0.5628295802102036, + 0.5495710222354202, + 0.2898290419900689, + 0.3806464140128351, + 0.42438479686935043, + 0.26302929456432816, + 0.3319160739120177, + 0.3343041094323742, + 0.1584601944801178, + 0.46472610732298003, + 0.31480312623482776, + 0.2939467059484362, + 0.5205183942870506, + 0.24686426269871514, + 0.2053993086674685, + 0.4223008795703671, + 0.3807309557518312, + 0.2885317024673626, + 0.605335241955018, + 0.365220146897691, + 0.29278145163745983, + 0.27669989284336616, + 0.4211314066585971, + 0.2985784292873111, + 0.44900455057584626, + 0.2866114197738733, + 0.4009933522455402, + 0.3930500660743039, + 0.37640158124223544, + 0.3337360324784115, + 0.3250590354985444, + 0.36523050086733866, + 0.3386515402604782, + 0.4299446837729385, + 0.29419040275790004, + 0.3754610794875578, + 0.30919947998377634, + 0.33828030455593655, + 0.38342643267703524, + 0.383966736942098, + 0.35249450062881554, + 0.42498740816347247, + 0.3144517200870946, + 0.31079183057557874, + 0.5093976072637617, + 0.2732962210621321, + 0.32213447229034975, + 0.31903650025871966, + 0.25958582313133016, + 0.31626050549985546, + 0.32316454673437767, + 0.26665381983719344, + 0.3037117859406053, + 0.28194885751909193, + 0.3211099046643772, + 0.32844477098289027, + 0.36615480511858267, + 0.3632918366462331, + 0.45976161965132406, + 0.244592493224313, + 0.38925785889505443, + 0.4565629922326056, + 0.2668411858477138, + 0.27120652871379386, + 0.5893608615494336, + 0.27853353022563765, + 0.25439215700950996, + 0.4524659403639548, + 0.43967269171663975, + 0.43932736706516606, + 0.36850145556941427, + 0.4888069418775122, + 0.44106465359662694, + 0.4548449476166496, + 0.37720920803767294, + 0.4474841182772107, + 0.4023087917971252, + 0.39876664867579736, + 0.4210794142316618, + 0.4620787641114399, + 0.43268474426180137, + 0.652891309455475, + 0.3135016338018826, + 0.5705464977710827, + 0.6400536679642316, + 0.6577138223044862, + 0.6611042684861598, + 0.49680737272824543, + 0.5195142867052344, + 0.39753554442729255, + 0.6759956313711911, + 0.42101508544285915, + 0.48566626517258205, + 0.3169543365941837, + 0.49344045380632856, + 0.43127296806142246, + 0.3827636236958485, + 0.440661686282662, + 0.31403304342609656, + 0.4194059089530477, + 0.27356633219090926, + 0.34434621008234007, + 0.3429824781578894, + 0.3370394899289199, + 0.3905184118707049, + 0.27130225188545876, + 0.32794796920376323, + 0.3519303271332479, + 0.3587724382248888, + 0.3232680457920206, + 0.35866628663355293, + 0.56360524186998, + 0.465488164158385, + 0.40504976430677986, + 0.47325329746135925, + 0.4584949396810214, + 0.29722594588980167, + 0.30217947335933776, + 0.43422511725002505, + 0.4023534990956249, + 0.31416876438208285, + 0.42453601304515737, + 0.5202418116742442, + 0.528703510676125, + 0.3861150001775634, + 0.2973681050916786, + 0.458562496485694, + 0.2737199069610605, + 0.31679721101451663, + 0.48508061631481214, + 0.3072390889742379, + 0.8233769455103535, + 0.5053316510021464, + 0.40093334035970274, + 0.38615549359399837, + 0.4424851406568341, + 0.4412535530145277, + 0.3854693639525693, + 0.4939602302299519, + 0.3482349083224647, + 0.5029037314212983, + 0.39262740008724195, + 0.45437373046539353, + 0.46150751640884813, + 0.4367357021381091, + 0.42655833173831137, + 0.34188559848427763, + 0.35998581606324437, + 0.5201333072838731, + 0.5157391978574518, + 0.4047787854685532, + 0.5777832509850428, + 0.4401254340572119, + 0.5420674819627362, + 0.3939415120581259, + 0.4349874948539422, + 0.4202371521712451, + 0.5538192030058263, + 0.5606074842038942, + 0.5274724866529016, + 0.5298053153098576, + 0.5075795436853919, + 0.5069844577964943, + 0.42362195593804486, + 0.3906946535770142, + 0.3634316467205458, + 0.2742267918251928, + 0.3691751673993669, + 0.3186996692566795, + 0.3195686447296471, + 0.352333766979662, + 0.4568179768951529, + 0.3342380652899826, + 0.29866973713291983, + 0.3442217381197905, + 0.54267430853712, + 0.6032196455930489, + 0.521406976907398, + 0.5617992843524288, + 0.3955604910927979, + 0.5628862156913219, + 0.3725670928546805, + 0.6247164339949575, + 0.4638307884435056, + 0.5485538115103832, + 0.4205528472353269, + 0.5328227430477088, + 0.5375693228008153, + 0.46885836606724474, + 0.4437645872016454, + 0.4957111785879889, + 0.4978829145498288, + 0.5077436027979518, + 0.3467548698206091, + 0.37247619481712746, + 0.3150644138191728, + 0.3571699914802288, + 0.4979375500482621, + 0.5652728413885436, + 0.31446341548635864, + 0.579428480583676, + 0.49891448071591865, + 0.47029104412712425, + 0.38477036587935876, + 0.4575492910511891, + 0.4745070461886513, + 0.42013848410697635, + 0.4046062633494409, + 0.8502650532701947, + 0.4300678870901562, + 0.4668549687262351, + 0.4852172604932328, + 0.5025511554124477, + 0.4684475969118222, + 0.4057062903693824, + 0.5868413298830226, + 0.4492811920400439, + 0.38759399818270074, + 0.44761105190369893, + 0.42986211810848235, + 0.4354060631860583, + 0.41210960145559833, + 0.41939260342499823, + 0.4320026442779013, + 0.43418188970956123, + 0.28905040272783994, + 0.4327622907519314, + 0.28999398183309155, + 0.3764557221105333, + 0.23587264562600338, + 0.43436176631065304, + 0.36106717862309384, + 0.2796507084852296, + 0.4945904696727182, + 0.2569406182467826, + 0.26251579216651727, + 0.5440210314769408, + 0.46818802569029006, + 0.4515333243032536, + 0.4627592318054179, + 0.40372778967002987, + 0.40457715101841846, + 0.6013269757691243, + 0.346162677983615, + 0.5444349185603007, + 0.36493897790011554, + 0.2690656058017508, + 0.21707740716610657, + 0.38918169326246155, + 0.4788763449089589, + 0.5341294123787632, + 0.5611811037620178, + 0.44728296059520556, + 0.5064940210803995, + 0.37091798747653737, + 0.48115469965120966, + 0.39906821640618506, + 0.3915085034885978, + 0.3945626724169801, + 0.37403073057862984, + 0.44858610525594583, + 0.3707376784595839, + 0.4082365018888236, + 0.4462982795665941, + 0.4288834967319819, + 0.4566100766250015, + 0.38930570418448485, + 0.38400540637208747, + 0.47951051921768695, + 0.4472448624178452, + 0.5216481755679963, + 0.5006051631549523, + 0.5527666659436384, + 0.5788476870105753, + 0.6653274281692027, + 0.6484990733056737, + 0.4755340620214814, + 0.517940695290087, + 0.4161479381795282, + 0.5181097798740345, + 0.3905652417772892, + 0.46959395251597547, + 0.388758107317548, + 0.5748956250682853, + 0.389127925274209, + 0.5466659955177696, + 0.6519777684955159, + 0.6223354741877364, + 0.5541319079577454, + 0.6391809305801837, + 0.5348622771316381, + 0.45256340196104866, + 0.39557681858450366, + 0.39862093242035024, + 0.29261042893237804, + 0.3597515707028373, + 0.4526559913277071, + 0.5331874583751849, + 0.4220880834262727, + 0.3529455311786388, + 0.2724944580202876, + 0.46110892147749083, + 0.44582721571565204, + 0.39327980924874417, + 0.3474350290501432, + 0.4986196158469982, + 0.37532681307089527, + 0.4266647719769361, + 0.3738114504752857, + 0.45617516232871513, + 0.4526563042382214, + 0.3231968922088813, + 0.37505924314532696, + 0.4421544050546616, + 0.3775104578906056, + 0.37404313807303996, + 0.3862749548096748, + 0.3700260915964491, + 0.33010953242294455, + 0.39305474430025333, + 0.42213764719535146, + 0.39300505507363637, + 0.2678404382506704, + 0.5666162547233874, + 0.3846370622818161, + 0.49829488414404727, + 0.5216974711759478, + 0.3607041840192924, + 0.48774457233653434, + 0.4892333183265376, + 0.48350565756927416, + 0.6272315356342775, + 0.4102124625691879, + 0.44577659956232957, + 0.4726299713296033, + 0.381772546489668, + 0.4775128746819263, + 0.45539545144681065, + 0.3205529145957849, + 0.4296073752053213, + 0.45864295206414013, + 0.4033934751681701, + 0.28191232046649406, + 0.41770356676806636, + 0.3145862705145477, + 0.39906412697052424, + 0.6229570618540062, + 0.5355421633839528, + 0.5447643310293047, + 0.5513610371797286, + 0.6275452114550524, + 0.5420579582997126, + 0.5919852949649997, + 0.5597914992173215, + 0.5974439602945915, + 0.41964826831886415, + 0.4112144816671145, + 0.4064974936254811, + 0.31929610705179845, + 0.274311779222423, + 0.341261813736924, + 0.34022152579205667, + 0.31228214506855906, + 0.28041840864354084, + 0.4197047035846993, + 0.2875095420756036, + 0.4469651524645012, + 0.3259864474366632, + 0.5159440114851612, + 0.570139667062017, + 0.5020996175837605, + 0.26971381872913985, + 0.3611172057926299, + 0.566796453464855, + 0.3106007590737668, + 0.3202423979119696, + 0.2781828513767859, + 0.8361870702145795, + 0.5180535508158206, + 0.41038382731420303, + 0.5696772120787911, + 0.5136298466217742, + 0.47916404244946426, + 0.40733211627196964, + 0.4722221161924357, + 0.5087034434532058, + 0.43021121823310393, + 0.5630262179317495, + 0.5155681346753749, + 0.2777507200183981, + 0.40439081737383603, + 0.2924574123029328, + 0.35731526429467564, + 0.2911450228249345, + 0.48795341217804783, + 0.40344628014903544, + 0.3283455647627484, + 0.45642402441850105, + 0.5096123311097932, + 0.3300003647336093, + 0.4604597268697535, + 0.2953572063578585, + 0.43139245623326256, + 0.4845408538493046, + 0.34166840884140126, + 0.3198824167682632, + 0.27056938565858096, + 0.5491475016699731, + 0.2894207526698432, + 0.3504753234308731, + 0.5005475698856063, + 0.3813562441871832, + 0.40601534512775705, + 0.30699001567958756, + 0.33252826529125573, + 0.39141020550215005, + 0.3805829587749968, + 0.4338423576241767, + 0.32332645075589583, + 0.4747548135954051, + 0.270670677001008, + 0.4891189535113995, + 0.40564081059305074, + 0.38829081325510223, + 0.454423506319354, + 0.3746033295289792, + 0.46407699166959976, + 0.3382605756764401, + 0.4876936648621028, + 0.3711551506157865, + 0.32355792764293057, + 0.4210995903060436, + 0.30943207240720166, + 0.4934620385259318, + 0.2470746585588792, + 0.6645633237519154, + 0.6696868681222485, + 0.564653341851222, + 0.5626722114017274, + 0.611403565922816, + 0.6066165041747049, + 0.2687374135538656, + 0.3511162511248527, + 0.31145543802301184, + 0.36268643119118726, + 0.31902924138294847, + 0.3340217574938314, + 0.36945139741007876, + 0.4044387713116344, + 0.4148928420153238, + 0.331914909958968, + 0.36389764199247954, + 0.30472837971297057, + 0.4621697189337692, + 0.25020909575903494, + 0.40743772573423886, + 0.42654213783340056, + 0.372099052398188, + 0.4714967138547415, + 0.3942769664143106, + 0.5087144815184601, + 0.2749723732126442, + 0.4722737291857277, + 0.37354878583998413, + 0.3601352020480595, + 0.3230622528911176, + 0.37247360912314326, + 0.3484459671400913, + 0.38990183724661415, + 0.48821223539429637, + 0.4489465232456644, + 0.5303424146962465, + 0.3433761005735552, + 0.38662836627541475, + 0.39856643606456643, + 0.4120673608987907, + 0.4712800637112631, + 0.409679810122564, + 0.2665324785207347, + 0.44986694661872845, + 0.399401093947047, + 0.38438076469882015, + 0.469358050857827, + 0.3592193486195493, + 0.286164345803389, + 0.2663346737629529, + 0.3342751270275796, + 0.3726051926684324, + 0.33724734379789795, + 0.26491796268334855, + 0.3482832666039315, + 0.5132071073850414, + 0.6002459932107538, + 0.5533835166983645, + 0.517867245349668, + 0.5382237310348856, + 0.4822108172449703, + 0.5550222167027296, + 0.5058868236592658, + 0.49652032174223343, + 0.34319137042920284, + 0.3809172131076034, + 0.48532720453901534, + 0.6025754701477378, + 0.18772844363215202, + 0.34362777372369807, + 0.3350488117926936, + 0.29708677794815835, + 0.3327814074315273, + 0.431239855664546, + 0.32864016739811946, + 0.4598914437362837, + 0.7431428785145225, + 0.6114312040691388, + 0.7761871333990688, + 0.7914283044947845, + 0.6436455973358902, + 0.7354224903436511, + 0.7539085756845806, + 0.36413578433601246, + 0.3938319052269927, + 0.410473343761152, + 0.5108731620294447, + 0.33809848970129835, + 0.4204054016477298, + 0.34953277577281794, + 0.37730157901558714, + 0.38110043112889425, + 0.3672117496642532, + 0.29963326209254054, + 0.4271513736716355, + 0.38672260843614487, + 0.3746055048557936, + 0.41367205501253707, + 0.31413040005526793, + 0.3194317042319957, + 0.31398766393167105, + 0.30935932078874845, + 0.2573282501691207, + 0.33855798931255043, + 0.26454937580561994, + 0.5079499516259259, + 0.48911862309829984, + 0.3414315257384625, + 0.29143947612464943, + 0.2472138445790216, + 0.5110664773154142, + 0.5805662403750903, + 0.6348636848915917, + 0.3309948399485635, + 0.36006705720710935, + 0.2568239603009025, + 0.46305342572655184, + 0.3176608953893725, + 0.30595830000904534, + 0.4737157125843942, + 0.5082536145321651, + 0.46120814831523005, + 0.31719237795560706, + 0.4328010085727809, + 0.440826438967815, + 0.4857761172067625, + 0.4440461448961979, + 0.4908663373544331, + 0.4003454889668285, + 0.44409503912858617, + 0.4408379178913001, + 0.4730321145999128, + 0.28997153750221755, + 0.3508802672883244, + 0.4474892011690265, + 0.40991407559190085, + 0.5808978733300814, + 0.30263458670243687, + 0.5488506746308388, + 0.7054258546614496, + 0.5617319359211025, + 0.3716265157683498, + 0.31680183513429855, + 0.3350045320935791, + 0.2789434881286474, + 0.40036602796030857, + 0.49231191997650736, + 0.4417761198703313, + 0.3612152056361586, + 0.31013228866341214, + 0.47661793931533686, + 0.5059690558967921, + 0.5324246928587754, + 0.589900654058221, + 0.47913719559718615, + 0.5160750149817382, + 0.36254563367854314, + 0.4859597919532638, + 0.2802538351547144, + 0.29955370142360344, + 0.40510099551162676, + 0.3777745850364396, + 0.5327896517120783, + 0.5192655997067556, + 0.6037498775961686, + 0.6548648833697226, + 0.7507017947156456, + 0.5970163434386663, + 0.7653761433598576, + 0.43494467444769447, + 0.397330614838183, + 0.3990803003602122, + 0.4879255289431995, + 0.3313818291590371, + 0.4787168972467664, + 0.4573464806730954, + 0.4927014521645197, + 0.4692152941135653, + 0.4170765855627058, + 0.4267509333837099, + 0.4501755624731793, + 0.38603534140869716, + 0.4497327753318244, + 0.41566788095060425, + 0.4479110272999121, + 0.49580267551057294, + 0.5315062204758314, + 0.4979433738219191, + 0.3737754028514789, + 0.44640192950388635, + 0.4796912814510799, + 0.3328461473376756, + 0.4244734264490384, + 0.39613772929836705, + 0.4208456213886983, + 0.38485240985432695, + 0.42792681245244824, + 0.36881331593439415, + 0.4255483581923108, + 0.4431439675326713, + 0.4644539901556357, + 0.3103256278416466, + 0.46852139418355376, + 0.46122660532029963, + 0.4005231668798521, + 0.5253271142857536, + 0.5293972679126211, + 0.7466596370254074, + 0.2619876496410798, + 0.29710704405964405, + 0.3739478622429713, + 0.35337024516118903, + 0.26851850044984504, + 0.307884956973838, + 0.29535428940217023, + 0.6722427789547024, + 0.5653506691747407, + 0.6384034485131544, + 0.3376363977250311, + 0.36958584431970265, + 0.28911594024123033, + 0.4119947528300249, + 0.4144306712292582, + 0.23960883406518374, + 0.29277911413006125, + 0.41594420368697677, + 0.38715923547925707, + 0.4373500919405174, + 0.4129999179259089, + 0.41506471472802076, + 0.3749285975725931, + 0.29172422496315015, + 0.6870565340022879, + 0.4748861029440787, + 0.6904004091659266, + 0.5737339941437233, + 0.27068755282343215, + 0.5061945311080407, + 0.48301801669794736, + 0.4974113149229941, + 0.4674369037041004, + 0.38197542194056566, + 0.4709503807337745, + 0.4846020446073411, + 0.5011750666641941, + 0.4636609685172284, + 0.342416507871148, + 0.4239950610415826, + 0.28452075145063094, + 0.3258995743511355, + 0.6336488457359492, + 0.4935115229280981, + 0.5818776920490314, + 0.4695743955499942, + 0.45660722614140525, + 0.4339881520280753, + 0.39713711103761856, + 0.4885317351623682, + 0.4038200759678553, + 0.25261810236012633, + 0.3866566554055523, + 0.4674698617290781, + 0.25704532687961484, + 0.7607609156127296, + 0.41297093175829713, + 0.5679311508636172, + 0.7278019333868527, + 0.6823577447788624, + 0.3690930383063303, + 0.8105793241415767, + 0.4848434970067931, + 0.6593348631195564, + 0.3843894759974573, + 0.4753984644626931, + 0.36947536375600964, + 0.41737748543211506, + 0.49388789467651706, + 0.5469487398939107, + 0.5292173322176955, + 0.35605698194466184, + 0.38153905045892306, + 0.39747208310701837, + 0.40297844080566586, + 0.5524827564447675, + 0.31376653021327866, + 0.5074662340948659, + 0.4274510270164295, + 0.4495461667205207, + 0.5172914559762352, + 0.582724039611789, + 0.4616421881726485, + 0.3927514073183455, + 0.4343664814572309, + 0.5160861723852401, + 0.49979930531520206, + 0.4021211001801499, + 0.3426564748631431, + 0.4686761133548331, + 0.5291689381220998, + 0.5277188025741791, + 0.40210095093610665, + 0.579361306127932, + 0.5889423180667571, + 0.42173819424952635, + 0.498698555462801, + 0.5288247166939073, + 0.4952225221545199, + 0.44882668337330345, + 0.35683148220874555, + 0.29261945299342473, + 0.3440886132276007, + 0.3781136162724426, + 0.3126857610821109, + 0.3991111595917788, + 0.32928485289642323, + 0.3748318129718603, + 0.37444966356488263, + 0.3163200515564319, + 0.33086696435286717, + 0.3083312537300921, + 0.36201265468265176, + 0.4095241850604592, + 0.32971300817859256, + 0.2832072459049772, + 0.39472891980397257, + 0.5060505543953424, + 0.6326031229035106, + 0.28888322276244494, + 0.5167981908325129, + 0.5367005778936488, + 0.537291341298935, + 0.5501362380507139, + 0.4156158192055572, + 0.46055713989797, + 0.37658080414434486, + 0.4984327470158686, + 0.41280320458430286, + 0.6182307537016769, + 0.47022530408723767, + 0.49916997770988025, + 0.28795887618408395, + 0.3177256164025259, + 0.3886958227281402, + 0.467142146487749, + 0.36909584631648945, + 0.555330346317468, + 0.33155295195445533, + 0.6067543221451627, + 0.4961420522640465, + 0.5109339890525645, + 0.5060667797049692, + 0.4963868723978021, + 0.4148631155489368, + 0.5265183693593587, + 0.5210232719631417, + 0.5169917855235359, + 0.5155173045138984, + 0.4290659127577296, + 0.3334585956153898, + 0.3037399835484943, + 0.36867105320443516, + 0.2428287019193123, + 0.38329794242591775, + 0.43281820548684025, + 0.42215723246532594, + 0.407280249332062, + 0.4400067079870614, + 0.4091490776249868, + 0.48659597311875197, + 0.4732587955230346, + 0.33221915928382784, + 0.46696932097562643, + 0.5468519868035493, + 0.3509555430844014, + 0.29688845867258645, + 0.3671480995310316, + 0.39966381609635754, + 0.49770569269329473, + 0.3260372465924245, + 0.45474046703651894, + 0.26341581959398624, + 0.44753081611609064, + 0.48672540338340053, + 0.5274736663289928, + 0.2970412810631696, + 0.311288930576936, + 0.3194648948889046, + 0.2696577437028447, + 0.24669565163607476, + 0.3530852315865635, + 0.4434481540698736, + 0.27169921583540085, + 0.37235758829142834, + 0.4348706302334144, + 0.31344755438142474, + 0.3640502498946778, + 0.44394647837111967, + 0.532072431752263, + 0.27632953689609874, + 0.44532382409588417, + 0.39525538995237414, + 0.3224664088130923, + 0.32177139824402584, + 0.45981401369479336, + 0.4266648789751903, + 0.26306068365925844, + 0.4816573908552263, + 0.4110226061390944, + 0.3743226454837113, + 0.5117644054831785, + 0.39725570644862124, + 0.32194594710377983, + 0.33365014820114464, + 0.3171108118719583, + 0.3880580021670711, + 0.3639857349556349, + 0.38362229292767247, + 0.4307733888234682, + 0.40452831022032215, + 0.5131539404159141, + 0.43947965123679156, + 0.33734956779226083, + 0.4064913168818565, + 0.44731490089794457, + 0.43334012403852534, + 0.45144392842463626, + 0.5145405075554629, + 0.4606574174350366, + 0.5198348257415959, + 0.5513638412245601, + 0.5755968249954595, + 0.45048039639634135, + 0.38120594458473794, + 0.24588040483913975, + 0.4127839217700007, + 0.4750536407726904, + 0.4264865825600861, + 0.4415167473392452, + 0.4578279872341326, + 0.31256030718147787, + 0.4656814782963988, + 0.3752281918411639, + 0.27240325540233373, + 0.40837001924430627, + 0.4195001377992321, + 0.450642725902569, + 0.45677102180255574, + 0.30355741800672364, + 0.565567015327873, + 0.5084256037122463, + 0.48335448556178334, + 0.5239758814884666, + 0.6027614859170725, + 0.46935103866775973, + 0.30535208235666206, + 0.3850255457483094, + 0.38572161006881894, + 0.3728856831229572, + 0.3698354308317803, + 0.33117160047480365, + 0.3564859285097526, + 0.4033002333823855, + 0.29825708265342643 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.20303897614186278, + "valid_tracks": 4, + "per_track_pearson": [ + 0.24404050062891863, + 0.18979849997313536, + 0.2255444008754548, + 0.15277250308994225 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.28048366883296943, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2927660765103106, + 0.2721139970780286, + 0.28133554212370365, + 0.2967011748388399, + 0.2845263315667872, + 0.29938824200217395, + 0.2721895621982371, + 0.24362795696492262, + 0.27938970103854843, + 0.2705581266887339, + 0.2882888319766783, + 0.26759489364041583, + 0.2608809506591594, + 0.25193240581356324, + 0.2785104117980056, + 0.2949787939306581, + 0.27218873943103394, + 0.2670915207278411, + 0.2870558464994719, + 0.2708091984255448, + 0.2898669491226738, + 0.28197823437553543, + 0.28312889699867605, + 0.27155412879028307, + 0.2755660241585208, + 0.28453493120778356, + 0.2681405173931981, + 0.2996894754219908, + 0.27637500278192606, + 0.27829117782814583, + 0.2713056594993157, + 0.2919361108750996, + 0.2756308509229557, + 0.292665338357637, + 0.24426094320228134, + 0.28643327066248236, + 0.27045043247053163, + 0.2703956615829585, + 0.28132705660727253, + 0.2655944237662728, + 0.26325097314786294, + 0.2711196238521038, + 0.26816983711399806, + 0.2720133396660558, + 0.27177223517452015, + 0.26531790330901384, + 0.28057439819814034, + 0.2801440461294562, + 0.2852456782902342, + 0.28845756211263557, + 0.2806063870417231, + 0.27100664793495116, + 0.2772233790278373, + 0.2766725633432416, + 0.29104424786695565, + 0.25581781414639515, + 0.2822882357802728, + 0.28502931856869523, + 0.27986255103281205, + 0.2743760404750547, + 0.2643910354889973, + 0.273509100934536, + 0.28417374890159813, + 0.2846238718101628, + 0.2882226915798268, + 0.2761662999545708, + 0.30066367666939153, + 0.263221892570772, + 0.2801153611779926, + 0.2854353597246337, + 0.2796958842728697, + 0.2916402043096955, + 0.2725974579433363, + 0.2919543218976326, + 0.28629535610668505, + 0.3026202413358845, + 0.26566851987931156, + 0.28472775403641565, + 0.2835208303278231, + 0.2849319634536048, + 0.3014291695634362, + 0.27075520521450513, + 0.28913235931400977, + 0.2724059171770404, + 0.2822191868902089, + 0.2795764426171132, + 0.2640163898106108, + 0.28096982228012823, + 0.2734317093207227, + 0.2966265592594698, + 0.2850742417980326, + 0.2876338164265247, + 0.28421096276961644, + 0.27042861504069343, + 0.2748769706609767, + 0.30180914331565645, + 0.27768890800769624, + 0.28413323999340134, + 0.2646382689272549, + 0.27997048826573034, + 0.26068997013369405, + 0.29881651728327985, + 0.29437807171081326, + 0.2949349318313774, + 0.289710996826879, + 0.2896382024184023, + 0.26274042211107257, + 0.28213434743470617, + 0.29695479286748844, + 0.28957246239541284, + 0.2822857364452035, + 0.2936194964935258, + 0.28941145507715305, + 0.2916901158255843, + 0.2937496710958549, + 0.29375017873350306, + 0.2549602992622316, + 0.28263845862711495, + 0.28105774011170115, + 0.27637752805533666, + 0.26997684696702623, + 0.2826598697047777, + 0.27095002326444806, + 0.2771805402696457, + 0.25234891750324207, + null, + 0.278579274452312, + 0.27186532693533133, + 0.26884041674212633, + 0.29208646518593573, + 0.26940214746127306, + 0.27297921508146766, + 0.2887345407388609, + 0.28665021451927325, + 0.29860148641081175, + 0.30063262952660136, + null, + 0.27975389740211376, + 0.2758780627265488, + 0.2558964782862872, + 0.2622834564623829, + 0.2752669666414298, + 0.29460391675756487, + 0.24611864080177248, + 0.26829720896291576, + 0.2581992302743398, + 0.2607735323564376, + 0.25908252408989596, + 0.24896009345046546, + 0.2837801702144034, + 0.27044458144190064, + 0.2978636932144483, + 0.2859689239317472, + 0.30268197392926294, + 0.2784261111227856, + 0.2859653122515765, + 0.2834094105524998, + 0.25630090545267137, + 0.29393554919985665, + 0.2579764925887698, + 0.26842398421876923, + 0.2797486258964035, + 0.2961362373603979, + 0.29200306680658167, + 0.2754187695343082, + 0.27791217839571436, + 0.2817670288720972, + 0.28375980200389944, + 0.2522202327663996, + 0.2639156894276222, + 0.26961965035864344, + 0.2817442639695712, + 0.28612224206643844, + 0.2755088046642691, + 0.28789604654696915, + 0.2633976598767171, + 0.29950325348218554, + 0.29978651754526964, + 0.2746946150587278, + 0.2585496325492187, + 0.2872007290085049, + 0.2813576723952542, + 0.24990407783184512, + 0.28330593585784425, + 0.27815371164570235, + 0.25746765370836766, + 0.2981052797609088, + 0.2845293869637757, + 0.30275454367438304, + 0.26304574620253673, + 0.29091329496583324, + 0.3029225790463113, + 0.29480822256440753, + 0.2920658259951796, + 0.2921979048091582, + 0.2876957040106413, + 0.2959081102311436, + 0.2633232683243, + 0.30120560557755405, + 0.26831310578471507, + null, + 0.2662693363849417, + 0.27638630133092407, + 0.26941472878742867, + 0.2719013392870307, + 0.2706061367766245, + 0.26390437721077675, + 0.27278786768544117, + null, + null, + 0.2555402169256119, + 0.27757076446702633, + null, + 0.28158309147395644, + 0.24876491512742474, + 0.2724977942886914, + 0.26145194354308204, + 0.3056149680325427, + 0.2909303196340696, + 0.2610096981119219, + null, + 0.29596843092500436, + 0.27857308222831917, + null, + 0.26173009264089137, + null, + 0.2643526127588152, + 0.2888011928014402, + 0.2834561151024767, + 0.2832073442000969, + null, + 0.2636143549288496, + 0.281734777789062, + 0.26666399934187573, + null, + 0.27105657450023674, + null, + 0.26237245313198765, + null, + 0.2673551267742644, + 0.2782258997912483, + null, + 0.2737045544088369, + 0.2724216457539176, + null, + null, + 0.2927250075144747, + 0.26293679703787204, + 0.26082948182250393, + 0.2643334950514555, + null, + 0.28676472586164975, + null, + 0.284346376260307, + 0.2818565487642087, + null, + 0.20230324451584175, + 0.2685516655590385, + 0.2609827939541797, + null, + 0.25866787274391423, + 0.2865193248008661, + null, + 0.28199793644101795, + 0.28105645449219174, + 0.2717557271684251, + null, + null, + null, + null, + 0.27664918444278713, + null, + null, + 0.2693309229767057, + 0.2737749482203894, + 0.2735295441299488, + null, + 0.2826567050251228, + 0.2643395517678382, + null, + null, + 0.2742707758562524, + 0.2688267702887117, + 0.2820833759548521, + 0.2915457498078725, + 0.2874282536123562, + 0.29161058563547365, + 0.26973179741654296, + 0.2897439203263095, + 0.28062304374284736, + 0.29023602996152376, + 0.2541831077302542, + 0.29701945173231425, + null, + 0.2805831618809108, + 0.286608304367465, + 0.2776952095793623, + 0.2850941091007752, + 0.2843917411964927, + 0.27378655325737034, + 0.27608419156777, + 0.27183759205159136, + 0.2748390669907365, + 0.28955050746859523, + 0.2838777615667222, + 0.28557305409250305, + null, + 0.28293212344565744, + 0.27200839357014867, + 0.2594858718819212, + null, + 0.2671231669825222, + 0.19564407867605624, + null, + 0.28202577674777174, + null, + 0.27470032025996005, + 0.2759160592689721, + 0.2760938137890464, + 0.2651007191610619, + 0.26557891736407196, + 0.2641902220572971, + 0.24518508070332365, + null, + null, + 0.2842276054006015, + 0.2574275541078257, + null, + 0.29121074984343137, + null, + 0.2851568024775044, + 0.2778998630251041, + 0.2811999390906755, + 0.2716042899110435, + null, + null, + null, + null, + 0.29337824050241385, + null, + 0.27707633768155104, + 0.27226768939951745, + null, + null, + 0.28273370939461184, + 0.30185565910376516, + 0.29576084280661574, + null, + 0.28796468591577273, + 0.28715540697981173, + 0.2873707949592094, + null, + 0.26815678269622467, + null, + null, + 0.2852374037802311, + null, + 0.2940992219887311, + null, + null, + 0.2870441370633895, + 0.2700466359913459, + 0.27170264644038045, + 0.2733426129616842, + 0.26832941151362166, + null, + 0.2843797683571403, + 0.31889359326892097, + 0.3037316022680583, + 0.2521265015009247, + 0.25316227096441907, + 0.30935970739299984, + 0.2835479779754656, + 0.2662451230223744, + 0.29336145136035197, + 0.2914724826257243, + 0.2871638041946804, + 0.2920165622283853, + 0.29255133796670296, + 0.2857844581005674, + 0.2858998285549299, + 0.2412477291939754, + 0.2590021704046879, + 0.28401347639104385, + 0.3033256696943613, + 0.31643097159492567, + 0.3227611656601144, + 0.2990843329844308, + 0.2747471505490925, + 0.25330931814336377, + 0.23952913629684827, + 0.30426837752046004, + 0.32008080130496563, + 0.3032687363407109, + 0.32056095398539586, + 0.2547050931239014, + 0.25508444528092517, + 0.24754334095649597, + 0.2567587718087018, + 0.24621400000949217, + 0.2846506005383263, + 0.2363722679502153, + 0.2897693122907627, + 0.22542250400936392, + 0.30258990088463344, + 0.28746225913540585, + 0.26664984617993315, + 0.2797851946384009, + 0.24965677306212028, + 0.25517862008011494, + 0.2704208260860645, + 0.2923241848502626, + 0.24903489021147868, + 0.26560699211181504, + 0.25299667868233655, + 0.27859395982827506, + 0.2542939518464288, + 0.26422064357529723, + 0.24437159920558837, + 0.2791312970532842, + 0.26473492566074497, + 0.2479638624151927, + 0.27768354391007244, + 0.2982825288781345, + 0.2855986782890262, + 0.3016067947260859, + 0.2901289660539915, + 0.2892986754041812, + 0.32252701091708386, + 0.31681080338429335, + 0.3142921331643363, + 0.31738976663595475, + 0.3209982616335759, + 0.2843070981852178, + 0.2989998998085606, + 0.30658853843131834, + 0.25400291055553514, + 0.33351324516242725, + 0.2993274012562851, + 0.28114162354404787, + 0.2765974420539357, + 0.2934173505690109, + 0.2766319842526141, + 0.27619801211658407, + 0.32166044878312317, + 0.30502670886327043, + 0.30542192970007864, + 0.30013796949266763, + 0.302585634319374, + 0.2755497986623584, + 0.30499164130610634, + 0.2905751826606442, + 0.2845981007318409, + 0.2773444270775944, + 0.278898491670663, + 0.29229416494023497, + 0.3059289160287305, + 0.2858060742193283, + 0.2823630924042662, + 0.29378070296568953, + 0.2993555732464695, + 0.2979925305062755, + 0.2950117393450166, + 0.2907948964575095, + 0.320100501330187, + 0.3106989167301298, + 0.3204402572414637, + 0.2759755365425189, + 0.32918983942169067, + 0.31091505329317, + 0.3169115819475352, + 0.3069579567277499, + 0.3134726765006046, + 0.2592264466232192, + 0.30386655730065815, + 0.2729236387081344, + 0.281360101398533, + 0.27970368660916317, + 0.28147978692373804, + 0.27948250046777623, + 0.28607322407218516, + 0.28005381410304453, + 0.3150517297364169, + 0.28995875467088283, + 0.27753893119826606, + 0.25333678940829163, + 0.2968551085069213, + 0.2893759267845135, + 0.3021071006841185, + 0.30369851986502866, + 0.31777425259995323, + 0.27717328982643363, + null, + 0.2997695285974585, + 0.2651939209078536, + 0.27036104219302626, + 0.2704384630762419, + 0.29790869751555826, + 0.3081855416622152, + 0.309239991290469, + 0.3166040778010386, + 0.3044915897042605, + 0.2799469907144045, + null, + 0.3027118998637682, + 0.3067295872715948, + 0.3210527178340594, + 0.3016828434902144, + 0.2904204003611861, + 0.3005177769342876, + 0.2781496800289075, + 0.28176696448317223, + 0.3030107268478759, + 0.28956549551799826, + 0.30461203363742545, + 0.28137976403480647, + 0.3183091124429488, + 0.2751249517042726, + 0.3018264780589262, + 0.30046356804272756, + 0.2790886777944952, + 0.2534896921689081, + 0.24763716991505286, + 0.2531022920988768, + 0.29971200352918664, + 0.2905278771492757, + 0.28758870549014315, + 0.2804559371843344, + 0.2813311550693416, + 0.3039816579555068, + 0.26555591316381144, + 0.30815320352144415, + 0.28860766413322386, + 0.30303273143274406, + 0.30445975666506697, + 0.2788095897063503, + 0.2644772085129725, + 0.2932454242139512, + 0.3134446208807867, + 0.31457363448462505, + 0.28223528860404506, + 0.2648803149352835, + 0.28556066044789397, + 0.3070703645156966, + 0.3120957548780266, + 0.28808508030682883, + 0.24822615554660268, + 0.2935105117507071, + 0.29898375402655286, + 0.25646585029069147, + 0.2981395472451542, + 0.3051729487362862, + 0.287808264954901, + 0.3089638690834549, + 0.2557109052929672, + 0.2814216196025212, + 0.29052102085453385, + 0.27608653001506817, + 0.2876403941648981, + 0.3136101822420438, + 0.2603581663327862, + 0.26059263697253765, + 0.24829456139434425, + 0.2686084203062753, + 0.29337765067262533, + 0.3161682141060825, + 0.2955934962446422, + null, + 0.29055583585585937, + 0.2856796369771542, + 0.25979479584638016, + 0.28188080272156035, + 0.2993235784505806, + 0.3113916780964195, + 0.26470699374448736, + null, + null, + 0.3012956612263253, + 0.2872664232790692, + null, + 0.2661263246839258, + 0.2419858895018158, + 0.2831424363809504, + 0.2461114369349561, + 0.26194671893097576, + 0.2579885228631007, + 0.2965453061768928, + null, + 0.29864003130690464, + 0.29473543593484847, + null, + 0.24650577842886384, + null, + 0.2689180386325467, + 0.296390940522619, + 0.28643545081740235, + 0.2615140202227406, + null, + 0.25143424189584884, + 0.26411377462556224, + 0.27673332420150243, + null, + 0.22712298316302187, + null, + 0.2648992138538525, + null, + 0.2544175549384939, + 0.2608892465383517, + null, + 0.25008017276007016, + 0.29677975947928825, + null, + null, + 0.29505610805349813, + 0.286217989218794, + 0.26662641316962643, + 0.24585202821654517, + null, + 0.25901087991251753, + null, + 0.26154901813866305, + 0.264667616919346, + null, + 0.2292270207595904, + 0.25082188188689575, + 0.2504886858961225, + null, + 0.2634234222228462, + 0.2588733857729165, + null, + 0.3143827954554551, + 0.2975843900909078, + 0.3134323737335338, + null, + null, + null, + null, + 0.31253065161627575, + null, + null, + 0.29216315639771895, + 0.2878343801861941, + 0.3035686781059759, + null, + 0.3174789034465343, + 0.3131871468446434, + null, + null, + 0.2919657682116867, + 0.27562337771189804, + 0.300913835362684, + 0.2617207500781729, + 0.2621950634656635, + 0.27397889166580414, + 0.24576828658202207, + 0.26546620621192224, + 0.2424852343098119, + 0.2618784750944824, + 0.2724163189992058, + 0.2561346747437039, + null, + 0.2681053067121895, + 0.2587238812503895, + 0.31100440164675786, + 0.2948548264860583, + 0.28713813982508535, + 0.29025606240853374, + 0.2816772961440689, + 0.28870703749332677, + 0.3028736500199391, + 0.26034490777643593, + 0.2591532411956809, + 0.2503115815329209, + null, + 0.26066912174925905, + 0.31708673258241116, + 0.31620115041130853, + null, + 0.2586533963398708, + 0.2780078269284293, + null, + 0.2686112053776524, + null, + 0.2947523064504993, + 0.27851145050714216, + 0.2881712494718725, + 0.27006062163319344, + 0.3026445365966732, + 0.24561600172902034, + 0.22865943150537885, + null, + null, + 0.28821473111222023, + 0.30078263483580253, + null, + 0.2708345483269708, + null, + 0.2790905625201898, + 0.2729818411653068, + 0.2613040640465481, + 0.3006682691052434, + null, + null, + null, + null, + 0.27504545374281814, + null, + 0.28074574962971194, + 0.30010866705688566, + null, + null, + 0.29996346390699086, + 0.2570622145912148, + 0.250033099332434, + null, + 0.25549129115932306, + 0.2571704951070108, + 0.26779165859135245, + null, + 0.3025490407698007, + null, + null, + 0.27749361963995256, + null, + 0.23345527810821018, + null, + null, + 0.2649384192438135, + 0.3028008982921232, + 0.3057254041747847, + 0.2974463582266634, + 0.30152702835465633, + null, + 0.2800731123061684 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861, + 0.6937712344280861 + ] + }, + "junctions": { + "mean_pearson": 0.2240537907436797, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2824461642353725, + 0.2993167044328379, + 0.28911622966794254, + 0.18126692223312765, + 0.39512697755346865, + 0.30407413474367817, + 0.012471228258718344, + 0.25622526207505397, + 0.2591423074959086, + 0.2631032289817984, + 0.21809653235102566, + 0.20046210567741438, + 0.08555284084955293, + 0.17719345187380403, + 0.20948125006762308, + 0.21016541425097293, + 0.23811141520396273, + 0.17630643777483399, + 0.19047895552601815, + 0.21724238334253962, + 0.25911338251124827, + 0.28458082881699454, + 0.14489611061902016, + 0.22674988620228492, + 0.22538655613520236, + 0.241400558076924, + 0.20324707195460176, + 0.27968482834063774, + 0.12239870387217161, + 0.2879253243001583, + 0.27070815909637774, + 0.32840800840895934, + 0.23647672778563208, + 0.28887865001251195, + 0.256479404882463, + 0.2799987287057979, + 0.25527990379293125, + 0.22407382261777642, + 0.28405712839942365, + 0.12112964573403914, + 0.19689031842839014, + 0.20664627645546374, + 0.20477128935087743, + 0.18481106874070607, + 0.1849055899700621, + 0.1501777440474809, + 0.13191316482991125, + 0.20465876483738757, + 0.15873984174273145, + 0.170646151662555, + 0.1834149420875143, + 0.19901421277541811, + 0.17035435861848797, + 0.24969104949905302, + 0.2155592208999413, + 0.2241292802009203, + 0.16560069722060666, + 0.1679274855025525, + 0.22749426209130028, + 0.214261363978728, + 0.11956675855109027, + 0.20777889975472588, + 0.21126052720115987, + 0.05302146171964449, + 0.10496220053919081, + 0.33240559808871234, + 0.2661461349391503, + 0.2944575586977847, + 0.1381273621487804, + 0.18707626245342263, + 0.1312111904246842, + 0.27737749807568873, + 0.17612237581063792, + 0.12398620466363826, + 0.3426647344807325, + 0.3962112041758353, + 0.19529207102984739, + 0.1882204453827454, + 0.1512852359622233, + 0.35557541204773224, + 0.24302816999738822, + 0.2717509331850244, + 0.2132939622872358, + 0.3222235202161386, + 0.26332476662695825, + 0.2844662120773463, + 0.2832028208260864, + 0.30577870198414103, + 0.28988456905854837, + 0.2776066044918514, + 0.30618342784721037, + 0.2766828239080145, + 0.18161583773250212, + 0.26476368554346025, + 0.2786841886440985, + 0.2679268159720965, + 0.27442565705333827, + 0.27455119113996873, + 0.26686317571356466, + 0.22826750740094373, + 0.1924195606232738, + 0.2516605806900299, + 0.3245868093950266, + 0.15351175280242824, + 0.1819371055611128, + 0.12261573082424744, + 0.07322096247355983, + 0.2626941138796848, + 0.2872753439822564, + 0.25636355677445116, + 0.1933993037068354, + 0.27746184875315577, + 0.28688055002513974, + 0.38576428665645396, + 0.28592955805703085, + 0.1465771987070091, + 0.1854380729117799, + 0.2457401218704906, + 0.17340339169251, + 0.2380802366589161, + 0.22076155156282198, + 0.18930779579500326, + 0.22120702300763637, + 0.21132967384362553, + 0.11644662061390346, + null, + 0.2673005454857849, + 0.26840335255544656, + 0.16385768369783577, + 0.22858676944150827, + 0.27567742191743927, + 0.22750241321045525, + 0.1446349265562069, + 0.36977343142916114, + 0.3499065169340458, + 0.15926021113479807, + null, + 0.23368377944091434, + 0.28701429848188736, + 0.2040865262402263, + 0.2176526497180297, + 0.28493467896860936, + 0.2684773367027385, + 0.32176333512943694, + 0.24867765411531495, + 0.21501110047302502, + 0.25645747580705686, + 0.23520457850327176, + 0.1932798226128312, + 0.15699273617451853, + 0.23126978444257365, + 0.3587085338988465, + 0.20522799960043672, + 0.296678606240697, + 0.38290106694769493, + 0.3409295505453182, + 0.26482912101781997, + 0.30761801690672225, + 0.28000499353932917, + 0.3194132911055824, + 0.23944912856791928, + 0.33576796208872944, + 0.1952371476085414, + 0.3083039410650497, + 0.12410437513929674, + 0.3206555645060305, + 0.3195601327466587, + 0.23644992385148259, + 0.17901634998930693, + 0.1509041268020434, + 0.2467694590420642, + 0.26010394097165934, + 0.3080855274050613, + 0.17199090599346004, + 0.20008221365591097, + 0.2464634118250196, + 0.27410559760063424, + 0.19432204347170745, + 0.212464394525661, + 0.18092620191372225, + 0.1650659214180188, + 0.1353260548836617, + 0.06997275979257278, + 0.26487173514970264, + 0.27868832494074147, + 0.26448261491208164, + 0.2249690998506194, + 0.19131432723731276, + 0.2481626391800114, + 0.26334733615752953, + 0.2724201985607786, + 0.3847031339029097, + 0.15740804060797534, + 0.13300399782871325, + 0.16748110523177684, + 0.14463472624724372, + 0.16468482200646173, + 0.15911450456050738, + 0.23069195917828036, + 0.3784769248829497, + null, + 0.10266939752245842, + 0.21300398851295435, + 0.2643625563174798, + 0.30987605396345047, + 0.1254766627976798, + 0.08939504835629168, + 0.19279059929687384, + null, + null, + 0.20305675463689324, + 0.29903683575421813, + null, + 0.15008630251150754, + 0.24269052030486243, + 0.21816925974801965, + 0.18210004137222016, + 0.11497699075059266, + 0.15053080127875024, + 0.09561672427245733, + null, + 0.27885058361847487, + 0.25704457579315315, + null, + 0.16281202598360886, + null, + 0.31239035010564553, + 0.15565476996510952, + 0.19579712892725762, + 0.2544552725738029, + null, + 0.11607420640834928, + 0.13374632880833695, + 0.22812257657113097, + null, + 0.1839291410293589, + null, + 0.2711307118371949, + null, + 0.10903192000041002, + 0.20821905395514131, + null, + 0.17163998958362378, + 0.2427076689401821, + null, + null, + 0.22659206696134265, + 0.2602384730196844, + 0.09052529053724193, + 0.09637738185648201, + null, + 0.36834855220283064, + null, + 0.20295205908095262, + 0.15105232067389307, + null, + 0.2104296069637429, + 0.25798799310919435, + 0.35818616401686915, + null, + 0.18635663656547086, + 0.23050719778296247, + null, + 0.3798071003773277, + 0.33948899421014817, + 0.36965737049611397, + null, + null, + null, + null, + 0.2253399406617692, + null, + null, + 0.20123014324957605, + 0.26619830058036015, + 0.3666255212140457, + null, + 0.18574054978094007, + 0.3412996361383495, + null, + null, + 0.30135586122840025, + 0.1507668733120532, + 0.21786558472159895, + 0.13407899903396203, + 0.15956829209581183, + 0.10970699126061786, + 0.20845871487699758, + 0.17658872533178804, + 0.14550156686657612, + 0.09203139707375584, + 0.313893641957374, + 0.2577447621493191, + null, + 0.25625786510453247, + 0.29655559570543266, + 0.053524336514237596, + 0.14514757164315192, + 0.0897369604371139, + 0.0552569147462417, + 0.21766816706841224, + 0.15133118142426372, + 0.23127866463378402, + 0.17808035048664003, + 0.25710237416516346, + 0.21930562868685952, + null, + 0.18614676066573638, + 0.20263711249691663, + 0.2126145891236034, + null, + 0.14389857152671653, + 0.19997082953622036, + null, + 0.23599256390383241, + null, + 0.13316216716980686, + 0.16068475443367486, + 0.2605346334450377, + 0.09450667904888031, + 0.17923335698615372, + 0.11695379523939903, + 0.13395384927491077, + null, + null, + 0.18845779331418186, + 0.1636350325200194, + null, + 0.18637952135755695, + null, + 0.2775915961077477, + 0.1946055223233524, + 0.12286587715402725, + 0.21979248286699934, + null, + null, + null, + null, + 0.10188267763360997, + null, + 0.08085733467390453, + 0.30514118687147984, + null, + null, + 0.26649693784675565, + 0.21082084696986475, + 0.16470552360446533, + null, + 0.3771023708329736, + 0.3136888721772168, + 0.17355004256582923, + null, + 0.33645512615897916, + null, + null, + 0.2530964726691911, + null, + 0.18124418064301917, + null, + null, + 0.26028697967595005, + 0.2614222941195829, + 0.2110810251881899, + 0.22570716739735985, + 0.199444904390449, + null, + 0.21029093294003467, + 0.18877863760950858, + 0.1683525435045144, + 0.31929778150346017, + 0.10843063787370204, + 0.3216751619770746, + 0.2465966595267891, + 0.1571235406138378, + 0.16680223056635587, + 0.2861145568042369, + 0.17353570055247888, + 0.25345552583612646, + 0.2991999257086431, + 0.22052396756635517, + 0.18671616033094793, + 0.15324964710170424, + 0.2852871741178276, + 0.2570070868212009, + 0.34445047261701917, + 0.115588659219032, + 0.26681067606396874, + 0.19577211283242996, + 0.2045987529877483, + 0.07743355907704685, + 0.23627221783873606, + 0.33541454588526687, + 0.329500365228332, + 0.3350669787026491, + 0.3349225911303318, + 0.24892238901708255, + 0.19377691955247284, + 0.19435475027402943, + 0.35486081187807994, + 0.227862971024791, + 0.24390542251408898, + 0.0624613376335234, + 0.3270256308540748, + 0.032622358471708156, + 0.19812161455832408, + 0.15163498014520715, + 0.31131260062336713, + 0.2490188518758334, + 0.019811052804197513, + 0.09876732690296643, + 0.23667165545993793, + 0.15943009729512236, + 0.15596379920169265, + 0.16090427631454352, + 0.19132351051905552, + 0.23716925619147716, + 0.2045721979798251, + 0.20159989394374317, + 0.10629406230385433, + 0.15366674036464065, + 0.07413654317894464, + 0.043782452516218394, + 0.10845602795955944, + 0.25906698347876767, + 0.2708145909847692, + 0.1915214048660593, + 0.2388315578387086, + 0.33201446519648986, + 0.29777766531999994, + 0.15555164817403058, + 0.224471172249441, + 0.19059073770498952, + 0.1685087063945003, + 0.12656826808702412, + 0.22267485650376562, + 0.12968101912418398, + 0.2651991772787135, + 0.33618108760236043, + 0.15814501098836178, + 0.18532802065226814, + 0.2208914874306654, + 0.08884881679413534, + 0.08024764243550131, + 0.305733447689078, + 0.3053017535209414, + 0.3272837877495389, + 0.19671104685236798, + 0.22892319234289618, + 0.2923712127255235, + 0.31061733515482565, + 0.15207844051892766, + 0.25818252876644837, + 0.23520390118086582, + 0.1961109493271364, + 0.26977298128123617, + 0.17366749952566962, + 0.25321783966343364, + 0.30736646073199525, + 0.2022489449917708, + 0.18117201316780654, + 0.16818000526219626, + 0.11748619345766206, + 0.1564224645127089, + 0.23769008225244376, + 0.20066090581522322, + 0.32585646862475504, + 0.21051433520015142, + 0.14002013216455939, + 0.24279571165341368, + 0.16194477447118935, + 0.1922447142398607, + 0.18955883635100904, + 0.2627334323534956, + 0.049883346090775584, + 0.2250748454498676, + 0.2236037709637277, + 0.08699303927098352, + 0.06007222082622346, + 0.14443177526174877, + 0.154465893562206, + 0.17867759628701663, + 0.1608485590283016, + 0.08060865711064645, + 0.32492710565780253, + 0.15533729719607678, + 0.11750220233422444, + 0.12131551563334396, + 0.32373738309371647, + 0.17200834791857264, + 0.13678889834317529, + 0.16768381672226368, + 0.2782643576365804, + null, + 0.366532198358971, + 0.3205210024411202, + 0.24875945069534058, + 0.42732534285395174, + 0.18925260402254102, + 0.08968453536490412, + 0.1380994090156205, + 0.3859062516642586, + 0.1774177775092803, + 0.14011594252850393, + null, + 0.1426889345556619, + 0.12447610316229457, + 0.10219861091709827, + 0.0720834716500037, + 0.19160240230085945, + 0.14955899538748266, + 0.15904183030915714, + 0.2793242034989631, + 0.07600440672015636, + 0.22597389468033727, + 0.26160549748344414, + 0.13660741974350293, + 0.1965661915484332, + 0.18455025515240314, + 0.34369264550392054, + 0.31517252730684786, + 0.28522379313078183, + 0.37897094473849297, + 0.30334162552982413, + 0.22744474144593718, + 0.2070195843966492, + 0.13850963291103657, + 0.04450016836825654, + 0.13227129507272326, + 0.1478274711724512, + 0.21824936162522426, + 0.1912584754670576, + 0.2506682207814725, + 0.2156188134629105, + 0.12285092580089403, + 0.14660526748489894, + 0.30896497064387396, + 0.13708854344678012, + 0.2186396153900472, + 0.28028721314529625, + 0.2236461856333961, + 0.07916787127020396, + 0.267823877439179, + 0.17188072971506044, + 0.29841955059566044, + 0.1541429255708039, + 0.10442768162743318, + 0.18121349644460297, + 0.23087619025970413, + 0.13433177707476102, + 0.2057631318334384, + 0.20617711826777507, + 0.34669009076289886, + 0.1316415253105998, + 0.23932834491758012, + 0.11609998494831385, + 0.20497074630006198, + 0.23464942056391058, + 0.2418970164561091, + 0.20276415644985143, + 0.12118932229973407, + 0.20818825682620387, + 0.28485667596127634, + 0.31136637506974885, + 0.26374752526230966, + 0.19135732207703315, + 0.19640210113052617, + 0.32794914664099767, + null, + 0.4226286940882423, + 0.23240913655862047, + 0.43458705543872855, + 0.36356659936732094, + 0.3545994909585473, + 0.2126359666741415, + 0.2457180254580169, + null, + null, + 0.21298328566661748, + 0.2499271682975163, + null, + 0.13719102535271366, + 0.23474878668782098, + 0.19437820252150467, + 0.2640059338350188, + 0.20047665187380062, + 0.28002290543062835, + 0.18739528546340467, + null, + 0.3179696211618675, + 0.2732926633178716, + null, + 0.3449500181034301, + null, + 0.42628280444655375, + 0.2511210483841074, + 0.17310705032607093, + 0.18590515108999334, + null, + 0.24925774720956972, + 0.16599590140991774, + 0.2873643711287458, + null, + 0.32114416438444826, + null, + 0.23685033498832034, + null, + 0.2875666737033515, + 0.34591954037168354, + null, + 0.3078386899997385, + 0.34062422982790647, + null, + null, + 0.35466629775048014, + 0.24937677887861107, + 0.04056355709429299, + 0.18443324578331055, + null, + 0.30005071471751765, + null, + 0.1591319891679374, + 0.2962370588392288, + null, + 0.27658976094193816, + 0.2791478645367696, + 0.19328217300168216, + null, + 0.3216320984141282, + 0.4252088431462733, + null, + 0.14270633793636897, + 0.2299358272435639, + 0.24931801698720246, + null, + null, + null, + null, + 0.2194696553832783, + null, + null, + 0.3710294462205733, + 0.2772221742035162, + 0.20018017148613246, + null, + 0.08301447422391693, + 0.2572143857461947, + null, + null, + 0.263427725048995, + 0.02515145501468169, + 0.25100377727410317, + 0.14758419104904555, + 0.1564074721019983, + 0.2901882749977938, + 0.0015216706030614026, + 0.26313132298400993, + 0.09095009257449546, + 0.3983058905676905, + 0.35482272830441497, + 0.28949673945306814, + null, + 0.08033694698201876, + 0.2863885099050951, + 0.24899295862707008, + 0.26118204354030355, + 0.1406417117374512, + 0.2381703365634153, + 0.24074705645232417, + 0.23324477635506063, + 0.26758405308783245, + 0.299896267170186, + 0.24037457263748438, + 0.1530028864061584, + null, + 0.21070255451773318, + 0.3642603131161859, + 0.25273360548469764, + null, + 0.2919199076612892, + 0.32064115657831016, + null, + 0.2533471102359787, + null, + 0.35274241598958395, + 0.3892817931142173, + 0.08394478703729497, + 0.29676314218921035, + 0.30310655234483763, + 0.20607470357470967, + 0.3344302029560856, + null, + null, + 0.2850982551873031, + 0.3054572625743474, + null, + 0.2830406271329694, + null, + 0.3577842192404553, + 0.333558255128818, + 0.3190062392509337, + 0.2854510120246089, + null, + null, + null, + null, + 0.320767608517843, + null, + 0.3545582289882456, + 0.3568320351050793, + null, + null, + 0.26581801841170405, + 0.05252302153991619, + 0.2643915959499457, + null, + 0.14754046586966527, + 0.22953922166003618, + 0.2516835257905133, + null, + 0.2684339837594022, + null, + null, + 0.2804140004601978, + null, + 0.21811389434816175, + null, + null, + 0.2823092424793934, + 0.3018851135045513, + 0.401290223927975, + 0.35890667998167824, + 0.3135748493268343, + null, + 0.28509454166774867 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.9006622516556292, + 0.9243421052631579, + 0.9074733096085409, + 0.8892405063291139, + 0.9205298013245033, + 0.8813559322033898, + 0.9014084507042254, + 0.8888888888888888, + 0.907051282051282, + 0.8971631205673759, + 0.8922558922558923, + 0.8915254237288136, + 0.8961038961038961, + 0.90625, + 0.92, + 0.8848684210526315, + 0.8636363636363636, + 0.8846153846153846, + 0.9044585987261147, + 0.9150326797385621, + 0.9064748201438849, + 0.9, + 0.8798701298701299, + 0.895910780669145, + 0.9074733096085409, + 0.9177631578947368, + 0.9230769230769231, + 0.8983050847457628, + 0.9121621621621622, + 0.906896551724138, + 0.9006849315068494, + 0.912621359223301, + 0.9029850746268657, + 0.8875379939209727, + 0.9010989010989011, + 0.8990536277602523, + 0.9007633587786259, + 0.890282131661442, + 0.8461538461538461, + 0.9026845637583892, + 0.9010238907849829, + 0.9243986254295533, + 0.9045801526717557, + 0.9222222222222223, + 0.8950819672131147, + 0.8885135135135135, + 0.9261992619926199, + 0.899641577060932, + 0.8829787234042553, + 0.9013605442176871, + 0.9051094890510949, + 0.9117647058823529, + 0.875, + 0.9081632653061225, + 0.8880597014925373, + 0.9081632653061225, + 0.889273356401384, + 0.9205298013245033, + 0.886986301369863, + 0.8767605633802817, + 0.9137931034482759, + 0.9245901639344263, + 0.9139072847682119, + 0.8907849829351536, + 0.889261744966443, + 0.9272151898734177, + 0.8851351351351351, + 0.9161073825503355, + 0.8907849829351536, + 0.9117647058823529, + 0.896797153024911, + 0.9096774193548387, + 0.898989898989899, + 0.9271523178807947, + 0.910394265232975, + 0.911864406779661, + 0.9003215434083601, + 0.9096989966555183, + 0.9172413793103448, + 0.9149659863945578, + 0.9006622516556292, + 0.9177215189873418, + 0.9041095890410958, + 0.889967637540453, + 0.8947368421052632, + 0.8978328173374613, + 0.9075342465753424, + 0.9084967320261438, + 0.9071428571428571, + 0.9082278481012658, + 0.9128919860627178, + 0.9106529209621993, + 0.9017543859649123, + 0.8945578231292517, + 0.8986486486486487, + 0.9166666666666666, + 0.9152542372881356, + 0.889273356401384, + 0.9072847682119205, + 0.9050632911392406, + 0.8733333333333333, + 0.8918032786885246, + 0.900355871886121, + 0.9041533546325878, + 0.9249146757679181, + 0.9333333333333333, + 0.9026217228464419, + 0.9130434782608695, + 0.9069767441860465, + 0.9256756756756757, + 0.8964285714285715, + 0.8993288590604027, + 0.9090909090909091, + 0.8983050847457628, + 0.9047619047619048, + 0.8952702702702703, + 0.896551724137931, + 0.8716216216216216, + 0.8700361010830325, + 0.9044368600682594, + 0.8964285714285715, + 0.9084745762711864, + 0.8897058823529411, + 0.9144736842105263, + 0.8785425101214575, + 1.0, + 0.9065743944636678, + 0.8780487804878049, + 0.9166666666666666, + 0.917910447761194, + 0.894927536231884, + 0.906896551724138, + 0.9024390243902439, + 0.8947368421052632, + 0.9017543859649123, + 0.9096989966555183, + 1.0, + 0.9288135593220339, + 0.9029850746268657, + 0.9177631578947368, + 0.9051094890510949, + 0.8880597014925373, + 0.9227941176470589, + 0.8689655172413793, + 0.887719298245614, + 0.8702290076335878, + 0.8976377952755905, + 0.8993055555555556, + 0.875, + 0.8798586572438163, + 0.8740740740740741, + 0.900709219858156, + 0.8937728937728938, + 0.9285714285714286, + 0.9172413793103448, + 0.9072164948453608, + 0.8863636363636364, + 0.9314079422382672, + 0.8869257950530035, + 0.916058394160584, + 0.8977272727272727, + 0.8888888888888888, + 0.9090909090909091, + 0.9024390243902439, + 0.8905660377358491, + 0.8848920863309353, + 0.903114186851211, + 0.9337748344370861, + 0.9018867924528302, + 0.887719298245614, + 0.8901098901098901, + 0.8737201365187713, + 0.9210526315789473, + 0.9143835616438356, + 0.8884615384615384, + 0.9260700389105059, + 0.894927536231884, + 0.89198606271777, + 0.875, + 0.8916083916083916, + 0.8954703832752613, + 0.8805970149253731, + 0.9040590405904059, + 0.9032258064516129, + 0.9049295774647887, + 0.9261992619926199, + 0.9143835616438356, + 0.8888888888888888, + 0.8781362007168458, + 0.9278688524590164, + 0.8993055555555556, + 0.9264705882352942, + 0.8928571428571429, + 0.8697183098591549, + 0.8850174216027874, + 0.8586572438162544, + 0.8529411764705882, + 0.8958333333333334, + 0.9150326797385621, + 0.9, + 0.9922480620155039, + 0.8914473684210527, + 0.8813559322033898, + 0.8870431893687708, + 0.8851351351351351, + 0.9131832797427653, + 0.901060070671378, + 0.8717105263157895, + 1.0, + 0.9924242424242424, + 0.9042904290429042, + 0.9090909090909091, + 0.9923076923076923, + 0.9, + 0.8754578754578755, + 0.8759124087591241, + 0.8805460750853242, + 0.8940397350993378, + 0.8689655172413793, + 0.889261744966443, + 1.0, + 0.8850931677018633, + 0.8762886597938144, + 1.0, + 0.87109375, + 1.0, + 0.8945454545454545, + 0.887719298245614, + 0.897887323943662, + 0.8888888888888888, + 0.9927536231884058, + 0.8913043478260869, + 0.8838028169014085, + 0.8701754385964913, + 1.0, + 0.8787878787878788, + 1.0, + 0.8636363636363636, + 1.0, + 0.88, + 0.8926174496644296, + 1.0, + 0.8698884758364313, + 0.8786885245901639, + 1.0, + 1.0, + 0.8794326241134752, + 0.8921568627450981, + 0.8725490196078431, + 0.8821548821548821, + 1.0, + 0.8657243816254417, + 0.9844961240310077, + 0.8701298701298701, + 0.8865979381443299, + 0.9861111111111112, + 0.8884615384615384, + 0.864951768488746, + 0.8727272727272727, + 0.9927536231884058, + 0.8571428571428571, + 0.8996655518394648, + 0.9917355371900827, + 0.9049180327868852, + 0.8666666666666667, + 0.8915254237288136, + 1.0, + 1.0, + 0.9922480620155039, + 0.9916666666666667, + 0.8870431893687708, + 1.0, + 0.9921259842519685, + 0.9047619047619048, + 0.900709219858156, + 0.8878205128205128, + 1.0, + 0.9010238907849829, + 0.9041095890410958, + 1.0, + 0.9846153846153847, + 0.8815331010452961, + 0.8840579710144928, + 0.8795986622073578, + 0.8851351351351351, + 0.863481228668942, + 0.8843537414965986, + 0.89419795221843, + 0.8719723183391004, + 0.8813559322033898, + 0.8718861209964412, + 0.89171974522293, + 0.8708487084870848, + 1.0, + 0.8912280701754386, + 0.8783783783783784, + 0.8785714285714286, + 0.8888888888888888, + 0.9047619047619048, + 0.9166666666666666, + 0.9017543859649123, + 0.8724137931034482, + 0.8786885245901639, + 0.8851351351351351, + 0.8913043478260869, + 0.8703071672354948, + 1.0, + 0.887459807073955, + 0.8918918918918919, + 0.907051282051282, + 1.0, + 0.8849840255591054, + 0.899641577060932, + 1.0, + 0.8969072164948454, + 0.9850746268656716, + 0.8974358974358975, + 0.8888888888888888, + 0.8877887788778878, + 0.8840579710144928, + 0.9115646258503401, + 0.8732876712328768, + 0.8636363636363636, + 0.9858156028368794, + 1.0, + 0.8627450980392157, + 0.8809523809523809, + 1.0, + 0.8664259927797834, + 1.0, + 0.8870431893687708, + 0.8791946308724832, + 0.8690909090909091, + 0.8803986710963455, + 1.0, + 1.0, + 1.0, + 0.9841269841269841, + 0.8561643835616438, + 1.0, + 0.8858131487889274, + 0.926984126984127, + 1.0, + 1.0, + 0.897887323943662, + 0.8903654485049833, + 0.8925925925925926, + 1.0, + 0.8972602739726028, + 0.8888888888888888, + 0.8581560283687943, + 0.9919354838709677, + 0.9267515923566879, + 1.0, + 0.9922480620155039, + 0.8839590443686007, + 1.0, + 0.8737864077669902, + 1.0, + 1.0, + 0.8841698841698842, + 0.9025974025974026, + 0.8900343642611683, + 0.8831615120274914, + 0.8626760563380281, + 1.0, + 0.8675496688741722, + 0.9097744360902256, + 0.9122807017543859, + 0.9094202898550725, + 0.9214285714285714, + 0.9272151898734177, + 0.896797153024911, + 0.9163763066202091, + 0.9018867924528302, + 0.9215017064846417, + 0.9244604316546763, + 0.9238754325259516, + 0.8953068592057761, + 0.9006849315068494, + 0.9143835616438356, + 0.8996282527881041, + 0.8717948717948718, + 0.9084507042253521, + 0.8962962962962963, + 0.9071428571428571, + 0.8976109215017065, + 0.9090909090909091, + 0.9045936395759717, + 0.9297658862876255, + 0.9210526315789473, + 0.9035714285714286, + 0.9309090909090909, + 0.9105960264900662, + 0.9094076655052264, + 0.9032258064516129, + 0.9011857707509882, + 0.9087719298245615, + 0.9007352941176471, + 0.9153225806451613, + 0.9184397163120568, + 0.9304029304029304, + 0.9326241134751773, + 0.9083665338645418, + 0.8971631205673759, + 0.9020979020979021, + 0.9028776978417267, + 0.8943661971830986, + 0.9343065693430657, + 0.911660777385159, + 0.9215686274509803, + 0.9278350515463918, + 0.9244604316546763, + 0.9210526315789473, + 0.9198473282442748, + 0.909433962264151, + 0.926923076923077, + 0.9236947791164659, + 0.9442379182156134, + 0.9053030303030303, + 0.9181494661921709, + 0.9135338345864662, + 0.9107806691449815, + 0.9460431654676259, + 0.9283154121863799, + 0.896797153024911, + 0.894927536231884, + 0.9024390243902439, + 0.9138576779026217, + 0.9419795221843004, + 0.9230769230769231, + 0.9169675090252708, + 0.9117647058823529, + 0.9071428571428571, + 0.9073359073359073, + 0.9239543726235742, + 0.910394265232975, + 0.8905109489051095, + 0.894927536231884, + 0.9220338983050848, + 0.910394265232975, + 0.899641577060932, + 0.9130434782608695, + 0.9172413793103448, + 0.9045936395759717, + 0.9028776978417267, + 0.9018867924528302, + 0.903114186851211, + 0.8972602739726028, + 0.8935361216730038, + 0.9074074074074074, + 0.925531914893617, + 0.9191176470588235, + 0.9122137404580153, + 0.9166666666666666, + 0.8905109489051095, + 0.9084249084249084, + 0.9059233449477352, + 0.8896797153024911, + 0.906896551724138, + 0.9182156133828996, + 0.9087719298245615, + 0.9139784946236559, + 0.9238754325259516, + 0.8996282527881041, + 0.9006849315068494, + 0.899641577060932, + 0.9003831417624522, + 0.8978102189781022, + 0.9166666666666666, + 0.925, + 0.9166666666666666, + 0.9236111111111112, + 0.9169811320754717, + 0.9037037037037037, + 0.9122807017543859, + 0.9039145907473309, + 0.9084507042253521, + 0.9176029962546817, + 0.9172185430463576, + 0.9068100358422939, + 0.8853046594982079, + 0.9076923076923077, + 0.9187279151943463, + 0.9153846153846154, + 0.8814229249011858, + 0.915129151291513, + 0.9122807017543859, + 0.9175627240143369, + 0.9010989010989011, + 0.9151943462897526, + 0.9070631970260223, + 1.0, + 0.9361702127659575, + 0.9142857142857143, + 0.9290780141843972, + 0.9090909090909091, + 0.927797833935018, + 0.9265734265734266, + 0.9151943462897526, + 0.9210526315789473, + 0.9024390243902439, + 0.9007633587786259, + 1.0, + 0.9337979094076655, + 0.8933823529411765, + 0.9440559440559441, + 0.9512195121951219, + 0.9107806691449815, + 0.9042553191489362, + 0.9090909090909091, + 0.9113475177304965, + 0.9182156133828996, + 0.914179104477612, + 0.9026217228464419, + 0.9205776173285198, + 0.9157509157509157, + 0.8884615384615384, + 0.9077490774907749, + 0.9283154121863799, + 0.9188191881918819, + 0.9320754716981132, + 0.9241379310344827, + 0.9267399267399268, + 0.9323308270676691, + 0.9256756756756757, + 0.9222222222222223, + 0.9239130434782609, + 0.8937728937728938, + 0.912621359223301, + 0.8986013986013986, + 0.9176029962546817, + 0.9173228346456693, + 0.8972602739726028, + 0.942652329749104, + 0.9183673469387755, + 0.9182156133828996, + 0.927536231884058, + 0.9111111111111111, + 0.8953488372093024, + 0.9222614840989399, + 0.9077490774907749, + 0.9471544715447154, + 0.900709219858156, + 0.9107142857142857, + 0.9265734265734266, + 0.9051094890510949, + 0.9233449477351916, + 0.9144981412639405, + 0.9264705882352942, + 0.9047619047619048, + 0.9154929577464789, + 0.9259259259259259, + 0.895910780669145, + 0.920863309352518, + 0.8873239436619719, + 0.9214285714285714, + 0.9037800687285223, + 0.9114391143911439, + 0.911660777385159, + 0.8929889298892989, + 0.8888888888888888, + 0.89568345323741, + 0.8873239436619719, + 0.9038461538461539, + 0.9193548387096774, + 0.9078498293515358, + 0.9829059829059829, + 0.8900709219858156, + 0.8888888888888888, + 0.8917910447761194, + 0.8901098901098901, + 0.9212328767123288, + 0.8877551020408163, + 0.8904593639575972, + 0.9920634920634921, + 0.9833333333333333, + 0.9247648902821317, + 0.9185667752442996, + 1.0, + 0.9087719298245615, + 0.889273356401384, + 0.8923076923076924, + 0.9044368600682594, + 0.8989547038327527, + 0.896797153024911, + 0.9151943462897526, + 0.9921875, + 0.910394265232975, + 0.8925925925925926, + 0.9923076923076923, + 0.9029850746268657, + 1.0, + 0.9016949152542373, + 0.8932384341637011, + 0.9087837837837838, + 0.8982456140350877, + 1.0, + 0.9084249084249084, + 0.9228070175438596, + 0.8823529411764706, + 0.9844961240310077, + 0.9145907473309609, + 1.0, + 0.8869257950530035, + 1.0, + 0.9138576779026217, + 0.8945454545454545, + 0.9924812030075187, + 0.9018181818181819, + 0.9175627240143369, + 1.0, + 0.968, + 0.9206896551724137, + 0.901060070671378, + 0.8881578947368421, + 0.9023569023569024, + 0.9915254237288136, + 0.8910505836575876, + 0.9841269841269841, + 0.8936170212765957, + 0.8843283582089553, + 0.9920634920634921, + 0.9040590405904059, + 0.9020979020979021, + 0.8847583643122676, + 0.9922480620155039, + 0.8985507246376812, + 0.9049295774647887, + 1.0, + 0.9010238907849829, + 0.903010033444816, + 0.910394265232975, + 0.9852941176470589, + 0.9919354838709677, + 0.9920634920634921, + 1.0, + 0.8996539792387543, + 0.9908256880733946, + 0.9908256880733946, + 0.900709219858156, + 0.8945578231292517, + 0.9054054054054054, + 0.9922480620155039, + 0.9090909090909091, + 0.890728476821192, + 1.0, + 1.0, + 0.9204152249134948, + 0.9140893470790378, + 0.8945454545454545, + 0.9075907590759076, + 0.8912280701754386, + 0.8868613138686131, + 0.8860294117647058, + 0.9081632653061225, + 0.8867924528301887, + 0.9018867924528302, + 0.9084745762711864, + 0.8971631205673759, + 1.0, + 0.9305555555555556, + 0.910394265232975, + 0.9115646258503401, + 0.8982456140350877, + 0.9265734265734266, + 0.9020979020979021, + 0.9006849315068494, + 0.9010989010989011, + 0.9093959731543624, + 0.9151943462897526, + 0.8962962962962963, + 0.9054545454545454, + 1.0, + 0.8964285714285715, + 0.9035714285714286, + 0.91, + 1.0, + 0.9204152249134948, + 0.9130434782608695, + 0.9672131147540983, + 0.9114754098360656, + 1.0, + 0.9169435215946844, + 0.9217081850533808, + 0.9222972972972973, + 0.9215686274509803, + 0.8938356164383562, + 0.8843283582089553, + 0.8821292775665399, + 0.9918032786885246, + 0.991869918699187, + 0.8943661971830986, + 0.8888888888888888, + 0.9922480620155039, + 0.896797153024911, + 1.0, + 0.8976109215017065, + 0.8961937716262975, + 0.9020979020979021, + 0.8933823529411765, + 0.9781021897810219, + 0.9921259842519685, + 0.9915254237288136, + 1.0, + 0.9049295774647887, + 0.991304347826087, + 0.8884615384615384, + 0.9158249158249159, + 0.967479674796748, + 0.9846153846153847, + 0.8924731182795699, + 0.9192982456140351, + 0.9035714285714286, + 1.0, + 0.9247311827956989, + 0.9163763066202091, + 0.8976109215017065, + 1.0, + 0.922077922077922, + 0.983739837398374, + 0.9834710743801653, + 0.9054545454545454, + 0.9833333333333333, + 0.9056603773584906, + 1.0, + 1.0, + 0.8819188191881919, + 0.9065743944636678, + 0.9144736842105263, + 0.8986013986013986, + 0.9016949152542373, + 0.9915966386554622, + 0.9016949152542373 + ] + }, + "contact_maps": { + "mean_pearson": 0.1348292448886617, + "valid_tracks": 28, + "per_track_pearson": [ + 0.1480469600540382, + 0.13587073463442906, + 0.12985699737883036, + 0.04583933639668546, + 0.08361201863275683, + 0.08139833866982622, + 0.14520614930163683, + 0.11516511269788769, + 0.2450126987941672, + 0.18836399039982132, + 0.06404968605242306, + 0.11806815763723906, + 0.21358056304898168, + 0.05160177684832484, + 0.09815504988245578, + 0.1856335589463887, + 0.0560958998832771, + 0.18275321159413482, + 0.1569047471584472, + 0.16237147790310713, + 0.07534072803883315, + 0.08617589286717828, + 0.07955839764264477, + 0.18655053635865737, + 0.17938588460079047, + 0.19393938089940574, + 0.13112025181539358, + 0.23556131874476585 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.3450137695882094, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..14936e98960921f7b219e7b1197ea5c9d1214170 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S015/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5851419159369071, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6858629037200465, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.16535341782125368, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7579046379406581, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3230399618448669, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.2424778355181878, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.19698733159616444, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.3510159899244791, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.16760028985755937, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.23552216690812636, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.7072691711793893, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.13437076063427897, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.1815651064372292, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6027d6b978ff4d897bd09804a48192b746e2fd40 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/README.md @@ -0,0 +1,98 @@ +# SeqFunc context15 wider-context continuation with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,265,544 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Continuation of context09 from 44,000 through 53,700 training examples with unchanged model, balanced-splice objective, batch one and original four-epoch learning-rate schedule. Context09 extended the locally trained 40,000-example model from a 701-bp to a 10,941-bp local receptive field with two identity-initialized residual blocks. The matched local01 control sees the same complete interval order and learning rates. Optimizer, RNG and sampler state are restored from our own checkpoint; no pretrained weights are used. At saved example 50,600, loader workers increased from2 to4 after an I/O-bound utilization check; model, optimizer, interval order, objective and LR schedule remain unchanged. Checkpoint SHA256 0cb79bd2dbe25eb650a65222425ef06591f816de0c2b9e6322b14abf5c2abe4c. + +This checkpoint records **53,700 updates**, 1 complete training permutations and +12,680 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 4-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +500 warmup training examples. The objective is `balanced_splice`. +Other family losses follow the reference. Recorded accumulated training-loop wall time is +209.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +## Measured validation results + +Family-mean Pearson: **0.366055** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.531940 | +| dnase | 0.635852 | +| procap | 0.446963 | +| cage | 0.235722 | +| rna_seq | 0.391042 | +| chip_tf | 0.443555 | +| chip_histone | 0.447471 | +| splice_sites | 0.193183 | +| splice_usage | 0.313293 | +| junctions | 0.245515 | +| contact_maps | 0.142068 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.6236 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6202 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.2737 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.6588 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3474 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2440 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2476 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3711 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1541 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2437 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.7463 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1431 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2114 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..65036e132a92042fa6dc0ce304f5876d3cb9d842 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/context15/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..94a570dc6fd8ec055c1690e8e910f5beeb22b997 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/package_manifest.json @@ -0,0 +1,34 @@ +{ + "parameters": 13265544, + "source_checkpoint_sha256": "648c34b7a72ec2e5762743d528e0438d2c4d1e7515334265ea8d8ea71691b970", + "files": { + "README.md": "8679f6a09d4ef07356addc19b18a627693c4d386b178664180073a41e535166f", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "9831aa062f8214ce09f9ae54bf027f96d2a9f8c1ccf6938d9e9b2355b3638971", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "reference/alphagenome_50m/layers.py": "26a116a72aa3f0e307a3606163547edbdecb7965f7fdcd96bc5f805cd21b3352", + "reference/alphagenome_50m/model.py": "bcd41eba9c2f952250f76aac8812a9dd995ee206791fbe640c751fa073428b8b", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "a71ed8c14bb3dee86bdba5967e4152192bd35dad78d32934a1aaab17a4719835", + "training_provenance.json": "ace2792f596ff5b9b56c8c46cb3c918809fe8e991891e7d1e720042b99fd7eaa", + "decoding_policy.json": "6bec4bff9dfa561892044ddd7e1b82d348480f7397be0d81a6bdfa3e28e0a4dc", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "validation_tracks.json": "df42b46a1d2bbe52cebcc8b691167a5c31438b786c86b5069e9b25b8a51210f8", + "validation_variants.json": "ae09c403e65427098e13c0ab49e6d28ba6b2527d406f60508876de122d302bfd" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/dataset.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..af31f86a5589a16f0007e6aca5ca7a7bd552d6a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/layers.py @@ -0,0 +1,124 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + self.attention(x, attention_bias) + return x + self.mlp(x) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3c2effcf45b2aa0eec7e31aa3f668107d7ff90 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/model.py @@ -0,0 +1,430 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e5a5cee0af43ebd5a92833bed8d5dcc8b264a8d3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_config.yaml @@ -0,0 +1,29 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..38086af671c60e7cd9bd07e80c2f76f994c413e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/training_provenance.json @@ -0,0 +1,91 @@ +{ + "schema": "seqfunc-single/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6 + }, + "epoch": 1, + "offset": 12680, + "global_step": 53700, + "examples": 41020, + "schedule_epochs": 4, + "objective": "balanced_splice", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/context15", + "config": "/workspace/research/experiments/context15/config.yaml", + "resume": "/workspace/research/experiments/context15/progress.pt", + "objective": "balanced_splice", + "epochs": 4, + "stop_at": 53700, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 500, + "save_every": 100, + "keep_every": 10000, + "log_every": 100 + }, + "elapsed_seconds": 12576.76766523812, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "device": "NVIDIA H100 80GB HBM3", + "world_size": 1 + }, + "intervention": { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + }, + "interventions": [ + { + "kind": "extended_local_context", + "parent_sha256": "72ef8003d6d4a54f8b583bbc01b7300ecbd141a5a4fdb6598f43e190d9dd5eff", + "step": 40000, + "blocks_before": 4, + "blocks_after": 6, + "receptive_field_bp": 10941, + "parent_parameters_optimizer_moments_rng_preserved": true, + "new_blocks_initially_identity": true, + "optimizer_ids_remapped_by_parameter_name": true + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..4f8fa519d9909c9745c4b853b24323a68a26ff9f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5319402128547635, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6557230396933296, + 0.6761758628574831, + 0.4218739615377023, + 0.6019903664965628, + 0.427800599612794, + 0.42456724648225186, + 0.4430771284275712, + 0.5254022058760045, + 0.5368005096891313, + 0.5965815186365716, + 0.5323755377078627, + 0.5041077140108507, + 0.3544522215016494, + 0.48060765955393386, + 0.6427385318213762, + 0.4456669641527201, + 0.5031047663532744, + 0.4645365210741006, + 0.5027727307970686, + 0.4846544014098977, + 0.45936022193509185, + 0.41978756787416244, + 0.4869538072855864, + 0.4745237675735944, + 0.46254169659193356, + 0.4479279927967282, + 0.43032915217648954, + 0.4534282070860328, + 0.46170430903596466, + 0.5126192347373861, + 0.4383572417468503, + 0.46648390493112096, + 0.43080002634125786, + 0.35490636631820127, + 0.3235676473066472, + 0.5270221121697425, + 0.4667507306545283, + 0.5722197985896277, + 0.4627569930992898, + 0.4497280399625353, + 0.5251936692710608, + 0.5239059061052619, + 0.44757121970341135, + 0.42472972343514404, + 0.5114487016503017, + 0.5035453341319648, + 0.3867489441577972, + 0.4356184516578271, + 0.4282779383877523, + 0.4762107501817519, + 0.45102082758714873, + 0.40980807498789035, + 0.49547328901801796, + 0.5594184431631346, + 0.41042146389278145, + 0.6617109117662592, + 0.7568217531465165, + 0.5029185522070662, + 0.5630155790302989, + 0.5739395057492468, + 0.6998714051634811, + 0.43952598731297043, + 0.6714832379302467, + 0.5936679836117332, + 0.5942738169734019, + 0.46067128801505913, + 0.5052287109584526, + 0.5932433138599249, + 0.7416740861939446, + 0.6663261170196255, + 0.487237512863913, + 0.4328391175561354, + 0.4009906704154259, + 0.4858262583021083, + 0.45698543617530013, + 0.44178301324729696, + 0.4560391205236997, + 0.4943556170950002, + 0.5318412161203128, + 0.4463651364026838, + 0.5214216490349769, + 0.548513229391082, + 0.4689285248144755, + 0.44178592257171634, + 0.46757125733413785, + 0.4570270020379543, + 0.4369566151792549, + 0.43208923721441406, + 0.40907704077657714, + 0.4623521588426563, + 0.45796175662985894, + 0.4582116248107489, + 0.4499064698946058, + 0.4622862785588928, + 0.52605371805666, + 0.3493768154036497, + 0.396095714711924, + 0.38694155113168843, + 0.43886807335912675, + 0.4592683171342139, + 0.41144668575305576, + 0.4167817480457173, + 0.45452350545481685, + 0.4233526999775967, + 0.49229679353791955, + 0.49926909861391444, + 0.4902312980868821, + 0.46987536481732667, + 0.49049764080102515, + 0.49221577812751394, + 0.5452367490646066, + 0.47431622929775635, + 0.4709174637945653, + 0.395700075090323, + 0.4208116786451012, + 0.3929659175352726, + 0.42999367137934363, + 0.44146253331906143, + 0.44605219514569205, + 0.4286402193455868, + 0.6184609518266584, + 0.6202416730041236, + 0.6888579064006755, + 0.6854997384597543, + 0.6661113595590887, + 0.5457858998038707, + 0.8212781861082484, + 0.6744800976147652, + 0.7212594298539512, + 0.6470022333152389, + 0.6933249721806645, + 0.5322918243757764, + 0.557889747332791, + 0.5332849122936122, + 0.6788310111787025, + 0.6891676115347142, + 0.6608518863537027, + 0.5011292786612217, + 0.40075071372493964, + 0.7722164975534046, + 0.7414680530789404, + 0.7571375907537392, + 0.7495065699871432, + 0.5518772908967775, + 0.821178631382313, + 0.7309959008636165, + 0.6841955325885754, + 0.6876807556644104, + 0.8035427119487025, + 0.767464789715178, + 0.5823933128258959, + 0.590898764775407, + 0.6712192105275977, + 0.7038592704874967, + 0.7165946656413102, + 0.5720958036511258, + 0.6997442615956995, + 0.561159807055323, + 0.5693191453721557, + 0.6494034589163604, + 0.5404244325989396, + 0.7289889151736424, + 0.7414073968756112, + 0.6006335123192977, + 0.5661538471392236, + 0.5434069934056367, + 0.7004622967715091 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.635852294696373, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5439051436928095, + 0.9188541262946383, + 0.7813086620733867, + 0.4984192688960664, + 0.48800264641427643, + 0.502491794347101, + 0.93533087679236, + 0.7117338467638836, + 0.42350849436644494, + 0.3315267727998384, + 0.8788199592176594, + 0.6903311020148062, + 0.415321025678689, + 0.49923162958374345, + 0.4747711946447414, + 0.8027818023230942, + 0.7857341706662534, + 0.8477637383187041, + 0.8578890271768569, + 0.8884232069387998, + 0.6758463657822295, + 0.6058430872537794, + 0.8998192738336298, + 0.5004958888607027, + 0.5206768851661658, + 0.7779192306116783, + 0.44774801337212783, + 0.6545143568330885, + 0.5007673069765234, + 0.5311962436523736, + 0.6174530189221148, + 0.44939867104584463, + 0.8389561163980402, + 0.6232635089117218, + 0.8786335517349458, + 0.8866192522386706, + 0.43347854326123647, + 0.8247035349907746, + 0.8744011887957687, + 0.8578799120678338, + 0.824689230876659, + 0.6668565924809079, + 0.7024843243342322, + 0.627336724852215, + 0.5094373754007401, + 0.633812085197159, + 0.7307114763017418, + 0.49800390258499944, + 0.581632267849633, + 0.3937363414435243, + 0.364382625105448, + 0.5540727032092623, + 0.4825022697067885, + 0.8270238199712238, + 0.4197075292240151, + 0.3963741880641212, + 0.44040179824198594, + 0.48212294492267077, + 0.45252261624093815, + 0.5159283014294684, + 0.4257918381179335, + 0.6076077768887767, + 0.4757608089259062, + 0.4630004656497802, + 0.4919271796316438, + 0.49892707467332176, + 0.5399281557208887, + 0.4587777497132562, + 0.36902588368057027, + 0.4124201186187749, + 0.5029790595590442, + 0.6307729279243778, + 0.48284788616517377, + 0.6201575192361346, + 0.3432710298129348, + 0.8088696642683617, + 0.8929151420979167, + 0.6510748718859295, + 0.43033648978624445, + 0.44640737467559327, + 0.5010008253357959, + 0.49099388438733227, + 0.46493714155645954, + 0.4713691578317445, + 0.5485763933152432, + 0.4879784271575963, + 0.5939409315509302, + 0.4960618490564477, + 0.6315333081844512, + 0.4582710765744151, + 0.6212158206654497, + 0.40428074793050933, + 0.7223846179013995, + 0.4606206815055604, + 0.7103493236748881, + 0.4894822964867905, + 0.7329827874376725, + 0.6982861045561918, + 0.35111374075468477, + 0.35687324377887353, + 0.45887012419747925, + 0.3660998568253373, + 0.3481270266877246, + 0.7136943532244209, + 0.6770272649243098, + 0.4058669345658758, + 0.45582491099864897, + 0.5786674404980398, + 0.8473998392771055, + 0.5522033777760307, + 0.5431493380027413, + 0.5712769514453828, + 0.4911530220182629, + 0.44054283223180457, + 0.7757177223501489, + 0.5458651239918756, + 0.4868457182089851, + 0.5458463276329799, + 0.6222330212367783, + 0.4434670626065397, + 0.5268806014143176, + 0.5029649697116001, + 0.5430046817882959, + 0.533184062343191, + 0.6026094275213043, + 0.5508253689398959, + 0.622015394987301, + 0.6799968289542813, + 0.6947537868604285, + 0.6104918717246236, + 0.4580692112762369, + 0.6067048828700093, + 0.6257166836469362, + 0.7894886740562234, + 0.44419560573483474, + 0.31997103534757987, + 0.5820736743620487, + 0.47081988691111065, + 0.6385465989970053, + 0.6014003652382166, + 0.6113079246312981, + 0.5564146026947797, + 0.4645444101627355, + 0.7746931406217826, + 0.6070166006556117, + 0.7612843914355181, + 0.38693702983185235, + 0.3700082332177491, + 0.393114656727807, + 0.339469612435886, + 0.8917904743206946, + 0.32247202522653107, + 0.8961132328532204, + 0.7403618895198618, + 0.46933600923069013, + 0.33726424874640354, + 0.349439205688831, + 0.36481081884680233, + 0.639959368900194, + 0.735078560356949, + 0.7329128418271631, + 0.6594956326111928, + 0.5610557369392097, + 0.46736102399224433, + 0.6812215025393629, + 0.6122816117467049, + 0.5087490410901929, + 0.6295254668700487, + 0.4536970363948329, + 0.4843241765819893, + 0.39698555870835217, + 0.8285994311775864, + 0.442385781579187, + 0.4679757823004914, + 0.6549753332741708, + 0.6927626167368047, + 0.4965831282392008, + 0.3469292417617062, + 0.8730063441548883, + 0.3672830733919599, + 0.7672882838607756, + 0.4674432855319962, + 0.590424931916964, + 0.7174020463536663, + 0.45431451742041074, + 0.5661379771441051, + 0.5911975990868988, + 0.8176013493140597, + 0.4155737657270168, + 0.6091556661479535, + 0.5527956492141621, + 0.7156043274219868, + 0.8940959636219625, + 0.4178877143881283, + 0.583945805406047, + 0.4719505299602729, + 0.6268178080021619, + 0.7048373700680735, + 0.563172048250355, + 0.44537711812116604, + 0.8684049258198546, + 0.8810071140723662, + 0.7519604321746799, + 0.8332922363808435, + 0.8437652194808765, + 0.8339340529885039, + 0.8458617233490121, + 0.36103712245768865, + 0.3790096122393707, + 0.5091009702635306, + 0.4605819875731501, + 0.4500893207183798, + 0.6013535966962296, + 0.8620002667545165, + 0.8462152911182454, + 0.6760653564966705, + 0.8253655838666196, + 0.825503227145697, + 0.8557297105505245, + 0.9030912258827686, + 0.5103844703248924, + 0.5857036890140643, + 0.6679681371674653, + 0.6070146562112234, + 0.8221271279558825, + 0.569581437641818, + 0.9081868737647278, + 0.8805032017544587, + 0.8698547677163732, + 0.8989487581439912, + 0.8908787237159818, + 0.8717369071008987, + 0.829863532718145, + 0.8564382310934756, + 0.9046798749038413, + 0.44493000021769125, + 0.9224113032798871, + 0.8759516591011564, + 0.8189853619753068, + 0.594773861130171, + 0.8701167704364648, + 0.5150630458433965, + 0.7481505537056238, + 0.8719461749348273, + 0.5847114074647028, + 0.5716354051210922, + 0.6882337609671657, + 0.5220316157736139, + 0.5325633334625142, + 0.7290907690361592, + 0.881158325534253, + 0.7341584064028659, + 0.8118244912845353, + 0.6219097469450462, + 0.8628354881602563, + 0.804324323431442, + 0.8352638365711282, + 0.9016512514942313, + 0.8292445363831932, + 0.8951319358163561, + 0.8684550329916736, + 0.5314470480781747, + 0.8739068464440276, + 0.5616341855529929, + 0.5340973652225043, + 0.878644845330432, + 0.9024137138225955, + 0.877269506991091, + 0.5897859360516804, + 0.622740446714364, + 0.66470481806328, + 0.795018712951509, + 0.8888935378227318, + 0.7094424766483789, + 0.9060298034069926, + 0.9148003258190878, + 0.6157719964864714, + 0.32304925090759623, + 0.5187363698196413, + 0.896630304854565, + 0.4397511020411347, + 0.6148266375978831, + 0.8806874355487516, + 0.8281678795700003, + 0.6248706248816619, + 0.8631325139667835, + 0.8924728912146928, + 0.8865945273086526, + 0.8755252753274123, + 0.8635238085622555, + 0.8914267861939866, + 0.8542246206715218, + 0.9253195409808234, + 0.7806711952224007, + 0.8302511935806937, + 0.906197730345009, + 0.553593306857866, + 0.5625588906310989, + 0.551049642787432, + 0.5311722346347039, + 0.9048938433601355, + 0.7988394136183821, + 0.846692226158935, + 0.7945299319757956, + 0.764472025453458 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.44696318020775255, + "valid_tracks": 12, + "per_track_pearson": [ + 0.42892164789484155, + 0.4800946946221443, + 0.3825190460423405, + 0.4685390811497714, + 0.473457943515536, + 0.46821247112380293, + 0.4148719942184768, + 0.47303552382982167, + 0.3832652618110247, + 0.460521133026216, + 0.4557292565245923, + 0.4743901087344629 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.2357216070590708, + "valid_tracks": 546, + "per_track_pearson": [ + 0.21290788314650344, + 0.26086943256463524, + 0.23410339434410943, + 0.1442139296834477, + 0.08147914158811308, + 0.23305506547066393, + 0.12210515185357168, + 0.15450686414636242, + 0.2482629337289014, + 0.20126303523610498, + 0.2508768254105765, + 0.13694693829336688, + 0.1597742414545425, + 0.1329591903932345, + 0.10208322376975856, + 0.1821825460422286, + 0.20252650720310517, + 0.25025692499938035, + 0.26881214758908045, + 0.24162406084378055, + 0.20335648741317497, + 0.20563553671168888, + 0.3153119972731003, + 0.17631738136304495, + 0.1754907735042938, + 0.23822517124214326, + 0.23427967786340284, + 0.23940535759862558, + 0.2559279764268811, + 0.24286330442821666, + 0.1684351953559208, + 0.2685900265309831, + 0.19465597863366108, + 0.14784530330161422, + 0.2743049376415285, + 0.24315414177512565, + 0.2780836692772466, + 0.28551744367233106, + 0.27184652521343533, + 0.29031378299298743, + 0.3144301176917757, + 0.25779226987775766, + 0.23269690667524487, + 0.2409922997959703, + 0.14935343375536106, + 0.2007851552183081, + 0.21598761790449675, + 0.2359101168905296, + 0.14315290071309741, + 0.1841332590093953, + 0.2523056935151009, + 0.25411891144886894, + 0.2733657630357383, + 0.19431663332543067, + 0.2985083580430256, + 0.3123182967427338, + 0.1747675455460314, + 0.30165218420189616, + 0.21289875496104016, + 0.280638693482841, + 0.28694161764345705, + 0.19095423982803, + 0.253294227542341, + 0.19251992946098193, + 0.26683929152312214, + 0.28728860450814503, + 0.29542759855468714, + 0.19351663453448623, + 0.127748476634271, + 0.11814654721571781, + 0.22854392315846062, + 0.26369815209845493, + 0.2276037234859154, + 0.23995270289919024, + 0.26455861843528194, + 0.22105157964945654, + 0.2656194747407049, + 0.23898806306065298, + 0.2661374497973049, + 0.2745517770248489, + 0.22935136148629107, + 0.1390642120857306, + 0.24862237062528214, + 0.3064204145798915, + 0.1385072444457713, + 0.2528348228702733, + 0.21397410050650467, + 0.2988273958272699, + 0.1786667265382443, + 0.2616586804341382, + 0.19134481602587952, + 0.22829893507532745, + 0.23843898449517714, + 0.11973068354276872, + 0.2616883981469619, + 0.22088257522562624, + 0.24062985507027865, + 0.24980292284074954, + 0.2337576080646053, + 0.14180451761840246, + 0.22670721282949094, + 0.23664625161510786, + 0.2570767794810981, + 0.3258735869078506, + 0.2866872523495394, + 0.277814467205814, + 0.19451452383231005, + 0.24735276747838786, + 0.19524346322575653, + 0.20999883287146617, + 0.22115244680960627, + 0.2163107684283799, + 0.239155354349656, + 0.23736821244874265, + 0.2005996393099343, + 0.24536804550906666, + 0.23753773789758234, + 0.2483417770954475, + 0.2326986060976566, + 0.27636370571451246, + 0.26750393461041005, + 0.26523705159657857, + 0.15860913357160353, + 0.17922437366952385, + 0.21438171301680956, + 0.2541894051277251, + 0.23203174439964974, + 0.2528296919567053, + 0.3344676426288417, + 0.2647168206984186, + 0.2679183202876471, + 0.1864283862842239, + 0.2567677174724358, + 0.2914059684975533, + 0.3230321962853496, + 0.22752947244208763, + 0.2849160705206517, + 0.21234202094795046, + 0.13217242203150883, + 0.26024489812390117, + 0.29336959113237177, + 0.24101127148896867, + 0.2626965572203313, + 0.2080628719944386, + 0.21865481939842854, + 0.23604802318634682, + 0.25586708773839945, + 0.3073351886329163, + 0.22176412648239366, + 0.20566934389144503, + 0.26237197860567935, + 0.19939676501715198, + 0.23996707896871428, + 0.24926083415125405, + 0.23009206627970846, + 0.1924884358524836, + 0.2859656087902664, + 0.304007917461477, + 0.19196470581478947, + 0.17956492154118237, + 0.15685762614140553, + 0.2127370726268046, + 0.21745376826290325, + 0.25674421412828347, + 0.23352232127742062, + 0.18888158798392105, + 0.2363591309910867, + 0.17216267533223414, + 0.17568919488505275, + 0.2807235808466082, + 0.2389788733520023, + 0.18972612960633287, + 0.23623989654764432, + 0.3461806460945983, + 0.2433324510711705, + 0.16641849476796905, + 0.15846636087903487, + 0.22484022959769906, + 0.27050099032479363, + 0.30372015609884107, + 0.2984488015796289, + 0.30352663697651916, + 0.1616312133573091, + 0.27713452772500896, + 0.23981055384595068, + 0.19063546858218733, + 0.21266493973700373, + 0.16775451887479642, + 0.32456704989161445, + 0.24775754842755468, + 0.20241032754121027, + 0.1486940423931313, + 0.14765839412825196, + 0.22187762812675385, + 0.24066978640912695, + 0.20523102467140078, + 0.23747576141193952, + 0.23617080388168024, + 0.17489284073123276, + 0.23307255188559448, + 0.12998387625645144, + 0.12062514524846935, + 0.1277431508504526, + 0.12090658169101032, + 0.14116081850262815, + 0.2231020081272475, + 0.22839456054033744, + 0.18871807759882386, + 0.1226712503824748, + 0.20559974181031132, + 0.30147385000831195, + 0.2728537287914202, + 0.29427088752747604, + 0.2841910796901845, + 0.2573191885517044, + 0.23142946132411704, + 0.30795069573801037, + 0.3184831782599654, + 0.2695597505550881, + 0.2937502652917397, + 0.2651430982141527, + 0.19979675767773716, + 0.2735060891271132, + 0.24594613137202964, + 0.304518265732892, + 0.327686255645008, + 0.26433036977579444, + 0.23881171842588222, + 0.2517635744656735, + 0.27981798744242603, + 0.17347425710195355, + 0.17078986974630936, + 0.30412246845689567, + 0.2263114966412711, + 0.2032718024715401, + 0.2332366995576137, + 0.31533111074647885, + 0.18560183827627008, + 0.25318801981716543, + 0.24493241522940476, + 0.1251056573296991, + 0.17844786806998084, + 0.16359796487233405, + 0.2693320301943265, + 0.20222135287629597, + 0.25510897501710844, + 0.0998480967617346, + 0.3038067111080792, + 0.25394277638112617, + 0.27954425861146875, + 0.2710208159120444, + 0.16892588312326415, + 0.31781244781661644, + 0.14599169614481117, + 0.27357233252397717, + 0.23203749861797263, + 0.348313948556892, + 0.29902772760490876, + 0.30639284213218293, + 0.3702714916267437, + 0.2334127161968585, + 0.25840017303389917, + 0.13889982451686178, + 0.24608553165989452, + 0.26062591393871454, + 0.18190540685586443, + 0.14531697979842187, + 0.14571995470457907, + 0.17084557711387058, + 0.2573169780169138, + 0.3208284822055961, + 0.1615758129231643, + 0.13777251969300486, + 0.21703554941649122, + 0.3059705389784143, + 0.2535059535613684, + 0.16837752262711406, + 0.07692624954836422, + 0.25905621368663184, + 0.15756496257291425, + 0.18024756793229488, + 0.26574959998933406, + 0.2140879192708323, + 0.2596728634370876, + 0.15413499062107577, + 0.1791396607604906, + 0.15465041981422797, + 0.1354474621938998, + 0.16456026704398863, + 0.20831104107311596, + 0.26139124921855333, + 0.2751813122802737, + 0.24877119554147867, + 0.21092837769728415, + 0.2193384042845612, + 0.31925494651105235, + 0.1675061103828578, + 0.1860121605006318, + 0.2715841421184485, + 0.23704844705893607, + 0.24353208440984375, + 0.26786087485483473, + 0.25505206280604786, + 0.17037931259732608, + 0.2750515813137193, + 0.18057463778296942, + 0.15188971480711944, + 0.28044168414543486, + 0.2597694622922504, + 0.31668935083652544, + 0.30452072742200453, + 0.32115647957932225, + 0.3012676569316516, + 0.3205748043571311, + 0.26667948424287513, + 0.24878841429523435, + 0.25639811252675077, + 0.18505107526494236, + 0.227465671684308, + 0.25021601017551853, + 0.28124437834868815, + 0.1489538647862187, + 0.1955052749903435, + 0.2572954371796263, + 0.27276497276854444, + 0.2812462836865391, + 0.20318075416204726, + 0.3055511641643496, + 0.31021831577538295, + 0.19536267174890176, + 0.3107521501786113, + 0.21339478362009726, + 0.300818935762531, + 0.28854768379435713, + 0.19382105486315404, + 0.2616056091403593, + 0.21348665494867677, + 0.2785326214712633, + 0.2904536171858191, + 0.29598111215828943, + 0.19424292278818844, + 0.14085722251749758, + 0.1207808767798865, + 0.23140674997032884, + 0.27780649470635393, + 0.24169451685713683, + 0.2433627727325351, + 0.2750899415468389, + 0.23311593925963364, + 0.2641075193699519, + 0.24571834312366025, + 0.2694907193250894, + 0.2855512852472781, + 0.244738155294646, + 0.14588413278075926, + 0.2640620402446968, + 0.32112570109636934, + 0.14694623710514837, + 0.2589227422358436, + 0.2574688813614358, + 0.3195894577412187, + 0.18336361693425682, + 0.28552887759059264, + 0.2055761734612157, + 0.24658788410428742, + 0.25366910639126805, + 0.12773045151972656, + 0.2654562409412076, + 0.2181258723814508, + 0.2434612571628777, + 0.27702793361308947, + 0.23578732084826712, + 0.14295653199619973, + 0.2366748893858545, + 0.2556679759167307, + 0.26573379414249576, + 0.326726085421071, + 0.2851859851634601, + 0.287243581532298, + 0.2005166109696389, + 0.2533949777606716, + 0.1943413316303632, + 0.2094977632300672, + 0.21976614239823672, + 0.2256716975807752, + 0.2501293397306674, + 0.2580893671476238, + 0.2028267073102433, + 0.2591994150886512, + 0.23767031947200926, + 0.25149919052893144, + 0.23489970038305838, + 0.2826825595552453, + 0.2703530084860752, + 0.26576666025193796, + 0.18666053050738293, + 0.2094292903902413, + 0.2357721992690156, + 0.2718573705985961, + 0.23715120753464927, + 0.2732368879048443, + 0.34368713105758536, + 0.2944225190546336, + 0.27288216888922245, + 0.19252003290868758, + 0.2544613555172581, + 0.29180735611343495, + 0.3256746707612584, + 0.247303099353107, + 0.30172916776469855, + 0.23507791946613463, + 0.14937655930021057, + 0.27719226541415987, + 0.28716642606588394, + 0.23569688813305206, + 0.2895127859526591, + 0.21165599808049168, + 0.22714384439872015, + 0.23968480294278455, + 0.24748397359550808, + 0.30970179781239604, + 0.25247459009471007, + 0.21698235050462336, + 0.27340229061237936, + 0.21631781976853212, + 0.25207499071338074, + 0.2572977887135832, + 0.23795564242130987, + 0.22382103250535382, + 0.28916076380056976, + 0.3162222142939593, + 0.19749219469735793, + 0.20815296265545113, + 0.1731463592704396, + 0.20871718908990033, + 0.23038272362716644, + 0.2894941910274185, + 0.2644214789039891, + 0.19534732670624105, + 0.25083714102854493, + 0.2014984645655947, + 0.15912286630025543, + 0.31700683913749883, + 0.24996259888021163, + 0.21578888328046889, + 0.23515773380543756, + 0.35887068377947673, + 0.256487076487638, + 0.1536977433587734, + 0.17338079193799247, + 0.2512541697885908, + 0.24877288648294912, + 0.3257404584578654, + 0.3126156161568296, + 0.3254628338749846, + 0.15281657081558458, + 0.3006165871697473, + 0.27309180591011845, + 0.2286191064865749, + 0.2547258728923987, + 0.19005967620508316, + 0.317544529815052, + 0.2600919806378989, + 0.23157313240313174, + 0.19152508002592286, + 0.13472194918650304, + 0.18333898855797243, + 0.2742824078722019, + 0.23016815869634777, + 0.26991006201990314, + 0.2603969692489789, + 0.20041768273616128, + 0.23450047309413347, + 0.11480264494284854, + 0.08857816927852614, + 0.10835251178759889, + 0.1031986723602807, + 0.11112158456269079, + 0.25288448869389285, + 0.20144888221428262, + 0.1956766479328386, + 0.11592365674426501, + 0.22341803160753215, + 0.32171115813189527, + 0.3252486331417515, + 0.3300852218910294, + 0.3154434092795136, + 0.2978934876591563, + 0.25292885163946643, + 0.32357276666929424, + 0.3385477696621407, + 0.28965384484573936, + 0.3202050178313366, + 0.2943404474714668, + 0.23590343385059537, + 0.30450238985297295, + 0.27273515644112895, + 0.3438862098089296, + 0.34119533378717604, + 0.27141189515144765, + 0.26449708398752847, + 0.262151729037085, + 0.29148305726943974, + 0.14716914995731245, + 0.13791890000210563, + 0.3347521268292367, + 0.24375581561517293, + 0.23338254373600667, + 0.24919293985566557, + 0.34254349452936267, + 0.22073292807263, + 0.264657731807183, + 0.2508101949079358, + 0.11905192798992165, + 0.16956617008269492, + 0.1499119705035641, + 0.29336388730942514, + 0.20932789534669047, + 0.2757751971859954, + 0.11189888861530692, + 0.324523224581585, + 0.2757695872104182, + 0.3026412794486124, + 0.30073908804872457, + 0.20090070410408223, + 0.34804254498854026, + 0.1340427464308683, + 0.28457530634945283, + 0.2168134224221887, + 0.35113347454756055, + 0.31359791489661115, + 0.33124050920852566, + 0.37740405221502754, + 0.25536206699532865, + 0.261579193404191, + 0.1252837155373551, + 0.26290470777918856, + 0.27896881381473454, + 0.22345625794597507, + 0.1309504448238071, + 0.13223813573766027, + 0.16022937915604366, + 0.27363021059281933, + 0.3338795054468341, + 0.14507637197574355, + 0.11396870012983669 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.3910424259940242, + "valid_tracks": 613, + "per_track_pearson": [ + 0.3294750291832576, + 0.4272911507487793, + 0.38529061138522186, + 0.41291558246320254, + 0.42650613779811564, + 0.2304331839055166, + 0.2789193482553719, + 0.419520333845257, + 0.40207669630126086, + 0.4513663524049629, + 0.3918098866082702, + 0.3559975834575913, + 0.31064409467463744, + 0.40261289527275546, + 0.3626417351312882, + 0.38450468620394646, + 0.4757787450859073, + 0.38067470366188544, + 0.39615051567451676, + 0.3622395287002943, + 0.3168240539374141, + 0.41992634160274067, + 0.3160020888251661, + 0.2928749870375198, + 0.29178741252137785, + 0.4019537912875075, + 0.3704079854492818, + 0.4129983894600805, + 0.37981393103337735, + 0.39216957550820936, + 0.3580000436119569, + 0.3990680091125449, + 0.43227641609127937, + 0.3410720083780666, + 0.4288482846312778, + 0.3371568382854876, + 0.4118840284525566, + 0.3971828974377138, + 0.3642990717892403, + 0.3981259894705163, + 0.3954578533591446, + 0.37352592065916235, + 0.41400279938047463, + 0.39721661093181954, + 0.3345748961269642, + 0.27466333645069396, + 0.31187621478026517, + 0.40416864458855234, + 0.34501714735470607, + 0.4145220430052065, + 0.41449192152033554, + 0.45526768032722975, + 0.2670225281361277, + 0.3482778425431127, + 0.28167953586967487, + 0.3627254203655336, + 0.3438725314991441, + 0.40123425328819745, + 0.4513729393999733, + 0.42996231595474416, + 0.42927739631739414, + 0.41128575824301356, + 0.43440691349201804, + 0.2755737699680101, + 0.40738694630857974, + 0.3968725487629148, + 0.4295849749008481, + 0.4339751224687522, + 0.4183310956586669, + 0.4313539146490844, + 0.41385534645994343, + 0.4236742183278897, + 0.3800631206615701, + 0.4130167158134884, + 0.4018593941015277, + 0.40299925366946754, + 0.42413626040237973, + 0.42660798703606034, + 0.36311492803625767, + 0.42655371383250257, + 0.3895482855797482, + 0.3162707869661623, + 0.4649948950941738, + 0.4714820781166497, + 0.38999608501510774, + 0.3944637764210421, + 0.4136631891109243, + 0.3209183290997244, + 0.38962264883586667, + 0.3229032577224718, + 0.3168092406620232, + 0.43276534515337933, + 0.4083676890477054, + 0.3936648179847172, + 0.3943214742340628, + 0.4196686436270727, + 0.4218984724131137, + 0.39888904883024645, + 0.3614053972727339, + 0.31206159461201616, + 0.28090582787978186, + 0.2797747703549142, + 0.31559572290987636, + 0.41278889090883275, + 0.32129471237191337, + 0.3228877733371502, + 0.4208040086056274, + 0.3705102171794568, + 0.43527857902236083, + 0.35314872030267797, + 0.38171564627624543, + 0.3866319503441275, + 0.344701013430192, + 0.3687090252963884, + 0.4019960431873822, + 0.40377345051404245, + 0.4397959307778388, + 0.3431872229676861, + 0.34396219639464076, + 0.3592531272674592, + 0.36971788914501874, + 0.3581662875170419, + 0.3299439194871856, + 0.38071152469944086, + 0.4411359042930862, + 0.26461328898723113, + 0.2932233044938931, + 0.2611737890222189, + 0.26305954884227184, + 0.33710032608830315, + 0.44071740021602657, + 0.3184340441893157, + 0.3303066552303376, + 0.3350026890259036, + 0.34108196927586476, + 0.4464469281447742, + 0.4092266855820395, + 0.4544712698286325, + 0.3619670536946673, + 0.41597911851750136, + 0.3126232370514934, + 0.3626580343490571, + 0.4452133088299386, + 0.35933540528510444, + 0.2753286286847308, + 0.31194252066080963, + 0.3130660365952466, + 0.3432201871782092, + 0.3122319503914776, + 0.3321998293601537, + 0.32344814842026837, + 0.3326991653752861, + 0.3930134155682837, + 0.29477288523751405, + 0.3527125589264708, + 0.29706668807224473, + 0.35408172088086043, + 0.33922279383525933, + 0.39456286446479233, + 0.3805559223893481, + 0.44091298695697906, + 0.35555250498167795, + 0.38048342782285505, + 0.33788974942142047, + 0.3348897368774098, + 0.3345238121857356, + 0.3401191208844234, + 0.291680486324732, + 0.34261350552227765, + 0.34061828469138483, + 0.4365314269594112, + 0.3877661778112299, + 0.4754955865262027, + 0.34202998040129756, + 0.30500521174491035, + 0.3472965415680156, + 0.34269841320248534, + 0.4800497875925393, + 0.30308771295219894, + 0.32849038937263464, + 0.2952053337680374, + 0.4221242474632081, + 0.28088606201480054, + 0.4224651443986064, + 0.25769583775802335, + 0.35569442156138836, + 0.41454632864640717, + 0.2632007012315294, + 0.35268161676921106, + 0.367539685490066, + 0.4325751733232697, + 0.3842466389984629, + 0.36350413831686207, + 0.3557261984165388, + 0.42080162870685445, + 0.3900018758940436, + 0.3827638799992706, + 0.23564233379273777, + 0.38245152803536653, + 0.38607376578551983, + 0.3435188944356992, + 0.4296346773028414, + 0.4576524297996135, + 0.4271609635817266, + 0.4125506563234115, + 0.40601198748209516, + 0.41574489396479464, + 0.2029795870991257, + 0.40870868774458013, + 0.4046066008006697, + 0.3547674392227065, + 0.3479610553891443, + 0.39392256819963184, + 0.4041637396709899, + 0.3525222790508411, + 0.428473023061952, + 0.38258235480928493, + 0.30397473407148545, + 0.41254599397377173, + 0.4230039503496177, + 0.4525022867043645, + 0.46787151661108345, + 0.40611591228061267, + 0.3983825930305662, + 0.3909465939852046, + 0.408303855564976, + 0.3907944347607543, + 0.41184493354932117, + 0.30003060470724985, + 0.45478766234048795, + 0.37400058851610574, + 0.4162103450869112, + 0.4330677674326672, + 0.2818691688622986, + 0.4450994242794678, + 0.2665771823580343, + 0.22077359166234162, + 0.43265124943718625, + 0.42855757406811856, + 0.41005478665340533, + 0.4282301762864606, + 0.42801193311615127, + 0.4298389883720192, + 0.43754910737619884, + 0.230042141405663, + 0.38986520875151387, + 0.38128415246309644, + 0.40029915592870785, + 0.37294993311739705, + 0.35963607893339933, + 0.31618997379486624, + 0.3819645839413241, + 0.3575322989030037, + 0.35788662412275574, + 0.41028647148463554, + 0.40863693110473814, + 0.3190270630915423, + 0.4261150277446929, + 0.35262581411783267, + 0.42507346385601774, + 0.3237142490942928, + 0.3433246857514746, + 0.4406333564150964, + 0.45560848405722687, + 0.3986865791114701, + 0.43437031792485276, + 0.4322150536096786, + 0.3106045387216614, + 0.39903440149796054, + 0.4163252958207148, + 0.27672545367987866, + 0.31696564196826066, + 0.5187027472694848, + 0.49795888117220133, + 0.5039395556022909, + 0.477266827314596, + 0.25389498720730425, + 0.26601011977977773, + 0.5019123589647422, + 0.4769617526879703, + 0.49465995865232426, + 0.46573710282911535, + 0.38415989125587396, + 0.3798738451323289, + 0.40363685259904286, + 0.357286293435719, + 0.37894977317155143, + 0.4793639320414922, + 0.4303092313539999, + 0.5117469149953433, + 0.4038661558916541, + 0.319295800535078, + 0.5191154786564903, + 0.31401394451016557, + 0.2762013041961484, + 0.30483560446603, + 0.4807874078926881, + 0.5127767814196221, + 0.5193812814049317, + 0.46051809653609616, + 0.45776143472086356, + 0.3920960044002832, + 0.4643879506990244, + 0.3658732901057422, + 0.4590669368443827, + 0.4983895097716373, + 0.46630507325200815, + 0.5014390163075416, + 0.49242364871365457, + 0.4833588056631826, + 0.4577617701738058, + 0.46211116256908946, + 0.45766114607619535, + 0.48225075843487614, + 0.44230282271609594, + 0.345666520461487, + 0.31227713983332256, + 0.36140286969669744, + 0.44653658506264243, + 0.3565950825694647, + 0.47728736745115335, + 0.4996253540924809, + 0.4873261605446262, + 0.2760246557417554, + 0.36491025034594277, + 0.2633672846414481, + 0.38755377600021895, + 0.4032474936121443, + 0.4753393091657558, + 0.5099608809591276, + 0.4635988408197186, + 0.48949375460505934, + 0.5164042722873388, + 0.5283526435590523, + 0.28142952639658825, + 0.5026382467380771, + 0.5128613356658045, + 0.5309160012657481, + 0.5386018329253976, + 0.5222166093981452, + 0.5357606605695686, + 0.48247167505163485, + 0.5142735213065354, + 0.45626737285440794, + 0.5215539968233812, + 0.4909637961606595, + 0.4999692609082671, + 0.5204961760586317, + 0.5312983203733405, + 0.3490341973463757, + 0.4714263185736847, + 0.4490655433960324, + 0.3755114620400772, + 0.5225841063958797, + 0.5469372212934841, + 0.4630913257467681, + 0.47836615305958297, + 0.44873490973749447, + 0.305066013843388, + 0.43863178824503263, + 0.34431610064797696, + 0.3580129018106785, + 0.4967266750209015, + 0.4545389739486437, + 0.43595974499323575, + 0.4206399791170048, + 0.4663354572680887, + 0.47730305006093776, + 0.4362091709642095, + 0.40641590099802816, + 0.39032095762476116, + 0.2977855967548239, + 0.2841196777498866, + 0.3334820956078158, + 0.5282301058511114, + 0.3173521557721921, + 0.34110481009797977, + 0.5125408421253511, + 0.3752231355024043, + 0.5106698264743921, + 0.36893362775145017, + 0.4445919619178189, + 0.456512191606786, + 0.33978981686238313, + 0.4565001514898977, + 0.5122254229057026, + 0.41124321788443374, + 0.4649897313478171, + 0.35322657660390483, + 0.37124286934447076, + 0.38625143868208495, + 0.497672176554276, + 0.41683977886739704, + 0.40031853241866344, + 0.4469544134128467, + 0.4984525433816009, + 0.26408555556017693, + 0.2918882723800409, + 0.28382081967130324, + 0.2736093823217488, + 0.4051570350527003, + 0.5020067682166796, + 0.32631049075689367, + 0.3618439088170795, + 0.46825953171047807, + 0.41300023956016085, + 0.5321013940862217, + 0.4433525886623162, + 0.5202811276072093, + 0.3453403579874465, + 0.4439575197764729, + 0.3052248070492451, + 0.411702244306596, + 0.47463633148716017, + 0.35061556813680994, + 0.29033468919832656, + 0.3521761417759765, + 0.39861810453950725, + 0.4480096071751111, + 0.4649141812837211, + 0.33756480292516405, + 0.3927034711307527, + 0.3452658311374709, + 0.46124397216429835, + 0.339108589655902, + 0.3505519235512249, + 0.33587651819886105, + 0.4711513837197286, + 0.3135419049608515, + 0.48512655524902104, + 0.4139742812195636, + 0.4716519442171468, + 0.34322626132783285, + 0.39324128912797124, + 0.26447150636009614, + 0.27737598737064917, + 0.2688645375083296, + 0.2691748029111917, + 0.26796019791912823, + 0.39966844633950194, + 0.376841461478396, + 0.5379327435727224, + 0.44983024548768824, + 0.5588967972056039, + 0.38060745525745826, + 0.3488092398337359, + 0.39333860841884694, + 0.3353787387413314, + 0.5294518468099552, + 0.2741350048701113, + 0.37690789369070954, + 0.2801629511113799, + 0.49552413636535386, + 0.23984521987654017, + 0.4053992703876803, + 0.23866633342859575, + 0.37929427389122483, + 0.4817060297410441, + 0.30202915451920453, + 0.4057618600733526, + 0.3525764708809833, + 0.43629612641044974, + 0.4847239979791925, + 0.38925382844433204, + 0.3822343585251786, + 0.46739658563228614, + 0.46174612390345554, + 0.46198352113066954, + 0.22373861333160452, + 0.44883979140749913, + 0.4941607694875179, + 0.3247071553096442, + 0.5152504833299477, + 0.46910796639010766, + 0.4780820058855299, + 0.48196325767692083, + 0.45242025933784247, + 0.4320023663728143, + 0.19993867655922673, + 0.5051636319737836, + 0.41026898624877123, + 0.467700084784156, + 0.4571492993970512, + 0.5187173992636666, + 0.49739047259354957, + 0.44802979115957664, + 0.5374893697583032, + 0.49421018149785256, + 0.36381165266468735, + 0.5285083989205703, + 0.5033399314291845, + 0.43172000551472767, + 0.5299306081588229, + 0.3923363290336114, + 0.36669277957012614, + 0.3648594508133745, + 0.3815908416591329, + 0.3260556024447931, + 0.39721706490066666, + 0.24621043650175653, + 0.5282698372902854, + 0.3332043369255343, + 0.4172889167038411, + 0.5353630991423967, + 0.2628389473192031, + 0.5228733199719048, + 0.2453933807934759, + 0.21527460467998363, + 0.5071658882237173, + 0.48240964627276955, + 0.4327984154097014, + 0.447155154462156, + 0.4998482097582823, + 0.5086935308016691, + 0.5324489940761176, + 0.27484638403998635, + 0.4436324557030781, + 0.4121880132601087, + 0.479077874463538, + 0.42115542870018247, + 0.41257511994857377, + 0.36555488274578396, + 0.43185065441848625, + 0.4097624067272069, + 0.41132918006681923, + 0.47998690589795656, + 0.4255787708259503, + 0.2899741357698095, + 0.5010510790534736, + 0.4231085437183412, + 0.5199803796509003, + 0.2750383888669589, + 0.27592073367202413, + 0.46698396053446456, + 0.48483287800608393, + 0.46663513458684985, + 0.47624285560060775, + 0.5253549170828447, + 0.2734382068911301, + 0.4840171857980262, + 0.48175982081297913, + 0.258053468218789, + 0.36165351856112926, + 0.15799224542039336, + 0.3473735110388229, + 0.4166952873258761, + 0.34336061178699645, + 0.3721877305080471, + 0.3283083158130912, + 0.3654381864441722, + 0.3752228136822243, + 0.34261297526395074, + 0.2956001168946528, + 0.3011325671250633, + 0.2990794846448506, + 0.2962363096853208, + 0.2889474498720095, + 0.29698158009498865, + 0.29299835333574265, + 0.27721116049295647, + 0.26463852014560996, + 0.34811386055368143, + 0.3322510240117469, + 0.34539237091475916, + 0.3317852557520724, + 0.3370544380471692, + 0.35911254366092077, + null, + 0.3868230029066658, + 0.42512901237256273, + null, + 0.22171200583422973, + 0.45202744100969106, + 0.21548281574723094, + 0.3898127100619267, + 0.4138263727458015, + 0.42941373670209876, + 0.4456946493350275, + 0.4405990991442446, + 0.4463736766656481, + 0.4160260116286289, + 0.3769299895059113, + 0.3721442423036811, + 0.3069421534455305, + 0.4384657466369096, + 0.28671399914948453, + 0.3666843455687096, + 0.3370797615430367, + null, + 0.36972909309279817, + null, + null, + 0.34369835398846216, + null, + 0.3724066142677958, + 0.31255463384636933, + null, + 0.40044242013256365, + null, + null, + null, + null, + null, + null, + null, + 0.37311084149403645, + null, + null, + null, + null, + 0.3111495168952739, + null, + 0.3612981666747794, + null, + 0.38318880777537934, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.3802956008833064, + 0.27503311112648793, + 0.35426958593635693, + 0.37817163403669957, + 0.31625873923600867, + 0.3865874108113899, + null, + 0.4144903938399182, + null, + 0.3513581467501031, + null, + null, + 0.3621507854810188, + 0.3907245944642325, + 0.3309034363650762, + 0.33172363529976656, + null, + null, + 0.33788538249057243, + null, + null, + 0.3613672924604631, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.35615007832373263, + 0.34450188328878495, + 0.3697643827745114, + 0.3257385924381302, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4435554552421489, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.23160680658736416, + 0.3503968708021975, + 0.3113741750446158, + 0.4738280296304989, + 0.4071210918066691, + 0.2884136965664943, + 0.39214000286839107, + 0.44408556558183077, + 0.3780155697404455, + 0.38655582322077475, + 0.40628363395592737, + 0.39914654868616173, + 0.2902320811132377, + 0.44503676550809146, + 0.3821908547423639, + 0.3859961485696473, + 0.43056111307925704, + 0.3651601474595661, + 0.4899223098633218, + 0.38556645062323686, + 0.3890480875549222, + 0.3500615331539497, + 0.32391830785826453, + 0.5096768963982407, + 0.3615918790440407, + 0.3956568505178837, + 0.31043148165106393, + 0.38271080460643186, + 0.40212939944419623, + 0.4103110230801147, + 0.4157101826387445, + 0.42226758389883684, + 0.5003986475948229, + 0.29170543893061285, + 0.6685150001091698, + 0.2865491656535232, + 0.23787699493872883, + 0.1818259544630441, + 0.3390332866326001, + 0.23172899607436698, + 0.2807418564845627, + 0.4720851026532605, + 0.3118211386520434, + 0.41403590748885, + 0.3034991226941798, + 0.40301975300186244, + 0.3086993378552323, + 0.2251768329018028, + 0.4759043082824806, + 0.44612417686011985, + 0.39678736095329203, + 0.37081719758548615, + 0.38591335746418626, + 0.40041341336990727, + 0.3656732240501801, + 0.4085361892799906, + 0.331819686833487, + 0.468374342097908, + 0.4645970016531154, + 0.47241227809850844, + 0.3052385597084712, + 0.21133180040798963, + 0.4045803571577976, + 0.4319894967113871, + 0.28200499755036684, + 0.3578744259271537, + 0.35202159299025315, + 0.47635172565097766, + 0.5023907355818242, + 0.43758900218058777, + 0.3764695448767549, + 0.7163383373744451, + 0.36230340198293437, + 0.3804419968408317, + 0.3783572043738231, + 0.44203906195451315, + 0.3117781404880756, + 0.6521817708904075, + 0.41974561997986576, + 0.42636502717751773, + 0.4085995829170554, + 0.4289922690286574, + 0.49601657169650654, + 0.44691176904331925, + 0.35618175606759445, + 0.31387845415324656, + 0.4648699820563558, + 0.4376087231532399, + 0.8067275268801898, + 0.48263645847597575, + 0.38910717103234277, + 0.359300467079495, + 0.4310525796075943, + 0.3724705028188501, + 0.4155360393072702, + 0.5528195185668753, + 0.35286441349999287, + 0.35651856018216466, + 0.3744776543125712, + 0.3246546592257004, + 0.26368602623795223, + 0.3860806995584232, + 0.34129246382290285, + 0.33031283854297466, + 0.37632960484992745, + 0.23742043645709887, + 0.29731422445044453, + 0.3037995420124164, + 0.38350871409958315, + 0.38399049846912375, + 0.24646874246516876, + 0.3353468409612961, + 0.3171528059971812, + 0.36762053958496205, + 0.36964557304471135, + 0.4463657853654287, + 0.38884685767914395, + 0.3928616182087714, + 0.3660545788988461, + 0.41068982631514556, + 0.3610401500344481, + 0.4135461501534346, + 0.3892998246206853, + 0.30504788944903083, + 0.3715354145102185, + 0.3259352176377013, + 0.34546784025573585, + 0.32978889563769226, + 0.3518722661479799, + 0.5385561763717814, + 0.39234977160915707, + 0.33329691310833326, + 0.40730150766132256, + 0.6109837894387072, + 0.31367177081958875, + 0.37329459781970964, + 0.3672917728572025, + 0.3169349494583613, + 0.34032786689574573, + 0.3219648482874097, + 0.3433918540694565, + 0.4452058405638117, + 0.5461349400698245, + 0.34215107742632195, + 0.4049331304948215, + 0.4840437193282127, + 0.4392459619142566, + 0.2816878224361127, + 0.41987190146002, + 0.4216556845897794, + 0.4168179927426568, + 0.34251656128174285, + 0.4220708663691575, + 0.4694205660408356, + 0.4161763872393945, + 0.4071901277732335, + 0.30640104182288813, + 0.34709521575058105, + 0.3547676133047723, + 0.42595796098252203, + 0.3840479049122849, + 0.36260909959228016, + 0.3963605126311851, + 0.3923853227656795, + 0.38621943616873883, + 0.37365215060780915, + 0.35580116214665386, + 0.46380979475872414, + 0.29224563966566275, + 0.38555772315345627, + 0.377722938285431, + 0.3290864152813328, + 0.4922458052767681, + 0.3413697999112548, + 0.31043485161057627, + 0.4056556059527023, + 0.2858254981131426, + 0.4128081089013162, + 0.40459423928994, + 0.32735366520582365, + 0.39259578810013607, + 0.3917706829716059, + 0.3257808540643508, + 0.38855010452528055, + 0.3290772974816966, + 0.38140311013668754, + 0.4268282074129639, + 0.2966248362855629, + 0.3237859330645732, + 0.397019228579593, + 0.31230309193439826, + 0.3513519474556378, + 0.42770988255436465, + 0.448661209288154, + 0.33652580977653607, + 0.3324121654350274, + 0.33135465146812876, + 0.3279056190469008, + 0.2727811906471417, + 0.4496401067199794, + 0.3506669920432175, + 0.33702578328955934, + 0.3695365868181938, + 0.3246968219123833, + 0.4479173052779917, + 0.3597417325723117, + 0.42267731866091274, + 0.3664022457661663, + 0.3121034083271653, + 0.49231335717377683, + 0.3385537718475184, + 0.41721377535882337, + 0.3548793269020873, + 0.39271748653130856, + 0.8535336176603423, + 0.42706251917279003, + 0.3980488482411088, + 0.4023211714659932, + 0.4609940909805492, + 0.5444962360065866, + 0.6120565560653416, + 0.627829869448748, + 0.4255650102979557, + 0.3957322307917738, + 0.47740346829720476, + 0.7343193378278792, + 0.34004883304528716, + 0.365920494986277, + 0.5452273093508826, + 0.5215304210047198, + 0.5642230994562686, + 0.4498252431418569, + 0.3206527694257487, + 0.3946413795544455, + 0.6004881376069929, + 0.44381697658757674, + 0.34291670633676735, + 0.6154744183493368, + 0.5511132922176496, + 0.45482970104852977, + 0.490745360754379, + 0.7119279889674561, + 0.47396363419811494, + 0.5080897517689189, + 0.4525687205937766, + 0.7625882184358682, + 0.3325075224971891, + 0.6426473897364499, + 0.4023503853257397, + 0.5506250092464291, + 0.4354201270801058, + 0.39223516662053354, + 0.7026933316194721, + 0.51101697972243, + 0.3235609331951117, + 0.38024707758010773, + 0.6461755946894945, + 0.3743668781962948, + 0.45070031239560326, + 0.47716276445522426, + 0.36861201915257213, + 0.47514313409701847, + 0.48731067853406934, + 0.42368475348448537, + 0.39752146278636274, + 0.47251516995204373, + 0.3611346216995652, + 0.33599652389081464, + 0.630943466097053, + 0.3797385299276921, + 0.4556383135402138, + 0.5706116133703673, + 0.3724942628610444, + 0.6616656430603384, + 0.3546316969992096, + 0.35689205208965114, + 0.5602900852425993, + 0.4875931204238483, + 0.585601428669886, + 0.7310170735181761, + 0.6684910866791179, + 0.41414752859781667, + 0.47158048959833265, + 0.5539859345988518, + 0.33350374726008636, + 0.3458411339484652, + 0.5663211011603867, + 0.5347539674005587, + 0.7309436680030701, + 0.4724193984819731, + 0.36907278697844836, + 0.4679812563004765, + 0.3804113429693768, + 0.47858232375934734, + 0.7056524591816187, + 0.6281295276872215, + 0.5027438531620303, + 0.5324834242253296, + 0.4995178475329389, + 0.5183190643571013, + 0.7393172927522661, + 0.30924785362861185, + 0.4443644788485529, + 0.2994373975507524, + 0.6039617158273366, + 0.3603522058214447, + 0.7378976121131633, + 0.42087711685621526, + 0.4260316367812786, + 0.6957509014239511, + 0.5213525808646193, + 0.3341732948563589, + 0.5673099085330006, + 0.4113560207472309, + 0.5125539012086259, + 0.5031898180522849, + 0.3431365638334081, + 0.499041998298248, + 0.3981047764135566, + 0.4633722371298783, + 0.38738739695844976, + 0.32104427264308344, + 0.4133128907050462, + 0.5245002256188486, + 0.5849039709308054, + 0.663300591581192, + 0.4154732646941666, + 0.3078238273997791, + 0.4662101450661219, + 0.4699778787685069, + 0.4312761002053624, + 0.48960861070598677, + 0.4819977205830483, + 0.381334412944245, + 0.3355127082096429, + 0.5103681637029288, + 0.766693643612494, + 0.5333870889132704, + 0.6051561829630975, + 0.5931838737433449, + 0.7682677266461412, + 0.5704922915346172, + 0.3402954220806777, + 0.5879237213870998, + 0.38653288385749157, + 0.5268385340817554, + 0.46671380051958145, + 0.4013184550890284, + 0.49964748637966017, + 0.7827473770781449, + 0.23546853493516812, + 0.46632617073185983, + 0.47678143214130675, + 0.6260911671474692, + 0.38203296124780084, + 0.2985897808082649, + 0.4305868946740513, + 0.7233227808503714, + 0.3926453625080226, + 0.445769229220908, + 0.5099845709818244, + 0.4526506649986435, + 0.41624007460744106, + 0.5031402118344639, + 0.6163328080191406, + 0.4663869374545118, + 0.4042109578643776, + 0.5517825688528261, + 0.45858335091549235, + 0.5727562835372011, + 0.5474306400026339, + 0.5346104512935358, + 0.5519543512267312, + 0.32409982110024943, + 0.32739913317999964, + 0.5391803375440691, + 0.31651781794780115, + 0.6953189007166306, + 0.31467313373243216, + 0.5647082333519078, + 0.5740386901736618, + 0.6875670030485893, + 0.3083153242058931, + 0.3591152738657879, + 0.3476526307850824, + 0.3652775169101317, + 0.3113615567838651, + 0.2993365825331225, + 0.6186319816101041, + 0.37017046662247616, + 0.6832295582667985, + 0.5827453163226451, + 0.45003846198941194, + 0.5798609857887257, + 0.38792506862882903, + 0.552529436084448, + 0.5403360276386224, + 0.47215479453175124, + 0.3920099240762418, + 0.46660105555653647, + 0.49362780790428623, + 0.52770044860483, + 0.3973570565572042, + 0.34502206181854506, + 0.6246059597703343, + 0.47858469644392804, + 0.659552883053559, + 0.6414759997427997, + 0.5134961002343089, + 0.32235371923432554, + 0.3511758392516666, + 0.3627604940782127, + 0.46510906134760355, + 0.7461964945152553, + 0.474144323236561, + 0.3820082713488447, + 0.7435449398547076, + 0.4444869115167925, + 0.6272719699222655, + 0.5321803854733809, + 0.42375989545391246, + 0.5037632665725263, + 0.46214649916927164, + 0.46152922933340085, + 0.5797044979549308, + 0.5409556454892392, + 0.5287189669685382, + 0.4437746276064786, + 0.46776096370734627, + 0.4829889317404001, + 0.49123158790250476, + 0.768583035147426, + 0.4918325752121045, + 0.4772484920590883, + 0.4440062494198354, + 0.2953601848293295, + 0.6656718292323697, + 0.40751317796382375, + 0.33465464983276005, + 0.5176727459666917, + 0.4414903467928723, + 0.34194415772869574, + 0.47955370941175823, + 0.4350236587197104, + 0.7791453458168355, + 0.3192886213693863, + 0.39624268654293016, + 0.34187725263191243, + 0.43407967110432816, + 0.4015880402042058, + 0.3105626927768555, + 0.40917408057324456, + 0.28123541503289573, + 0.4232425885136744, + 0.7375777388007648, + 0.3177195495480675, + 0.4055525314261398, + 0.497950296807053, + 0.41886307338088097, + 0.39306391307127264, + 0.4784081698034289, + 0.3555369085151173, + 0.5247364028063748, + 0.5284795251459408, + 0.23562202000652205, + 0.6322397186164203, + 0.6888522217469103, + 0.5031732636872235, + 0.4817812622745893, + 0.5268726962844968, + 0.3977033846536846, + 0.3947831466717517, + 0.4820262370413543, + 0.7797825508228272, + 0.3802056899545012, + 0.6037121286423691, + 0.3481356046830687, + 0.458033325905713, + 0.47972230435694707, + 0.49209558884832194, + 0.39136169461811515, + 0.32286604464401825, + 0.3291821762810637, + 0.3284870470121423, + 0.5216773299738373, + 0.4511963889032364, + 0.3073440376987119, + 0.4064856331517162, + 0.4548160487650487, + 0.301932296421826, + 0.5551475023012108, + 0.4896405100347853, + 0.40370343517530266, + 0.456092528263478, + 0.3119701796910145, + 0.42619001350428065, + 0.499372133926343, + 0.5805323520357406, + 0.43527859920098133, + 0.7785127318844167, + 0.34122658621081786, + 0.5561395907702449, + 0.7373576303745125, + 0.6334378361129271, + 0.28104367432065325, + 0.49782577562058994, + 0.4205397125750765, + 0.34706947800490406, + 0.36727237799127704, + 0.3769086452599425, + 0.4653520893849246, + 0.4986972386851751, + 0.3557783169405904, + 0.6118188785253527, + 0.7657313620207816, + 0.4391781955303812, + 0.43238396075029417, + 0.44095827215636124, + 0.4961792143462805, + 0.516360480717186, + 0.49443970630750156, + 0.4911113644340608, + 0.3342682718254912, + 0.6555839465032233, + 0.4789918316131817, + 0.5550867017796671, + 0.713062700940891, + 0.31964697948983123, + 0.5929806194698993, + 0.38925984348155834, + 0.45687846036241353, + 0.3473639953811083, + 0.5176148078819267, + 0.4738379781477525, + 0.3302149919509575, + 0.42905745392267247, + 0.4080259694106176, + 0.5391188778464588, + 0.591347847869133, + 0.4519120302530776, + 0.4648415671565295, + 0.4132473181056727, + 0.7407717954421608, + 0.46049042591579176, + 0.5199331131297529, + 0.4637319403004284, + 0.46848041512664407, + 0.6114217501313379, + 0.3194400047477395, + 0.5665228666139536, + 0.3823410941931528, + 0.5402781348136333, + 0.42272394249935824, + 0.34557427776684796, + 0.34867206143737767, + 0.4713921911763662, + 0.3414357872388451, + 0.5081099194812897, + 0.7342812750334555, + 0.7240197252191265, + 0.39115672343937974, + 0.3445190914382632, + 0.702206611297448, + 0.4475780740445299, + 0.3849191188109173, + 0.5156111832760261, + 0.36066383133058555, + 0.5974852693387397, + 0.4349689905737199, + 0.565670403122565, + 0.5895129322273818, + 0.4438366224942456, + 0.5129120664226574, + 0.6567616349576947, + 0.5080750780687106, + 0.3630904802156831, + 0.41601456771620465, + 0.3098617605726119, + 0.5381642511987124, + 0.3475600588241702, + 0.5182960950182636, + 0.4970286654161118, + 0.3703308792468694, + 0.7203329236769704, + 0.3275954846878838, + 0.7248152377845622, + 0.44123404217990125, + 0.5409861034166547, + 0.7418915692647605, + 0.48058482323854673, + 0.5525991898586124, + 0.38434436579184017, + 0.7053814659162226, + 0.41802072775074944, + 0.32759553514463313, + 0.5599991599319586, + 0.519001070346598, + 0.6402226345922147, + 0.548163403786313, + 0.39518968858199816, + 0.485833015605196, + 0.29064882116882734, + 0.45611247450881487, + 0.5653094187920259, + 0.5103323262761964, + 0.35730604660266285, + 0.5415054941498347, + 0.5818597978323846, + 0.567867315582477, + 0.4364771566235278, + 0.3624713452116022, + 0.5076562963815983, + 0.46619965608576935, + 0.5565543615404599, + 0.4912215851836303, + 0.34784821238570607, + 0.4099767295201153, + 0.5060222254119351, + 0.40284444702643796, + 0.7344895681721658, + 0.5150964123708116, + 0.4488386301488926, + 0.4964014414656338, + 0.39772162757814505, + 0.43737624104098005, + 0.5407410970575308, + 0.5502469935309902, + 0.7115774014231127, + 0.5362451440374412, + 0.5758746205910342, + 0.5287595926811675, + 0.48433237215832464, + 0.5747258121564826, + 0.39229477840677324, + 0.6341649226962492, + 0.5968577446931276, + 0.724820677852561, + 0.5189203137885047, + 0.5460996693212905, + 0.5418213701198973, + 0.41614022271473255, + 0.537640774805774, + 0.34303282284324366, + 0.6804528682598506, + 0.4815600094188316, + 0.7367898467302302, + 0.7571102571867122, + 0.47809262977631345, + 0.4474008092002636, + 0.49774009578686973, + 0.3536588956515842, + 0.5003938135150188, + 0.7591805336943355, + 0.4879117588821305, + 0.5093442467666964, + 0.5878931893450918, + 0.5058233854355583, + 0.3518479225949627, + 0.6110279043627822, + 0.43132370445179896, + 0.39466941817453394, + 0.7330449295200117, + 0.4154142041786257, + 0.44105480527533053, + 0.3996762686527623, + 0.48700521018162224, + 0.5622785665255888, + 0.6241724605110521, + 0.6816240931007578, + 0.625761213731918, + 0.5900245917143777, + 0.4608660748938635, + 0.6068141692581981, + 0.5585453720683488, + 0.7816339220517458, + 0.7411496678059414, + 0.3558897224144031, + 0.3782110615046246, + 0.34197630189317196, + 0.5218724061459213, + 0.419035260364338, + 0.788292015283041, + 0.3825508208915576, + 0.42430780655176775, + 0.5683908219205046, + 0.4868342504374841, + 0.5661042084654458, + 0.6263498078803031, + 0.4747837657739172, + 0.6191136834711455, + 0.3857784146587309, + 0.6863932490686189, + 0.7799089841810956, + 0.7647782633997858, + 0.5818754704571061, + 0.4332897696949317, + 0.5536435121030816, + 0.7425658504167478, + 0.6929174096795649, + 0.6279994110112286, + 0.4895450796060075, + 0.4507128230817797, + 0.49525665841446165, + 0.7367302570825642, + 0.4223667099277728, + 0.4342519754514232, + 0.31900116887252167, + 0.602445301437505, + 0.30847043232047944, + 0.41716868888586667, + 0.6699526647229656, + 0.5420560016710765, + 0.6861049491103484, + 0.5492137022553689, + 0.5185823423762859, + 0.6705957414162456, + 0.6370753593067071, + 0.5669184212744627, + 0.3858703586029249, + 0.6487017207974954, + 0.5255998596005877, + 0.4085823211285388, + 0.3576818147826079, + 0.483957488006942, + 0.6409490060639855, + 0.5133087812760997, + 0.6158230505988463, + 0.57872527315254, + 0.4266399446724297, + 0.41206926148158957, + 0.6908102602955306, + 0.5017149880985523, + 0.6037878034661953, + 0.7265345472073947, + 0.42059746523852665, + 0.3752965251548628, + 0.37118107650241416, + 0.3784295941786187, + 0.7152899057511974, + 0.7380905614356612, + 0.5680796244383742, + 0.42623537771084963, + 0.4684018663415742, + 0.3963657771484219, + 0.7798508148350319, + 0.35422328342812187, + 0.48727689978644567, + 0.3438891858163313, + 0.5086208080300868, + 0.37791984844130183, + 0.567682080788584, + 0.4882425826778412, + 0.33339282482368304, + 0.700650092588324, + 0.5151221272891139, + 0.4270499296812903, + 0.34492970044408505, + 0.36268176574197086, + 0.4120101657972787, + 0.4269901089159654, + 0.3790004783112567, + 0.3733921904180548, + 0.40960480626389134, + 0.3904475600636491, + 0.4190759735877325, + 0.34276643425625014, + 0.4501830992984502, + 0.2151001248643672, + 0.2776108875475287, + 0.3363953574575074, + 0.3940203492103348, + 0.4436158455483347, + 0.43272856053269443, + 0.34283872674552784, + 0.3465036560607168, + 0.3364841414504689, + 0.6776286522533794, + 0.3375922057060573, + 0.3189287972791057, + 0.783717635650921, + 0.3713871581811494, + 0.4245187581554409, + 0.3562326170850693, + 0.45366462739447155, + 0.28790608022146175, + 0.40687582673601386, + 0.36342118097803805, + 0.4898067982131693, + 0.26798633931232274, + 0.4193239220628643, + 0.40576111305167334, + 0.5124072769328645, + 0.39530192969792705, + 0.47694165076523265, + 0.31334502894028804, + 0.45933655645086746, + 0.4158082476771598, + 0.36029678726620423, + 0.47580666885972045, + 0.37476218768640185, + 0.478505270397385, + 0.38878154923237895, + 0.37362141635757923, + 0.8077775542167291, + 0.30878589119384703, + 0.44226232136006033, + 0.3716755151235278, + 0.4141757160584483, + 0.33179618782217524, + 0.3847641901539348, + 0.4641549509816506, + 0.5542189964081705, + 0.34991975342635234, + 0.3901054040490199, + 0.2959526865973192, + 0.4161788368284856, + 0.4335380075205274, + 0.43195231083946906, + 0.4649944860223889, + 0.3725989256283447, + 0.35469025181659375, + 0.4737255401167261, + 0.409295966966612, + 0.45573452993152336, + 0.3071654355697168, + 0.32816148413274804, + 0.4861420759484288, + 0.4607455992584978, + 0.41140727023615864, + 0.37280011589515405, + 0.37296526225718457, + 0.8362540724552263, + 0.3358179153755515, + 0.33852615166092487, + 0.3597057143376978, + 0.3192844110243351, + 0.3489242100891665, + 0.3505100319733331, + 0.30900583993944614, + 0.3719606380669184, + 0.3559391329195934, + 0.4451759168289584, + 0.39440431321707237, + 0.4429782348860467, + 0.4694666500927957, + 0.4186179706045343, + 0.3871296064848765, + 0.5621020027301937, + 0.40289989782506147, + 0.4246552051536711, + 0.5016941195388089, + 0.4560676323833047, + 0.3759787207596597, + 0.30380702428119716, + 0.38321875336325234, + 0.5046497245932831, + 0.4745090709718926, + 0.3321774536812098, + 0.3740341346393401, + 0.39894625825688873, + 0.4650090684432545, + 0.34498013243667935, + 0.3611867021891713, + 0.4968154126369215, + 0.5377649537783401, + 0.321316874569692, + 0.430181115043074, + 0.33303834828325707, + 0.34342679685596644, + 0.3451141811650824, + 0.7917422420270573, + 0.42554945855379167, + 0.45090945843630603, + 0.3226778661318694, + 0.3820463198727164, + 0.3822673534256335, + 0.7039796019738771, + 0.42598494999386605, + 0.3991241384382569, + 0.4526934583720973, + 0.4238689144711686, + 0.4230803673501471, + 0.45978606249970416, + 0.4949953429408805, + 0.2875981053089973, + 0.3562472095840945, + 0.45014575009999813, + 0.43337220792054215, + 0.4180321110131439, + 0.4173085686790136, + 0.3910396491589309, + 0.32457719115136124, + 0.3614659217913528, + 0.41089713054426763, + 0.38498189064620236, + 0.6114317542007488, + 0.36184454267236266, + 0.3833496168830655, + 0.31719793980321853, + 0.5690868906938024, + 0.40005824287126085, + 0.3349430707183067, + 0.40095652145470495, + 0.4181869157107914, + 0.42274235700177776, + 0.33618872501323116, + 0.5517680963923203, + 0.4727995176867773, + 0.32325508688909743, + 0.42299466074996894, + 0.4498104296861884, + 0.43864640311949626, + 0.3859566568769971, + 0.4703176308955736, + 0.3431320183315655, + 0.6590345751010931, + 0.39566388087382287, + 0.46003168951944207, + 0.4689996600075477, + 0.40323547468878135, + 0.42038980153268063, + 0.39993564880686877, + 0.3127037851878529, + 0.39740544269127226, + 0.5988453665070502, + 0.3843710100405115, + 0.3406864106426327, + 0.5110263155134759, + 0.42667324143876134, + 0.42143176857465076, + 0.6224652639758551, + 0.3921727574475116, + 0.4576051534829195, + 0.3908691764697356, + 0.39891238035824816, + 0.4868313053904724, + 0.37229516398890966, + 0.4767296036161038, + 0.4193361771669627, + 0.45810744267594816, + 0.3331841998418582, + 0.44193856159960876, + 0.3873257141085243, + 0.3343427994291558, + 0.323312878794595, + 0.35165711081597895, + 0.43885755660204473, + 0.5419353203415657, + 0.45727292477975967, + 0.4522733783074451, + 0.47325727127831285, + 0.3568023626712684, + 0.37827391947798983, + 0.4842270112932058, + 0.372382877490991, + 0.40621056424771923, + 0.48644051611683475, + 0.3951738842570255, + 0.44894167741816726, + 0.505962505051569, + 0.4104979802983936, + 0.3865689204493746, + 0.5155449306717599, + 0.3894211676381696, + 0.4437856167797105, + 0.3225330847024283, + 0.5025370419485489, + 0.6032273041393247, + 0.7917233735757669, + 0.4324643782514968, + 0.4002978378764119, + 0.4331047342948573, + 0.3689510592623481, + 0.3630129062944776, + 0.3712781668890252, + 0.4504783851632735, + 0.42440665108217424, + 0.38272176318923834, + 0.45289987127800674, + 0.44142112158875685, + 0.44901040950570065, + 0.34234349522830226, + 0.33139078338237105, + 0.445779177264259, + 0.4161978751725479, + 0.39093199855553845, + 0.4013395491393546, + 0.40064016215395687, + 0.3770322978378726, + 0.2863945073257072, + 0.4430804131711278, + 0.4443798548592961, + 0.36863300427825996, + 0.40474863500729086, + 0.31190796659089176, + 0.40518095596109477, + 0.27328042468078706, + 0.5316797013981044, + 0.37103521106772497, + 0.33624271674602657, + 0.42834827240714307, + 0.39938893833845035, + 0.3378976560954554, + 0.6023458082258472, + 0.507355884628843, + 0.31580238269109373, + 0.479201660132906, + 0.42109666765324394, + 0.42732616517229804, + 0.4164377400133161, + 0.2917219885913112, + 0.374936673445159, + 0.5063540314777816, + 0.3426110419418752, + 0.43855536466840955, + 0.34063798624104064, + 0.3796819402915403, + 0.38893290840037564, + 0.3488700112150465, + 0.4286508072709135, + 0.43633349639367575, + 0.3817665724691077, + 0.350427628030661, + 0.42675297965064113, + 0.33572404521904237, + 0.4005444264866777, + 0.39158380890121136, + 0.4882455258538891, + 0.5366279782458152, + 0.4423448263817839, + 0.41802342425986766, + 0.43470453590463376, + 0.4453805209485307, + 0.44048546168105157, + 0.5272015457245852, + 0.40590569889910577, + 0.32302762018061726, + 0.5034352282199805, + 0.4069271422062722, + 0.4560624306242949, + 0.4477135068577776, + 0.4073651018601204, + 0.4979265990718401, + 0.33408699316027846, + 0.3847334642518896, + 0.3243570673047979, + 0.32614845727172354, + 0.3552333427652428, + 0.40936245787291126, + 0.34756507141904813, + 0.42202622911488474, + 0.5309611339641174, + 0.4307156031591867, + 0.49346427820856, + 0.7796615965412942, + 0.5606413174703532, + 0.37083888048372865, + 0.41917288728100677, + 0.4540896045593454, + 0.43321014885715714, + 0.3195404751637516, + 0.5817923920785555, + 0.47603179795966316, + 0.40479849417555985, + 0.4668791544126194, + 0.3904569577355421, + 0.49523962185823467, + 0.5108743596702113, + 0.4046690326355498, + 0.537035600635679, + 0.49867247312830204, + 0.38109355497830055, + 0.45196599479997673, + 0.5070302892638824, + 0.3777667466222953, + 0.396763423581249, + 0.32173580329257884, + 0.5638276281292487, + 0.5156793498152578, + 0.8220255230614608, + 0.5371875341640941, + 0.443279561235988, + 0.2837269486894715, + 0.475296914592795, + 0.43348652484129624, + 0.6073699013900612, + 0.34963304519861993, + 0.5373162643917258, + 0.3358891318075286, + 0.36985617973226653, + 0.3405138907041574, + 0.32809802720388326, + 0.4295666131867307, + 0.32569758745358884, + 0.43290771437757253, + 0.45036874019644424, + 0.40317625716245914, + 0.3328510279644479, + 0.33570815905309387, + 0.41871770529964136, + 0.32503525917167236, + 0.5390422314560392, + 0.4312328686517065, + 0.49947871298981955, + 0.4241908136388978, + 0.36409548388498353, + 0.5596701274751099, + 0.47816659514366194, + 0.36954938706811613, + 0.5943477927626863, + 0.30603831529993103, + 0.449718525415199, + 0.36519707889067077, + 0.34589576050757453, + 0.3270775675245901, + 0.37943981046029635, + 0.48163204672455356, + 0.3406990736101659, + 0.6928238346394168, + 0.4472651341089469, + 0.34421544733726417, + 0.3897815401885673, + 0.34480522295724275, + 0.5810108865013403, + 0.29922510984133543, + 0.377109786459844, + 0.3869295019803472, + 0.43493298260445645, + 0.3330123394487318, + 0.4111104387349588, + 0.31331028052150406, + 0.4373162022075126, + 0.7665722663291955, + 0.452724332763133, + 0.3835239032801704, + 0.4016362908055528, + 0.39971963826814866, + 0.3111752661631619, + 0.3759557322647443, + 0.4532894530811518, + 0.7514529700291195, + 0.5512483524620059, + 0.4601985804857202, + 0.3955089981887119, + 0.35967112153938857, + 0.4172064511908413, + 0.3095974426998448, + 0.43755676492932166, + 0.5157550257455426, + 0.3332324842216375, + 0.3878057260820182, + 0.38292017583266263, + 0.4715007493428059, + 0.34578886440786233, + 0.5419759204441963, + 0.4358407277216774, + 0.31174855812296587, + 0.4173241849595179, + 0.428700055522063, + 0.5299589490507507, + 0.5342936997575981, + 0.24488623031878742, + 0.7259736914152318, + 0.712233674591801, + 0.20786079710142738, + 0.3055062800350475, + 0.38973425845276183, + 0.4525343450924614, + 0.27219781853787456, + 0.3796382224181283, + 0.46079855919113516, + 0.4728692527803419, + 0.46745179503577305, + 0.36994210579575915, + 0.366458614341368, + 0.3543911765843637, + 0.3195112446400683, + 0.43264938625633675, + 0.37975210412881544, + 0.38160733075664777, + 0.50469592543147, + 0.41774910461473225, + 0.5285171283451581, + 0.4432651895607594, + 0.6076402474138273, + 0.43431794263688556, + 0.4326595641543163, + 0.49318708569800424, + 0.4518617312377553, + 0.33297733977246585, + 0.6286431913514257, + 0.612289540817615, + 0.4435923072331397, + 0.3690531327842449, + 0.38486969548359656, + 0.5311623403345878, + 0.37954231691054624, + 0.45553214692462923, + 0.43655776627218, + 0.4761875754108746, + 0.4394604168344226, + 0.38222150901139373, + 0.38873692267367815, + 0.5440799266403709, + 0.4886667044686446, + 0.4654986629047661, + 0.4861896051854698, + 0.3906513121494534, + 0.2539968761257354, + 0.2840078788608533, + 0.3288407863543753, + 0.5250396919290771, + 0.46230626457244123, + 0.29730968468633906, + 0.4530100295268104, + 0.3579680796550404, + 0.33775007501119303, + 0.3724337393741205, + 0.393225094722891, + 0.3628406113833954, + 0.44047055479427527, + 0.62730782622811, + 0.34711109394914413, + 0.43632649765540416, + 0.5386338724429673, + 0.36171928448587465, + 0.4571817899971399, + 0.5954809660799841, + 0.4566634401608159, + 0.4556076008729364, + 0.45065915009424545, + 0.41736097276738765, + 0.2860720105003508, + 0.36398564546749307, + 0.4890831791152497, + 0.5027565259897869, + 0.4774855940008674, + 0.3301006881144404, + 0.464489619791773, + 0.4700457958605112, + 0.5103420852182049, + 0.4068126148672508, + 0.46909952632100754, + 0.38626789629525465, + 0.4694316183871093, + 0.4450453360853342, + 0.41602355353401066, + 0.32657180688251536, + 0.45770086262695464, + 0.4679888637336243, + 0.2696818084818934, + 0.36852467620466883, + 0.4394443211040588, + 0.405944576269107, + 0.41209913606121934, + 0.4915788988800451, + 0.4977078397183947, + 0.4469142688866873, + 0.47355269823943197, + 0.35967052121286236, + 0.4647041875852305, + 0.42052974910116503, + 0.3947680173375142, + 0.41133313529047366, + 0.45210599985261823, + 0.4112577051028021, + 0.4344919425534994, + 0.4260229166256482, + 0.7238286814324287, + 0.547668361447513, + 0.5069712797294104, + 0.5362370593495569, + 0.49052154097187756, + 0.45082121241636786, + 0.5413364967074665, + 0.3622260409291573, + 0.3955595860255068, + 0.4686669238864892, + 0.33851090385961496, + 0.444662528026786, + 0.5009708193369977, + 0.4390256573042069, + 0.44571958900845143, + 0.46256197336584737, + 0.32713814248076345, + 0.5384496920837829, + 0.49106017085172127, + 0.37923513120663505, + 0.430055968022236, + 0.3835914061645724, + 0.35213666891561624, + 0.4553134580535227, + 0.4430904003324767, + 0.42183241599637866, + 0.4585545592044197, + 0.3889175237145118, + 0.4864980611216659, + 0.36448975925621757, + 0.38978835646093907, + 0.43412843017612207, + 0.46298446762751033, + 0.44737901792168366, + 0.4694735348796678, + 0.40411641600687465, + 0.34818486018891526, + 0.35409085467014256, + 0.4384925623505503, + 0.41053116611243173, + 0.4469156062358075, + 0.47618776944574126, + 0.426689139577591, + 0.4106136287114883, + 0.3570026747985741, + 0.40682529714216104, + 0.5291691684442582, + 0.39721680817185934, + 0.2076050053695922, + 0.36040720554458905, + 0.42280299146206585, + 0.3125855231116414, + 0.3021599769037157, + 0.45862333238016967, + 0.4568423159311948, + 0.2943487074219821, + 0.3245385257698862, + 0.37712375754237526, + 0.3686400084678995, + 0.5301085393765423, + 0.3049381784323027, + 0.3065475558856549, + 0.32717972979388377, + 0.34073910235954996, + 0.2549530130047488, + 0.31998669981543554, + 0.3000236079453723, + 0.32670271619577806, + 0.3022102059558457, + 0.42196528589641613, + 0.3255045484291704, + 0.3663706579773084, + 0.3703503117166673, + 0.43319877225505476, + 0.32146651058647796, + 0.3971501239262446, + 0.37260473128451055, + 0.3781946949595626, + 0.3775584079438628, + 0.3990514900811756, + 0.38124660746163136, + 0.6127588874300053, + 0.32799109142322147, + 0.3964859022066472, + 0.5634342660319375, + 0.48805717624562006, + 0.37318073347768177, + 0.4699274139092702, + 0.42253636015095336, + 0.463879929196852, + 0.47646731424184857, + 0.35642259536272036, + 0.5836084185288595, + 0.43269358577178396, + 0.4295946939169165, + 0.36641581400409823, + 0.4368694235312864, + 0.39410037054862307, + 0.4626194185087171, + 0.48242115262301066, + 0.45284070403030474, + 0.4704205507436612, + 0.5717233863570561, + 0.44725450397302424, + 0.41228557593527043, + 0.48043048569756464, + 0.5437679805801566, + 0.5299178733106805, + 0.44628810932742846, + 0.5724482029210533, + 0.5638473092593336, + 0.3887275480510879, + 0.3900519427977533, + 0.4415948712356452, + 0.35081597669552367, + 0.6468958256896474, + 0.48932739689655347, + 0.42962935113274725, + 0.35682711991503185, + 0.4656810417445643, + 0.4053608790119245, + 0.5231065727984668, + 0.44122232595307675, + 0.4267808794339811, + 0.520475395455081, + 0.31755817480434784, + 0.41896182123376924, + 0.5435675350442685, + 0.34442612812898477, + 0.42612734497281674, + 0.3294187022640725, + 0.36969112555901745, + 0.3936015887994572, + 0.3978223319656746, + 0.46573639459246574, + 0.3459239562768432, + 0.33258471806271966, + 0.43363402756564556, + 0.39953160860094256, + 0.37676029471108163, + 0.4503424626039004, + 0.4070168088503801, + 0.42439167510782455, + 0.5160594280675314, + 0.5625189186647075, + 0.4646417467858897, + 0.36156075021175144, + 0.4731266025217908, + 0.44331616443132443, + 0.39385671756961915, + 0.4290370134000336, + 0.3273201047457699, + 0.33194334163506345, + 0.3072536729699807, + 0.3700940856859907, + 0.3444785114096529, + 0.6527356261963165, + 0.43395291438974715, + 0.41859183627799856, + 0.30107498836298213, + 0.317875036687294, + 0.3224513389105086, + 0.3560107575751208, + 0.3219227364120235, + 0.3332331035373198, + 0.3180623465928299, + 0.3826577262882489, + 0.3911413658190668, + 0.42401739571154723, + 0.4112620244404961, + 0.3714678086520601, + 0.5026831210145184, + 0.498043165715655, + 0.37995034200964384, + 0.4255324186375242, + 0.4343609854184185, + 0.39036025071105845, + 0.36084439557660275, + 0.513402965938869, + 0.448074276920182, + 0.5828812567949311, + 0.45072269227752454, + 0.44488124230104825, + 0.44616619628576865, + 0.45188229536002655, + 0.33367307534676033, + 0.6039017472021733, + 0.4075927591063498, + 0.2308430206308366, + 0.46848146604877877, + 0.5805614533435393, + 0.3686714927721435, + 0.4569324041626812, + 0.754917281496109, + 0.4439409189998689, + 0.5150043904962034, + 0.2929380804288487, + 0.48542569597379837, + 0.3095227456360997, + 0.3332959793073851, + 0.3289000106894339, + 0.2840288292563241, + 0.44787420431536046, + 0.21211104922461027, + 0.29765536587256675, + 0.45475802584585845, + 0.4066377008137516, + 0.4025717539800431, + 0.3914534616035686, + 0.3380649480289261, + 0.4246597317373791, + 0.5221890747763114, + 0.5414088917253554, + 0.35985618628252297, + 0.4522291064627775, + 0.4139045087397441, + 0.7331914120392983, + 0.4005200978371369, + 0.559814510427876, + 0.7378485117404396, + 0.7160989405697871, + 0.4540513117600516, + 0.3802566863071513, + 0.31202351635376707, + 0.45612265249259876, + 0.4983632279951526, + 0.49451517587142824, + 0.5734847418206094, + 0.42722775023964604, + 0.5521864472833116, + 0.4389398798340069, + 0.39183171518528676, + 0.7029170670908652, + 0.6831396552170188, + 0.47530585775546286, + 0.41977270591926336, + 0.47148476264361777, + 0.4224725412638386, + 0.45383542158125617, + 0.6438792586881622, + 0.6692169914866914, + 0.42648860390102294, + 0.4367217423917794, + 0.4388520187305763, + 0.3709489078230254, + 0.49413296196128037, + 0.3435179885494535, + 0.48510369770585016, + 0.5063067437866401, + 0.34736532468130454, + 0.4458871627232168, + 0.4075508822091868, + 0.34212110543826685, + 0.5069375746194479, + 0.4614123849415296, + 0.4058225813237372, + 0.3747917548999921, + 0.48083752478343145, + 0.42792468604320233, + 0.3460081350265431, + 0.34620184813350013, + 0.26821237277832016, + 0.24633325596427139, + 0.28129257640061534, + 0.4508022193465168, + 0.4915140518414872, + 0.35218664590993565, + 0.31995367242777556, + 0.35117286928231806, + 0.4361006247028076, + 0.4337497016889775, + 0.49493789142092853, + 0.35996704007373886, + 0.4279689792585102, + 0.5205075644753115, + 0.381209243764422, + 0.5262408137832187, + 0.4779187821242003, + 0.431794340387619, + 0.4097968657319585, + 0.3644135914168046, + 0.24157660739180423, + 0.2927443117934803, + 0.27866838885622164, + 0.5185389948165308, + 0.43992888789106566, + 0.42448372363319514, + 0.4917180993038627, + 0.5124452377539065, + 0.3241098877021836, + 0.4272151899286269, + 0.3701233230439643, + 0.677201316661334, + 0.44031941516544765, + 0.4587532373760937, + 0.36688092611723294, + 0.45377163409706556, + 0.4896935465910857, + 0.43678839420934656, + 0.4088781926414411, + 0.4880755570697468, + 0.4527042203416598, + 0.5608608347419556, + 0.42118108183070785, + 0.48995539324522147, + 0.4478117020393532, + 0.4278244076976962, + 0.4059959563544932, + 0.47239302986696424, + 0.43134694558381603, + 0.48966954341596386, + 0.3519389279204396, + 0.32894933927976894, + 0.32969667608542796, + 0.4085587712309686, + 0.45115952273449117, + 0.3748593188118374, + 0.4377948763271397, + 0.4748895254686275, + 0.4263526527788455, + 0.4314009152826369, + 0.43387628841501974, + 0.2692779690485587, + 0.42041185287602334, + 0.2721356462004788, + 0.43245119415651245, + 0.22608062784712943, + 0.6546898192037918, + 0.3213048038333488, + 0.4543614103777934, + 0.5593207456669113, + 0.477902302557107, + 0.2752838211389989, + 0.4303721592464015, + 0.34100013513754823, + 0.7735381670272227, + 0.280839648324268, + 0.522137445554439, + 0.5031902034211075, + 0.5174068323814617, + 0.6315233754609451, + 0.4350331482189334, + 0.5802066573760651, + 0.4561499584255655, + 0.39078792574104276, + 0.33593761978435693, + 0.4557114002888026, + 0.3162232210578762, + 0.7150980553044791 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.44747137722249597, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.4943240822289969, + 0.7613354205633704, + 0.7359782045145502, + 0.8193564177984787, + 0.7424462426431265, + 0.6546697589656184, + 0.6505253884848208, + 0.6663607140132811, + 0.41948272247957546, + 0.7084356050466734, + 0.33256614997716033, + 0.49180892360395106, + 0.3663645227625286, + 0.32599445960890794, + 0.3259655917844326, + 0.3124303596752278, + 0.3667180475865258, + 0.56469105942786, + 0.45995914088343137, + 0.4241770877335112, + 0.3372195952301577, + 0.6690642987484694, + 0.36562090598713054, + 0.5940513403120012, + 0.3750268667785445, + 0.3192942112677732, + 0.4073991956109953, + 0.45947488310583706, + 0.5750702920837668, + 0.4956474326300593, + 0.30011005064526286, + 0.38889207663754083, + 0.3612542747707062, + 0.3576219692177776, + 0.46016842407928904, + 0.36508469981542596, + 0.3765219131444544, + 0.5322524152316382, + 0.5614861071601762, + 0.6292604098740865, + 0.7909750356119096, + 0.7824627043751097, + 0.7748989028480403, + 0.6605586499599663, + 0.40313002733408243, + 0.4261267603431372, + 0.7828684083625015, + 0.37901610782885486, + 0.3790617674759249, + 0.5158302146358534, + 0.412802550169105, + 0.7782452667064839, + 0.7954641607319224, + 0.40676216910946245, + 0.3691090274311418, + 0.37858430139025306, + 0.3753365972903351, + 0.42661072117157606, + 0.41057005973967714, + 0.5292058360755928, + 0.47652917263116534, + 0.4435690506548738, + 0.44035409579987594, + 0.5139760231321846, + 0.4941221371781742, + 0.4059913726193043, + 0.3845383810404804, + 0.5624597730311323, + 0.3899594305440894, + 0.45346388372552515, + 0.48578610990439414, + 0.46947346468596585, + 0.448995130065922, + 0.4378778491278776, + 0.4277046230647621, + 0.45118815573201926, + 0.3026680509479546, + 0.30167524324876993, + 0.38710557434525583, + 0.34320249529356234, + 0.4434433090845558, + 0.40807979145719775, + 0.3645440034402061, + 0.3642867922995653, + 0.40948093472203595, + 0.4902987777631892, + 0.41930785513497193, + 0.34056880417423735, + 0.35581983101349, + 0.3938907616183417, + 0.3510105481798697, + 0.3531052075571653, + 0.5104301707115637, + 0.5094045413307907, + 0.3421114523183382, + 0.5755702463915834, + 0.3771676504569708, + 0.5541254877401808, + 0.35961241514220305, + 0.47488887496320764, + 0.457746579439105, + 0.5251089881090004, + 0.2783807464996432, + 0.41781163104112207, + 0.3255380088700472, + 0.3476976476882882, + 0.4826179380762536, + 0.4133646816505591, + 0.34572684098769074, + 0.811256086350568, + 0.8045195598317789, + 0.8097476448162774, + 0.8257843868822459, + 0.5852046998210549, + 0.6890530400299848, + 0.7117501666847319, + 0.7567583496949268, + 0.7114952077226538, + 0.5884996677052015, + 0.6565399753193423, + 0.5760039382464407, + 0.773777402915356, + 0.6590261478504086, + 0.4335718340981404, + 0.33715305316474253, + 0.300921134834188, + 0.34113553269463515, + 0.557846376573922, + 0.3596534705375118, + 0.29458137898118164, + 0.4273586546454573, + 0.3879840980757346, + 0.3778547496382915, + 0.3733026925192267, + 0.41665157858064145, + 0.4264650400449828, + 0.40836488169468255, + 0.345044650310332, + 0.3870008584622328, + 0.403316191860168, + 0.3566332867075522, + 0.31984721476110073, + 0.344397302709461, + 0.599478280587158, + 0.3167638724658607, + 0.5705642520592049, + 0.36741701647786357, + 0.44460510767452316, + 0.3397302415978827, + 0.5645342595903777, + 0.3209618202828471, + 0.5036385666153727, + 0.3289418546686589, + 0.2824557623395466, + 0.37286418314368885, + 0.3384174870127032, + 0.27938123309304397, + 0.19649250506451763, + 0.3797321479037006, + 0.3554040192344314, + 0.494878654500461, + 0.46005814758885183, + 0.5131461894870362, + 0.482753161491802, + 0.3698685934435555, + 0.4740024040184778, + 0.4388296142191277, + 0.4429178531037721, + 0.4230253331919449, + 0.45854187444077726, + 0.39762157282120264, + 0.24926887021701002, + 0.4267384998770383, + 0.3617891591898189, + 0.26904857994180004, + 0.2779573042942314, + 0.5027903666606717, + 0.5124673740563211, + 0.5710580590520498, + 0.48406676239151736, + 0.3032854096948632, + 0.5062781401873757, + 0.4961500357106744, + 0.5260354072843892, + 0.3063492852114893, + 0.5275784288844322, + 0.4515625760499227, + 0.5266271024250312, + 0.256496495048747, + 0.39985145734288013, + 0.35441343398962516, + 0.2734521284465333, + 0.5099042905082727, + 0.4446857511077561, + 0.5605325688104241, + 0.33853678315660973, + 0.5908314491356071, + 0.3090159642864736, + 0.32857475481469584, + 0.40641846700431544, + 0.33434589447115537, + 0.2821729678817962, + 0.30252904974536, + 0.3751167589577728, + 0.6321245050160507, + 0.300497109042505, + 0.2677393178084518, + 0.4217729879549787, + 0.4238083360889634, + 0.40671015482322, + 0.46918138500357276, + 0.4216305640213678, + 0.30701980242223953, + 0.3810814891245691, + 0.36354543773899706, + 0.5317186604705213, + 0.5812179050356475, + 0.5770418774175219, + 0.567400908623286, + 0.30524088793199877, + 0.4197327543858232, + 0.4413944214526331, + 0.2844482347283132, + 0.34421253223997006, + 0.36119748029741794, + 0.16564673845637506, + 0.4861984926346831, + 0.32572299511884784, + 0.3135938020710979, + 0.530007769020386, + 0.26880788726785604, + 0.22466639890702325, + 0.45061511764631024, + 0.39661412882922925, + 0.3089461325720905, + 0.6210908078334734, + 0.42874749770438586, + 0.30616289493465565, + 0.28702778211336516, + 0.43950762080962275, + 0.32134269839710844, + 0.45914858391050045, + 0.30421483059581056, + 0.42251388344607294, + 0.41372005775803405, + 0.3904527910817449, + 0.35408072440665367, + 0.34054026526477, + 0.40005511613031314, + 0.35350114593586923, + 0.43565104529114085, + 0.3243415005835852, + 0.3945713815290544, + 0.34334696234033546, + 0.35429310263660957, + 0.39788141807881255, + 0.39921576440353573, + 0.3716719090249633, + 0.4371089534708503, + 0.3592758906774982, + 0.32738917680498913, + 0.5257768828247199, + 0.3002765218849268, + 0.32885251330261084, + 0.3404658919552132, + 0.29524666737788974, + 0.34586814858588044, + 0.3864670913276746, + 0.27781337545531964, + 0.33597794827691974, + 0.3197164164580381, + 0.3478341839457343, + 0.35849035830499776, + 0.397462928658073, + 0.3810501690156495, + 0.4741837257160267, + 0.27108139074138343, + 0.402640571371055, + 0.47053628009923787, + 0.2917610957979841, + 0.29589871179397914, + 0.6137930255158638, + 0.34805775672731776, + 0.27538051394239266, + 0.4879788517731687, + 0.4632022337878289, + 0.4509434812986833, + 0.39695128302305893, + 0.5067565683682757, + 0.47908974329840875, + 0.5046705304492668, + 0.4220360150886119, + 0.4776560032718451, + 0.4099818318340202, + 0.41207583769539413, + 0.45969101020683056, + 0.46393066575131414, + 0.45825965022078974, + 0.6573923129029904, + 0.32916000785604277, + 0.5850072209152661, + 0.6458277171955126, + 0.6732946047428991, + 0.6753735958720939, + 0.515685769569003, + 0.5367607430292083, + 0.41215343637957935, + 0.6759315218962353, + 0.4395712371216133, + 0.49908231303127193, + 0.3364191630766252, + 0.5110573173927496, + 0.4394444097527288, + 0.4087501506309037, + 0.45555516640322385, + 0.3418631471277181, + 0.4298099903578944, + 0.29517671103392235, + 0.3715495869692586, + 0.3622819359082228, + 0.35000815556544507, + 0.3980527508227856, + 0.28499046073470297, + 0.3441858677678087, + 0.3742796590875275, + 0.3838897990130841, + 0.35716510632681026, + 0.37936155107757213, + 0.5861421586354407, + 0.48248136212649234, + 0.43467475359237256, + 0.5019884157597545, + 0.4639172306300346, + 0.32385934809284544, + 0.31297191831931104, + 0.458775541211977, + 0.40482134763645217, + 0.3342372500891571, + 0.43947915869943277, + 0.5388463407064521, + 0.5503842262361234, + 0.40559744006581117, + 0.3249496035012579, + 0.4760865627659033, + 0.31408132959491025, + 0.33785700912057104, + 0.4988056965263239, + 0.32930643680053656, + 0.8248081151959282, + 0.51928194608551, + 0.4173210319102729, + 0.4074230023951289, + 0.4560351616121451, + 0.45209466043522134, + 0.39894934076231814, + 0.5116297220404791, + 0.37073236939687765, + 0.5267643148592882, + 0.41981453347372233, + 0.47006792486017285, + 0.49210116687182215, + 0.45641838095137044, + 0.4542696602238975, + 0.3735143179554588, + 0.39932956843204076, + 0.5508064204757278, + 0.5397154228339012, + 0.4236924104980205, + 0.5900111755115267, + 0.452674252330371, + 0.5576338288821415, + 0.41917141951784953, + 0.46168476777792583, + 0.4341697114770977, + 0.5682646103984186, + 0.5733277872731657, + 0.5443832363847207, + 0.5401692814702547, + 0.5285778856017873, + 0.5272767919354903, + 0.44852284486346694, + 0.40385673261636207, + 0.3906145395586267, + 0.31286041473532805, + 0.37669840622091, + 0.33200370426803283, + 0.34313700679721165, + 0.37531921769601845, + 0.486919755195818, + 0.3592908640126979, + 0.3265013350584926, + 0.36580951021641067, + 0.5563580517562178, + 0.6175970325673108, + 0.5410347220063118, + 0.5713566513653943, + 0.4090452296515695, + 0.5847202856837914, + 0.4017324170913721, + 0.6305940062197994, + 0.4758942603838114, + 0.5741650021213152, + 0.43900560599830735, + 0.5508403605504341, + 0.5546487421329425, + 0.48646197805611485, + 0.47098995075674743, + 0.5235350039513035, + 0.5355264579690043, + 0.5320387030758236, + 0.35948347689005755, + 0.3833021261362986, + 0.3295797013469174, + 0.3778221166149244, + 0.5150697946534523, + 0.5718639094505024, + 0.3242886255411796, + 0.5897348216324207, + 0.5192122109479165, + 0.48290025088053773, + 0.40093447338750504, + 0.4734430807495525, + 0.5040600608044643, + 0.4221462961395497, + 0.4391498159008187, + 0.8478034948443975, + 0.44599883032600396, + 0.4882372976241207, + 0.5000484083745252, + 0.5343515349363382, + 0.47313342360013627, + 0.40952740693627315, + 0.5896403022410852, + 0.4598717321446777, + 0.39521231261257805, + 0.45485068204995666, + 0.4437851132826027, + 0.442914700882427, + 0.4243356790255703, + 0.4470871374699428, + 0.43347297543542535, + 0.44363340752067626, + 0.31215561506755013, + 0.4442892840421578, + 0.3140296404834278, + 0.3890323328016019, + 0.25906111998552717, + 0.45000308383108784, + 0.3760441645487383, + 0.305129317047536, + 0.5226875869568273, + 0.3135779479962512, + 0.28011890237460946, + 0.5521372329819522, + 0.4751745301146295, + 0.4553894754834662, + 0.4781498821720516, + 0.4111215914638423, + 0.41516160583135886, + 0.619240509470175, + 0.37288739738637283, + 0.5490485026718062, + 0.38198608124529404, + 0.29365358730361624, + 0.24955255962175124, + 0.3999530005948166, + 0.4889071502608325, + 0.5467062518107745, + 0.5735941297765599, + 0.4660003030516408, + 0.5281315524877841, + 0.40769011792358223, + 0.4981209653830596, + 0.42277104713226754, + 0.40223691608831746, + 0.411269259896802, + 0.3760206122680783, + 0.4531097318606193, + 0.3933109922976773, + 0.4330584080609639, + 0.4706201151883734, + 0.4489006882228955, + 0.4521833527605631, + 0.3835065503928498, + 0.380775948599226, + 0.489514925173071, + 0.4761778374589228, + 0.5322467319324891, + 0.535153609089796, + 0.5676277529587058, + 0.5939064682804029, + 0.6679448829325898, + 0.6514684321689, + 0.48046398277932517, + 0.5249908165747629, + 0.43641003408828005, + 0.5283497384134372, + 0.4059398303648802, + 0.5035059798800667, + 0.4150460218579582, + 0.581196245307697, + 0.39699395324473463, + 0.5590900320918218, + 0.6557094988674304, + 0.6250057394588365, + 0.5574812650084953, + 0.6438903179244594, + 0.5423168692060029, + 0.46561930551282094, + 0.39294391720814126, + 0.41909931257289873, + 0.3098423549663159, + 0.36395687016354905, + 0.47590923282367326, + 0.5516833117916999, + 0.44660639391818396, + 0.37104379547972166, + 0.3099671202371521, + 0.4790884846029036, + 0.4563436213668038, + 0.4038790494168439, + 0.39817619107166785, + 0.5054618947033606, + 0.38808487783612533, + 0.4371069569519764, + 0.3877274397697534, + 0.4732276833804283, + 0.46283203301694104, + 0.3426177699575473, + 0.3973111095836398, + 0.4557736304363148, + 0.39184716762246324, + 0.3890633621933441, + 0.40342130023878653, + 0.383323513261796, + 0.3615887972518683, + 0.4142826251977608, + 0.44873168889353654, + 0.4052593003621144, + 0.29469868050700365, + 0.5755518797776806, + 0.3894697506890984, + 0.5039417096374248, + 0.5314681876803793, + 0.3735109887028749, + 0.4982430642116276, + 0.5120889835353963, + 0.4947616610652592, + 0.6377597333318626, + 0.43731529719169515, + 0.46599777041124285, + 0.4847057424326893, + 0.4404369110827112, + 0.5110330920214556, + 0.4776635557662816, + 0.34502874225188696, + 0.43860787331973355, + 0.4654416548307147, + 0.4187412905143801, + 0.30577521040830763, + 0.4505493943509438, + 0.37057775426223405, + 0.4323915033907728, + 0.641920632170248, + 0.5470983987435953, + 0.5562866666827799, + 0.5579651788381783, + 0.6368814382651241, + 0.5594780140655861, + 0.6119202432468064, + 0.5850012530865825, + 0.6137386548139213, + 0.41589952130909824, + 0.40259981881671836, + 0.4030934638077011, + 0.3370034801622172, + 0.29439110664070495, + 0.3535388856331644, + 0.354860087139635, + 0.3428313189381945, + 0.28972701670927653, + 0.4349585270741336, + 0.29444890620779807, + 0.4587033236476059, + 0.3404540829393501, + 0.5233378697641722, + 0.5805580618500371, + 0.5054900363027475, + 0.27863257656065943, + 0.38318608061637477, + 0.5708114779553464, + 0.32452563806558365, + 0.3695255944090959, + 0.2948589210191687, + 0.8305296569740329, + 0.5301408449301442, + 0.42404355747558187, + 0.5799069536145425, + 0.5221902050509595, + 0.49089933431404237, + 0.4244996195729673, + 0.4930405656733775, + 0.5328557404140747, + 0.4465804201066792, + 0.5718698992692812, + 0.5296798857509312, + 0.29095426511372785, + 0.422660935665811, + 0.3042885031649772, + 0.37065065588233975, + 0.3077835515108131, + 0.4973585865385675, + 0.4135071088906106, + 0.3522813772288084, + 0.47392893440671613, + 0.5228642135004671, + 0.3522098743273372, + 0.48319391462992745, + 0.34378485112770735, + 0.4475638048770634, + 0.4931839762238418, + 0.37448012658590396, + 0.3310146849509008, + 0.2964377990108673, + 0.5667451595825543, + 0.3375368522113066, + 0.36523622291999414, + 0.513841670788865, + 0.39147120686906106, + 0.4222315757827838, + 0.3296074337467971, + 0.3503352564836033, + 0.40413737242504805, + 0.4059070383273494, + 0.45080016210281787, + 0.3530864621745989, + 0.4853868844440838, + 0.2878044193205066, + 0.49956444505370007, + 0.42206718150706574, + 0.41420750905852993, + 0.4708369425352304, + 0.4014637312938291, + 0.47279440900259195, + 0.3584308845987069, + 0.49553916996703745, + 0.38577676605144157, + 0.3573159636745515, + 0.4387760048403126, + 0.3477547053896588, + 0.5057994057120364, + 0.26745236940448125, + 0.6657858541951931, + 0.6710990492893785, + 0.5683133517565507, + 0.5640266748889018, + 0.6138808823992726, + 0.6112606483530265, + 0.28682365896324347, + 0.3651083391585362, + 0.3230015554754297, + 0.377826736344736, + 0.34269096291449086, + 0.3453112030318706, + 0.3864037818863056, + 0.4243714568439493, + 0.43033242346801487, + 0.35852817864067826, + 0.37866439827864845, + 0.3127120974500003, + 0.4708240561596018, + 0.2525736227076867, + 0.4157494945009328, + 0.44461464099281606, + 0.40032316495434866, + 0.4852121602715616, + 0.4256141842484514, + 0.5262623910319385, + 0.29107058423398613, + 0.4928443316020681, + 0.3971980871030335, + 0.37723302562332633, + 0.35067586995584443, + 0.3989357968145883, + 0.38458332849287363, + 0.41238716037448075, + 0.49133139158063815, + 0.46466408953375465, + 0.5392573108277381, + 0.35932609996528475, + 0.39127173759146083, + 0.4151530731249191, + 0.4277977306803256, + 0.47918939940328265, + 0.4603015280803944, + 0.30067597042465777, + 0.47509746502013256, + 0.4390572753974653, + 0.42114071091394534, + 0.5082293778845614, + 0.3697518994004251, + 0.30596674878936453, + 0.27036473367031083, + 0.3510094614187468, + 0.40611716073552023, + 0.38187939587951897, + 0.3074269930857864, + 0.36398262830105044, + 0.5292555532687032, + 0.6027237810336081, + 0.5765906128260796, + 0.5301838160280999, + 0.5494724638041695, + 0.5027176810249206, + 0.5694457640080262, + 0.5151260071772259, + 0.5315002608760826, + 0.35770970337056707, + 0.40404727458641176, + 0.5139681887254196, + 0.6236480065392601, + 0.20606622903633404, + 0.3675217623715632, + 0.37062198718764927, + 0.34830933345466447, + 0.36095435490121724, + 0.468190998275287, + 0.3736518381370023, + 0.48874429627409627, + 0.746657914847991, + 0.6128897260451617, + 0.7822697750472134, + 0.7980211099119741, + 0.6495766228370776, + 0.7435342856584469, + 0.7608222194674492, + 0.38889440311003093, + 0.41319678807229326, + 0.4205575332789684, + 0.5279755796049022, + 0.3550052788105108, + 0.43894128697299684, + 0.3590456673475673, + 0.40405398806411, + 0.3972762189184029, + 0.3947674139340163, + 0.32129215034132563, + 0.43704965800178, + 0.40524588351196406, + 0.3953595989019839, + 0.4252197388273041, + 0.3285672267318044, + 0.33692173511421825, + 0.3204503172062036, + 0.32891078445594063, + 0.27466731324991045, + 0.35262273596257104, + 0.2850512448812807, + 0.5173423923314654, + 0.5105438182030488, + 0.352455607070503, + 0.3084008547638719, + 0.26815039659039214, + 0.5220311029415345, + 0.5941803313260452, + 0.6373132879955923, + 0.34728017249138243, + 0.3702791263477362, + 0.2682988657263907, + 0.48515746376880753, + 0.3209499075410649, + 0.326707341925244, + 0.4835909387829611, + 0.5374698819756192, + 0.4732236607656122, + 0.32584773970351166, + 0.44283591666550587, + 0.462052822533785, + 0.5018194126483879, + 0.4629239909543051, + 0.5074967353786034, + 0.4140810505581084, + 0.47967444000781034, + 0.4574838360857132, + 0.5302331329874556, + 0.3229430491410914, + 0.3741431813955233, + 0.48193257453497734, + 0.4275579045202512, + 0.6265391043736362, + 0.33512624384838985, + 0.5521493486427482, + 0.7085827472866455, + 0.5705110467769504, + 0.37912691216151373, + 0.341609683142238, + 0.3472737519175294, + 0.30720082653377856, + 0.41191333975253935, + 0.5175717240763186, + 0.4603352321348408, + 0.39722962605834516, + 0.3296946883131765, + 0.4990039437775548, + 0.5102186633708554, + 0.5397265881278047, + 0.5944355171803367, + 0.5145040477352927, + 0.5172192799759269, + 0.39137254885437794, + 0.5288673550491417, + 0.3006283077779668, + 0.31946284778672207, + 0.4281922661515106, + 0.40224863863061294, + 0.5762532007191263, + 0.5824818120684364, + 0.609733896817765, + 0.658862605831358, + 0.7513670191344459, + 0.6038092932720533, + 0.7636283890132566, + 0.4689759226113446, + 0.40164577306008553, + 0.40627087581746313, + 0.5370196565869352, + 0.35348008999624714, + 0.4890928083766964, + 0.4725600284303728, + 0.5080398678489154, + 0.48748314458507325, + 0.4382235534325912, + 0.4406608083522236, + 0.47044339620715364, + 0.40836586486353216, + 0.5008628769700001, + 0.434540307084521, + 0.46048453514470217, + 0.5201618847571691, + 0.5866366599663512, + 0.5343847287700826, + 0.38704568649942467, + 0.4621057828755347, + 0.5044640188452024, + 0.351539744842955, + 0.43245939222920876, + 0.4111784918363235, + 0.43079178685554903, + 0.40204410066072055, + 0.43468929318389543, + 0.38953239720429883, + 0.438315672436589, + 0.47087287890221663, + 0.4641997099749397, + 0.31715431007795936, + 0.475228578622864, + 0.49591854415327125, + 0.40559051836062526, + 0.5388068906112264, + 0.582399443718415, + 0.7452150091147126, + 0.2951337560358203, + 0.3056320532929555, + 0.405234069715321, + 0.3714941413649786, + 0.2876334370075067, + 0.33994089198489763, + 0.3176567739015929, + 0.6738060380305051, + 0.5812377459586561, + 0.6473045409086771, + 0.35721801966821093, + 0.3907826493363919, + 0.31285183021185853, + 0.4346053351964335, + 0.43226537450940794, + 0.26441758668408544, + 0.3081724760974087, + 0.4500049620038047, + 0.3980747608408351, + 0.45110037210038334, + 0.47520206015850613, + 0.4242992701746332, + 0.4022170100828054, + 0.30457583466018767, + 0.6860816850101316, + 0.49848151156526654, + 0.6927093051618338, + 0.5828311692073912, + 0.29172830018150436, + 0.5213266023829964, + 0.49552965871051374, + 0.5172238620921454, + 0.4886610303475394, + 0.39746351117325635, + 0.49153107778888255, + 0.500123713527892, + 0.5222287784440351, + 0.4871888371863805, + 0.36367294875409145, + 0.45273926527665304, + 0.29852905327156953, + 0.3631443883746569, + 0.6383939501431378, + 0.5352499616165418, + 0.5850011373711146, + 0.4767751582513115, + 0.4727803082695359, + 0.4437632066864728, + 0.41268635039426205, + 0.5072823298709148, + 0.4191449297634499, + 0.27487128877367933, + 0.3907437402227114, + 0.48849873987097137, + 0.27603880380250717, + 0.7620330684376108, + 0.4216744460174698, + 0.5761721265759447, + 0.7281945896036705, + 0.6839496550434421, + 0.3855008171378721, + 0.8071482329322978, + 0.49032583165250976, + 0.6641536107300243, + 0.39811451386968494, + 0.5184904897030846, + 0.4089914590240218, + 0.4588156513247989, + 0.5320375429414043, + 0.6197281767617914, + 0.5648252404675491, + 0.38800145047876844, + 0.42084246954482113, + 0.4155549931300633, + 0.4201014125624232, + 0.6242729109886856, + 0.33626904502515564, + 0.5493278211169177, + 0.46812336699854634, + 0.4852974263680721, + 0.5484093380641236, + 0.63620013432091, + 0.5064324275263611, + 0.4076005811374656, + 0.4587613965415569, + 0.5295322534471398, + 0.5242780165060139, + 0.416493017732902, + 0.36798130454466965, + 0.47370801125754064, + 0.5357691121062537, + 0.5361182612120047, + 0.4129351240762213, + 0.5836097256299732, + 0.5935820592288551, + 0.43496445206938616, + 0.5135925180265045, + 0.5422010270637748, + 0.520153841695429, + 0.46510525419859455, + 0.38550458954657363, + 0.31955696847658055, + 0.37371605450130224, + 0.4043124651830002, + 0.3419253295124875, + 0.43186606915757536, + 0.35249870966579944, + 0.3941897358360858, + 0.39339699842064807, + 0.34535728373361885, + 0.3542557607698863, + 0.32493117676802497, + 0.38746171291130016, + 0.430606701899743, + 0.3606338997702118, + 0.30546169175996923, + 0.40380280625339066, + 0.522639871822092, + 0.6391522828260632, + 0.3102663944892946, + 0.5345139371555413, + 0.5486540649413475, + 0.556434072082137, + 0.574466426633779, + 0.43063071171329714, + 0.5009448021127726, + 0.3813226311177827, + 0.5480447954229776, + 0.42997741557783753, + 0.6244407110838618, + 0.47631403172930675, + 0.5178113188695751, + 0.3111446663505666, + 0.34817349429362343, + 0.40126089655385394, + 0.48030027557090194, + 0.3989346026014255, + 0.5641773235838878, + 0.34606601533584186, + 0.6084865197587401, + 0.5116647473959881, + 0.5208928483436769, + 0.5257519030585456, + 0.5175358896076342, + 0.4315161966895178, + 0.5438798575452628, + 0.5324234680663116, + 0.5357260192431408, + 0.5362225630305012, + 0.4452671990326923, + 0.34792223467645556, + 0.3259162706962396, + 0.3807535190766167, + 0.2605124924078675, + 0.3979746892993058, + 0.4545153212295897, + 0.43493467693049165, + 0.42696381585429516, + 0.4562138041541331, + 0.42058883404030883, + 0.4951785767815874, + 0.49015827275153806, + 0.34954436555970203, + 0.48683064061211057, + 0.5671543073278896, + 0.3699862052035897, + 0.3200240601169241, + 0.3971447786889142, + 0.4322815609771293, + 0.5572112048013959, + 0.3494931365125743, + 0.5015085450603375, + 0.2823853587140404, + 0.48022288448253597, + 0.5182524079937219, + 0.5954451256980264, + 0.3155034813934646, + 0.32516140926949066, + 0.3276936333758204, + 0.2898562630868693, + 0.28295590419394867, + 0.3660786916160208, + 0.46664251415380514, + 0.29250617611889046, + 0.38766701371953904, + 0.4454541466294641, + 0.34050812686505133, + 0.4039369454916243, + 0.46110228472878967, + 0.6002543869568074, + 0.30140587201516006, + 0.45775761334071813, + 0.43164658846064136, + 0.33894623898757964, + 0.3423946072660355, + 0.4835556516260584, + 0.4791701107635424, + 0.28312840897424535, + 0.5116219252401271, + 0.4176547688375072, + 0.38649852777686644, + 0.5613765576806082, + 0.4205475762309871, + 0.34339315633824674, + 0.34147041919322985, + 0.3452336690491035, + 0.40547766291520543, + 0.39675691778799177, + 0.41698216108122443, + 0.4572579827740308, + 0.4292827081581028, + 0.5760967857331057, + 0.47704102716771574, + 0.37211013705250673, + 0.42667533193203105, + 0.47963438982554824, + 0.48264143728588343, + 0.4742050081852297, + 0.5657631923109945, + 0.4914277846936496, + 0.5300640468761006, + 0.5670295912079482, + 0.6548904278997726, + 0.4645316594538931, + 0.41446191980167246, + 0.2659263171735902, + 0.42531880146365514, + 0.49766342693533183, + 0.4807452815818924, + 0.4581622088141186, + 0.47986177291458065, + 0.33844236478362283, + 0.49289516745915035, + 0.4033623394057816, + 0.29073272286465796, + 0.4251122847077926, + 0.4370877856948802, + 0.4683560833946864, + 0.46921794087160124, + 0.3282746917202292, + 0.5784782991627687, + 0.5377049602732833, + 0.5044357161909823, + 0.5575588481968652, + 0.624669268010974, + 0.49248896077033544, + 0.3179543466522004, + 0.39724001631935146, + 0.3980935556745595, + 0.38562916738586805, + 0.3947588298104203, + 0.3523318643012306, + 0.383058264616565, + 0.4248599920468209, + 0.3270207182482311 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.19318253684998804, + "valid_tracks": 4, + "per_track_pearson": [ + 0.2249036718269706, + 0.16795561013555188, + 0.2206640064532959, + 0.15920685898413373 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.31329299365628194, + "valid_tracks": 626, + "per_track_pearson": [ + 0.296943069994194, + 0.2995941323584055, + 0.28090804867949015, + 0.28913560874885214, + 0.30454094296764195, + 0.30469669921633863, + 0.3013092851982351, + 0.26320370380018276, + 0.3165340011815078, + 0.3023966590265798, + 0.2949249942506623, + 0.30654089262064765, + 0.2977587733860121, + 0.2950407572213254, + 0.27432605440686453, + 0.28379873423954877, + 0.2960856562392638, + 0.3027985663249122, + 0.29908934820169947, + 0.29788957835752033, + 0.30872664989340737, + 0.31666561970153934, + 0.29886803399712497, + 0.2577663254041615, + 0.2955732160146075, + 0.3012805484666387, + 0.2895926461432173, + 0.31251659319630787, + 0.2773398488233017, + 0.2803820176481476, + 0.2863508778385311, + 0.2825203111235546, + 0.2669840078278314, + 0.30428269212661446, + 0.2553667428533677, + 0.2898560029926243, + 0.24961724907730168, + 0.2980177206449979, + 0.29339804178284784, + 0.29221786232295127, + 0.28644529891525455, + 0.25102213178962957, + 0.24079067612411806, + 0.2591070847553164, + 0.28337696312477134, + 0.27052551132420577, + 0.25445306205721413, + 0.27097007704101306, + 0.25544805689024697, + 0.27302841694299856, + 0.25225268670725337, + 0.2504822480530001, + 0.27572144608991767, + 0.273444544601543, + 0.2710344298070388, + 0.2784615951595728, + 0.28627419527819614, + 0.31160894545532775, + 0.2986362419989186, + 0.2928544843604201, + 0.2677837788534483, + 0.29656800503595326, + 0.3066719235670261, + 0.29286213008517903, + 0.29421200707719847, + 0.2881042880641922, + 0.30996800232593597, + 0.2776932731606627, + 0.28552864947348816, + 0.2762393209129945, + 0.29299527159274363, + 0.30892411191776176, + 0.2983223082974887, + 0.2920110605494682, + 0.2963811073779051, + 0.30210572969215, + 0.2935843968975853, + 0.2950184645162669, + 0.295900454173602, + 0.294000667745115, + 0.3176564047736167, + 0.30321374115990096, + 0.28644952356694775, + 0.27932337987934003, + 0.2859796285528919, + 0.30172410741128625, + 0.29261088782048705, + 0.2950426837751965, + 0.3034995065279725, + 0.3047471522622305, + 0.2909546852638688, + 0.2840021598820607, + 0.2926514644708001, + 0.2882519005006681, + 0.28679841385639704, + 0.30143445190952844, + 0.29575073715179173, + 0.30265050286785233, + 0.2929537766755617, + 0.2989809159889806, + 0.2740280352466941, + 0.29541190970662046, + 0.29410967101048835, + 0.3067738002855457, + 0.30379458195824766, + 0.2795659539127949, + 0.2535219681994062, + 0.27979599507604824, + 0.2930061240679644, + 0.29884841486490615, + 0.28906313537580985, + 0.3011806646017833, + 0.3098050422921563, + 0.2945850327283566, + 0.2772439068846432, + 0.2937367548935822, + 0.2828169278033207, + 0.2786133830108788, + 0.28417250458959936, + 0.2822856548495069, + 0.31591188907333007, + 0.2980850960420469, + 0.27996632673251204, + 0.3000878363331113, + 0.2587542229518603, + null, + 0.2963423914889407, + 0.29415921296132314, + 0.2618203680947921, + 0.2731100591079258, + 0.273221265162549, + 0.29227601793697455, + 0.28777782445472827, + 0.2812182677585642, + 0.3000475789541818, + 0.2954696649182692, + null, + 0.30578396188486984, + 0.30332413046169654, + 0.23829612026536315, + 0.26172264314050003, + 0.2900182210861282, + 0.2797808507543059, + 0.2939031033142445, + 0.2829864618189861, + 0.2764027334651994, + 0.2612009507785105, + 0.299286036366051, + 0.2747785579271002, + 0.2850349754375085, + 0.26337713988301376, + 0.29564954789282183, + 0.2928686363266562, + 0.2892757760010957, + 0.27940299051772216, + 0.27551873082375944, + 0.26553733779732575, + 0.27963917316271786, + 0.295383943009593, + 0.2694630051980976, + 0.2705524962018105, + 0.26683113775043943, + 0.3231360184662569, + 0.30246113255298496, + 0.2829893605989357, + 0.3011333638662268, + 0.29749048995718447, + 0.2797973698923818, + 0.26138303565584914, + 0.26078086826327435, + 0.28757165837901016, + 0.2991547019862427, + 0.27884491014171925, + 0.2680904371046959, + 0.28786142374624024, + 0.23726197079836456, + 0.30198978740077087, + 0.30515482365567814, + 0.28643943647499737, + 0.2948213747903359, + 0.2789544470400309, + 0.2788325328593846, + 0.23642685729973972, + 0.3097956972572573, + 0.2911514657539045, + 0.2271484398183309, + 0.3046144973811546, + 0.2938045010785394, + 0.30739207076159997, + 0.27893898867356554, + 0.3109283025524095, + 0.26880629498731934, + 0.29475767708777795, + 0.2941204799001996, + 0.2884014282734279, + 0.29080138717810855, + 0.28062517818288263, + 0.27547819813125685, + 0.3230662076366892, + 0.30784563957696176, + null, + 0.2986323333944661, + 0.2907299369442339, + 0.29032028686109146, + 0.2993685373315832, + 0.31538250392153067, + 0.3071201836136402, + 0.29092139510812526, + null, + null, + 0.30225679024837165, + 0.3118514835628, + null, + 0.30141741553987944, + 0.27567619612184746, + 0.2826587998580448, + 0.2862087299507278, + 0.29669255617660617, + 0.2947657148850676, + 0.28040269277427726, + null, + 0.3050267348499069, + 0.2856694496266657, + null, + 0.2718309920762695, + null, + 0.2715150134220077, + 0.29608029805008884, + 0.30608769847676537, + 0.3006861776013108, + null, + 0.28403032867853334, + 0.2793112654224673, + 0.28421065446526617, + null, + 0.2826562699322315, + null, + 0.28196430031973563, + null, + 0.27688081454637054, + 0.2855949293135629, + null, + 0.2849780426323489, + 0.29472359571038576, + null, + null, + 0.30159192398276824, + 0.288445209088943, + 0.28683533176120274, + 0.2821262866420028, + null, + 0.2847241102492975, + null, + 0.28555150406927365, + 0.29723254577399405, + null, + 0.21391168263659321, + 0.29286895285908887, + 0.30684913331668195, + null, + 0.2841029903057329, + 0.3019107708286463, + null, + 0.29978760954861344, + 0.29889911558524873, + 0.2942448465341642, + null, + null, + null, + null, + 0.30830041908294425, + null, + null, + 0.3088862519691556, + 0.3063953485622109, + 0.30898332824772523, + null, + 0.29763916975445737, + 0.29645977902492454, + null, + null, + 0.2952981541989083, + 0.29916681563554487, + 0.2877984818161152, + 0.28814364535751985, + 0.28959552832945207, + 0.2819723586805036, + 0.2709284842382977, + 0.2875799930724047, + 0.28372782839445543, + 0.288014576223825, + 0.2920031177688957, + 0.292357657482431, + null, + 0.2867583990385951, + 0.29738411750410426, + 0.2822157020578329, + 0.28386215999863607, + 0.2741847638887359, + 0.2857156939504572, + 0.29663089172884005, + 0.29122827690808417, + 0.2927643272306572, + 0.2991247557773898, + 0.29368157547244317, + 0.2930067771283295, + null, + 0.28590595452469425, + 0.30541393268807476, + 0.30211839644603056, + null, + 0.2966106246742912, + 0.24585216455652037, + null, + 0.3118779054804624, + null, + 0.2989626407881828, + 0.2866552123187031, + 0.2943012645196077, + 0.2835559808644892, + 0.3018595990584427, + 0.2808630795053754, + 0.2695450813544629, + null, + null, + 0.29955196034126347, + 0.280697030829402, + null, + 0.2923413206376651, + null, + 0.2873936524817728, + 0.29454900169561193, + 0.2922909396602594, + 0.29308026704053847, + null, + null, + null, + null, + 0.28484264945613613, + null, + 0.29488510816712143, + 0.31966479025595446, + null, + null, + 0.29394426740413804, + 0.29353125052177653, + 0.2757146163179292, + null, + 0.287540653214646, + 0.30500900351736293, + 0.29656464315012143, + null, + 0.30903528548989456, + null, + null, + 0.2960999122516771, + null, + 0.2876017123415378, + null, + null, + 0.2909524254206637, + 0.28390968008983486, + 0.29321984713278537, + 0.2935018712913187, + 0.2930390943350422, + null, + 0.2826664391297742, + 0.39207973068616486, + 0.35671535402828464, + 0.29775885141573827, + 0.30716780335500515, + 0.3666579174875221, + 0.34321971545249047, + 0.3361426626470012, + 0.3469808013525377, + 0.3488581985716905, + 0.3644001772717707, + 0.37034063860746913, + 0.34547004241668233, + 0.32885157935071285, + 0.3365027702219885, + 0.2801454160755144, + 0.28450758419714167, + 0.3455927782389383, + 0.3628062453574356, + 0.38238489112893587, + 0.38485450945901223, + 0.3628762109237452, + 0.33286670323495837, + 0.29406509504979195, + 0.2888010562921251, + 0.3593935555014653, + 0.34268712425889275, + 0.3516089287627468, + 0.3768226797787877, + 0.29297023472037714, + 0.2845215819257048, + 0.3060541448900691, + 0.30701276671895483, + 0.29836185299311385, + 0.3373943684485848, + 0.2720836995898253, + 0.333340918318946, + 0.25797763009989544, + 0.3840845650891141, + 0.3539561300549584, + 0.31531939959953403, + 0.3243269203680502, + 0.29834480457244306, + 0.29738541555079506, + 0.31428457145284694, + 0.3518204699282073, + 0.2823930506889008, + 0.30782125786456616, + 0.2969368717055251, + 0.31443212879650106, + 0.3126328579839772, + 0.30735605961616247, + 0.2943022291059537, + 0.3226186117067737, + 0.32173017708187174, + 0.3053514701030905, + 0.3153821223842926, + 0.3640448298269055, + 0.3487502237181048, + 0.35517689162325244, + 0.3685027289273138, + 0.36779599821999664, + 0.39311106037835924, + 0.3808628543149715, + 0.36324053992597444, + 0.3615895585396201, + 0.35280880466139863, + 0.3469864075184575, + 0.36050594653340207, + 0.3634346116975233, + 0.29948085308114, + 0.384744907081528, + 0.3655596093706943, + 0.33922729365705234, + 0.33164068860163975, + 0.3349309281509583, + 0.33938111099592344, + 0.3406753582778314, + 0.36014779927385016, + 0.34229457632834887, + 0.3586257477200161, + 0.36302963037904745, + 0.3549773143361215, + 0.3360787470101946, + 0.36002271757386556, + 0.3406133095772638, + 0.35619573043389574, + 0.3286934150522211, + 0.33997738901731445, + 0.33985478282438536, + 0.3630231749813855, + 0.3433104176289532, + 0.3336575093494523, + 0.3493696682127022, + 0.3586224964865919, + 0.3535820386802548, + 0.35739168687599476, + 0.3394726449418064, + 0.3927091398152049, + 0.37272838077456044, + 0.3925060902244571, + 0.34773723198222206, + 0.3799412233485383, + 0.35507044944273103, + 0.3448789414925364, + 0.36006159249315983, + 0.3595431662798253, + 0.30966498353934396, + 0.3410325314736077, + 0.31699100941994934, + 0.32235834586389683, + 0.33304592529184557, + 0.3194948422523325, + 0.3417447540358006, + 0.32981164017725273, + 0.32658575638710396, + 0.35959891672963895, + 0.33814004282883753, + 0.34836281841223343, + 0.3012250365439267, + 0.3397762836244792, + 0.3637606263824659, + 0.35076959705335536, + 0.3509147597107876, + 0.3425187814311313, + 0.34968863758971375, + null, + 0.34711042526399416, + 0.32920897975719327, + 0.34285505479212086, + 0.35515485277370157, + 0.34605350241180977, + 0.3518484925624954, + 0.35578276154557703, + 0.3652519185487909, + 0.3666908607721857, + 0.33695447830169145, + null, + 0.34556367959516004, + 0.36443778573536095, + 0.3468883476416953, + 0.34872551690002773, + 0.3487816965109886, + 0.3565884149884289, + 0.3322267897546386, + 0.36910354256892364, + 0.3708740143310698, + 0.3529712134839452, + 0.3532168402243649, + 0.386531985382581, + 0.3858859946878321, + 0.3548383593040747, + 0.3673428048776282, + 0.3508554379688682, + 0.3389115745013777, + 0.30148236909959314, + 0.30833213362335654, + 0.31378951148534706, + 0.35818184341776516, + 0.3553755908481032, + 0.3411159570578988, + 0.33522687825771164, + 0.3504440889032328, + 0.3661363525853945, + 0.3341993172526342, + 0.3369372215881243, + 0.32992585499382093, + 0.3810840946928633, + 0.34581965085767563, + 0.3517808068959198, + 0.3523215524053357, + 0.34292935681895537, + 0.37076128426471194, + 0.3701220295239795, + 0.3428630018492363, + 0.3164766161752791, + 0.3418797088455738, + 0.36338565930243616, + 0.37572514284249187, + 0.35680562185244935, + 0.29897633163604764, + 0.3394038969172278, + 0.3422706559588517, + 0.3277727924649776, + 0.35296628159179294, + 0.3478963917338435, + 0.35163147306450593, + 0.3754299428590747, + 0.3231722217630835, + 0.35977934436105485, + 0.3227076254368297, + 0.34206009489791483, + 0.36208235928665844, + 0.3633983229537285, + 0.3250174418518685, + 0.3161766648316352, + 0.3126122089940434, + 0.3324741728674783, + 0.34191485954053347, + 0.3811117669100894, + 0.34784401296667283, + null, + 0.3419150690995189, + 0.3378676052157987, + 0.31666089876233855, + 0.33123143638447705, + 0.3472021607232034, + 0.36173303707549437, + 0.3105465399469226, + null, + null, + 0.34337416765294076, + 0.33236583152740984, + null, + 0.33454278020599404, + 0.29303852388008667, + 0.3343981254107602, + 0.2973432783669248, + 0.3165437636371218, + 0.320310773188047, + 0.37903402514861495, + null, + 0.35880143502326944, + 0.36623632563415226, + null, + 0.31208095840408856, + null, + 0.32687658742537484, + 0.3436333288250237, + 0.316228756391299, + 0.30833818215187964, + null, + 0.2851480759999816, + 0.3214366928349016, + 0.3135698294494845, + null, + 0.2752641606402349, + null, + 0.30710984463111013, + null, + 0.3238823474956166, + 0.3167641287444934, + null, + 0.2869638619076747, + 0.3729650522687956, + null, + null, + 0.35005400578656226, + 0.32205038420728366, + 0.2974924774406919, + 0.28108592630427576, + null, + 0.3011518038578812, + null, + 0.3143162226386586, + 0.3078457201151547, + null, + 0.2816688980410467, + 0.2968138319382073, + 0.3085448233387221, + null, + 0.3108595516903014, + 0.3083167224859554, + null, + 0.3823302779648257, + 0.3740678175044022, + 0.38308996291728337, + null, + null, + null, + null, + 0.3866167086567256, + null, + null, + 0.3538617377124581, + 0.3381171742160209, + 0.37944580620392815, + null, + 0.38567449145587146, + 0.38321774940129905, + null, + null, + 0.3520494102723337, + 0.3226773438127313, + 0.3590621093686975, + 0.3120553375932859, + 0.3231779266925698, + 0.3315837414585327, + 0.29394820573072783, + 0.32174329911651495, + 0.2934102230819224, + 0.31827040703677983, + 0.31365770682663424, + 0.30687779723110603, + null, + 0.3181248420084489, + 0.3114782032494372, + 0.3586886599368192, + 0.3476180410672872, + 0.32921072078213887, + 0.36383265713171237, + 0.33839658473366796, + 0.35163331468903886, + 0.3601055845329823, + 0.3305117424070978, + 0.3150037309669744, + 0.307449094253216, + null, + 0.29956403748456156, + 0.3800951385377848, + 0.37697740025757015, + null, + 0.30368337973787307, + 0.33065186649522704, + null, + 0.32876670508282374, + null, + 0.334491633409099, + 0.3204043709100778, + 0.33223629434669955, + 0.3191783232703369, + 0.3592533896246249, + 0.3043083987730132, + 0.2788667727088059, + null, + null, + 0.3379081228433185, + 0.37621478975639533, + null, + 0.31836007733389465, + null, + 0.3270554923906042, + 0.31431528342335924, + 0.31959297677909904, + 0.37560432793705373, + null, + null, + null, + null, + 0.3404996714212476, + null, + 0.3342920506950779, + 0.35239989295456736, + null, + null, + 0.3353039935021805, + 0.2985131384041745, + 0.2838781095722088, + null, + 0.3197018977972253, + 0.32118275709619876, + 0.3048524948561567, + null, + 0.35232158619774845, + null, + null, + 0.3369063416327336, + null, + 0.28562214333723857, + null, + null, + 0.3099991337175372, + 0.365519362030537, + 0.37812283053211915, + 0.3665808140206745, + 0.3735591840806361, + null, + 0.32587085336514043 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071, + 0.691355228388071 + ] + }, + "junctions": { + "mean_pearson": 0.2455147203691296, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2651147299410885, + 0.2881022626805533, + 0.25407720896229147, + 0.17987429353647338, + 0.2807583105601266, + 0.28506291909791787, + 0.05557072672680863, + 0.24780263709758635, + 0.2377701692332523, + 0.22171039297703057, + 0.3051699333874042, + 0.16288886918608927, + 0.0885141286924459, + 0.15003698667796736, + 0.3172694730611322, + 0.20939251097188988, + 0.272734864986471, + 0.18729982751946742, + 0.20277057586063613, + 0.2456386827069588, + 0.2517756605709164, + 0.2009583532728903, + 0.17289121261286677, + 0.27505247818005896, + 0.20023467817681675, + 0.1346393634864296, + 0.19995892451605027, + 0.19345981264142464, + 0.23832871855748147, + 0.3333224344147466, + 0.2881456320034428, + 0.35581318667857564, + 0.2914747410959827, + 0.2613061178397234, + 0.22715538987270184, + 0.29335827681896964, + 0.2510333391310191, + 0.267750758131493, + 0.23853477341432966, + 0.21261001108316843, + 0.2832951337250668, + 0.23287944028435723, + 0.3293230795685537, + 0.3280548007429679, + 0.17824334944607803, + 0.17365517976743783, + 0.19396504124153655, + 0.1872908068064721, + 0.26142631976481856, + 0.2266053269500736, + 0.2982932148377143, + 0.14429476871477193, + 0.30511391136407906, + 0.24884188860400303, + 0.29408969375649036, + 0.2952799640887202, + 0.22074988058310555, + 0.1785731280129136, + 0.21228901552660026, + 0.2404211689156759, + 0.12441441109352835, + 0.13055344177311198, + 0.10151036552679817, + 0.1251584761875002, + 0.1310214082955215, + 0.3085768174662427, + 0.24061657983990564, + 0.35978219139905326, + 0.22750685574629623, + 0.2945583664076725, + 0.06246824232926207, + 0.2800106832085144, + 0.10513634161410712, + 0.11766191471283233, + 0.31916795895309846, + 0.32407743009844986, + 0.23498331799341143, + 0.1855423066973512, + 0.16094940949628664, + 0.3156243589772838, + 0.24354080632480593, + 0.2370354340863339, + 0.22082843379351633, + 0.25406492771760325, + 0.2686369774636051, + 0.25529806971205904, + 0.24439923386432955, + 0.26246019719583225, + 0.2531573206427063, + 0.2707318654200306, + 0.2333461436430968, + 0.24508145166498935, + 0.1973674168672344, + 0.23439151667287825, + 0.3573977671016301, + 0.21362902016593824, + 0.3065684189682246, + 0.27379006075215534, + 0.1906706518510612, + 0.2874184786185483, + 0.2773791902503348, + 0.2997355481113776, + 0.2589148845933363, + 0.19744982495074279, + 0.19405509497682377, + 0.1237863418174428, + 0.08016884154339812, + 0.28109355298625055, + 0.27177215771578267, + 0.18831783912066288, + 0.2350094372892838, + 0.2420171824540672, + 0.27350509488508307, + 0.3083083057306101, + 0.21349562116459572, + 0.18157425382584658, + 0.18300597448506112, + 0.25222606788989715, + 0.22316400923331375, + 0.30370209845809015, + 0.24456537130816255, + 0.2223524312605476, + 0.33861777377729696, + 0.142698575657864, + 0.1213758993039908, + null, + 0.27301451999559223, + 0.16668990340361794, + 0.11060893167718895, + 0.19414251171787517, + 0.20723427973353517, + 0.3320964676576818, + 0.204907051057112, + 0.3134829936480258, + 0.3100963207483806, + 0.13301216310282016, + null, + 0.34940906155971396, + 0.22732846445452334, + 0.25384989914869177, + 0.19492855933262915, + 0.24547691288344634, + 0.28618298564176964, + 0.23547301964630185, + 0.25573524867517466, + 0.23365307815881806, + 0.2571880243006618, + 0.3225601708397006, + 0.2240997327194981, + 0.24794716099971692, + 0.21480617989079148, + 0.2734923751168993, + 0.20057145971065587, + 0.22961581074544163, + 0.2964768584050746, + 0.32977693435672517, + 0.26641751770441485, + 0.22870546753487306, + 0.19053720673074864, + 0.18491605068735195, + 0.20576976172041006, + 0.25772271648423234, + 0.2305135244068355, + 0.302530787153451, + 0.1509551072681173, + 0.3185303068298319, + 0.2660616980968928, + 0.25773112441331775, + 0.20671788755037745, + 0.16396695901533745, + 0.34579792012562194, + 0.3315359889921547, + 0.31167613454568965, + 0.1850668580188929, + 0.22475747337280566, + 0.24805459927576226, + 0.23618953911752025, + 0.2201647879850615, + 0.2581872537397712, + 0.2716573264521677, + 0.25882303256898254, + 0.1888857238563287, + 0.19013078654691473, + 0.34046122078937113, + 0.34086550005742355, + 0.19472346292761544, + 0.28086436833810724, + 0.3989386085106029, + 0.3116002879214052, + 0.272781671714892, + 0.29247325111218603, + 0.35451072618235163, + 0.18234287590932857, + 0.18857447238890024, + 0.25762419743245213, + 0.22997149683775406, + 0.28210411362187127, + 0.1837856211508684, + 0.25640991908037813, + 0.2927768340488887, + null, + 0.1519293248440331, + 0.2800450522389822, + 0.24215126442044205, + 0.2579679202312961, + 0.15834275516164562, + 0.10953739195480315, + 0.16839120053435114, + null, + null, + 0.20046910174517585, + 0.3035290127275535, + null, + 0.2550081539046337, + 0.3083987947532707, + 0.16964286223515165, + 0.12610412517257785, + 0.15975185315139886, + 0.23923120460698086, + 0.04499054526471618, + null, + 0.2892179036195482, + 0.280028717647287, + null, + 0.21812243867334385, + null, + 0.28586670375398626, + 0.19941186054226875, + 0.15947258614063928, + 0.18948677203324082, + null, + 0.21166946570797615, + 0.12608751007885363, + 0.21549060567852826, + null, + 0.19483886962221994, + null, + 0.2823912224925999, + null, + 0.18845847839026636, + 0.20699530628775242, + null, + 0.20288157175057228, + 0.24617750039614072, + null, + null, + 0.2514806603490743, + 0.2005626447107165, + 0.12784671721391488, + 0.1347497720677781, + null, + 0.3022981847344918, + null, + 0.24115096233053482, + 0.18270459396976033, + null, + 0.24800779268801926, + 0.23526029710600882, + 0.3765751790747132, + null, + 0.2382543998955624, + 0.20260482605254537, + null, + 0.2470230656173449, + 0.3149205198517726, + 0.2634715418622825, + null, + null, + null, + null, + 0.19235536540592552, + null, + null, + 0.1692028017737254, + 0.14926857619911818, + 0.27303332981240447, + null, + 0.15919447306482978, + 0.2582874001480929, + null, + null, + 0.2876338396829203, + 0.19776833097714008, + 0.24685125519795986, + 0.18023353468569314, + 0.17514201273351865, + 0.15305103678836532, + 0.32301630786497726, + 0.15554005379965685, + 0.2824935327739868, + 0.14010137741942574, + 0.2638278748027086, + 0.24244437666182342, + null, + 0.2366009363910005, + 0.28121025155686885, + 0.09742935173522319, + 0.21013924244173918, + 0.22454642854503268, + 0.11014364752781933, + 0.16759477951914487, + 0.13894244771520528, + 0.26075807487141095, + 0.18679427267859977, + 0.2881320865683647, + 0.285167245800492, + null, + 0.17969245991338828, + 0.20430657819370612, + 0.1787750833509491, + null, + 0.14475202245601126, + 0.17275149822002234, + null, + 0.3703007095890708, + null, + 0.15999028578220575, + 0.16153134734937577, + 0.23732184813878204, + 0.11152212052437334, + 0.12355848945306672, + 0.23406031698644866, + 0.18798456883726206, + null, + null, + 0.19382500404174927, + 0.2045427597728264, + null, + 0.20224502334081826, + null, + 0.30460484524014364, + 0.19931033369905107, + 0.1472368985457242, + 0.21104911627816236, + null, + null, + null, + null, + 0.14938250948010026, + null, + 0.12743620870078476, + 0.27715109754651357, + null, + null, + 0.25689622904383125, + 0.23334653601830083, + 0.17289505972162006, + null, + 0.3344623889581043, + 0.2638979400577647, + 0.18685127460150375, + null, + 0.21260085414347518, + null, + null, + 0.30375604185272365, + null, + 0.1416965202601532, + null, + null, + 0.19382388783652432, + 0.23925579446027853, + 0.20282868179516061, + 0.23704095121549265, + 0.21891254978715993, + null, + 0.15507474741248164, + 0.23796130777425262, + 0.191956792140737, + 0.31189267535341403, + 0.31118154376435153, + 0.3036890726169889, + 0.34165431063049484, + 0.2781827331336294, + 0.31179465694123626, + 0.28252995182151713, + 0.17894617188625353, + 0.2808491247281045, + 0.32058283026335177, + 0.1941896770067995, + 0.25364203168236865, + 0.21911950299280838, + 0.4109340400031911, + 0.09501529821764454, + 0.27497591517591113, + 0.20639130272731845, + 0.30386897360592047, + 0.2920352604709266, + 0.3369224800615459, + 0.13120107485435037, + 0.27512600027449974, + 0.31630685637798495, + 0.29596881108795275, + 0.18855872152343803, + 0.2871804443267541, + 0.3191962060351431, + 0.22171546536918488, + 0.2734108000845164, + 0.34461469442165127, + 0.2310681417317368, + 0.21808047793196772, + 0.24348578568080453, + 0.19946184665183647, + 0.1324084124106281, + 0.2645294142486599, + 0.26859542533116976, + 0.2596059999488655, + 0.24201312802970315, + 0.03985048745910567, + 0.286311089275562, + 0.26548483026125014, + 0.10931268191577857, + 0.1280510985691937, + 0.20063745175594597, + 0.18758783857695335, + 0.22799416440710887, + 0.30844241993263516, + 0.2000240844181721, + 0.3024355265827133, + 0.2550266959505602, + 0.23057532359689642, + 0.21797167922600746, + 0.23492743708494424, + 0.2919711436696268, + 0.2206386832248253, + 0.25246061476317294, + 0.21305144996382125, + 0.3688863607722304, + 0.1061779397588394, + 0.04966179052803081, + 0.2689098174047919, + 0.14626144042011058, + 0.3302982415237456, + 0.22766559339331796, + 0.23809472627318706, + 0.20763341676265412, + 0.26417672749268273, + 0.12160212411469531, + 0.17936910213555438, + 0.240662576895159, + 0.2438314978062503, + 0.301840431628035, + 0.3167337019063715, + 0.2788410418080733, + 0.2515067144005142, + 0.3019415173914763, + 0.18346934664433662, + 0.22063236379906942, + 0.28686830810290054, + 0.21082536076030076, + 0.1223824636939947, + 0.3107358367320932, + 0.2458969221631822, + 0.2775478847093137, + 0.23346172730637296, + 0.22081200465635492, + 0.15212167357826553, + 0.2285763901257717, + 0.17139694563988728, + 0.22157673739933548, + 0.24037515392359957, + 0.24280181832089073, + 0.3524414663352252, + 0.18650969918722704, + 0.33977662480887544, + 0.31371486654130787, + 0.3132749935180927, + 0.13289548001871201, + 0.375848479571431, + 0.29963630555884035, + 0.17257840780139294, + 0.19765169824210438, + 0.23331046039340309, + 0.018367997624646908, + 0.1868709630228443, + 0.2756561048336295, + 0.24234145431668466, + 0.08269948704518734, + 0.24309630588280665, + 0.3100812015700732, + 0.21042591187436535, + 0.2246976775640247, + 0.15556405667490017, + 0.2544113951376714, + 0.1744470358010502, + 0.18432387287440252, + 0.2470496230204313, + 0.2637713841011118, + 0.12045900737731509, + 0.27063644080503213, + 0.22315371633590803, + 0.3454260029781783, + null, + 0.33025181493758854, + 0.31660684800342187, + 0.30567321356011434, + 0.3998326845948632, + 0.13548043335869922, + 0.26760166990843387, + 0.31639769157657194, + 0.3563271641073284, + 0.22706934982886934, + 0.27515488052068426, + null, + 0.24973748780389332, + 0.24746017795670588, + 0.3236652176537927, + 0.3283176627008057, + 0.24527924433636633, + 0.2080804900836883, + 0.2779541684697903, + 0.3343387672045333, + 0.1954012224238602, + 0.2237413543233617, + 0.2941386182382701, + 0.11634422807826443, + 0.16869390135651371, + 0.26677165223712535, + 0.34183906481039333, + 0.19592774680838948, + 0.5259194595423378, + 0.46148757273680824, + 0.4153821105076721, + 0.3562329704654142, + 0.33071378688914516, + 0.2914647094176005, + 0.3028622415894755, + 0.31268048948528054, + 0.3413239403635956, + 0.32428141437035995, + 0.39539336385366225, + 0.4147136428824568, + 0.285380618201804, + 0.2391624363629693, + 0.12953991092730924, + 0.12078778434410131, + 0.16098819341177198, + 0.1559220748645899, + 0.3357469253111871, + 0.2516590028016604, + 0.10783973841685383, + 0.41431897002640367, + 0.2797974453977346, + 0.4959910182329252, + 0.31979144645419855, + 0.26736567275300194, + 0.25090585652985214, + 0.19093571588371144, + 0.22471565604867316, + 0.23113302430562285, + 0.2711060748076917, + 0.2335850516365487, + 0.3161054942175191, + 0.389765208723256, + 0.21813415985792034, + 0.2866750166184193, + 0.23835949836795198, + 0.40471748763665943, + 0.27417285950640774, + 0.18730444728961146, + 0.24377309878835524, + 0.3142609623785544, + 0.2843974226320995, + 0.33750956241830105, + 0.1609616993051365, + 0.2958738991403875, + 0.22350079397048894, + null, + 0.32368993041624977, + 0.3458629360653439, + 0.34172317087517007, + 0.41169832619020513, + 0.1202341231929967, + 0.14674056000482433, + 0.3041316235171465, + null, + null, + 0.24338553188108258, + 0.35976095456483553, + null, + 0.17905234056948044, + 0.33083954977876867, + 0.22138007912015314, + 0.22547327616343565, + 0.288010775288277, + 0.34502459073265335, + 0.2400432825467889, + null, + 0.32903574871662755, + 0.2461811924056508, + null, + 0.36167213873011883, + null, + 0.5253615875301372, + 0.25546261406644466, + 0.13416906602702552, + 0.19072281635485344, + null, + 0.2547437133534117, + 0.2548458648461543, + 0.30357883371010863, + null, + 0.5213228063213872, + null, + 0.24965119534127342, + null, + 0.2772087481209718, + 0.25707900741618034, + null, + 0.3427530428324221, + 0.2993962736682929, + null, + null, + 0.15469412450436124, + 0.27300403650066096, + 0.07367579171638598, + 0.2351060928555203, + null, + 0.27229448900260966, + null, + 0.15104574682932814, + 0.3623116686275771, + null, + 0.2234488780158718, + 0.3081656241874028, + 0.20721720838019658, + null, + 0.24708412888014036, + 0.47996511035516526, + null, + 0.2950269724700699, + 0.38633739589158506, + 0.36065973609783114, + null, + null, + null, + null, + 0.3157484564185155, + null, + null, + 0.3384412451153035, + 0.20884492024836956, + 0.35750451476159995, + null, + 0.18038233736550152, + 0.3558489822041806, + null, + null, + 0.23659298253320657, + 0.03751304648748621, + 0.2491086470459785, + 0.17548681361693416, + 0.1948978095602265, + 0.2121824098883705, + 0.16159586052650837, + 0.2403035923946989, + 0.21595755359349575, + 0.3149570467515923, + 0.3076268287571656, + 0.23602685831300116, + null, + 0.12879640207989052, + 0.278521911783998, + 0.20724724741316694, + 0.3261413293357426, + 0.170776077870036, + 0.3763168073304765, + 0.278066218446584, + 0.2036982087060039, + 0.33068638982451976, + 0.37359212730957697, + 0.2369587975856711, + 0.201062532108227, + null, + 0.22798457285159324, + 0.4283085341872954, + 0.356821665951396, + null, + 0.2808305077605326, + 0.3113478178844292, + null, + 0.30020935136000376, + null, + 0.27620499245946967, + 0.3307798680068576, + 0.25612806752091294, + 0.3543057306739189, + 0.18324199447337183, + 0.40836704301555116, + 0.37906550574052084, + null, + null, + 0.20097600400497284, + 0.3174281360002555, + null, + 0.2278024174858298, + null, + 0.2057987270400815, + 0.2532003857005587, + 0.3609457449928214, + 0.24675432658141497, + null, + null, + null, + null, + 0.2607844220198329, + null, + 0.3110368480048229, + 0.27831626329045855, + null, + null, + 0.30730300129403465, + 0.13204189512793074, + 0.28972440555014894, + null, + 0.2674486091301377, + 0.2879897111650045, + 0.23315490072157583, + null, + 0.2279333226178483, + null, + null, + 0.2303462822237199, + null, + 0.24202898327382602, + null, + null, + 0.2223090755663457, + 0.3254366636151731, + 0.4101745437325181, + 0.42233174545800733, + 0.28609139983133214, + null, + 0.25436802864533214 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.8675496688741722, + 0.8881578947368421, + 0.8647686832740213, + 0.8670886075949367, + 0.9205298013245033, + 0.8610169491525423, + 0.8626760563380281, + 0.8576388888888888, + 0.8878205128205128, + 0.8652482269503546, + 0.8518518518518519, + 0.8610169491525423, + 0.8701298701298701, + 0.871875, + 0.8866666666666667, + 0.881578947368421, + 0.8496503496503497, + 0.8782051282051282, + 0.8694267515923567, + 0.8627450980392157, + 0.8669064748201439, + 0.8709677419354839, + 0.8701298701298701, + 0.8438661710037175, + 0.8683274021352313, + 0.9013157894736842, + 0.8878205128205128, + 0.8949152542372881, + 0.8682432432432432, + 0.8758620689655172, + 0.8493150684931506, + 0.8770226537216829, + 0.8582089552238806, + 0.878419452887538, + 0.8754578754578755, + 0.889589905362776, + 0.8549618320610687, + 0.8714733542319749, + 0.8391608391608392, + 0.8859060402684564, + 0.8668941979522184, + 0.8969072164948454, + 0.8740458015267175, + 0.8777777777777778, + 0.9016393442622951, + 0.875, + 0.8819188191881919, + 0.8637992831541219, + 0.8439716312056738, + 0.8809523809523809, + 0.8686131386861314, + 0.8713235294117647, + 0.8522727272727273, + 0.8571428571428571, + 0.8470149253731343, + 0.891156462585034, + 0.8788927335640139, + 0.890728476821192, + 0.8561643835616438, + 0.8556338028169014, + 0.8551724137931035, + 0.8918032786885246, + 0.9006622516556292, + 0.8737201365187713, + 0.8791946308724832, + 0.9082278481012658, + 0.8682432432432432, + 0.8825503355704698, + 0.8873720136518771, + 0.8790849673202614, + 0.8825622775800712, + 0.8612903225806452, + 0.8720538720538721, + 0.8940397350993378, + 0.8673835125448028, + 0.8813559322033898, + 0.8713826366559485, + 0.8896321070234113, + 0.8896551724137931, + 0.8979591836734694, + 0.8807947019867549, + 0.8639240506329114, + 0.886986301369863, + 0.8640776699029126, + 0.8736842105263158, + 0.9009287925696594, + 0.8732876712328768, + 0.8790849673202614, + 0.8714285714285714, + 0.8987341772151899, + 0.8710801393728222, + 0.872852233676976, + 0.8807017543859649, + 0.8673469387755102, + 0.8648648648648649, + 0.8819444444444444, + 0.8711864406779661, + 0.8546712802768166, + 0.8741721854304636, + 0.8575949367088608, + 0.8633333333333333, + 0.8721311475409836, + 0.8825622775800712, + 0.8881789137380192, + 0.89419795221843, + 0.8833333333333333, + 0.8689138576779026, + 0.8862876254180602, + 0.9069767441860465, + 0.8851351351351351, + 0.8785714285714286, + 0.8859060402684564, + 0.8896103896103896, + 0.8677966101694915, + 0.8775510204081632, + 0.875, + 0.8620689655172413, + 0.8581081081081081, + 0.8592057761732852, + 0.8805460750853242, + 0.875, + 0.888135593220339, + 0.8713235294117647, + 0.8947368421052632, + 0.8299595141700404, + 0.9919354838709677, + 0.8927335640138409, + 0.8780487804878049, + 0.8977272727272727, + 0.8656716417910447, + 0.8768115942028986, + 0.8931034482758621, + 0.8780487804878049, + 0.8533834586466166, + 0.8736842105263158, + 0.862876254180602, + 1.0, + 0.8847457627118644, + 0.8544776119402985, + 0.9013157894736842, + 0.8795620437956204, + 0.835820895522388, + 0.8566176470588235, + 0.8620689655172413, + 0.8526315789473684, + 0.8587786259541985, + 0.8622047244094488, + 0.8611111111111112, + 0.8529411764705882, + 0.8692579505300353, + 0.8555555555555555, + 0.8687943262411347, + 0.8571428571428571, + 0.9123376623376623, + 0.8862068965517241, + 0.8797250859106529, + 0.8522727272727273, + 0.8916967509025271, + 0.8586572438162544, + 0.864963503649635, + 0.8560606060606061, + 0.8611111111111112, + 0.8996865203761756, + 0.8466898954703833, + 0.8566037735849057, + 0.8597122302158273, + 0.8788927335640139, + 0.9072847682119205, + 0.8528301886792453, + 0.8456140350877193, + 0.8534798534798534, + 0.8771331058020477, + 0.8759398496240601, + 0.8698630136986302, + 0.8807692307692307, + 0.8599221789883269, + 0.8659420289855072, + 0.8815331010452961, + 0.8639705882352942, + 0.8636363636363636, + 0.8501742160278746, + 0.8694029850746269, + 0.8708487084870848, + 0.8566308243727598, + 0.8802816901408451, + 0.8892988929889298, + 0.8767123287671232, + 0.8720538720538721, + 0.8458781362007168, + 0.8918032786885246, + 0.875, + 0.9007352941176471, + 0.8785714285714286, + 0.8450704225352113, + 0.8536585365853658, + 0.8445229681978799, + 0.8382352941176471, + 0.8715277777777778, + 0.8954248366013072, + 0.89, + 0.9922480620155039, + 0.9013157894736842, + 0.8610169491525423, + 0.867109634551495, + 0.8952702702702703, + 0.8778135048231511, + 0.8833922261484098, + 0.8552631578947368, + 1.0, + 1.0, + 0.9042904290429042, + 0.9025974025974026, + 0.9846153846153847, + 0.8793103448275862, + 0.8424908424908425, + 0.8503649635036497, + 0.8532423208191127, + 0.8609271523178808, + 0.8448275862068966, + 0.87248322147651, + 0.991869918699187, + 0.8913043478260869, + 0.8762886597938144, + 0.9849624060150376, + 0.8203125, + 1.0, + 0.8436363636363636, + 0.8807017543859649, + 0.8556338028169014, + 0.8585858585858586, + 0.9782608695652174, + 0.8586956521739131, + 0.8415492957746479, + 0.8771929824561403, + 1.0, + 0.8552188552188552, + 1.0, + 0.8636363636363636, + 0.9915254237288136, + 0.8766666666666667, + 0.8624161073825504, + 0.9927007299270073, + 0.8513011152416357, + 0.8655737704918033, + 0.9922480620155039, + 1.0, + 0.8829787234042553, + 0.8823529411764706, + 0.8562091503267973, + 0.8619528619528619, + 0.992, + 0.8621908127208481, + 0.9844961240310077, + 0.8766233766233766, + 0.8694158075601375, + 0.9861111111111112, + 0.8423076923076923, + 0.8488745980707395, + 0.8363636363636363, + 0.9927536231884058, + 0.8428571428571429, + 0.8595317725752508, + 1.0, + 0.8885245901639345, + 0.8596491228070176, + 0.8983050847457628, + 0.9925373134328358, + 1.0, + 0.9922480620155039, + 1.0, + 0.8803986710963455, + 1.0, + 1.0, + 0.8775510204081632, + 0.8723404255319149, + 0.8717948717948718, + 1.0, + 0.8737201365187713, + 0.8732876712328768, + 1.0, + 0.9923076923076923, + 0.8885017421602788, + 0.8768115942028986, + 0.8896321070234113, + 0.8547297297297297, + 0.863481228668942, + 0.8571428571428571, + 0.8737201365187713, + 0.8719723183391004, + 0.8440677966101695, + 0.8540925266903915, + 0.8694267515923567, + 0.8560885608856088, + 1.0, + 0.8771929824561403, + 0.875, + 0.8571428571428571, + 0.8585858585858586, + 0.8741496598639455, + 0.9027777777777778, + 0.8596491228070176, + 0.8586206896551725, + 0.8885245901639345, + 0.8716216216216216, + 0.8695652173913043, + 0.8703071672354948, + 0.991869918699187, + 0.887459807073955, + 0.8885135135135135, + 0.907051282051282, + 0.9906542056074766, + 0.8626198083067093, + 0.8637992831541219, + 1.0, + 0.8797250859106529, + 0.9850746268656716, + 0.8782051282051282, + 0.8611111111111112, + 0.8712871287128713, + 0.8478260869565217, + 0.8979591836734694, + 0.8458904109589042, + 0.8496503496503497, + 0.9858156028368794, + 1.0, + 0.8660130718954249, + 0.8673469387755102, + 0.9925373134328358, + 0.8411552346570397, + 0.9923076923076923, + 0.8571428571428571, + 0.8590604026845637, + 0.84, + 0.8604651162790697, + 0.9928571428571429, + 1.0, + 0.992, + 1.0, + 0.8493150684931506, + 1.0, + 0.8650519031141869, + 0.9047619047619048, + 1.0, + 1.0, + 0.8802816901408451, + 0.893687707641196, + 0.8703703703703703, + 0.992, + 0.8835616438356164, + 0.8754208754208754, + 0.8297872340425532, + 0.9919354838709677, + 0.9140127388535032, + 0.9925373134328358, + 0.9922480620155039, + 0.8873720136518771, + 1.0, + 0.8673139158576052, + 1.0, + 0.9922480620155039, + 0.8378378378378378, + 0.8701298701298701, + 0.8591065292096219, + 0.8625429553264605, + 0.8591549295774648, + 1.0, + 0.8609271523178808, + 0.9097744360902256, + 0.9157894736842105, + 0.9202898550724637, + 0.9214285714285714, + 0.9367088607594937, + 0.9039145907473309, + 0.9024390243902439, + 0.8905660377358491, + 0.9249146757679181, + 0.9244604316546763, + 0.9238754325259516, + 0.8916967509025271, + 0.8938356164383562, + 0.9143835616438356, + 0.9107806691449815, + 0.8901098901098901, + 0.897887323943662, + 0.9185185185185185, + 0.9178571428571428, + 0.9078498293515358, + 0.9125874125874126, + 0.9151943462897526, + 0.9331103678929766, + 0.9210526315789473, + 0.9142857142857143, + 0.9345454545454546, + 0.9139072847682119, + 0.9163763066202091, + 0.9175627240143369, + 0.8972332015810277, + 0.9052631578947369, + 0.9080882352941176, + 0.9233870967741935, + 0.9326241134751773, + 0.9304029304029304, + 0.9290780141843972, + 0.9203187250996016, + 0.900709219858156, + 0.8986013986013986, + 0.8992805755395683, + 0.8908450704225352, + 0.9343065693430657, + 0.9222614840989399, + 0.9176470588235294, + 0.9450171821305842, + 0.9244604316546763, + 0.9210526315789473, + 0.9236641221374046, + 0.9207547169811321, + 0.9307692307692308, + 0.9357429718875502, + 0.9479553903345725, + 0.9053030303030303, + 0.9252669039145908, + 0.9285714285714286, + 0.9144981412639405, + 0.9460431654676259, + 0.921146953405018, + 0.9039145907473309, + 0.8804347826086957, + 0.8989547038327527, + 0.9138576779026217, + 0.9453924914675768, + 0.9267399267399268, + 0.9061371841155235, + 0.9154411764705882, + 0.9142857142857143, + 0.8957528957528957, + 0.8935361216730038, + 0.9068100358422939, + 0.9233576642335767, + 0.8985507246376812, + 0.9186440677966101, + 0.9032258064516129, + 0.8924731182795699, + 0.9130434782608695, + 0.9206896551724137, + 0.911660777385159, + 0.9028776978417267, + 0.9132075471698113, + 0.916955017301038, + 0.8938356164383562, + 0.9011406844106464, + 0.9185185185185185, + 0.9184397163120568, + 0.9154411764705882, + 0.916030534351145, + 0.9201388888888888, + 0.8978102189781022, + 0.8974358974358975, + 0.9128919860627178, + 0.896797153024911, + 0.9172413793103448, + 0.9107806691449815, + 0.8982456140350877, + 0.9032258064516129, + 0.9273356401384083, + 0.8773234200743495, + 0.9075342465753424, + 0.9068100358422939, + 0.9157088122605364, + 0.9087591240875912, + 0.9094202898550725, + 0.9214285714285714, + 0.9166666666666666, + 0.9236111111111112, + 0.9283018867924528, + 0.9074074074074074, + 0.9052631578947369, + 0.9145907473309609, + 0.9049295774647887, + 0.9063670411985019, + 0.9105960264900662, + 0.8960573476702509, + 0.8924731182795699, + 0.9, + 0.911660777385159, + 0.9192307692307692, + 0.9051383399209486, + 0.9040590405904059, + 0.9122807017543859, + 0.8960573476702509, + 0.9084249084249084, + 0.9222614840989399, + 0.9107806691449815, + 1.0, + 0.9326241134751773, + 0.9142857142857143, + 0.9361702127659575, + 0.8972332015810277, + 0.924187725631769, + 0.9335664335664335, + 0.9257950530035336, + 0.9210526315789473, + 0.9163763066202091, + 0.8893129770992366, + 1.0, + 0.9198606271777003, + 0.8970588235294118, + 0.9405594405594405, + 0.9372822299651568, + 0.9107806691449815, + 0.9042553191489362, + 0.9195804195804196, + 0.9113475177304965, + 0.8884758364312267, + 0.9029850746268657, + 0.9101123595505618, + 0.9169675090252708, + 0.9047619047619048, + 0.8884615384615384, + 0.9040590405904059, + 0.9175627240143369, + 0.9372693726937269, + 0.939622641509434, + 0.9413793103448276, + 0.945054945054945, + 0.9285714285714286, + 0.9290540540540541, + 0.9222222222222223, + 0.9094202898550725, + 0.8974358974358975, + 0.912621359223301, + 0.8986013986013986, + 0.9138576779026217, + 0.9133858267716536, + 0.8938356164383562, + 0.9354838709677419, + 0.9061224489795918, + 0.8996282527881041, + 0.9166666666666666, + 0.9037037037037037, + 0.9069767441860465, + 0.9257950530035336, + 0.8892988929889298, + 0.9308943089430894, + 0.9078014184397163, + 0.9107142857142857, + 0.9230769230769231, + 0.9087591240875912, + 0.926829268292683, + 0.9182156133828996, + 0.9154411764705882, + 0.8974358974358975, + 0.9154929577464789, + 0.9300411522633745, + 0.895910780669145, + 0.9100719424460432, + 0.8838028169014085, + 0.925, + 0.9072164948453608, + 0.9114391143911439, + 0.9151943462897526, + 0.8929889298892989, + 0.8960573476702509, + 0.8992805755395683, + 0.9014084507042254, + 0.8846153846153846, + 0.9258064516129032, + 0.9112627986348123, + 1.0, + 0.9219858156028369, + 0.9057239057239057, + 0.8880597014925373, + 0.8974358974358975, + 0.9383561643835616, + 0.8979591836734694, + 0.9045936395759717, + 0.9920634920634921, + 0.9916666666666667, + 0.9278996865203761, + 0.9283387622149837, + 1.0, + 0.9228070175438596, + 0.889273356401384, + 0.9, + 0.9146757679180887, + 0.926829268292683, + 0.896797153024911, + 0.9151943462897526, + 0.9921875, + 0.9139784946236559, + 0.9037037037037037, + 0.9923076923076923, + 0.9253731343283582, + 1.0, + 0.9050847457627119, + 0.9217081850533808, + 0.918918918918919, + 0.9192982456140351, + 1.0, + 0.9230769230769231, + 0.9017543859649123, + 0.8860294117647058, + 0.9844961240310077, + 0.9145907473309609, + 0.9908256880733946, + 0.9045936395759717, + 0.9761904761904762, + 0.9176029962546817, + 0.9127272727272727, + 0.9924812030075187, + 0.9163636363636364, + 0.9032258064516129, + 1.0, + 0.976, + 0.9275862068965517, + 0.9081272084805654, + 0.9013157894736842, + 0.8956228956228957, + 1.0, + 0.8949416342412452, + 0.9841269841269841, + 0.9078014184397163, + 0.9104477611940298, + 1.0, + 0.8966789667896679, + 0.8916083916083916, + 0.8884758364312267, + 1.0, + 0.9130434782608695, + 0.9049295774647887, + 1.0, + 0.89419795221843, + 0.9096989966555183, + 0.9032258064516129, + 0.9926470588235294, + 0.9838709677419355, + 1.0, + 1.0, + 0.9100346020761245, + 0.9908256880733946, + 1.0, + 0.925531914893617, + 0.9115646258503401, + 0.9121621621621622, + 0.9922480620155039, + 0.916083916083916, + 0.9072847682119205, + 1.0, + 1.0, + 0.9134948096885813, + 0.9175257731958762, + 0.8909090909090909, + 0.9075907590759076, + 0.9192982456140351, + 0.9014598540145985, + 0.8933823529411765, + 0.9217687074829932, + 0.8943396226415095, + 0.9169811320754717, + 0.9254237288135593, + 0.9042553191489362, + 1.0, + 0.9340277777777778, + 0.9068100358422939, + 0.9081632653061225, + 0.9087719298245615, + 0.9125874125874126, + 0.916083916083916, + 0.910958904109589, + 0.9047619047619048, + 0.9228187919463087, + 0.911660777385159, + 0.9, + 0.9018181818181819, + 0.984251968503937, + 0.9178571428571428, + 0.9107142857142857, + 0.9266666666666666, + 1.0, + 0.9134948096885813, + 0.903010033444816, + 0.9754098360655737, + 0.9147540983606557, + 1.0, + 0.920265780730897, + 0.9039145907473309, + 0.9222972972972973, + 0.9058823529411765, + 0.9075342465753424, + 0.8880597014925373, + 0.8935361216730038, + 1.0, + 0.983739837398374, + 0.9119718309859155, + 0.8853046594982079, + 1.0, + 0.9181494661921709, + 1.0, + 0.9078498293515358, + 0.8996539792387543, + 0.9125874125874126, + 0.8897058823529411, + 0.9854014598540146, + 0.984251968503937, + 1.0, + 1.0, + 0.9190140845070423, + 1.0, + 0.8923076923076924, + 0.9225589225589226, + 0.975609756097561, + 0.9923076923076923, + 0.9068100358422939, + 0.9157894736842105, + 0.9142857142857143, + 1.0, + 0.9283154121863799, + 0.926829268292683, + 0.9078498293515358, + 0.9813084112149533, + 0.922077922077922, + 0.991869918699187, + 1.0, + 0.9236363636363636, + 0.9916666666666667, + 0.9056603773584906, + 1.0, + 1.0, + 0.9114391143911439, + 0.8996539792387543, + 0.9111842105263158, + 0.9125874125874126, + 0.888135593220339, + 1.0, + 0.9152542372881356 + ] + }, + "contact_maps": { + "mean_pearson": 0.1420678144607972, + "valid_tracks": 28, + "per_track_pearson": [ + 0.15755259002402194, + 0.13955489136590044, + 0.13375706407052598, + 0.07101721431501325, + 0.03670202354353835, + 0.08992090424958994, + 0.158368568438216, + 0.10885801586801583, + 0.2512557134320003, + 0.20043397057119824, + 0.07621827595528435, + 0.13080810250093494, + 0.21398634396782365, + 0.04875759651502359, + 0.10535619613930325, + 0.17562689409445936, + 0.06952819920648878, + 0.19313567866848788, + 0.1717703620363298, + 0.17239142612501174, + 0.09076879147286175, + 0.10055607567827658, + 0.09654416969582522, + 0.19974864429973652, + 0.18690825002544834, + 0.20592357992559285, + 0.15081628011781112, + 0.24163298259960184 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.3660549653284387, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..317decd5e33d88db570a82ff4e5ee8ef6512fc0c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S017/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.6235653817482589, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6201982594839738, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.2736899144878897, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.6587885387622007, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.34741817669845254, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.24403793569973142, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.24759813707886796, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.37106175211191283, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.15414392797681442, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.24366279462636162, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.7463370498921266, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.14311524030698292, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.21138539273780313, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/README.md new file mode 100644 index 0000000000000000000000000000000000000000..9ad9edec579acacfb1d1d920e717d361117875a3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/README.md @@ -0,0 +1,85 @@ +# SeqFunc scale13-small after 1 complete epochs with transcript-direction junction decoding + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **41,020 training examples in 10,255 optimizer updates**, 1 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +176.9 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +## Measured validation results + +Family-mean Pearson: **0.389847** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.539431 | +| dnase | 0.630721 | +| procap | 0.492437 | +| cage | 0.256694 | +| rna_seq | 0.517268 | +| chip_tf | 0.415247 | +| chip_histone | 0.438092 | +| splice_sites | 0.047388 | +| splice_usage | 0.393156 | +| junctions | 0.287302 | +| contact_maps | 0.270578 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5538 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5865 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.2304 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7565 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3773 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2146 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2988 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.2006 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1909 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2282 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5850 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2132 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2805 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..b91d9d7a3a86f14d40ac35e0b6a5e7076524db13 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch01/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..e9ee7986fa11ca00ffcde800a932dad6e1bf0666 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/package_manifest.json @@ -0,0 +1,47 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "files": { + "README.md": "82102588368aeecb3b676a2f6949325d4e5c2a9081bac7c0ea3cd138e0372c46", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "0595a0feea648c3d64e4428515828b4eb404a55ed1e2dfb27e2ff73d126e03c8", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "53c826d3593df4090e6832f77df2d21043c67e7f7fb22a5a2f51ef8dcbefc865", + "decoding_policy.json": "536c326d6752057fa48b9ad38c7120de5c9e6362068a233d999166f09f9dbc75", + "validation_tracks.json": "02e9479853d773336122f10616769d32d73866b6b5b300ded3e367516b6e1d28", + "validation_variants.json": "0b3e36ff456063327edeaff045da9126c9b1f1f0b26d323ed76c0fa6696ec5b4" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..dc0f0b20833fc99253b7dcf6023fd5e7a61b1128 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/predict.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..155ef4138af508a739d6af8d7cd283308fb7b4a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/dataset.py @@ -0,0 +1,293 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..810530f44b9dc68922a0b0b575320531240b21cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/training_provenance.json @@ -0,0 +1,119 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 1, + "offset": 0, + "global_step": 41020, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/progress.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 10612.629490868014, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 10255, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..d064069f070ee6d0415e214e3ea010b8aeccf448 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5394307915995341, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6568063875932839, + 0.6782587851826644, + 0.43185156334772395, + 0.5830929215868315, + 0.4411298648855118, + 0.4405995119653321, + 0.4526644367765145, + 0.5310417308213592, + 0.5336875754981915, + 0.5879222085433559, + 0.537655930192812, + 0.5125453457830168, + 0.3696234095978372, + 0.4914880739536804, + 0.6357453209879593, + 0.46599973150249235, + 0.514586765352444, + 0.48430395028583983, + 0.5127969196036514, + 0.49965476956950455, + 0.46861513138003186, + 0.43814458715558485, + 0.5073004390265184, + 0.4846478335853026, + 0.48586127278143837, + 0.4638399436027636, + 0.44295517900027787, + 0.46308452245583703, + 0.47332236952145573, + 0.5361316769548682, + 0.4464797210841843, + 0.4759699886916573, + 0.4333934227456335, + 0.36592002920367334, + 0.3334875321997587, + 0.5411443310222529, + 0.4842288476899477, + 0.5818000478517144, + 0.46565093935999746, + 0.4561740040164456, + 0.540509953141977, + 0.5468465172594491, + 0.455957743829696, + 0.4402964644085608, + 0.5120027593103733, + 0.5242625567246787, + 0.40450252404237647, + 0.4510523900328651, + 0.4468658548314412, + 0.49222851732291967, + 0.4615271483763196, + 0.42745376633381404, + 0.5209518203650485, + 0.5527858442517165, + 0.4217144147172788, + 0.6583399670521609, + 0.7670185438542273, + 0.5174296432859146, + 0.5694057495615129, + 0.5793292921059975, + 0.7305868452377714, + 0.45899002215927404, + 0.6747865032390965, + 0.6126465374166141, + 0.6063523442694909, + 0.4713704862680611, + 0.5312107304591029, + 0.5938209304451715, + 0.7456295450948104, + 0.6714921018048648, + 0.490858398933958, + 0.4429540519629567, + 0.4208294760411361, + 0.4986513074395492, + 0.46626613703505165, + 0.45370267359835004, + 0.46299623239234683, + 0.5002450277541637, + 0.5384596468516567, + 0.4577655790483886, + 0.5343506259708877, + 0.5567796087440376, + 0.47431927336783475, + 0.44992445058885916, + 0.48545540952653415, + 0.47146410015367013, + 0.45759659984730394, + 0.44553444272044795, + 0.41956849292401943, + 0.46210728336255436, + 0.4708844165777036, + 0.46768778411320755, + 0.4538037239859896, + 0.4665003211785806, + 0.5279431603996255, + 0.3556901352689567, + 0.41141798856322503, + 0.4040225200191718, + 0.45516897291424774, + 0.4707887478541568, + 0.424661425088321, + 0.4261868798940377, + 0.4672545724517948, + 0.432877226509591, + 0.5063821038107246, + 0.5091399749419332, + 0.49834303542289343, + 0.4807147604027043, + 0.5046103819225711, + 0.5017602939915442, + 0.5638228496685673, + 0.48538622608476667, + 0.4814339673425138, + 0.40924878638313866, + 0.43710238836820337, + 0.40159726511029775, + 0.4419562241797228, + 0.44948946455756644, + 0.4552421554058822, + 0.4395056674124156, + 0.615967203922434, + 0.6309418117702875, + 0.69620466850319, + 0.6779407700056409, + 0.6680253798696257, + 0.5415353938366567, + 0.8145206124292106, + 0.6767889160919279, + 0.7158822250594054, + 0.6488027719139996, + 0.6941753629847784, + 0.5344904551443149, + 0.5589933257737165, + 0.5342380130349218, + 0.6778887791524351, + 0.6999329581128304, + 0.6630490037741111, + 0.5127562506894642, + 0.39924532991562117, + 0.7684668827935929, + 0.7426499455509098, + 0.7577410939740734, + 0.7526029156832429, + 0.5436271629419501, + 0.8063836007313439, + 0.7213574521979109, + 0.6890777468302454, + 0.6857339966319776, + 0.7965111497541165, + 0.7612881299754917, + 0.5782774751271568, + 0.5915737491431233, + 0.6629633508225091, + 0.697194522405753, + 0.7128806374667979, + 0.5687790181223528, + 0.698679291752057, + 0.5658874834994675, + 0.568374354852192, + 0.6398948451609556, + 0.5473012593279241, + 0.7234792083983568, + 0.7376542088790192, + 0.6091430594580814, + 0.5583124535146406, + 0.5530381015782472, + 0.6972630909144314 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.6307214214259819, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5343543157288424, + 0.913932167898437, + 0.7710239905748589, + 0.4922903327467077, + 0.49184374918389484, + 0.5074628741852909, + 0.9245217247664741, + 0.693834676795347, + 0.4262479771590429, + 0.3344951981722611, + 0.8657377142203836, + 0.6812230579060337, + 0.40873543455738826, + 0.4901382888746148, + 0.4756192201451656, + 0.7970599052058963, + 0.7842234344113788, + 0.842962948203666, + 0.855465552833304, + 0.8840201107248165, + 0.6725824980060344, + 0.5946138857229347, + 0.8914317551210388, + 0.5111203071972883, + 0.5213807992925911, + 0.7715965623396498, + 0.4502962612982721, + 0.661030400339423, + 0.5091951611853786, + 0.5177080987586157, + 0.6140814717817994, + 0.450743512120933, + 0.8367918017944361, + 0.6224258030228239, + 0.8739320472368693, + 0.8813323385296999, + 0.4360830380271497, + 0.8212668257375921, + 0.8681312867022543, + 0.853478967869918, + 0.8198453562799393, + 0.6660879312198681, + 0.6980383803228727, + 0.6167690717906033, + 0.5095739588391128, + 0.6335139867312553, + 0.727782648712162, + 0.48067923333129886, + 0.5754500709861512, + 0.3926393001929363, + 0.36503450413315497, + 0.5492704002693611, + 0.47871112025798634, + 0.8194206448749274, + 0.42381672737682496, + 0.39170820039457216, + 0.44479412688615433, + 0.49025584603656713, + 0.46070473480845886, + 0.5087317192094738, + 0.43158599057794605, + 0.5926182067823048, + 0.4775216722499359, + 0.46697774116252, + 0.49376416434013076, + 0.5018768836344822, + 0.5373829360400454, + 0.45604200293372876, + 0.3710931530521904, + 0.41082287193257694, + 0.49357966750583243, + 0.625258616271321, + 0.48312055780589447, + 0.6042583007487872, + 0.3432317838412035, + 0.799581953233565, + 0.8842225872998617, + 0.6425925780452968, + 0.4252758632197314, + 0.44271200518964, + 0.5025806473817354, + 0.4857599412283257, + 0.45514648334361857, + 0.46357480218104674, + 0.5566053102781527, + 0.48576753175927256, + 0.5938226775761845, + 0.48909425375025967, + 0.6252873084115365, + 0.47434083476363303, + 0.6154662069822614, + 0.40164726052983935, + 0.7173061304340489, + 0.4572347223731034, + 0.7061634917050993, + 0.4908065442936714, + 0.7242493518492296, + 0.6907245647098852, + 0.35826991742658604, + 0.3662067770197589, + 0.4574348803631164, + 0.36289824164750284, + 0.34870742630657686, + 0.7081808838365746, + 0.6712548737195702, + 0.4000215060504816, + 0.46347773024824984, + 0.5774907296563128, + 0.8423550686633443, + 0.53733730406149, + 0.5385043270906827, + 0.558144973924187, + 0.49263415983200737, + 0.4323105855710034, + 0.7723087584071018, + 0.5428330114548985, + 0.4787247508852067, + 0.5345058151358517, + 0.6368710920893276, + 0.4361619293805519, + 0.5134652156028834, + 0.48107206097930205, + 0.5442590675130572, + 0.5337616662743399, + 0.5909169297513764, + 0.5428984491324169, + 0.6240445423486658, + 0.6510626386696347, + 0.6944937198205312, + 0.5990327005240397, + 0.45571025263998677, + 0.5976482820510204, + 0.617316749315173, + 0.7820825572534498, + 0.43927877529864057, + 0.32138894893993786, + 0.5753520846517645, + 0.4657862595801792, + 0.633888291366312, + 0.5996711971789981, + 0.6069740842268861, + 0.5548379453690823, + 0.473285263928278, + 0.7664837549673444, + 0.6050321132101376, + 0.7560975672001085, + 0.3889702409762628, + 0.3668096510686739, + 0.3890112783658817, + 0.3539524350787504, + 0.8858746839379396, + 0.3206581354764425, + 0.8833819079228769, + 0.7300837522803626, + 0.45549947136636576, + 0.3483977242009274, + 0.3510293607153961, + 0.3648867796778575, + 0.6389892790497921, + 0.7240459446435495, + 0.721419672546245, + 0.6510025803719783, + 0.553683064120451, + 0.4702803227340975, + 0.6758799160192798, + 0.6057094703623292, + 0.484641305310411, + 0.6202375088008139, + 0.452170129865745, + 0.47829679080362103, + 0.3999522423541187, + 0.822355075833945, + 0.4457755750302672, + 0.4455664184097475, + 0.6351743292705356, + 0.688345827554367, + 0.5066430066050069, + 0.3382673833570333, + 0.8680994209606188, + 0.36942563295347614, + 0.7589136168970431, + 0.4875865001384727, + 0.5829504264154499, + 0.7091961305111081, + 0.44622849723129027, + 0.5496687120304035, + 0.5866129823128787, + 0.8196914026687158, + 0.40788683423306227, + 0.6050577500747628, + 0.544731908862392, + 0.687183806367007, + 0.8860010554580816, + 0.4256114977209997, + 0.5874165029307811, + 0.48230023384591025, + 0.6149864403865627, + 0.6993075408417767, + 0.5527423814160365, + 0.4513884183492155, + 0.8605899682091153, + 0.8697788163307781, + 0.741738290569185, + 0.8232040694816156, + 0.8351842680597032, + 0.8239358853246249, + 0.8407306468397893, + 0.3579757265462674, + 0.37805995454087155, + 0.5005333734683605, + 0.4556847492576298, + 0.44356930486166496, + 0.5884007951050868, + 0.8495882509485668, + 0.8365957361349777, + 0.6669565517600445, + 0.8175363303981835, + 0.8105167172895456, + 0.8433133048956373, + 0.8938699804852441, + 0.5157849484362681, + 0.5828322799489215, + 0.6629509954295669, + 0.6132072779011716, + 0.8126816812146717, + 0.5691893919216436, + 0.9029318518681846, + 0.8745741385119701, + 0.8590557419507556, + 0.8918537366849685, + 0.8824365410782878, + 0.8651439206746019, + 0.8239349693740609, + 0.8477403352838594, + 0.894635177036424, + 0.44051875273225594, + 0.9164993751954764, + 0.8668225464784947, + 0.8075984861684768, + 0.5936446837997025, + 0.8613086650057633, + 0.5141032676451817, + 0.7376491307192377, + 0.8599137993313474, + 0.577932981766797, + 0.5672186286160437, + 0.6809064489883637, + 0.516888170238399, + 0.5396819731216049, + 0.7159183853372297, + 0.8782928650631946, + 0.7280162878519996, + 0.8077799521457762, + 0.6139524824009537, + 0.8541166264718371, + 0.7974595107463764, + 0.8297204795974068, + 0.8948343565041829, + 0.8202201380158015, + 0.8867230730053663, + 0.8548308592756291, + 0.525921766869949, + 0.8646326991096974, + 0.5595250747988392, + 0.5286247621032005, + 0.8620302484660032, + 0.890834457975205, + 0.8663655712269369, + 0.582997200624464, + 0.618662076586956, + 0.653686850066655, + 0.7827539197397307, + 0.8817705637224622, + 0.6988878970789301, + 0.8929218719905325, + 0.9038566668768565, + 0.6056133868159647, + 0.32653882915560356, + 0.5175249446759725, + 0.887061042838238, + 0.4376412983316149, + 0.6075262810520489, + 0.8756357843911036, + 0.8156295740470169, + 0.6167666349131979, + 0.8509110989246945, + 0.8840783295815479, + 0.8759621839110417, + 0.8672337490334927, + 0.8554268831511307, + 0.8792653674382622, + 0.8483000869860053, + 0.9138211385899008, + 0.7680147535774807, + 0.8248720362886595, + 0.8933468861258065, + 0.5471051027799351, + 0.5542088495321453, + 0.5451123771114108, + 0.5258786541589121, + 0.8950843241289194, + 0.7894873921228436, + 0.8351123830446195, + 0.7872042069854852, + 0.7581348182187376 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.4924369905316472, + "valid_tracks": 12, + "per_track_pearson": [ + 0.45994834550212027, + 0.5165991315577544, + 0.4661725998079823, + 0.5070403944807768, + 0.5040910880114275, + 0.5287185682295517, + 0.4394755945744672, + 0.5019379036811955, + 0.4670197812459195, + 0.5034208441082331, + 0.47753200647813837, + 0.5372876287022001 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.2566936337791563, + "valid_tracks": 546, + "per_track_pearson": [ + 0.24079413248518725, + 0.30135262163323884, + 0.262669822762357, + 0.1635815269443197, + 0.0882235841006401, + 0.26575061667910177, + 0.15112409282698477, + 0.17668252410538304, + 0.27437264008402784, + 0.22958508043840808, + 0.2830159506448719, + 0.15754891334640206, + 0.17960090326032296, + 0.1494391211761189, + 0.12373912276622225, + 0.17178028632146683, + 0.2188082904514429, + 0.2978026880258185, + 0.29430787563804606, + 0.2783958113449603, + 0.22913608037455035, + 0.22840173313202733, + 0.3502420116795119, + 0.17514826393725952, + 0.20991040775697778, + 0.27797436769233574, + 0.2557284587507735, + 0.2696092873872962, + 0.2807099963388501, + 0.28479956855257915, + 0.1857554084780936, + 0.32027082446953037, + 0.18682910284969947, + 0.17233618520432553, + 0.31326898117595536, + 0.2640482074582295, + 0.3173682170704043, + 0.34311830497368295, + 0.32907843939074394, + 0.3115089175439312, + 0.35078554853579685, + 0.2836303963049096, + 0.26387881181073897, + 0.2563060671260773, + 0.17484787562765103, + 0.2441883962617676, + 0.25576377242321674, + 0.27023512791295856, + 0.16111613512335962, + 0.20057881584179393, + 0.30535999162687816, + 0.282155518177094, + 0.2966714028042927, + 0.21094269198520701, + 0.3470703455768461, + 0.34270203956774176, + 0.1929047795040335, + 0.35086599371229427, + 0.23749841391918755, + 0.34424393089917377, + 0.3103596274136563, + 0.20846938428929884, + 0.2762235883713524, + 0.21569842397023195, + 0.2867843976419106, + 0.33407954240326676, + 0.3282046396233167, + 0.21092805879551624, + 0.14997207161229925, + 0.12574915279419743, + 0.25387402957050803, + 0.29237174290378254, + 0.25957005439918, + 0.25576751940006165, + 0.2910653794592504, + 0.2427883471741269, + 0.29031468166914187, + 0.2665561776008333, + 0.30793713085851954, + 0.302815754228068, + 0.2627148950072898, + 0.1469922155740767, + 0.2744141284251255, + 0.3562726836559775, + 0.1481483714493154, + 0.265068881384569, + 0.23877456867131894, + 0.3410456232771925, + 0.207424341674316, + 0.28887221310602523, + 0.21952262511030207, + 0.2690810630279838, + 0.2677937312965424, + 0.13300039836810176, + 0.2884199496671783, + 0.2441540141287035, + 0.27776101623208455, + 0.2756881577798522, + 0.25809583860326113, + 0.1532018937816255, + 0.25091569739627956, + 0.2696314446425335, + 0.29565891800501226, + 0.3491068023702593, + 0.31831454346319277, + 0.30163835155981966, + 0.22696500522852261, + 0.2892267280020694, + 0.22663522306880107, + 0.23115294996254962, + 0.24417188553570926, + 0.23898703967416937, + 0.2664584289230822, + 0.2604943226381977, + 0.2172435153426876, + 0.28394144112942926, + 0.2650904070553038, + 0.2723522382739176, + 0.2594960423371462, + 0.29946549407723644, + 0.2967799166380468, + 0.28892532146599087, + 0.18977434299163917, + 0.21668692133257172, + 0.24942702563145489, + 0.28337228314823326, + 0.26735908438967, + 0.28080022452222897, + 0.3624084412392518, + 0.3225067956372307, + 0.3046829935630958, + 0.220211446384226, + 0.30158581332583495, + 0.3399289863963341, + 0.3619956369512898, + 0.2612469364569484, + 0.3282236074359418, + 0.2614765568300504, + 0.1536163937361434, + 0.3098286050825549, + 0.3188029272633255, + 0.2680770591233613, + 0.30068773364812174, + 0.22803909459342184, + 0.2496113619969351, + 0.270812462897189, + 0.27241110389805656, + 0.3736559880120426, + 0.24444720206835052, + 0.24541071509473714, + 0.3285777092072438, + 0.24044694528548696, + 0.29532882151466616, + 0.29031389701269145, + 0.2684247161774354, + 0.27385806583421746, + 0.34195170162954114, + 0.3682146729105913, + 0.23385750035966543, + 0.2291649271773779, + 0.1888723387722842, + 0.24220127789076334, + 0.25717229606260156, + 0.2872137697120168, + 0.26274410750556654, + 0.2268054341388765, + 0.2698905472500097, + 0.2085438684752421, + 0.15285062055819945, + 0.325225071061801, + 0.271743828104815, + 0.2235284394336874, + 0.2749649778067554, + 0.3736157366720345, + 0.27243341801634985, + 0.1529510246728562, + 0.17640545170271188, + 0.24819169236155847, + 0.31341450449096886, + 0.32042242396530507, + 0.3206017423679301, + 0.3298857047941512, + 0.18202866382463942, + 0.30703661760733575, + 0.26872198056473906, + 0.22147133493400398, + 0.26863028384544296, + 0.18572722838537584, + 0.3621521335597168, + 0.28121300130788, + 0.23701805339793303, + 0.17536258039992766, + 0.17482408908734529, + 0.24961072410637952, + 0.2690704876936525, + 0.23516088549107958, + 0.26739552166875735, + 0.2796665003992606, + 0.19371147164965635, + 0.2727077983950676, + 0.1341502264749919, + 0.13469672113089032, + 0.12101239977866086, + 0.10861245204790089, + 0.16283069517432047, + 0.2587273269924546, + 0.25917134458800134, + 0.2247243385448137, + 0.11086468564266072, + 0.23206566336523957, + 0.31895678480733947, + 0.31183011651862336, + 0.3214047979294091, + 0.3011161962618119, + 0.28186618720808104, + 0.2497049951799422, + 0.32058892165804476, + 0.327943744347737, + 0.2903646203708518, + 0.32412140063532224, + 0.29974124112488487, + 0.23118558404563203, + 0.29276736398908737, + 0.28137599399638397, + 0.3402622221406527, + 0.3375826177951103, + 0.2856696186392496, + 0.28061096151476655, + 0.3044631427036624, + 0.329233091860567, + 0.2011367075979247, + 0.20439150565505668, + 0.34487762635551633, + 0.255588063274837, + 0.23027169338701584, + 0.25173712726514474, + 0.3526333426253374, + 0.2032047199077948, + 0.29296727794910343, + 0.2848455304827916, + 0.1148460126468668, + 0.21999012013642646, + 0.16281688282589563, + 0.30169640994817526, + 0.24341666557134345, + 0.28877326930358466, + 0.11652988718416336, + 0.3357082586485, + 0.2691116226925242, + 0.3191340199055214, + 0.30017760835086693, + 0.20121346206799703, + 0.365467047234304, + 0.14341865523086864, + 0.31361321696906574, + 0.19388524081484998, + 0.357909327462538, + 0.3211905706182193, + 0.32990991481375065, + 0.38150928048802524, + 0.2509476365586837, + 0.2900391955865857, + 0.12622536502535578, + 0.2792076666075345, + 0.3164102254215043, + 0.21402082091225555, + 0.12764086596426072, + 0.13259363446501457, + 0.15091463554111828, + 0.2688277680165596, + 0.33690981731580805, + 0.14351369378803724, + 0.11746245928798985, + 0.24195715896093628, + 0.3151161368386839, + 0.2638338585335448, + 0.17845473613617202, + 0.07458457373132021, + 0.27164858554325605, + 0.15849868344635018, + 0.1990091947889901, + 0.26890215751612917, + 0.2272209377720953, + 0.26884386627131945, + 0.1538877657153546, + 0.18994093577600976, + 0.16622286301742548, + 0.14013794764134455, + 0.16526818464503673, + 0.21930413481062777, + 0.2851104600182271, + 0.28670292391448277, + 0.28065217289053374, + 0.22746464015818105, + 0.23020154308337568, + 0.3316144411207771, + 0.1703248123200761, + 0.1993827720332153, + 0.31390326502854643, + 0.23706859572051628, + 0.25785211922161294, + 0.27727028756676697, + 0.2779867841171741, + 0.17208059526419117, + 0.3097469895231183, + 0.17521743309181828, + 0.1608044953444863, + 0.3065109299948886, + 0.2596660202057869, + 0.3474501373685932, + 0.35052529380813785, + 0.37251648006495314, + 0.3065795203730625, + 0.335133219506913, + 0.2693143553916457, + 0.2567939262407778, + 0.26909102912946925, + 0.19678305649260566, + 0.24891646482623989, + 0.27333329296982456, + 0.3275801773497707, + 0.1609767991185708, + 0.2087469584994305, + 0.28005898008341124, + 0.28411476423219995, + 0.2751718165844965, + 0.20363682837946576, + 0.34045207252758625, + 0.3155048593731087, + 0.2031943501646084, + 0.33862048960481245, + 0.2080443024248426, + 0.3540301353150644, + 0.3009966080189566, + 0.19984926993412405, + 0.26720967149416835, + 0.21885679287983936, + 0.2871687660439257, + 0.3143013012141053, + 0.30843995151004017, + 0.20882115589624664, + 0.1509412623721792, + 0.12995274091923809, + 0.2518110912324464, + 0.29625936959016014, + 0.2651299691749014, + 0.2549276168458399, + 0.28971724675698296, + 0.23176140041353702, + 0.2596013138798974, + 0.24182043168724443, + 0.29365054971548016, + 0.30284633133528244, + 0.2578803289030807, + 0.1529109148137871, + 0.2741197513548404, + 0.36315914140624944, + 0.15538531962505456, + 0.2811066945210919, + 0.27489554235097063, + 0.34468993064004866, + 0.19285406027794066, + 0.29435834855690857, + 0.2211825575596529, + 0.2680401464203765, + 0.27004004088524086, + 0.13607452635163392, + 0.27285263659515546, + 0.22417488737112054, + 0.26570625145739496, + 0.28860900125402866, + 0.24972863349131225, + 0.1490653382140917, + 0.25520780491672024, + 0.2624400959172932, + 0.29819905357373483, + 0.3314690678431127, + 0.2989752536259571, + 0.2969074622363315, + 0.22130039464324133, + 0.27376126268103007, + 0.21322400589563498, + 0.2225366459635585, + 0.23297552241985073, + 0.23378521405580271, + 0.261105623156991, + 0.2608502907833806, + 0.21405176122428635, + 0.2867864157880865, + 0.25238275646748876, + 0.261747673004631, + 0.2524586258089838, + 0.28935961527890663, + 0.2878982581538957, + 0.2847187826537373, + 0.19105437123741004, + 0.23191587031593192, + 0.25163424963703646, + 0.2924899459027051, + 0.25045322738757997, + 0.2888570225980087, + 0.3523045124722001, + 0.33970152989046315, + 0.29563979273064306, + 0.20582899417596354, + 0.27963597954796676, + 0.32359931196934155, + 0.3550886359754242, + 0.2497955896973299, + 0.3288114260231175, + 0.26730157257325277, + 0.16345464161954126, + 0.3128465019876067, + 0.2881091688191805, + 0.25293306242600755, + 0.31120094267855347, + 0.2146163960332468, + 0.24082137147014154, + 0.2565749280478092, + 0.25785668528302297, + 0.35637998034368434, + 0.28052532391090845, + 0.23781734322331413, + 0.3260889409400163, + 0.2517547859348116, + 0.2857973523299226, + 0.2837038030937281, + 0.2658680790182495, + 0.2814969794113462, + 0.3187216484565772, + 0.3534180667089768, + 0.21564861369210542, + 0.24028126476985814, + 0.1871126878325487, + 0.21802134382188648, + 0.2613744007968565, + 0.3080145377108629, + 0.28604978502472456, + 0.2137006779223682, + 0.2503394546435059, + 0.22653602435351944, + 0.1377895189095692, + 0.34320845719596726, + 0.27025249721231603, + 0.22480544692398943, + 0.26444620678871905, + 0.3617924868659705, + 0.27499266610721196, + 0.13587016458520182, + 0.18004954504918974, + 0.26751080193248816, + 0.276015583084656, + 0.3173786156760387, + 0.3242360646403823, + 0.32317191520728744, + 0.18719898168755242, + 0.31064079569646824, + 0.289522831372106, + 0.23371723600783187, + 0.30103470801870447, + 0.2076345706128419, + 0.3453663215723064, + 0.2719763658446778, + 0.2306229994460569, + 0.20405828585950567, + 0.15435550119719674, + 0.19945562140118095, + 0.29592653530882435, + 0.24687984468990168, + 0.28398406931977055, + 0.2773792822434558, + 0.2066395756650001, + 0.2723382477370396, + 0.11598916604415088, + 0.10411758847594349, + 0.09816998579442368, + 0.09540887206223805, + 0.12301270038675366, + 0.2664863877987237, + 0.21302994353249013, + 0.21492191028520757, + 0.10606719733953848, + 0.22754952177029175, + 0.3188485293601346, + 0.3537127623687728, + 0.35400640600597744, + 0.3276567545662311, + 0.31261030200539824, + 0.25998044768887835, + 0.3349348097165322, + 0.3307561020955533, + 0.2797983487168692, + 0.3301700788562566, + 0.3065107974880387, + 0.2641502734635647, + 0.3125891781718526, + 0.2945539250822049, + 0.37239649375747785, + 0.3361180777083504, + 0.29309228255048947, + 0.2833460246556456, + 0.29734344905486043, + 0.32685129626495085, + 0.16118793079071403, + 0.15497433252856835, + 0.36538666931833047, + 0.2563731000669495, + 0.24234329885415934, + 0.25023998248780965, + 0.36658510202449024, + 0.24719398473580298, + 0.2917043297610272, + 0.2786492047054853, + 0.11050015591121161, + 0.19440887589633193, + 0.15198784424571884, + 0.31114630843022584, + 0.2541943611282751, + 0.2951304657674909, + 0.11480119862237934, + 0.31899916033768194, + 0.27771424353556495, + 0.3285411991452898, + 0.3025065063821372, + 0.202739330345678, + 0.3946335595481437, + 0.14573202864842513, + 0.2971233319918931, + 0.17934400159273228, + 0.3377420093543397, + 0.3241794413412709, + 0.351305486255051, + 0.36069188566594323, + 0.2595187297139158, + 0.2823984224346779, + 0.11654856298526929, + 0.2644195904988238, + 0.3188490332752882, + 0.24272561691959035, + 0.11773713149211854, + 0.1258845692721771, + 0.14380593327676464, + 0.2700929202673903, + 0.319565847186528, + 0.12143384997984344, + 0.09913843823410852 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.5172675498096178, + "valid_tracks": 613, + "per_track_pearson": [ + 0.45470439479738917, + 0.6086120429963947, + 0.5182393231652582, + 0.5621743851302102, + 0.5894333194509228, + 0.3376106929067236, + 0.40831018433130384, + 0.5733800382279225, + 0.5613654408218689, + 0.5847627217056497, + 0.5085328597310085, + 0.5030573565343641, + 0.4333232736138895, + 0.5445326750540587, + 0.49986145905323487, + 0.5275580732971402, + 0.6062881571230377, + 0.5191378538124822, + 0.5606259094158292, + 0.5021895485931661, + 0.45945157590238694, + 0.6043767696400332, + 0.4398933176931892, + 0.41791403391197124, + 0.38322701150458116, + 0.6096584103441873, + 0.49169806538099137, + 0.549818326337245, + 0.5445668654532934, + 0.5656028039502332, + 0.5004639607468055, + 0.5438545537728828, + 0.551357266332181, + 0.499133229383373, + 0.5821904965262015, + 0.4814251503622181, + 0.5573997253969343, + 0.5552050840974165, + 0.5271615314400065, + 0.5431540305254798, + 0.5748167933092071, + 0.49438428461911943, + 0.5621047380763129, + 0.5885042657728874, + 0.4769901573255967, + 0.38959234719909114, + 0.4793325279527168, + 0.5794133489939547, + 0.512519765445788, + 0.5784548777640461, + 0.5751916933261595, + 0.5571917555996514, + 0.3890969132221538, + 0.4947511114371156, + 0.3621136080325254, + 0.5290494119003796, + 0.4965922905410572, + 0.5927291580557593, + 0.6245631383584871, + 0.6516561566232492, + 0.6513214412471107, + 0.5855953287109624, + 0.6120861965382444, + 0.407885902078973, + 0.5733252539693594, + 0.5883375775282612, + 0.6010740280737032, + 0.6226132953168639, + 0.5899770821337169, + 0.6437651666956818, + 0.5737297283896474, + 0.5957252881849795, + 0.5460946864491348, + 0.5794602351982743, + 0.5601287557047921, + 0.5635806669030735, + 0.5892608405175991, + 0.6104681696864726, + 0.5175932570143722, + 0.6316225435288781, + 0.5783874754783234, + 0.45413481511882053, + 0.6302819887375707, + 0.599646280342351, + 0.5285427298616341, + 0.5612757493849774, + 0.6145901301548962, + 0.4599390430977731, + 0.572026893165636, + 0.48085817426508815, + 0.4483166490897322, + 0.6320615959972978, + 0.6078735813351385, + 0.592867319665595, + 0.5847967473055917, + 0.6134668999271021, + 0.629000662733586, + 0.5914342982355978, + 0.5216697663113158, + 0.4536902997285064, + 0.4053418646756031, + 0.3992465842679246, + 0.4670246982010749, + 0.5952494540149483, + 0.4488732759685867, + 0.4684790909921331, + 0.6060974437742256, + 0.5190088370717619, + 0.605116455404589, + 0.5046392524479947, + 0.5671846156356976, + 0.5632502887513304, + 0.4837584566995142, + 0.5142287186075649, + 0.5982022652697357, + 0.5574018313041642, + 0.5767779096371196, + 0.5052375512237681, + 0.4936147113924656, + 0.5133026341099733, + 0.5763454193446472, + 0.5229073736420559, + 0.48715635720083444, + 0.5535782986344752, + 0.6286951248831436, + 0.3576388252405222, + 0.4062324125978019, + 0.36783027778921035, + 0.40064850219167086, + 0.48172437060496937, + 0.5968852917602089, + 0.4646311828597436, + 0.5019731466226836, + 0.5300301116085605, + 0.5262227563044866, + 0.5822608125242664, + 0.5650782524753103, + 0.6058948570006732, + 0.4762938510304224, + 0.5777189313817093, + 0.4472067337278739, + 0.545127718467862, + 0.5976162664331865, + 0.48604698026159565, + 0.3951584765092523, + 0.4602395194135049, + 0.4799740263985197, + 0.4950195149982312, + 0.47038326182469975, + 0.46963135788168026, + 0.47563705164964143, + 0.46616954669590416, + 0.5716426812917397, + 0.4287167334809963, + 0.47737410232250693, + 0.4450483334325239, + 0.5350035075915439, + 0.4593318809383125, + 0.5402293830707037, + 0.543842460190975, + 0.5816990721154124, + 0.5213772352188459, + 0.5376946989236153, + 0.4419632289330479, + 0.44281196477507334, + 0.44435885665359326, + 0.450672653970338, + 0.418020146339788, + 0.49641185808776944, + 0.48973152839589806, + 0.5475070188125991, + 0.49696764729533405, + 0.5758005348692631, + 0.48591708124596833, + 0.43935431899592836, + 0.48447633940054485, + 0.44696804646765875, + 0.5674850309780255, + 0.37619507727315693, + 0.40510861805139947, + 0.3860264454993423, + 0.537590332755586, + 0.39869613470851906, + 0.5296436759830024, + 0.33146018852981385, + 0.4869788769671926, + 0.5365072495058947, + 0.32762232768706057, + 0.46332902578565804, + 0.469701100575255, + 0.5710998994834132, + 0.5090953908200024, + 0.4910390080964614, + 0.4668328730216006, + 0.5195870845377778, + 0.47733729568943645, + 0.5016920585721553, + 0.27173241904925877, + 0.4941887600451814, + 0.5154622328222063, + 0.4251395373776619, + 0.5287137456963193, + 0.5662885079289988, + 0.5410004324986537, + 0.5357907686763492, + 0.5358312510693407, + 0.5314203179422424, + 0.2676790677567254, + 0.5377904714354658, + 0.5134717857739466, + 0.4544312518494591, + 0.4441993276838634, + 0.4917379365335809, + 0.5085775197881247, + 0.4796804093309318, + 0.5781341928470503, + 0.4839141782881327, + 0.43057581968494557, + 0.5190682597433495, + 0.5871551986330322, + 0.5719609327365108, + 0.5700502473783157, + 0.5299771023667744, + 0.5258218591682504, + 0.5253489892140035, + 0.521826791646353, + 0.5186265055607608, + 0.5259157255392057, + 0.40272441637500117, + 0.5707708558452329, + 0.49212960624160507, + 0.5250914575988739, + 0.5624712225282005, + 0.35313249113365214, + 0.6261498126665515, + 0.34534245737405694, + 0.29764661954341365, + 0.5600456173451782, + 0.5800390287762233, + 0.557730628726079, + 0.5645480043967717, + 0.5638641948576176, + 0.526870449999836, + 0.5341138512957928, + 0.2723942855981138, + 0.49601853480884955, + 0.5105532054555032, + 0.5663786599933839, + 0.4655175727572795, + 0.5071954244927775, + 0.44186454780802625, + 0.498969043485133, + 0.45888111505196016, + 0.46182496200492396, + 0.5422806623658262, + 0.5133419823500026, + 0.4296903080137033, + 0.5776596671512219, + 0.5185371677175331, + 0.5469637814469729, + 0.43276794478253966, + 0.47910468817651763, + 0.5925963992474711, + 0.6039499393935726, + 0.531888810680181, + 0.5430427674988525, + 0.5876838907004858, + 0.44862051401253344, + 0.5461892607867888, + 0.5441216792587135, + 0.36134763091683575, + 0.4520604577641234, + 0.6674229979089829, + 0.634915795179524, + 0.6286091736506886, + 0.6162317897409111, + 0.3331375227133031, + 0.3838561777867173, + 0.623120729011295, + 0.593891871496141, + 0.6009078329375462, + 0.5500973706797928, + 0.4872349101331905, + 0.49379484312298033, + 0.5096287937434957, + 0.4645511618259004, + 0.48618562755654265, + 0.5982456409767671, + 0.5668814462648962, + 0.6405428326043132, + 0.5182331761324368, + 0.47285714729648154, + 0.6601518754308733, + 0.45049825401104804, + 0.3873780861749716, + 0.40350799115383057, + 0.6255398835260357, + 0.6439834923363719, + 0.635523944873535, + 0.5554533222080859, + 0.5542998201621969, + 0.4955396362322747, + 0.5666843772167911, + 0.47897193584619213, + 0.5637022347697626, + 0.6090303104261359, + 0.5820528062825234, + 0.6342673377987194, + 0.6320730292288721, + 0.5984023418638544, + 0.5830894405995244, + 0.5774603660334922, + 0.5657843523376447, + 0.6070836809401917, + 0.5691705493591056, + 0.45216853193759154, + 0.44377766285038345, + 0.5137045964496757, + 0.6195314070176718, + 0.5225205421626166, + 0.6173972322308635, + 0.6447748672552578, + 0.6184530770236608, + 0.3644705515012438, + 0.47633734984909826, + 0.3447368614118441, + 0.532953095906313, + 0.5213253134346788, + 0.6028251749646547, + 0.6541706461822269, + 0.6127911016845259, + 0.6428431554216854, + 0.6528035901135779, + 0.6678861181044811, + 0.4077817329592523, + 0.6359028316988868, + 0.653975508610554, + 0.6745503858762072, + 0.6953986414058102, + 0.6657627565194784, + 0.6917668042722369, + 0.6150488833678313, + 0.6485051466238986, + 0.5689036639381386, + 0.6637804276712425, + 0.6192466516473594, + 0.6042424325466356, + 0.6524570311880896, + 0.6780133370634183, + 0.4624013241494904, + 0.6154536084314195, + 0.5756833501326418, + 0.4915514759486729, + 0.6527856588154245, + 0.6758415554414273, + 0.5776079521680524, + 0.5925296529206744, + 0.5766483697535555, + 0.44474501813052136, + 0.5833827144485123, + 0.49115007952427314, + 0.49443237803738455, + 0.6520347735128622, + 0.5856408117454949, + 0.5673090083816299, + 0.5428246097626603, + 0.5934534210432786, + 0.6100584314829142, + 0.5671760078258217, + 0.5152183315445393, + 0.492404559093461, + 0.3873257884416522, + 0.3722584758474051, + 0.4290959725212265, + 0.6670007842275761, + 0.4261637710693934, + 0.4442322258053096, + 0.6632049058474179, + 0.5083412586035997, + 0.6118544891055272, + 0.4949713991354249, + 0.5661619504146366, + 0.532409895156882, + 0.4614297336247577, + 0.6060427204273071, + 0.6723726620864667, + 0.5433743512146938, + 0.5799442693986097, + 0.487526970685889, + 0.46655070178259966, + 0.5289612139356371, + 0.6612841936902188, + 0.5313320731125485, + 0.49867686388927934, + 0.533331609754047, + 0.6309600566114989, + 0.3298824653131137, + 0.3715465865611128, + 0.3627420139068573, + 0.37386691282421053, + 0.5236548753936997, + 0.6327183827634079, + 0.4477626417926, + 0.49539266518861214, + 0.6185363780749421, + 0.5521013394250134, + 0.6374818283247635, + 0.5638497744065216, + 0.6545852413092682, + 0.47099895538485553, + 0.5728702172983826, + 0.4329578858630618, + 0.548485730116957, + 0.5931193795365751, + 0.4708398089314529, + 0.3843099930478617, + 0.4500998641217228, + 0.5181247891177939, + 0.5639308668400237, + 0.5887660744832592, + 0.4546110776009054, + 0.487757449158708, + 0.46026143921091894, + 0.5911789047949311, + 0.4428208289722924, + 0.4530489885182441, + 0.47516650609927186, + 0.5902287326791757, + 0.41985705916357846, + 0.6216464291887676, + 0.5485067282092139, + 0.5810220329915781, + 0.5232036972961863, + 0.5538853644044898, + 0.3417735184439226, + 0.3565531621368242, + 0.34998684714879624, + 0.35355216379689147, + 0.39505976951576977, + 0.5246097722383433, + 0.49669035542237244, + 0.6513520253168958, + 0.5824800127161102, + 0.6707057593640129, + 0.4977728816457889, + 0.48445468592876667, + 0.5098363863862446, + 0.4361624375588882, + 0.6249678268837199, + 0.3229056219911366, + 0.47947284514943084, + 0.3525211000798796, + 0.6240189812832126, + 0.37044482711020305, + 0.5245131552337333, + 0.3240498405628546, + 0.49139932989237606, + 0.6097248960491032, + 0.40384634628623683, + 0.5354045458825546, + 0.4177472576915904, + 0.5322825653421009, + 0.6101888218353005, + 0.5092210976136224, + 0.46478750897207305, + 0.5974436797513646, + 0.5966138527648434, + 0.5876269834436371, + 0.257035013245176, + 0.5959806117254989, + 0.6116290136784711, + 0.3999481256085025, + 0.6626342887011578, + 0.5509422803868497, + 0.6039407651983373, + 0.6133661300161313, + 0.589393525655985, + 0.5502742399560355, + 0.2517411777854869, + 0.6300416254444428, + 0.523920370095823, + 0.5825449691614467, + 0.5718638338854615, + 0.6543686478906748, + 0.6094705470339634, + 0.5422199326730509, + 0.6474275870494463, + 0.6202775845375512, + 0.4886340355755812, + 0.661046573292628, + 0.6170377158790217, + 0.5218574531119266, + 0.663000277510696, + 0.47994944447173554, + 0.458569108703414, + 0.45713748010679595, + 0.45360667362154494, + 0.4098578212917766, + 0.47725756357796517, + 0.3328663619308341, + 0.6598658387023978, + 0.4262156987199589, + 0.5034343942654402, + 0.6755390992675903, + 0.3367808058106394, + 0.6638018563066986, + 0.31188555814120955, + 0.3128361653583496, + 0.6124801208097861, + 0.6014984810208909, + 0.550394457543129, + 0.5588113489621029, + 0.6397129973855258, + 0.6102409444453034, + 0.6598011407295737, + 0.3023597225639455, + 0.5512339477546849, + 0.494153537466385, + 0.6032615616229229, + 0.5187327021811523, + 0.5348843671864182, + 0.49008386029028034, + 0.5408732012405705, + 0.521473045283936, + 0.5184955938913927, + 0.6018154775293321, + 0.502937705955059, + 0.3796479473119493, + 0.6383378378511304, + 0.5494395818441894, + 0.6689122124975588, + 0.3703583715732327, + 0.3799881170340328, + 0.6052923320207118, + 0.6033111526903451, + 0.5844274733548439, + 0.5848023337934665, + 0.6646235158993914, + 0.39851436825705233, + 0.6005206396958094, + 0.6032290445967583, + 0.35447616807577814, + 0.5405817047700892, + 0.22328111526135494, + 0.4373935586545736, + 0.5107449190313256, + 0.4220907193101693, + 0.49717426098002243, + 0.4165822427596677, + 0.5204290379734205, + 0.5334310604066106, + 0.43341734631945555, + 0.41499729269172364, + 0.4090108097462662, + 0.3905750154089375, + 0.3973142778465459, + 0.39023080458146253, + 0.38947652766659074, + 0.38776242192894217, + 0.3726965497099224, + 0.32385710136267815, + 0.5143506639499434, + 0.42833827146456355, + 0.4482500887976905, + 0.4336456605095952, + 0.45009530151707167, + 0.4395693186798603, + null, + 0.49116346791050247, + 0.4974743725485883, + null, + 0.3739355242875664, + 0.535730013201813, + 0.3501304667828904, + 0.5338382647756654, + 0.5592444879516733, + 0.5285763191714882, + 0.513543154167089, + 0.5169969159534631, + 0.5276934249510256, + 0.48495851308366716, + 0.44069664778728596, + 0.45910230598897106, + 0.4150798995412139, + 0.4996916175529785, + 0.41417961183153573, + 0.4611537886771919, + 0.44675419874115174, + null, + 0.4856584464822282, + null, + null, + 0.45263813095793637, + null, + 0.48839798331921974, + 0.44074347419294485, + null, + 0.5006272442354995, + null, + null, + null, + null, + null, + null, + null, + 0.48759622753506066, + null, + null, + null, + null, + 0.40607744315602434, + null, + 0.47887277113381477, + null, + 0.4982580384193301, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.501356847706744, + 0.38228495484957703, + 0.47128597931866917, + 0.49623014938158955, + 0.42791793125020605, + 0.5247585306281037, + null, + 0.5208325312907158, + null, + 0.471603990666241, + null, + null, + 0.5088767103289289, + 0.5165811072263379, + 0.4477872104586461, + 0.4369241279227172, + null, + null, + 0.46726350880031314, + null, + null, + 0.47861942009947733, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.48075798925218644, + 0.457585002447689, + 0.48922403617997073, + 0.43337292812011236, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4152474507607179, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.2204421252567642, + 0.3047676599513887, + 0.29641344522170204, + 0.43326289099606174, + 0.3656716096264701, + 0.2723948394493397, + 0.3484761671330741, + 0.3991573161341459, + 0.32689663962602844, + 0.33759903171834926, + 0.3930177760811931, + 0.37524656400760287, + 0.28144876956806647, + 0.3875950523980967, + 0.33715770866037703, + 0.34108381523467723, + 0.3634688568936561, + 0.30507409279365383, + 0.4617188016673426, + 0.33906032565566496, + 0.3383704965452034, + 0.32219683347698425, + 0.3035828913625036, + 0.4701339060774729, + 0.31429366520841906, + 0.3409094869462753, + 0.269520001109252, + 0.36101596487328463, + 0.3636565344708437, + 0.40309650279097353, + 0.37773525098904, + 0.38979354432516555, + 0.47136272205394114, + 0.2769578408658194, + 0.6209820938148917, + 0.2811348632599841, + 0.22495138266315412, + 0.17555852726470678, + 0.31457582582215204, + 0.21479573421343323, + 0.26754921544118293, + 0.4482501624194008, + 0.2989283106383826, + 0.3828544453295712, + 0.28603634257613936, + 0.378150927545295, + 0.2862276317713869, + 0.21438508613524052, + 0.4545367478112514, + 0.41237552383800574, + 0.3739667213287975, + 0.3601220118501994, + 0.3771642094118627, + 0.3964633968399699, + 0.4035330373687263, + 0.361594539506513, + 0.2825898761346134, + 0.39317520257799415, + 0.40661687316557754, + 0.4092618475270652, + 0.27989294925570346, + 0.20349880528875813, + 0.3848502231345379, + 0.3897498163667568, + 0.26502048105088494, + 0.3296536807636582, + 0.3359175089763676, + 0.43610707774093177, + 0.45096159394426094, + 0.4090420241271065, + 0.3505844709420184, + 0.6494430859513861, + 0.33172530382812737, + 0.3473685953676513, + 0.3454080272932317, + 0.4163663299272689, + 0.28411218402505384, + 0.6054953404561912, + 0.38645939830765536, + 0.3819981522266288, + 0.38811648432144424, + 0.39565739238820313, + 0.4600351549312595, + 0.38948940508673097, + 0.3196079132650519, + 0.2895424836253511, + 0.43404882655614346, + 0.37975061869710264, + 0.7828165837112537, + 0.4669056699993988, + 0.34884668677367425, + 0.33101328365968974, + 0.37437890007944513, + 0.3298154510350511, + 0.3806169327779415, + 0.5017700714557299, + 0.3324971576736854, + 0.32341768380124514, + 0.3369355350199138, + 0.29429440949944774, + 0.2443965145274796, + 0.36358852017383153, + 0.32871136809266693, + 0.320016218752466, + 0.366646386686796, + 0.20396027469817338, + 0.27717496447758516, + 0.27591902772714827, + 0.3781407019983623, + 0.3205790713821678, + 0.23201344114861058, + 0.3029629186079529, + 0.299831311598611, + 0.3183338510408409, + 0.3014903651430647, + 0.42822677829949424, + 0.37082835849399537, + 0.343593077223317, + 0.33504867709035674, + 0.39597852853059606, + 0.32499120375876517, + 0.38194295816801127, + 0.34393308874256867, + 0.2754682411032334, + 0.3432013468175661, + 0.30144846397396274, + 0.32599731888008926, + 0.30248705258380565, + 0.33597394353308807, + 0.5163103111227944, + 0.3510465325394487, + 0.3351441054891625, + 0.37286777369450963, + 0.5963328420908378, + 0.2902298834544838, + 0.32212213989424754, + 0.3338438793348348, + 0.28780903344434344, + 0.2934502716174061, + 0.3061625802315381, + 0.33844415522207677, + 0.41605542017341657, + 0.4866943328527582, + 0.3129276152625516, + 0.3549167356131354, + 0.4361336582427949, + 0.3809643814430543, + 0.24571911639837965, + 0.39783603131758694, + 0.4201954728326419, + 0.3837968444812761, + 0.3166521230337817, + 0.3676743280709077, + 0.4560395223347889, + 0.3689431367564142, + 0.3871903142779596, + 0.28738397363783, + 0.3310506971111059, + 0.3535708399306439, + 0.395593383979857, + 0.34487799743215136, + 0.33539001417303776, + 0.3528394365173586, + 0.3760636840031273, + 0.33279474842143714, + 0.3360011842935596, + 0.3133251099548616, + 0.411180693543648, + 0.26858237390197653, + 0.36699915071295336, + 0.3416018193218997, + 0.32465098866103165, + 0.46764886801524547, + 0.314894534668857, + 0.313040775782778, + 0.3817776503549596, + 0.2504412874809037, + 0.3789472737983958, + 0.36989790334929973, + 0.3112431019748992, + 0.3686501327289664, + 0.3346240136318977, + 0.27817750589834084, + 0.33048334384661104, + 0.3203890225693622, + 0.33822184871779953, + 0.39796525663843835, + 0.28515801524159173, + 0.3051079746707112, + 0.3821325441932829, + 0.30528990751092355, + 0.2905558649904348, + 0.40021758928901946, + 0.44822979063514, + 0.2937706687252837, + 0.30829167941627594, + 0.32523125183992, + 0.3165453953829288, + 0.25502859929154864, + 0.4229087290506778, + 0.3380924928500909, + 0.2836056724966107, + 0.3545614808347974, + 0.2880732956185968, + 0.41571743089015933, + 0.32607334310592995, + 0.38236471167422614, + 0.35287723945285426, + 0.2852180470516603, + 0.46765395496991335, + 0.31913741595737594, + 0.368983551636203, + 0.3016276859187938, + 0.3486381447665589, + 0.8039608192967902, + 0.37427803546482025, + 0.379687416881544, + 0.41546613311365105, + 0.4294353487241141, + 0.5351131526330827, + 0.5975710118208583, + 0.5975686032455957, + 0.3865711966991131, + 0.36307496162996983, + 0.453484620559309, + 0.7153674262140306, + 0.3343983698684345, + 0.3540381324405812, + 0.5241589418225223, + 0.4969500504260259, + 0.5252481461257492, + 0.40827328312380073, + 0.2681497848965431, + 0.3617158546363971, + 0.5700341327314405, + 0.4031067020525592, + 0.2748391846564744, + 0.5999707435334696, + 0.5279755928882454, + 0.42113287404269995, + 0.46628219165438817, + 0.691663558445776, + 0.4482441825497261, + 0.45514017269446616, + 0.4151449753976176, + 0.7506299115256256, + 0.3033181582921329, + 0.6223158970745292, + 0.3628405472785053, + 0.5314831206325042, + 0.41346624366358925, + 0.38047227426783864, + 0.6898992719005059, + 0.4797357119255497, + 0.30637399461025183, + 0.3595329427225433, + 0.6319314819071573, + 0.35199465912422345, + 0.4233841457549332, + 0.44197038495980867, + 0.3210763981497018, + 0.44102814974802956, + 0.4351999680651292, + 0.3925770432714626, + 0.3721955008566158, + 0.4588192539713658, + 0.3286756737229273, + 0.31569328636829014, + 0.615699744084761, + 0.36216646309295997, + 0.432277279328468, + 0.5214786057362519, + 0.3536919771254987, + 0.6461303219537843, + 0.3102492941003242, + 0.32404015217739396, + 0.5386146293199345, + 0.4553900680546747, + 0.5702766051334622, + 0.7162511437752871, + 0.6453070600220008, + 0.3763547371568457, + 0.4496928007427331, + 0.5217664964836216, + 0.3279501867949625, + 0.30936848081076296, + 0.5445196911625814, + 0.47960439420022544, + 0.6861654010178184, + 0.46307808197746225, + 0.34393699868052746, + 0.4249223876220908, + 0.3516531503802959, + 0.4491549226742638, + 0.6864561231439829, + 0.6024211196663952, + 0.46547465874190763, + 0.49954857823584387, + 0.46032249643832646, + 0.49954966172546067, + 0.7266850906835716, + 0.29496423158976154, + 0.42620671953329825, + 0.24449515233106298, + 0.5779287026908895, + 0.3480993050237296, + 0.7237856471884154, + 0.3797258667245585, + 0.41001415044399336, + 0.6721418058882292, + 0.48579595202620046, + 0.301512921197775, + 0.5460977053014436, + 0.37637392594175806, + 0.49189103132051176, + 0.4894672915491848, + 0.3256953011092015, + 0.4944480181299215, + 0.36251541782796787, + 0.432017383471534, + 0.36597123277193494, + 0.32390071056795655, + 0.37889312366481165, + 0.48494444017079824, + 0.5446600912109631, + 0.650839865035323, + 0.38807845045530265, + 0.28540329711858453, + 0.41697873941821706, + 0.43221476186968333, + 0.41644897076043635, + 0.45859719150631867, + 0.44256949849472504, + 0.3507217995768024, + 0.3209500602799055, + 0.4828629395018961, + 0.7545688782630295, + 0.5086455632756338, + 0.5743381770788776, + 0.578262616827669, + 0.749335780674715, + 0.5273536993872543, + 0.32676842869914696, + 0.561480877166751, + 0.3537012754415909, + 0.502809473300359, + 0.43298642126454084, + 0.3620468668360744, + 0.4768022216247422, + 0.7661694344209551, + 0.23982655150291993, + 0.4404563474172487, + 0.44206580200054435, + 0.6015554686638492, + 0.37346769614611897, + 0.29858830605127595, + 0.3981600251249365, + 0.7120145702103412, + 0.3623784454445176, + 0.41195829948498375, + 0.48688022645065016, + 0.4423389125121098, + 0.4106609365619861, + 0.46782166395113184, + 0.5947270458943555, + 0.44256770268338885, + 0.38200548716849814, + 0.5082037366630962, + 0.43042267496970493, + 0.5471461540288493, + 0.4855634610140965, + 0.4969885073686716, + 0.5105498753342942, + 0.3071543079786756, + 0.3258290328294947, + 0.4984177433165182, + 0.30183054117543995, + 0.6867068736693535, + 0.2968090332728888, + 0.5451759862671268, + 0.5301903084188687, + 0.6727658821910133, + 0.28455594667862144, + 0.3354777624215922, + 0.30837970985593904, + 0.3486502600692918, + 0.3048374107167979, + 0.28670825823096235, + 0.6004288083963231, + 0.3545494732520875, + 0.6701896280278443, + 0.5616347444575065, + 0.41869218767679717, + 0.5326336455423611, + 0.36059413436071136, + 0.49039070374860416, + 0.4938536482611556, + 0.46142694108557697, + 0.3569182411262675, + 0.44127190320810517, + 0.4639628840350882, + 0.5011455756436288, + 0.36994414630658073, + 0.32066007935833024, + 0.5853895613138623, + 0.4482628778320999, + 0.6444417418103209, + 0.5833551216980876, + 0.4898400481729438, + 0.30884177915684513, + 0.33219104705170033, + 0.33322082586251406, + 0.4033174381928346, + 0.7360241999819421, + 0.4515413676345123, + 0.35826919038490623, + 0.7261643120579903, + 0.40585361457138297, + 0.6026204755367274, + 0.5168317509294246, + 0.3994078054732111, + 0.46454075003797035, + 0.42161547394177856, + 0.4291750493718163, + 0.5677777764515902, + 0.5124100256472254, + 0.49616425164231903, + 0.40251939046223134, + 0.43242661908861774, + 0.43577936494188074, + 0.445484887590273, + 0.7541240238372934, + 0.44839035263966487, + 0.456157866180532, + 0.4009511988100948, + 0.2861833460905732, + 0.654249721917367, + 0.39866453344128633, + 0.3133322175458617, + 0.49862162312534586, + 0.3982007683367976, + 0.3069898808076219, + 0.4729870601440024, + 0.4037717347889222, + 0.7646424489328666, + 0.30297232263198004, + 0.3632299449306213, + 0.3105684785334082, + 0.4107574520067758, + 0.3627490816571434, + 0.29668681242268297, + 0.36454903524754073, + 0.26441123055752197, + 0.38358274699462364, + 0.7170317888603969, + 0.28228689764743486, + 0.3812743636816561, + 0.47825182724045495, + 0.3919098575508511, + 0.3746841912633012, + 0.45598629162771365, + 0.3398082128959304, + 0.4967044692316011, + 0.4877076464122874, + 0.2491308102155722, + 0.605076706624344, + 0.6571859766715303, + 0.47487180802649603, + 0.4540328380533485, + 0.5214621202255169, + 0.40105658697482266, + 0.4257276258665767, + 0.5343137066072103, + 0.7663592307284987, + 0.36326802686848236, + 0.5908839375772541, + 0.30661030350857127, + 0.43604538186872777, + 0.43898537598918264, + 0.4449305378703098, + 0.3631980576147287, + 0.25845496902826315, + 0.31519334976789515, + 0.35375225826450574, + 0.5489636093838328, + 0.42886970321999546, + 0.3061592081409922, + 0.38141019907837, + 0.4342244698495547, + 0.28128899348452313, + 0.5230679412656528, + 0.42701711278867716, + 0.3653758661377236, + 0.4210588384982104, + 0.29701681350231784, + 0.40351811654882286, + 0.4469482613886625, + 0.5624352415932395, + 0.41916104204142585, + 0.7675899708119204, + 0.3277160645992597, + 0.5437962012826167, + 0.7285711634023034, + 0.613552573144096, + 0.29533896218697403, + 0.4633757814976662, + 0.401298042873622, + 0.32256686173120475, + 0.3364526416608741, + 0.32831066857056806, + 0.42731411429212596, + 0.4819401082271756, + 0.34290657440708716, + 0.5921591010113546, + 0.7456092759686983, + 0.39520066903829837, + 0.3888451832269484, + 0.4057974659103224, + 0.45344881469739656, + 0.47118872047341886, + 0.4751647344665682, + 0.47310007543435273, + 0.3274658432263924, + 0.6273594803390591, + 0.45791836470869285, + 0.5236215334487404, + 0.6921246138528776, + 0.3007801865709651, + 0.5777089422264907, + 0.36390099858311414, + 0.42529950186389087, + 0.3341497362476102, + 0.4925071526958806, + 0.46607151413211567, + 0.30825545704297164, + 0.37991287972028676, + 0.3877574973045221, + 0.5274376950855493, + 0.558254249238649, + 0.4129129973890811, + 0.4276291101362281, + 0.38950708072738943, + 0.7222296409020547, + 0.3782706208890647, + 0.4548229601404204, + 0.4332650534053305, + 0.44089984628457674, + 0.5719823601625131, + 0.3153186340200399, + 0.5478940155797764, + 0.3488769321830768, + 0.5126175538102904, + 0.3915504584230997, + 0.31269806337586814, + 0.329136366679941, + 0.44072689201223464, + 0.3403493478154391, + 0.48203194416126094, + 0.7074844414809593, + 0.7062328148429655, + 0.36814273476947085, + 0.31414393200045665, + 0.6870380670806441, + 0.4234942029908764, + 0.3594249606199577, + 0.4782218207688047, + 0.337524189060528, + 0.5795855660598398, + 0.4160202858882618, + 0.5434940002212819, + 0.5530048805183909, + 0.3829077090343533, + 0.4818123236787283, + 0.6398272637855089, + 0.4905120832249186, + 0.33127869071729704, + 0.3848374772898655, + 0.27981793789734133, + 0.5070024219246541, + 0.3243076308974105, + 0.48663183483613903, + 0.47167162148350383, + 0.35474302491176407, + 0.7080888270569056, + 0.3082954130964696, + 0.7146339715924547, + 0.39735511900481946, + 0.5032594232810332, + 0.7352133927561307, + 0.442990376744397, + 0.5338831835186182, + 0.3063775238489387, + 0.687142361151034, + 0.3738490594217105, + 0.2996155049506239, + 0.5200574467981188, + 0.4907808594468602, + 0.614454995422657, + 0.5346124287584989, + 0.3573201719082778, + 0.4530564018334606, + 0.2794883487351321, + 0.4097267507496101, + 0.5497175072909094, + 0.48430825100780445, + 0.3252511328679001, + 0.5061237534631856, + 0.5643210590246616, + 0.5598488934298752, + 0.41896278873134846, + 0.3478870472597696, + 0.4886332347404569, + 0.43804310788712314, + 0.5345524092635469, + 0.4411630508298879, + 0.3358362299166402, + 0.3784175953849243, + 0.4771342246457, + 0.3730632462231289, + 0.7177351717502681, + 0.4955018040338343, + 0.42560545441503744, + 0.4675574765926011, + 0.36994107602688375, + 0.42293271496442675, + 0.5186305920983836, + 0.5281243772467474, + 0.6984834057082543, + 0.5116933465676226, + 0.5553432190140692, + 0.5021549374863346, + 0.4533625591363672, + 0.5586348194540413, + 0.376257800837327, + 0.6021700712074775, + 0.5772979596695444, + 0.7139970152758766, + 0.49459758659134917, + 0.5263178304299196, + 0.5210052209290079, + 0.37559369318057, + 0.5245704031047899, + 0.31013890268214156, + 0.6618285515233152, + 0.4482625153519122, + 0.7237392507826677, + 0.7366351147481061, + 0.42698645796629664, + 0.421986184831944, + 0.476093652401339, + 0.33078413103834603, + 0.46060200220063885, + 0.7444855813686788, + 0.4700811192089539, + 0.48336208390073804, + 0.5650877681083105, + 0.48425816205417044, + 0.3269666622929234, + 0.5928827035355443, + 0.4064415711895529, + 0.3604285469356264, + 0.7180916152139594, + 0.38770165466889267, + 0.403735895257757, + 0.37047289891023444, + 0.469263471575873, + 0.5423043968341809, + 0.5967520094983763, + 0.6630829911370147, + 0.6025943001768274, + 0.5701369700054321, + 0.4191804880135541, + 0.5922267629394897, + 0.526319049133968, + 0.7683402756590733, + 0.7146773761170997, + 0.31573737065012286, + 0.3459176689961901, + 0.3231307206468325, + 0.48633163915754146, + 0.387824043156546, + 0.7722664073851501, + 0.3493566483992234, + 0.3924238873921393, + 0.5508376244044488, + 0.45320819575061577, + 0.5221568903823073, + 0.6023422296034192, + 0.4275329054589624, + 0.607855932999667, + 0.35226864201402436, + 0.6558205590363726, + 0.7651139448334622, + 0.7530380431772319, + 0.5732530181296903, + 0.40667072024180406, + 0.5260905035129697, + 0.7223363454379901, + 0.6712190105378983, + 0.6029751176462609, + 0.4560659423076052, + 0.4393530246246372, + 0.4706163918077825, + 0.7230907320608458, + 0.39412781050569895, + 0.40599430118846064, + 0.2993483846565227, + 0.5714108813215845, + 0.2941692785827999, + 0.3723467948154408, + 0.6471680045762501, + 0.5249337786909882, + 0.6659717793345505, + 0.505278795248833, + 0.49688037909293037, + 0.6492901958043821, + 0.615220223157892, + 0.5494339351620983, + 0.3635155662653751, + 0.6292571403215904, + 0.4919747705925563, + 0.3819692157150424, + 0.32829789030189066, + 0.46857711029156124, + 0.6188044405041497, + 0.48043147464896896, + 0.587925343881948, + 0.5444625107323526, + 0.40103504440601695, + 0.380159530642265, + 0.6674040791799032, + 0.4746345955060816, + 0.5691677900479906, + 0.7125393569962206, + 0.39406456528309486, + 0.35162467113920115, + 0.34360291496389467, + 0.3391971503579808, + 0.7001948294226839, + 0.7236935109959104, + 0.5483047757668078, + 0.3993380233295659, + 0.43032258996105127, + 0.35790114188559596, + 0.7625439716725606, + 0.32368260021809836, + 0.4601420770273195, + 0.31527160700640544, + 0.4877243185492431, + 0.34427724412918254, + 0.5389631287783507, + 0.44592993337462333, + 0.32452527435285156, + 0.674369083991109, + 0.48291362995492754, + 0.39463493675821226, + 0.3314512752307115, + 0.33938440362520655, + 0.3849459984996843, + 0.39527752016850454, + 0.35458432402375517, + 0.3516851908435189, + 0.3842643036576539, + 0.3729967230810367, + 0.3912893500969667, + 0.32225391014029503, + 0.41814486988121086, + 0.19746213457319253, + 0.2355767564197019, + 0.28325773554202055, + 0.3733154667233615, + 0.38771507964192464, + 0.3833159700002406, + 0.2753366514822011, + 0.2837815371854143, + 0.2873770617442947, + 0.6393545610177105, + 0.29572224837701705, + 0.26681192301117984, + 0.7434629530890173, + 0.34298454892786756, + 0.37337667107839034, + 0.3080276185695119, + 0.4289074624834159, + 0.27526825809649014, + 0.37064245181939715, + 0.33455321703668417, + 0.4236200316300688, + 0.2363908733458969, + 0.39432087988285697, + 0.3855402717934651, + 0.47902887512530057, + 0.34120370326838295, + 0.45442195398900503, + 0.2911933213012055, + 0.4029500632575481, + 0.36324786842880724, + 0.3209863623484705, + 0.4543519986679493, + 0.3178044854275772, + 0.4543550680738498, + 0.3521439634830586, + 0.34051622005948784, + 0.796809320488956, + 0.2487118039243618, + 0.376381391968792, + 0.3501682210454005, + 0.3689435383988522, + 0.29649294466072235, + 0.32002802505728006, + 0.4362876532702063, + 0.5436547808128122, + 0.33195989490675987, + 0.36917002413818256, + 0.2627517411107884, + 0.39288462864666296, + 0.38030994261322676, + 0.3890144166171005, + 0.4255289037080552, + 0.3343674115466504, + 0.3268832305775563, + 0.4422783873700614, + 0.3598101521541799, + 0.4217371495261933, + 0.26414773640226535, + 0.29990221696326685, + 0.4503079519774661, + 0.43394268140408127, + 0.36700217481783914, + 0.327571030043184, + 0.3082310822551484, + 0.795506980751944, + 0.3037806659251681, + 0.28652969019978336, + 0.32711087884770496, + 0.27256896942982883, + 0.26536253801813336, + 0.2887521834746283, + 0.24555467958683377, + 0.3356721301289818, + 0.33481145414668656, + 0.4346079867449907, + 0.36051805073409104, + 0.410409940026662, + 0.4512854331909625, + 0.3825220770131738, + 0.365503483883987, + 0.539707403753796, + 0.39080014754075554, + 0.3902462245667389, + 0.47668616380219475, + 0.427559177435427, + 0.3396038981691679, + 0.2921424044904793, + 0.3609579020096545, + 0.46803901645706286, + 0.46645868698275444, + 0.3108177168973793, + 0.341220121257561, + 0.3694207581571192, + 0.4200916956373133, + 0.3430364624220406, + 0.34929697329946446, + 0.4620626205989624, + 0.4891234810557001, + 0.3080019384218316, + 0.41368082215800367, + 0.3066539129549986, + 0.30333065889377336, + 0.31385391417792774, + 0.7691745631031239, + 0.4035549886362866, + 0.41236777905237326, + 0.30685553386916226, + 0.360147105759346, + 0.34754654734992, + 0.6768272732486311, + 0.39002080684883084, + 0.3603716861955621, + 0.4324355462677576, + 0.39790911266365453, + 0.3700413509421322, + 0.4273091873169866, + 0.464487642137156, + 0.2715409759812449, + 0.3221375644715608, + 0.4360449809265519, + 0.39972941709700116, + 0.38467678130453214, + 0.37708591847223677, + 0.3716078901574308, + 0.29298543067471594, + 0.3614476740233313, + 0.37266165332643053, + 0.3429727150663843, + 0.5688191748427882, + 0.3390232731780352, + 0.38965822296427893, + 0.3019029563660969, + 0.5227357714216255, + 0.37537064705554984, + 0.33608742240360334, + 0.40044356209951715, + 0.36802324523432084, + 0.377391796159064, + 0.3129443083327292, + 0.5106833938914347, + 0.4287940545466708, + 0.3242922964823136, + 0.3690109834014575, + 0.4177500755810394, + 0.40626223966450714, + 0.35885423959163004, + 0.45501949754714477, + 0.3163412787347747, + 0.6361165033452201, + 0.3731956524966164, + 0.4473675221109028, + 0.43257620515386375, + 0.36376540789003353, + 0.3800661843279532, + 0.36544487593206654, + 0.30967978691216586, + 0.3806135560494165, + 0.5756661644468621, + 0.3637760172880171, + 0.33438228790493185, + 0.49200331132531056, + 0.4037392944054542, + 0.39030871062981587, + 0.5729862539874494, + 0.3408703794186437, + 0.4200742321749211, + 0.36082850393053395, + 0.35392069452366304, + 0.4480040634716913, + 0.32848633008376754, + 0.4499376212797294, + 0.3900055114167183, + 0.41244611646472445, + 0.3384597785919621, + 0.4078162181690358, + 0.37369307777359695, + 0.3261289961779944, + 0.29926576605910965, + 0.31472090944421477, + 0.41234469714835614, + 0.5205153239259487, + 0.4358015687393241, + 0.43028345927487643, + 0.4181519442606597, + 0.33975696830944624, + 0.318030509899532, + 0.4403709568707236, + 0.3533078358338905, + 0.38641736448124114, + 0.459755142181752, + 0.3650558650666239, + 0.397606736444824, + 0.49429511720840286, + 0.37009004307686466, + 0.36047927085206494, + 0.4885394534788855, + 0.36912081857822326, + 0.4084816330834422, + 0.29296422685363566, + 0.4569268058187015, + 0.5903173345373217, + 0.7802546458310345, + 0.40891078557184246, + 0.3658621249104308, + 0.4068540917475007, + 0.3279971927598194, + 0.3439774431109946, + 0.3409977551100336, + 0.4002035507618007, + 0.36966498933957925, + 0.36395061198191914, + 0.4235594286524433, + 0.3895155230537613, + 0.419039297187544, + 0.32516931175999186, + 0.3093630554856565, + 0.41595517596816356, + 0.3824328205711812, + 0.37108170217587877, + 0.3583926800494712, + 0.38291297575724687, + 0.3654072675602415, + 0.26974985457538636, + 0.3953588768820015, + 0.4246837809386442, + 0.35835604106028346, + 0.3812275291901526, + 0.2892423112765739, + 0.3646493747476635, + 0.26390184908639214, + 0.4846971081672641, + 0.3438963334786504, + 0.32054899100516676, + 0.38393369932405885, + 0.37649446074444654, + 0.30970776079352735, + 0.5748393639479656, + 0.46473098735700197, + 0.3204626567070116, + 0.4670275722066106, + 0.38653196228551673, + 0.3768448619360087, + 0.3900500280244622, + 0.2795450669743548, + 0.32512029169407464, + 0.455494906846183, + 0.3384036863576435, + 0.4205763213375178, + 0.3922602332327735, + 0.4398143252001134, + 0.462668113210531, + 0.35878856813172905, + 0.4800673460787416, + 0.44612494703125083, + 0.35606146965090013, + 0.3235825415460004, + 0.37885165673025056, + 0.3263890861431364, + 0.36924933973410967, + 0.3914767092082948, + 0.46492994485803174, + 0.49531802265192776, + 0.39020183993241725, + 0.3810685072680511, + 0.3935646404638152, + 0.42031995701968206, + 0.4234415363047833, + 0.5066207730515286, + 0.38305255050451403, + 0.2978554114355689, + 0.4788467314381212, + 0.3634999432985219, + 0.42396033478586337, + 0.41005971629146193, + 0.3820907057507093, + 0.4763566120377727, + 0.33115265252611276, + 0.3648214246843391, + 0.32284002077613244, + 0.3206586209175379, + 0.3397256025835932, + 0.36691544781768926, + 0.3152446945140662, + 0.3975432628652944, + 0.5168710618162273, + 0.39392725300946135, + 0.46699769409995234, + 0.7641735243377832, + 0.512699539601819, + 0.40468748233529955, + 0.39905898342730467, + 0.41196487856544656, + 0.36633078547530357, + 0.30202410074545916, + 0.555459535224265, + 0.43594283728263933, + 0.34660741880248286, + 0.4380888434988432, + 0.3574340282015475, + 0.47455163601008793, + 0.49315899746942277, + 0.36340675765000435, + 0.5165079989574254, + 0.4529013079969549, + 0.33367384512073434, + 0.4379379974534597, + 0.4577785884057065, + 0.34851441995621174, + 0.36569332366809537, + 0.31055844952434775, + 0.5469140890963701, + 0.481981898913872, + 0.8063324880609064, + 0.4942955770818224, + 0.4132676991055743, + 0.27469117262178705, + 0.45352980514176294, + 0.4225640311329727, + 0.5935102950727554, + 0.3254486794921555, + 0.4883263211007522, + 0.30268214399287113, + 0.3288903566671206, + 0.32053507614727084, + 0.32094056920405445, + 0.39668087307428834, + 0.3337085583117821, + 0.38965832348940743, + 0.4378683028850202, + 0.3637361990429764, + 0.31586809133247334, + 0.3222001198003825, + 0.396506523407835, + 0.30502748129531015, + 0.47398235594143767, + 0.4071326465148915, + 0.45434024172801724, + 0.3942048479519323, + 0.3546772682410658, + 0.5436233348076257, + 0.43173138342928735, + 0.3316591653938173, + 0.5719438562214255, + 0.299603333464955, + 0.4114533878619081, + 0.3427515894938951, + 0.3061645562532814, + 0.31683834000355576, + 0.3411542521488982, + 0.42478579705941427, + 0.3107177653305888, + 0.6735611153920102, + 0.40664119183425645, + 0.33760936039147493, + 0.35157773343174065, + 0.3129382392716604, + 0.5389901299274077, + 0.29175157413405267, + 0.33166729425354646, + 0.3660542064342987, + 0.4006800203769972, + 0.3105619047450366, + 0.3890869995174274, + 0.2878199941138888, + 0.3749386995542539, + 0.7543645045518739, + 0.4000921239128422, + 0.3593370440153219, + 0.36884985203445725, + 0.3809422623775203, + 0.28354974160499385, + 0.3503779540347295, + 0.40486687290407714, + 0.7387766263023768, + 0.5236524102719633, + 0.4052530366937998, + 0.3639880829677171, + 0.31550556036954175, + 0.3730730151408369, + 0.2836642167505047, + 0.38628642921155504, + 0.4665552119308829, + 0.271544116689859, + 0.36412824454436676, + 0.3554934493294949, + 0.43114374378767834, + 0.3459111816506402, + 0.4967335714190503, + 0.4575145784908494, + 0.2887578400162846, + 0.40026180959874486, + 0.3952380537129598, + 0.5037921462733288, + 0.5022153193330594, + 0.23409271875927273, + 0.6995863915803971, + 0.6915315314522684, + 0.19731442759211396, + 0.2695569913220823, + 0.3404135536310689, + 0.4194523464934354, + 0.25517409701610727, + 0.34080564862046436, + 0.44366491466504643, + 0.45681089584347484, + 0.4450046449914873, + 0.3292166929114293, + 0.32486865841285456, + 0.3302410937691016, + 0.30187813709738953, + 0.41350219755203105, + 0.3686217150483051, + 0.3580605337357289, + 0.4695460761410887, + 0.36989977479972186, + 0.48748138022885407, + 0.4198895885151441, + 0.5908915727732166, + 0.40721894031817735, + 0.39601498456103773, + 0.46279061974521546, + 0.42428621882377254, + 0.3159335681161139, + 0.592617517474171, + 0.556514084085356, + 0.41639349554559674, + 0.34953740762171004, + 0.3467748563456902, + 0.49471147825863576, + 0.3466502359413165, + 0.40526362337042604, + 0.38257759249238343, + 0.4562868914929232, + 0.4067797955357182, + 0.3533866782670801, + 0.33483840032405576, + 0.4966452052524814, + 0.4560649559338488, + 0.43726525054717374, + 0.4558899671601242, + 0.3471130715168126, + 0.2493593281277661, + 0.2637734061496423, + 0.28374941506058365, + 0.4926196697528318, + 0.4324126667163817, + 0.2530890612488384, + 0.42960090015777846, + 0.339575402860067, + 0.3136814985727473, + 0.32020567717949294, + 0.35024034186734393, + 0.31020119964618503, + 0.41465402586864214, + 0.601571959071724, + 0.3190023937912474, + 0.40964319349213957, + 0.5483104100727062, + 0.37285664180124434, + 0.4836166518424721, + 0.5627619792556061, + 0.4276550088918046, + 0.43414242423955657, + 0.3959023934243961, + 0.3710877580650724, + 0.27458473320886273, + 0.34854628717472175, + 0.46038558848269473, + 0.48684928401810945, + 0.4496905211577225, + 0.2948478961778114, + 0.4328994594169152, + 0.4176394420102936, + 0.47410857521994487, + 0.3953615944799953, + 0.4512863805133671, + 0.36695796681080506, + 0.44475676341974524, + 0.41261942670641755, + 0.395381770863562, + 0.29186426095628876, + 0.430305455748191, + 0.45211757397362273, + 0.26849945111395906, + 0.3333791820637197, + 0.43109443692850047, + 0.3719697459804812, + 0.3579804801143127, + 0.47006562833861987, + 0.4865262506879973, + 0.4117488146525896, + 0.41841758082002495, + 0.31703360189803537, + 0.4205865436276052, + 0.3618567181092255, + 0.3486728970270259, + 0.3694559482101097, + 0.4193912676958342, + 0.36249547828439027, + 0.4440076630739227, + 0.4524443859615036, + 0.706598160622952, + 0.5298728984545411, + 0.48317210421872775, + 0.5178461276899776, + 0.4884428277214048, + 0.46969772543527916, + 0.5218928135215212, + 0.34050316498197525, + 0.3811274829872446, + 0.4487646300367744, + 0.3342104680223252, + 0.4365542438996478, + 0.4779558778367802, + 0.4232803792004021, + 0.43241871451760583, + 0.44274179390016233, + 0.3113570560758258, + 0.4969534608277242, + 0.477115818400115, + 0.3556389063286109, + 0.4138971075597677, + 0.36791726096305033, + 0.3422151007702608, + 0.4257806594760286, + 0.42171572210255204, + 0.40648829560684047, + 0.44006780173874227, + 0.37570566791742, + 0.4561089106708536, + 0.34524622044915165, + 0.3680455332094333, + 0.4143856879427378, + 0.48298493314079205, + 0.47155790139623366, + 0.4535669888121758, + 0.3823032797271646, + 0.3396898442678861, + 0.3435206002830866, + 0.41529428292788223, + 0.40127601875328506, + 0.43593283888909645, + 0.4547511879528027, + 0.4136022652613164, + 0.3894229694291516, + 0.3442022317804489, + 0.3968511904176178, + 0.4941442889360442, + 0.37375007018465867, + 0.19853443848284996, + 0.3274102498208671, + 0.4254969099890236, + 0.2920616430656969, + 0.2873865046342579, + 0.44428843719731864, + 0.4408730903512797, + 0.318108069989705, + 0.3027148104382746, + 0.3726574735882662, + 0.3627949408050728, + 0.4724751183518636, + 0.27057582510054395, + 0.27697046796057695, + 0.2919934438660098, + 0.3075338306344056, + 0.24680509146098348, + 0.2943305830063701, + 0.28491591750114054, + 0.29735722941206116, + 0.30226363701314024, + 0.40009371302451874, + 0.3275007264800203, + 0.3662137394286936, + 0.3695426184846259, + 0.46779567383518583, + 0.31899604423876643, + 0.37717375635843337, + 0.3686704222133905, + 0.3791281354873153, + 0.3719679185022022, + 0.387373960125974, + 0.3739401120934396, + 0.5569011043954011, + 0.29969278748659883, + 0.39928713383397363, + 0.5060638831542906, + 0.4478321294240159, + 0.3502017617936579, + 0.4243821758158708, + 0.38216292362835425, + 0.41788278330565215, + 0.4449452266689023, + 0.34418703547513146, + 0.5527143976289701, + 0.39870552869679077, + 0.39062645260978074, + 0.33754962687986373, + 0.409120641489456, + 0.3615925575459939, + 0.4240092889929725, + 0.44378940057835453, + 0.4161017357826406, + 0.4466586994119345, + 0.5486832844563716, + 0.4050759972113672, + 0.3766134782829054, + 0.48528366213828183, + 0.5565647328331622, + 0.4953933658607912, + 0.4259462147965665, + 0.511179487447531, + 0.5130929990555061, + 0.3639108761231927, + 0.3660299709720034, + 0.4058532425225985, + 0.3265657370346405, + 0.5808963416995385, + 0.4747331049886453, + 0.3920951830318297, + 0.3381877003639002, + 0.4453709990915049, + 0.380180194558692, + 0.4767313694733282, + 0.42368060058412216, + 0.36132951943806013, + 0.4997598980278582, + 0.2949961845923755, + 0.3851649653064746, + 0.4953428748761194, + 0.3238743986982255, + 0.384469723647055, + 0.2999905786739688, + 0.3428756557012122, + 0.3510769443227254, + 0.37386466386167033, + 0.4293916122853224, + 0.32585660476087525, + 0.3018588042366417, + 0.4057717055800653, + 0.37142005554436847, + 0.34880530441328317, + 0.41891604945653566, + 0.43856931893778883, + 0.3991954345802629, + 0.49656062281226837, + 0.5456904947162023, + 0.4284299775783892, + 0.3296145455287334, + 0.4574765964955871, + 0.41494676755558174, + 0.37262439392243685, + 0.39874748186503767, + 0.3057325270362608, + 0.3023697220867061, + 0.2885714925568316, + 0.34455869732591193, + 0.3924534891952744, + 0.6298988289148181, + 0.4132006968637796, + 0.39275595351511267, + 0.2854516880682792, + 0.3383761538187618, + 0.30741643525888096, + 0.3834736117891659, + 0.3472798838647523, + 0.3635600818297866, + 0.32986999896049235, + 0.39699571626503277, + 0.35746076096718976, + 0.4121727831129331, + 0.37767149368952724, + 0.3389235852033243, + 0.4716925248914792, + 0.4675765199714692, + 0.351097097946533, + 0.3992944426698951, + 0.39649485887736, + 0.3599088076747547, + 0.333038427406023, + 0.49214151114513677, + 0.4246336154407684, + 0.5583267922859205, + 0.4238162989551173, + 0.41642188443437056, + 0.42647481048575175, + 0.4282979010500472, + 0.3222332883474201, + 0.5428620217668105, + 0.38286164867040906, + 0.22426675674305158, + 0.4229825941890866, + 0.5185568918621827, + 0.34096739613091703, + 0.4142880912266893, + 0.7040872749095196, + 0.4098532012880041, + 0.48452663490418957, + 0.2884187695575656, + 0.4394005058396537, + 0.2794066228817235, + 0.2554401231966602, + 0.29210138256100665, + 0.24309308635821028, + 0.4130201310085416, + 0.19931896919226924, + 0.27727553550368594, + 0.40305539403013163, + 0.4061177634814915, + 0.36963867566788616, + 0.37393266109247236, + 0.2883512407794648, + 0.393842359702826, + 0.47853187069163927, + 0.4978796425755025, + 0.3078419942807646, + 0.40713950523813636, + 0.3576641661221587, + 0.6934345402382095, + 0.34069589281835827, + 0.5303222505563134, + 0.718864122006158, + 0.6939679312747588, + 0.42266771854803387, + 0.3200217024432336, + 0.22425452550763356, + 0.4006142168337528, + 0.4515841908408154, + 0.4572749090909312, + 0.518510395745195, + 0.38435265801829993, + 0.5165159224186088, + 0.3551966085306322, + 0.3479490422164121, + 0.6686487409231194, + 0.6590066966843534, + 0.41284861333456147, + 0.3777875418922356, + 0.42009206404377736, + 0.3830855792223678, + 0.4108946686092629, + 0.620185587682747, + 0.6204099077556202, + 0.38828874047205897, + 0.39919145106534687, + 0.39425637696197136, + 0.3218806230646473, + 0.45604750621758416, + 0.30278714743067636, + 0.4272078507027649, + 0.4474140397764764, + 0.3180299129994058, + 0.3982168661807504, + 0.33371461240870803, + 0.2786427691199983, + 0.4856799449154567, + 0.3750733814922477, + 0.3639543937976508, + 0.3306146443468159, + 0.4333409637400149, + 0.37597929760627263, + 0.2999495084181294, + 0.30788791807309124, + 0.2669691130459741, + 0.2201304538035237, + 0.26162641015388194, + 0.3957250376159052, + 0.4343760565556111, + 0.2973903530933767, + 0.30335847160411467, + 0.29670301934377796, + 0.4277647154205028, + 0.3624168326495708, + 0.4459101763231269, + 0.31398634241607737, + 0.41396927457992755, + 0.46003767013604246, + 0.3285099639160904, + 0.44767391304084037, + 0.4156379780603309, + 0.388148277488129, + 0.38759012086982075, + 0.3143061293422608, + 0.20567271271489862, + 0.258001782144908, + 0.24408563730901348, + 0.46689028569014857, + 0.398356697212937, + 0.4046438812431246, + 0.40556005313244825, + 0.4261965862542195, + 0.3105920944642265, + 0.346434418771925, + 0.32052529137043634, + 0.6953898505834657, + 0.40638566721212466, + 0.42516169934445186, + 0.3312799691549406, + 0.4161508770660085, + 0.4478936599865948, + 0.395739644390549, + 0.36956773802400755, + 0.43474325952822945, + 0.4040100148389333, + 0.508227964159869, + 0.3815992237446506, + 0.448144106142208, + 0.40034128697408206, + 0.3941633609913282, + 0.3779791718161305, + 0.4329349618888023, + 0.39917802266757796, + 0.4499261731922465, + 0.29419162819499817, + 0.2925794682865192, + 0.29543016319283594, + 0.35815555178483244, + 0.3966013596451325, + 0.3680980837876199, + 0.3883609033739294, + 0.4222804784348236, + 0.3755167872424227, + 0.4004839104919911, + 0.3960268203903299, + 0.25102495685268356, + 0.37370900584458827, + 0.28891048094814653, + 0.3904406936964342, + 0.20408381126726646, + 0.646674074229909, + 0.29293586126766286, + 0.38687014945858444, + 0.5057847411486989, + 0.4264746851111012, + 0.24667677751563413, + 0.38448784769297756, + 0.30927665484208694, + 0.7530698599352652, + 0.25998416491103127, + 0.4547671706024092, + 0.4645091215198322, + 0.47123143108063426, + 0.6182671554323756, + 0.3556238684096557, + 0.5141697875828173, + 0.4011501675979125, + 0.3301769844450033, + 0.3074311074980186, + 0.404994602184597, + 0.26766027566746387, + 0.6818631791425511 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.4380923519004805, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.5071537503179383, + 0.7303442004388413, + 0.7111772953343971, + 0.7997883599197062, + 0.6901929689849652, + 0.6246071940302766, + 0.6751975833334771, + 0.6358647238955839, + 0.41644273043219665, + 0.6852637697449694, + 0.3035437550979093, + 0.4657673997246748, + 0.3889061041537047, + 0.34096233876736626, + 0.36793098407301833, + 0.3172976066372702, + 0.36659537146357496, + 0.5299016543212391, + 0.5229900757032224, + 0.43106747976272203, + 0.3275455462621275, + 0.6551912567657957, + 0.42744233642091595, + 0.5900755198230913, + 0.36218501492348004, + 0.2952738310144194, + 0.356491853237325, + 0.43576008647836806, + 0.5658589213874726, + 0.4723787041864121, + 0.3169432550517828, + 0.3871588325310726, + 0.3902799234964158, + 0.35456221435943114, + 0.4475702488973763, + 0.3518855059176402, + 0.44186082598612436, + 0.5017879266705908, + 0.5536473293484016, + 0.6130739310694945, + 0.7715319657549172, + 0.7608224022668464, + 0.756568968972198, + 0.6267629020177967, + 0.4257391806056592, + 0.48399949879555004, + 0.7628783083122079, + 0.39305241897990684, + 0.34829570075200555, + 0.5083851524339227, + 0.37738956976466687, + 0.7575300179265712, + 0.7753122417948015, + 0.4328301547867765, + 0.4236555289219875, + 0.3813369411738669, + 0.35157066783847984, + 0.4150656354336731, + 0.37021922603696045, + 0.49344000693803713, + 0.4460288251031266, + 0.4143134247743356, + 0.42912375451364015, + 0.47335528259906584, + 0.5227741722806135, + 0.39398383472548515, + 0.38986603564759786, + 0.5386648440706798, + 0.38919918540032244, + 0.42184921135160286, + 0.4367441057040984, + 0.4492095492587769, + 0.42296695795301964, + 0.4690582825231216, + 0.3961308052557469, + 0.480611298990921, + 0.2985766536233869, + 0.34795069898692016, + 0.3874445711313059, + 0.3316333159423423, + 0.4162617551946398, + 0.5243939842085292, + 0.37114381539416386, + 0.37806860162296374, + 0.3565487160082775, + 0.4456995290736598, + 0.4127828346614536, + 0.394243440353612, + 0.33700564612275913, + 0.34882677824615815, + 0.30575841864155034, + 0.3392411567085216, + 0.48103491201087434, + 0.517128171404812, + 0.4033228927863192, + 0.5479504280791743, + 0.36578136618940504, + 0.5266504837265801, + 0.3904265390253606, + 0.4373496393567057, + 0.4170503055037349, + 0.5053497734010354, + 0.30871431129572174, + 0.4578236394490386, + 0.3365285595823368, + 0.35053308803635613, + 0.4633891347353969, + 0.499829693135135, + 0.3495142915452247, + 0.7886794182909556, + 0.7789431830122182, + 0.7881225170999637, + 0.8036130850688119, + 0.5702155895098598, + 0.6788407064313242, + 0.698691176927147, + 0.7386393649770037, + 0.6812029581261997, + 0.5702067230766299, + 0.6422821986889986, + 0.5539261757112743, + 0.7536245782386175, + 0.6411479945418118, + 0.41963899870801963, + 0.34176761982872855, + 0.3089152102234951, + 0.3423416312349731, + 0.5343057550726342, + 0.4105243144366794, + 0.2978988777630057, + 0.4333095185493314, + 0.4064861602211099, + 0.45532380191981875, + 0.3660205300500697, + 0.4054429940373651, + 0.43965171722858076, + 0.4351078272282069, + 0.3594291304856044, + 0.3877841494665175, + 0.3848436025540669, + 0.3606144893988119, + 0.31799142991851254, + 0.40016193377350145, + 0.5901754090472502, + 0.30766322054293616, + 0.5356906967287587, + 0.34441882504486415, + 0.4604726635056812, + 0.4109719342763127, + 0.5632628414833516, + 0.3160149722088859, + 0.4670045317807928, + 0.32336695971763585, + 0.3061107290540021, + 0.37689129474106847, + 0.3298444124871361, + 0.2746286623127146, + 0.19903366551597057, + 0.3519506510496654, + 0.36432299274414204, + 0.4705953298776676, + 0.4178079095987308, + 0.49111221408194267, + 0.4372712604547964, + 0.37655713509329974, + 0.43420799848370994, + 0.4438307540058761, + 0.4378638471967976, + 0.37891852408813376, + 0.4309835964543433, + 0.3853923262955033, + 0.29785332413956256, + 0.4381973870713016, + 0.4159583720211243, + 0.2787171232387686, + 0.28994729810019226, + 0.4667662658746958, + 0.4990028265227574, + 0.5560041773073197, + 0.45417560523959505, + 0.3234645151167566, + 0.4944003057975433, + 0.4675874686319201, + 0.4740412144139132, + 0.29820451053265185, + 0.5008629852036093, + 0.44133575832452127, + 0.47102894101656273, + 0.25823468553925966, + 0.40426635147641327, + 0.33021353509115287, + 0.26854265666635674, + 0.48221452823001876, + 0.4222368360084593, + 0.5032077637503439, + 0.36805090621507824, + 0.5686406313068578, + 0.32631753665134244, + 0.42497320417880996, + 0.4059857353743111, + 0.3306500241707595, + 0.2780683122293303, + 0.35762106796033233, + 0.371821770791213, + 0.5894118658171548, + 0.34444924812259936, + 0.29104247544513456, + 0.42900630187566413, + 0.5031646253364753, + 0.39995798150436956, + 0.4469445755637033, + 0.3867012273060017, + 0.32745700555698004, + 0.37602931463219785, + 0.37151546773039124, + 0.4928411647465978, + 0.558146663085274, + 0.5589796122960993, + 0.5298462809120298, + 0.3280241017351346, + 0.4910607079644619, + 0.4435419274495246, + 0.282931176750759, + 0.33718847395248813, + 0.3300061450874503, + 0.21946159911711638, + 0.5247606165161662, + 0.3359111034038883, + 0.30439596932124213, + 0.5117659187233913, + 0.2914964023266073, + 0.2126772476775515, + 0.49161276756246586, + 0.40428224201332, + 0.31434678946869055, + 0.5773014671904434, + 0.5143359144421012, + 0.3075920132935755, + 0.2756490502071592, + 0.4676858248809676, + 0.3258961385492602, + 0.44771710387654223, + 0.3186328748334055, + 0.4592517664726781, + 0.4370279418851905, + 0.39012555131356025, + 0.35555243133512643, + 0.3331505608758077, + 0.442889487468016, + 0.349094053861403, + 0.42528135505909753, + 0.33088138951915586, + 0.3999512109098441, + 0.3572575815555781, + 0.36193066841020755, + 0.41818047487817583, + 0.40264345709938903, + 0.371876300891344, + 0.41778284579309366, + 0.41673319711993095, + 0.32395970311606453, + 0.5133093599647897, + 0.3276884047910275, + 0.3361492673458661, + 0.3790015155311866, + 0.3076797336998408, + 0.34880999812461116, + 0.48265509954464814, + 0.2771182672103295, + 0.3412564766386855, + 0.3312874185206782, + 0.3471450381933815, + 0.35888045551402564, + 0.4031496262742694, + 0.3787919330763375, + 0.46876280353094624, + 0.2950854337799762, + 0.4128529693396898, + 0.4891498950288027, + 0.30656292010223973, + 0.30327997084206737, + 0.5737016321136849, + 0.456423131832297, + 0.27995490563658837, + 0.48603090378941705, + 0.43243546338474587, + 0.40516586059994425, + 0.3574370335024613, + 0.4501360499593816, + 0.44770927370983205, + 0.47068298826107907, + 0.4623102674935057, + 0.4685571385741549, + 0.39576893776061567, + 0.38020331125581475, + 0.5320425214575103, + 0.45006250128000663, + 0.4588986458744319, + 0.6327081340603478, + 0.3093752097160088, + 0.5808142818047284, + 0.6276790745461371, + 0.6902432608115682, + 0.6686981195874816, + 0.519551838469366, + 0.5343060642695889, + 0.3833799553444939, + 0.666992219707838, + 0.46795092310693076, + 0.48650370078731076, + 0.3091517237259276, + 0.5029820240793489, + 0.4047656887711436, + 0.3834747443419342, + 0.4758005568864486, + 0.34746088631288385, + 0.4125923034516772, + 0.29537798296054285, + 0.4280204408955231, + 0.36289218382649974, + 0.3471742268724119, + 0.37603172235394944, + 0.2616146235596275, + 0.3436706365689572, + 0.37750044544578726, + 0.38986516595528636, + 0.3686665622252802, + 0.3840516868881271, + 0.5776714544986377, + 0.46949263136060104, + 0.4325250527743418, + 0.46797950246044234, + 0.4168082794991486, + 0.3303188090341793, + 0.3092165361467487, + 0.514626848633013, + 0.38953850276293683, + 0.3377925892777891, + 0.4307132162842787, + 0.5084192480130909, + 0.5528713366783765, + 0.4132297254830303, + 0.33878456069831775, + 0.46466595920867215, + 0.3494847033634331, + 0.33958840694100323, + 0.46266748607182956, + 0.3315437382126177, + 0.8013503825096171, + 0.5262923494047408, + 0.43043426915320815, + 0.42995006844225087, + 0.4607679819659222, + 0.4571852737363265, + 0.41064293847775085, + 0.5256273870638376, + 0.39620899497874174, + 0.545676717357648, + 0.4541812456188432, + 0.4987006343314875, + 0.5423899763413229, + 0.47603280051911595, + 0.4838825795845594, + 0.3894126471236482, + 0.41091232024092406, + 0.5943720465072305, + 0.5694368138854804, + 0.43035814646154447, + 0.5909009993017548, + 0.43870198624334195, + 0.5571075279559696, + 0.4529085052339187, + 0.4914851243318909, + 0.4421367614072762, + 0.5226068027929317, + 0.5343221931057126, + 0.5528545377316039, + 0.5054981312846784, + 0.5033780771931922, + 0.49943688812837284, + 0.4554336444867851, + 0.3910742488898427, + 0.3627019858491899, + 0.29751709406213644, + 0.34550921918695066, + 0.3019393921580928, + 0.3133335367783916, + 0.3277280447572993, + 0.45655413223802355, + 0.32813780021633804, + 0.3364791509764256, + 0.37497108169064375, + 0.5196496708088871, + 0.5845218362692347, + 0.5480746113597056, + 0.5402809839612094, + 0.41708428516231383, + 0.5510033494906921, + 0.43891787688179357, + 0.6343739283545723, + 0.46262170296852784, + 0.5225784461572887, + 0.4166462287962966, + 0.4997459512023416, + 0.5192485424065294, + 0.45211390803990686, + 0.446155118530545, + 0.48476712005091477, + 0.5277774668782844, + 0.5032096625503897, + 0.3521759415872536, + 0.386690916490977, + 0.3364048741568724, + 0.3607084750348538, + 0.49035185330831105, + 0.5308922140748285, + 0.3114692013605617, + 0.5587928082300776, + 0.5230827293849581, + 0.44457933585233916, + 0.3790251262149183, + 0.4383721224116227, + 0.5079507447757493, + 0.47686938109486526, + 0.44550733189334246, + 0.8165642817985768, + 0.41337762362595115, + 0.4771693546660191, + 0.45840384080934526, + 0.5177315752605463, + 0.44799097248584924, + 0.3906781532050738, + 0.552375444953329, + 0.44397395193301337, + 0.39433433578343147, + 0.4291709730034067, + 0.438642947854515, + 0.43776804274058484, + 0.41047869350001254, + 0.4401617267930472, + 0.43199715795312754, + 0.46294699502252373, + 0.3207493688166056, + 0.46888920369746634, + 0.3224449023095376, + 0.3749898860240477, + 0.28099021522687123, + 0.4679169207218309, + 0.361635109354071, + 0.30905160148204214, + 0.4817502616586717, + 0.37340832212811353, + 0.2773208997480894, + 0.5132226009682919, + 0.47249393394751305, + 0.4247528674657809, + 0.4804241653703893, + 0.38758064052589813, + 0.39774260348107077, + 0.5821999296714352, + 0.41310231920159907, + 0.539600562916564, + 0.3600529165181628, + 0.30408920750440155, + 0.27068894287622863, + 0.39575786531541335, + 0.449663569045881, + 0.5271762141866284, + 0.52746512661299, + 0.45163551805079977, + 0.5014195262955382, + 0.44607241925535074, + 0.4695147227671901, + 0.3922809817832996, + 0.3992751315285513, + 0.4061772576125156, + 0.3447421891310033, + 0.4124400016230501, + 0.3671482408044033, + 0.39218470105959097, + 0.470825479376206, + 0.4099141166122985, + 0.4140278568217217, + 0.3333668495506374, + 0.3546148016588683, + 0.46401615331775636, + 0.4905232979399586, + 0.5012304871929986, + 0.5102840831492604, + 0.5353048055830121, + 0.5654033353690303, + 0.6465533264779844, + 0.6272378060403335, + 0.45543373918019064, + 0.5029411600221464, + 0.4438932433094476, + 0.5114833998856305, + 0.3995316666019562, + 0.45958490713196576, + 0.4346640900378728, + 0.5568867236495866, + 0.3547112444991868, + 0.5406224495041924, + 0.6260024533721039, + 0.5982917895869868, + 0.5525790239483155, + 0.6231385781661948, + 0.5265227966391126, + 0.4600484480854685, + 0.3713815316456365, + 0.40636212118933684, + 0.3274692195502157, + 0.3354664103597285, + 0.5091371997184424, + 0.5165437770195651, + 0.4131618348412966, + 0.32878106932085954, + 0.35871664129725733, + 0.44153802934377356, + 0.41493822222173093, + 0.3846027013263758, + 0.47620047124501713, + 0.489227228793638, + 0.3708633193001343, + 0.424072350174525, + 0.337470085593461, + 0.4378118699663137, + 0.45644314890410775, + 0.3404702219374335, + 0.36504940094347377, + 0.41024103583790483, + 0.358843084457172, + 0.3602458926137557, + 0.3822213265017314, + 0.34962684063138005, + 0.37301863904832594, + 0.3726525855088365, + 0.4662397429888515, + 0.3673465699458526, + 0.2957087679332739, + 0.5290355049703314, + 0.383969990515094, + 0.49124979505239, + 0.49236864571280575, + 0.3585877988713107, + 0.4657741500398225, + 0.5323216061811009, + 0.4640567007473736, + 0.5831218005532831, + 0.38530049901412045, + 0.4554415547732813, + 0.43223445135166577, + 0.43375817188725907, + 0.4836492859562779, + 0.4698384807591797, + 0.35708854040600135, + 0.4201154167936567, + 0.47086873724302203, + 0.4038730060244809, + 0.31157779046499734, + 0.42781262514973906, + 0.45114225138976705, + 0.4346500542334458, + 0.5787558170138503, + 0.49831622281463117, + 0.5112561076132006, + 0.5105219063864416, + 0.5780859552765553, + 0.5129612379759004, + 0.5596090701906606, + 0.5530434963582855, + 0.5554771960722152, + 0.3688772271162767, + 0.38814332462682793, + 0.35314313226345506, + 0.3275335353146627, + 0.30053192359620423, + 0.3295234651043389, + 0.35226653104307704, + 0.33969337547708794, + 0.2870829067118212, + 0.43584727025387926, + 0.28101938426618844, + 0.4263812815376171, + 0.3255568538105696, + 0.5025090932831057, + 0.5390714961801698, + 0.49245409475729407, + 0.31550740310098924, + 0.3894076384432042, + 0.5547884206747887, + 0.3375642229057855, + 0.45007947740292076, + 0.30632855796438613, + 0.7911696432486914, + 0.49983110207374293, + 0.3972445454121555, + 0.5522726317065254, + 0.49325620940669984, + 0.4585403351347915, + 0.40726514489651455, + 0.46432177782226985, + 0.5151363770154193, + 0.4297836795833246, + 0.5329311080864825, + 0.5026577836166275, + 0.29304448954015494, + 0.4513558299001373, + 0.29510004843221194, + 0.3543411595823721, + 0.3036358115496927, + 0.4862044552847034, + 0.4110066742111459, + 0.36980904551136534, + 0.46493152354316447, + 0.5030061949169232, + 0.36927694090258056, + 0.4592386273182072, + 0.41003403865054583, + 0.4404295483077064, + 0.4689418713060842, + 0.42790407052882584, + 0.33213933165556914, + 0.3160144445475924, + 0.5276587786780624, + 0.39177005560228684, + 0.36615076684469866, + 0.49581771215270176, + 0.39263563715066024, + 0.41247408287446513, + 0.3154104638766657, + 0.35326263965470234, + 0.3904878402926401, + 0.4336102344696831, + 0.43626259250072763, + 0.3244683275244054, + 0.4496629721904608, + 0.23909867546374036, + 0.45055786619149435, + 0.4215092895281938, + 0.44362057997482185, + 0.45866497766846454, + 0.3799743313092267, + 0.45733013593643146, + 0.32211941245229364, + 0.46167200502191075, + 0.38081805762372, + 0.40802446948149146, + 0.42200237246886174, + 0.3262954123117622, + 0.4716631587778172, + 0.21154697860045402, + 0.6528816492820435, + 0.6595564961273257, + 0.5542786810397008, + 0.5484991845604605, + 0.5862378615130995, + 0.5915132407417742, + 0.27710669890111095, + 0.36224113436334043, + 0.33946006011282076, + 0.3712700331551545, + 0.3256940447782285, + 0.33355100614076266, + 0.36851824729512817, + 0.4258291828206715, + 0.40960266573131826, + 0.33220847608614745, + 0.3592473252314847, + 0.26949712294007183, + 0.4548477895494771, + 0.24192910840865486, + 0.3877784934288872, + 0.4294237162503014, + 0.4299131509555377, + 0.46302201417374583, + 0.3903783925806002, + 0.4912237972482116, + 0.23790654128621125, + 0.4774685247645724, + 0.4045812136887522, + 0.37110459665394596, + 0.35151210483297335, + 0.3882926008434512, + 0.3922020686305118, + 0.40857406212433073, + 0.46908829108090255, + 0.48211357980001385, + 0.5207693298371632, + 0.33947789779052334, + 0.3804104882254126, + 0.38245769285885234, + 0.41764860360421496, + 0.4504490154046392, + 0.44812955926018455, + 0.3340015903430695, + 0.5097818099180487, + 0.42363214930190635, + 0.4141744383830031, + 0.5221062491338306, + 0.3570095810816332, + 0.2746098548269484, + 0.2623859563941719, + 0.33894562904027503, + 0.3844183081157817, + 0.3471667425701603, + 0.37064830714347774, + 0.36989766507684047, + 0.4744162540218658, + 0.5538327213329505, + 0.6049801196808753, + 0.5091566587294309, + 0.5359372043496364, + 0.4675014208912283, + 0.5216596689833657, + 0.4735168967793036, + 0.5189956158891679, + 0.35834519537358483, + 0.4372984509852134, + 0.5530644080911226, + 0.6157348518174218, + 0.1994524521052163, + 0.32936888149700294, + 0.3909905556788014, + 0.4288204004176188, + 0.39918907260044656, + 0.4738452687567325, + 0.3706061942205526, + 0.4648976147605503, + 0.7171460998310726, + 0.5962505157205628, + 0.7377522015873009, + 0.7531776198386224, + 0.6360290713206679, + 0.7148357903907985, + 0.7208337476327478, + 0.3634047146066997, + 0.37965723782222277, + 0.38206198695945004, + 0.4871448736029017, + 0.338035653800973, + 0.43769439787939496, + 0.35977997814053275, + 0.45028746806605313, + 0.39131426225398525, + 0.39091407549422685, + 0.2844645450914965, + 0.4325802253628927, + 0.3833635659175871, + 0.4328298460631777, + 0.4200208863789122, + 0.30391127476622354, + 0.34553372116649556, + 0.32422451240694566, + 0.3401611649975077, + 0.26684334196509085, + 0.35222592740637604, + 0.2947618460950303, + 0.5007099930104937, + 0.5348926083517277, + 0.35568114988665556, + 0.31044323592317946, + 0.22105041867653336, + 0.5114974544848908, + 0.606326907574535, + 0.6327335926541193, + 0.3614825532385281, + 0.35113017633272636, + 0.24542308046896935, + 0.45896455391738444, + 0.32843761343049477, + 0.3631668828440887, + 0.46584581846295386, + 0.5028418119094779, + 0.45426888764818657, + 0.3062021274076105, + 0.4404078377087884, + 0.4712812421578792, + 0.47892424021656904, + 0.43922145567936216, + 0.48266641050693765, + 0.37500027277940345, + 0.4188550273286806, + 0.4047835588562977, + 0.4556743819861154, + 0.3344200008120011, + 0.39430614377302897, + 0.537261697379169, + 0.4044528752830868, + 0.593464169633943, + 0.3072897138997731, + 0.531296531640606, + 0.6918866197540141, + 0.5141734165413641, + 0.38104759154978557, + 0.34230912714924555, + 0.3285178732732516, + 0.31016336938018474, + 0.40484312522154187, + 0.5419408713991992, + 0.43829603473751455, + 0.38627870698083033, + 0.2867209098041075, + 0.4504706553602236, + 0.46939911795463, + 0.5470444126929581, + 0.5865405843866632, + 0.4760072198506514, + 0.49316347165690577, + 0.3732698562980851, + 0.4677120380147982, + 0.2937628276233644, + 0.33996620534248784, + 0.4228064484064585, + 0.3606945200418921, + 0.5423219767948028, + 0.5349226878059715, + 0.5993102999172215, + 0.6521819338527677, + 0.7305139086120978, + 0.5815914890642025, + 0.7368266688481127, + 0.4415659062593429, + 0.3863085305746743, + 0.3729037862425406, + 0.4935626669123063, + 0.3551984256487354, + 0.48186443333861373, + 0.4774617234646858, + 0.5035680020038439, + 0.4559705626794558, + 0.43003452821917404, + 0.41397561684694095, + 0.4268468869952185, + 0.3627591682564956, + 0.46775223851789866, + 0.3964870246765116, + 0.4246619807488492, + 0.46711407010705713, + 0.521332856700764, + 0.5213901743927559, + 0.3421025197383161, + 0.4587663285629236, + 0.484815585405446, + 0.3269161730986669, + 0.4118424323947888, + 0.3960723863730815, + 0.39040166556539774, + 0.38365300801318986, + 0.40578182071048496, + 0.4055863985687967, + 0.3984361788062366, + 0.4319287172559871, + 0.39106542676597406, + 0.30906324449872935, + 0.4535130989265415, + 0.4388798202445851, + 0.3952218260231795, + 0.5083812912604108, + 0.533913390347054, + 0.7170309069770697, + 0.32532902873126424, + 0.2946836623551855, + 0.46151910672611324, + 0.35571936666127124, + 0.27126058019096594, + 0.31552881289231655, + 0.31781210506945584, + 0.6608983591324497, + 0.5963600100452585, + 0.6352935947610109, + 0.34443001927063255, + 0.3431816120168949, + 0.31097413248461175, + 0.4692744782516768, + 0.4018196911849834, + 0.24303774206459708, + 0.2535088592190478, + 0.4102952847837689, + 0.3732115429364981, + 0.4193093253963093, + 0.43674761556613245, + 0.4182455585283008, + 0.3773042142299743, + 0.2966118704862093, + 0.6691885496395072, + 0.528234351200994, + 0.6720812791159759, + 0.5646493023127215, + 0.25168330125199134, + 0.485486505681103, + 0.490684619461586, + 0.5021131214962598, + 0.4512682962833044, + 0.38712523722567627, + 0.4786659494341772, + 0.49580781713757655, + 0.510835808360194, + 0.45349594576698776, + 0.3611280532282639, + 0.4131485560314716, + 0.352279773612177, + 0.43086465750104735, + 0.6291574245187558, + 0.48223296917936587, + 0.558013016793724, + 0.4438995028420007, + 0.46631268222207417, + 0.437238118727815, + 0.4110338590932557, + 0.4678719536134689, + 0.40672016958786333, + 0.2766137725499892, + 0.3672736438879978, + 0.45084389616677384, + 0.2544800749047958, + 0.7388786297763231, + 0.41845325074127787, + 0.5848869740711672, + 0.7054045946692221, + 0.6660283438869928, + 0.3727815463223979, + 0.7768081803680819, + 0.48385314642494376, + 0.6475488044162603, + 0.3726050693473005, + 0.5083793613007883, + 0.46500741187459654, + 0.5393418134723549, + 0.5237001401834723, + 0.5875355034457633, + 0.5383436188113622, + 0.3841110882734072, + 0.46413375439103755, + 0.42033766965695324, + 0.3840524229576459, + 0.5907378061646789, + 0.30008070507146006, + 0.5275832062246263, + 0.4494894972027066, + 0.5502766738958117, + 0.5163438429134654, + 0.587282771018361, + 0.4911491153785247, + 0.38922428820290905, + 0.4868451719644663, + 0.5106277000058577, + 0.48160483533129544, + 0.391646877447347, + 0.3441925684596108, + 0.46284097202451635, + 0.5234348626842581, + 0.5320678196186605, + 0.407698952738472, + 0.5567011495222534, + 0.5636165688878955, + 0.411834572192123, + 0.5034858543460414, + 0.5137379381246799, + 0.4858880538824158, + 0.43612149047814597, + 0.36651285851961146, + 0.3511571658150146, + 0.4257190147935061, + 0.39216296336881573, + 0.32936687983554175, + 0.43320402615150566, + 0.33331002778660457, + 0.42128023302898143, + 0.3686047700914388, + 0.33021674057299405, + 0.3336513177103265, + 0.3190817928443213, + 0.42047603126017863, + 0.40984781229340017, + 0.34525661943778047, + 0.3147306834449092, + 0.4076800698065317, + 0.5496829638040037, + 0.6261731017574707, + 0.3162025912451851, + 0.5169558788121204, + 0.5335055538261494, + 0.5380438139372213, + 0.5229703649326569, + 0.4180091995017116, + 0.4486418645648761, + 0.3309924693753786, + 0.4810567728435638, + 0.3944832799241469, + 0.6220082971077278, + 0.44549907311270914, + 0.47191066091507716, + 0.28621769607539005, + 0.2801714110609607, + 0.38779964681598345, + 0.4630993418459828, + 0.449899012119432, + 0.5563547129557359, + 0.3359487231025367, + 0.5826117779393984, + 0.48298026961910906, + 0.5005479660078869, + 0.5101356864973001, + 0.4774940817372767, + 0.41443269620055745, + 0.508855917898916, + 0.5159929535614779, + 0.5197813212951922, + 0.4935978961779331, + 0.4329942564841226, + 0.34258841819978453, + 0.3536209154662225, + 0.3777871325305316, + 0.24035253925838831, + 0.38175719579269507, + 0.47001245406481573, + 0.4167549293106288, + 0.40302661172715976, + 0.42972002328114994, + 0.39615041045586985, + 0.4832553857293192, + 0.4966785290931583, + 0.3567979551993399, + 0.4635925132136372, + 0.5301520351999915, + 0.36858691924817893, + 0.34791069271446723, + 0.4462837582859017, + 0.44181844016409144, + 0.5327302319252173, + 0.2972789817030858, + 0.4877001943992797, + 0.3341909547945496, + 0.5271375006205311, + 0.5073438418880174, + 0.5612382277445486, + 0.25545850736360665, + 0.3055509310541546, + 0.3059149061666529, + 0.27452071063554345, + 0.30276262793339637, + 0.38465718502620105, + 0.4397441408728825, + 0.27649173325948534, + 0.3477247756987676, + 0.4240415773289457, + 0.4071118939835478, + 0.4768237929298391, + 0.40858967818648, + 0.5586468273194063, + 0.23446620403220905, + 0.44753655564749883, + 0.4182710533490872, + 0.3492349012857, + 0.3734450137405833, + 0.45421257562152456, + 0.45626451696008574, + 0.24237993869858088, + 0.49504567236671526, + 0.4035467386517429, + 0.4012259040176982, + 0.5321428874841646, + 0.3773458838150921, + 0.36288571119379853, + 0.3722819602715836, + 0.4034122600021642, + 0.40443136237035426, + 0.3867066064251713, + 0.4123259438967379, + 0.493222968626721, + 0.41156022362132844, + 0.5361704191889975, + 0.45022767729374497, + 0.40842605152512146, + 0.4503487903367108, + 0.46034445918141587, + 0.4508283685386061, + 0.43273202233412, + 0.5519558394210908, + 0.496957173965237, + 0.5109457609988112, + 0.5399673211463201, + 0.6139719379762081, + 0.4469576625606233, + 0.43981264579299495, + 0.31745351812367617, + 0.4190867384438824, + 0.48007594841073015, + 0.44781539069715326, + 0.43605492402951795, + 0.48651101277582975, + 0.3468831929748234, + 0.44091740998524265, + 0.3708060384534195, + 0.29360398993830406, + 0.42659382588909406, + 0.42252444256722144, + 0.4235162327033595, + 0.4266755861850604, + 0.33246510257724127, + 0.568602858683796, + 0.5682944835143123, + 0.5008248831379745, + 0.5152580881680614, + 0.5929991227873015, + 0.4461680965089051, + 0.31426729720924773, + 0.4056161360113599, + 0.38533392351848034, + 0.3720843354941819, + 0.36023781446398506, + 0.3651908745109379, + 0.4196060303910557, + 0.41553357382642, + 0.3148126546658201 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.0473881090948396, + "valid_tracks": 4, + "per_track_pearson": [ + 0.048597849662046756, + 0.045606562403327985, + 0.052420324368813366, + 0.042927699945170296 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.3931558119201712, + "valid_tracks": 626, + "per_track_pearson": [ + 0.4024934780013171, + 0.3859951089413366, + 0.3684920372509827, + 0.39663195675767193, + 0.37962549714837734, + 0.39526217178920936, + 0.3304763180065924, + 0.3686818195446865, + 0.38627699783032715, + 0.38868758499808137, + 0.3919460087055975, + 0.37734301031176304, + 0.3859131117250519, + 0.36929063808482476, + 0.39040522587263193, + 0.39074799828685824, + 0.38849014012891375, + 0.3853618322881984, + 0.39023622498095545, + 0.3834719298579545, + 0.3965793907987911, + 0.40121130672931254, + 0.38101293586308377, + 0.39056522527767495, + 0.389444223493002, + 0.3989235724725366, + 0.3789166312666395, + 0.3825840376903636, + 0.38414640783348974, + 0.40096259490155256, + 0.38799054637276126, + 0.3814337353095191, + 0.36674411327435014, + 0.3916852922774139, + 0.3772182945188398, + 0.3910958013990191, + 0.34885872600601175, + 0.395828814624532, + 0.3923318393631195, + 0.39209002299530443, + 0.3906520288975859, + 0.37256067851745805, + 0.36919260882913224, + 0.38378714761701227, + 0.3834786651920989, + 0.3573814700468925, + 0.38298908526569425, + 0.3765504488447353, + 0.37252981089114595, + 0.39161267974016883, + 0.3684688577655807, + 0.3847421713892663, + 0.3860294473211303, + 0.3946674145410774, + 0.3859872620977525, + 0.3884808564590675, + 0.39285726300488105, + 0.38515529824491396, + 0.39076709077300154, + 0.40049767397920655, + 0.3824184399281184, + 0.3970527938519153, + 0.3979597372656722, + 0.39601566382680603, + 0.393903121671145, + 0.3787724774351388, + 0.4018422496828273, + 0.40511345746704014, + 0.3908268571025375, + 0.387254914888227, + 0.39209127599810795, + 0.38627814780752584, + 0.38696018481025585, + 0.37884623500885595, + 0.40944700550709656, + 0.41370795927456744, + 0.38739590054583906, + 0.38186037767711534, + 0.3809862652338185, + 0.3883061712847788, + 0.40078011919952145, + 0.37933749716922427, + 0.37289120997936304, + 0.3755153081300919, + 0.3893551743243752, + 0.3857653315353533, + 0.39233034771422415, + 0.3727377912582213, + 0.38565955082072434, + 0.38758941645253686, + 0.386911568424947, + 0.3748733901000797, + 0.3890294561044698, + 0.4067611938587695, + 0.4146849408078642, + 0.40530384307780115, + 0.3869095044609192, + 0.3915373486636454, + 0.39263747584575814, + 0.3904621220622598, + 0.3762037047538343, + 0.4011499502166185, + 0.41443977214742855, + 0.3911231365694491, + 0.3874851401554918, + 0.39344092439795003, + 0.37869828675545636, + 0.3557029269557699, + 0.4113202504997093, + 0.3937405112353227, + 0.3836619398091634, + 0.4120747332910252, + 0.39250458952642664, + 0.40487449774772255, + 0.4060323535015029, + 0.4083138819294275, + 0.3795022067841108, + 0.3842716985615187, + 0.3730730074603633, + 0.40401498371631533, + 0.4132020513398636, + 0.39705487915336285, + 0.4013276629688325, + 0.393735418762377, + 0.3994312719029663, + null, + 0.39700152293247704, + 0.39848995906034257, + 0.407184667091541, + 0.4253749802974013, + 0.3979738061659461, + 0.3853669287387897, + 0.36951939963190283, + 0.39407143876566664, + 0.4067066026180332, + 0.40688477287861663, + null, + 0.3980030708313016, + 0.41410083895837013, + 0.37444601415933865, + 0.37849231694717544, + 0.4063144793504493, + 0.40083646552860497, + 0.40663713211585445, + 0.3953923307417052, + 0.3953741941555188, + 0.394697605401601, + 0.40159764166855544, + 0.3870568418523107, + 0.4001703509634546, + 0.41342232896092607, + 0.41778576024197994, + 0.3892740976806208, + 0.3997542925966364, + 0.38505703948336717, + 0.4155316010763627, + 0.40122106181997197, + 0.4202212926338895, + 0.39033399990503537, + 0.41596861146744557, + 0.39987738569496, + 0.39580193994158847, + 0.3965698470315111, + 0.39332333254867774, + 0.4040232200299036, + 0.40230906501244523, + 0.40051280683483476, + 0.39613415358125104, + 0.3939860909278426, + 0.40363580555321077, + 0.40099540528632166, + 0.3932341990182605, + 0.39711273458112484, + 0.39763362393699414, + 0.3986758145321152, + 0.3625308999918256, + 0.39655988919193663, + 0.39897865620008394, + 0.4002671557067998, + 0.4013231395032385, + 0.396298570711172, + 0.39675781813903493, + 0.40615943895890544, + 0.39255943566510143, + 0.3805595303626151, + 0.37910710328644065, + 0.3978106256594695, + 0.3959806725431847, + 0.3985001020187363, + 0.3887209987964682, + 0.388076328920821, + 0.37607794137605466, + 0.37405814045669944, + 0.38511167088001336, + 0.3810421451488756, + 0.38784006626061707, + 0.3814991584779412, + 0.3865114982612059, + 0.3965423418060105, + 0.3948187161835738, + null, + 0.39316330092922736, + 0.3668673991202374, + 0.36423119483423755, + 0.3814874513633464, + 0.39869773452260293, + 0.39551291758462365, + 0.3874367230326257, + null, + null, + 0.3742283438980504, + 0.40631349234371555, + null, + 0.38303751575073436, + 0.3633685421408934, + 0.36587737451855307, + 0.3746256293152335, + 0.38374276881884295, + 0.38605626189298686, + 0.378610927788397, + null, + 0.37791842904900086, + 0.3698345059378174, + null, + 0.3331668232450087, + null, + 0.3569089463176276, + 0.3784459452805526, + 0.3725270416227176, + 0.3728226601388155, + null, + 0.38193052329988636, + 0.3770377820512684, + 0.3663754437361032, + null, + 0.37833598895796505, + null, + 0.3393181986306073, + null, + 0.3959092733393789, + 0.3641176164802696, + null, + 0.35701416457084667, + 0.37793092121795624, + null, + null, + 0.38970633259191906, + 0.37105290376645766, + 0.36791595757740714, + 0.37845483783637324, + null, + 0.3802605717791605, + null, + 0.37969911926146177, + 0.3875418567298422, + null, + 0.32265533626913134, + 0.3720126983927875, + 0.37134472111223654, + null, + 0.3610553781745296, + 0.37882267299015676, + null, + 0.3756977129404946, + 0.37515754257786377, + 0.35841617564953454, + null, + null, + null, + null, + 0.3715447527575255, + null, + null, + 0.3990505258714853, + 0.386325263559756, + 0.35385730007775335, + null, + 0.3840032813728378, + 0.3624995145366094, + null, + null, + 0.3873922964774013, + 0.3687962617427367, + 0.3633958118804703, + 0.38619884849649233, + 0.3909065108362117, + 0.36756408417095227, + 0.3843335862833609, + 0.38201371706428466, + 0.39101284191411473, + 0.37901090631185125, + 0.3661587630431713, + 0.39163853982558927, + null, + 0.3852472442341151, + 0.3913385932329467, + 0.38277595330447917, + 0.3705967364674788, + 0.37507185705219137, + 0.3790131782138748, + 0.38353267362701127, + 0.36032794612916247, + 0.3626564560774833, + 0.37156826482741984, + 0.38860772093885054, + 0.35759101665081955, + null, + 0.372817667912919, + 0.3818631989640049, + 0.3704840110453208, + null, + 0.3605918602608812, + 0.33785190067456455, + null, + 0.37905510034286, + null, + 0.3771979376258449, + 0.3705310275486025, + 0.37543080682315744, + 0.3851977005313698, + 0.38196879509381226, + 0.37883438872038616, + 0.37127532997052964, + null, + null, + 0.37351341635362323, + 0.3841097201079587, + null, + 0.3784136386724772, + null, + 0.3697593058564021, + 0.3656527139908571, + 0.3788275844251376, + 0.38242482905499, + null, + null, + null, + null, + 0.39488112284231874, + null, + 0.36717323383034217, + 0.38525463633105717, + null, + null, + 0.3713914992072936, + 0.38747438525219535, + 0.3947847424148415, + null, + 0.36776163649721705, + 0.3833834328777001, + 0.3638329981684219, + null, + 0.3822900526858014, + null, + null, + 0.3817039451540178, + null, + 0.3801683066628273, + null, + null, + 0.392271812700005, + 0.3704749338825016, + 0.37178658033457235, + 0.3753534634186977, + 0.3646288375601971, + null, + 0.37304740554738125, + 0.4259024471063623, + 0.43183120842290323, + 0.3754506297241297, + 0.37910955210788394, + 0.39507488479969, + 0.41348134759679794, + 0.4062909386347983, + 0.4286298216881602, + 0.40328430681066574, + 0.42539476142307564, + 0.434639689934054, + 0.39727265695269093, + 0.38250374981812985, + 0.37461289823431715, + 0.37153659718883686, + 0.3606619452067514, + 0.40280328989112024, + 0.42801575876516107, + 0.4144392819193071, + 0.4186663785293182, + 0.40089312060634935, + 0.37462966919259433, + 0.3630460989729009, + 0.3664343172846979, + 0.4302219383318904, + 0.4198968703753314, + 0.41925368358350007, + 0.4245226231324305, + 0.36794957408464696, + 0.35938127243766543, + 0.3802986417741771, + 0.38167827151162614, + 0.37360009444676107, + 0.39417801275232933, + 0.365556961242761, + 0.4146548333659249, + 0.3589913169161384, + 0.43650127838127695, + 0.4349758886826781, + 0.3742110786498504, + 0.3824531442468725, + 0.3794724072950866, + 0.38831118058972547, + 0.40777932003001444, + 0.40856325882273975, + 0.37384619305808114, + 0.3994779227928874, + 0.38819200261073833, + 0.415469298635937, + 0.3885854648696293, + 0.4007617623978605, + 0.3837041915827697, + 0.4150017863185885, + 0.40958868706740736, + 0.38355544823493537, + 0.40751690715936095, + 0.4132512669494668, + 0.4179067646254559, + 0.41098804524116817, + 0.4352521259668825, + 0.429683598645283, + 0.430546378072111, + 0.4340440021071919, + 0.4280687109193819, + 0.4306082209686629, + 0.42743232257721575, + 0.4049093121892602, + 0.4388398224152152, + 0.4343186014654664, + 0.3611099571784454, + 0.4202541366694048, + 0.4414560434234108, + 0.40302600595423843, + 0.40504186391792013, + 0.40816692434464846, + 0.416388697866153, + 0.3959962761010032, + 0.4153769768842826, + 0.4008672897511371, + 0.4186108346569944, + 0.42574964116027925, + 0.40908307966108015, + 0.4060779469006036, + 0.43544695338168493, + 0.4159763026310026, + 0.41398525121325447, + 0.39284640677055394, + 0.4165935688017075, + 0.41491705747681873, + 0.4228289086161727, + 0.41043896877135744, + 0.4066341039892422, + 0.4112454059349172, + 0.41873017051101225, + 0.43319774084061263, + 0.4344061083780319, + 0.4032111462532076, + 0.44816449561877314, + 0.4192337826663578, + 0.45215515562777603, + 0.4357272911626424, + 0.4395104909340269, + 0.42897638773139884, + 0.41213081368097587, + 0.4258508436448159, + 0.4225304881516105, + 0.41173845461846975, + 0.41580604064701904, + 0.3956344947196566, + 0.4045980143076794, + 0.41814071602441377, + 0.3985590607251363, + 0.4076358021144764, + 0.41513843212567736, + 0.41271872631285733, + 0.43226239178362474, + 0.4179048175626108, + 0.42635361933534605, + 0.3766264363591421, + 0.4306501882976844, + 0.4141652701312915, + 0.4408282348599204, + 0.4141988734158003, + 0.4222918855386992, + 0.4493392666943805, + null, + 0.40824863623072594, + 0.38707921605771817, + 0.42058619635528827, + 0.44052192643443117, + 0.423190706285444, + 0.4266023722488101, + 0.43342105195662517, + 0.45528306737780777, + 0.44224178874312503, + 0.4327163862456054, + null, + 0.41589930847189577, + 0.44004172680533116, + 0.43195288955380834, + 0.41348181788789373, + 0.4262979529421032, + 0.43362839389917635, + 0.4105300574767248, + 0.4506474992149608, + 0.4506059835082251, + 0.4445286720743856, + 0.43126546072960736, + 0.4611381505900625, + 0.45045258916208536, + 0.4521703815263274, + 0.4417759224155656, + 0.4275354278768189, + 0.39556403555045916, + 0.3765721659921307, + 0.38032353295793203, + 0.3766096089891502, + 0.4313292709719224, + 0.42022429127506217, + 0.4282481949393087, + 0.4258857357016606, + 0.43726159671783194, + 0.4137314809506342, + 0.39174024768765225, + 0.4236717747954226, + 0.4011434430201827, + 0.43462474520217653, + 0.42541277008174044, + 0.4418096173811931, + 0.44631694092690044, + 0.4288003639488888, + 0.4351719163529055, + 0.44107076432072934, + 0.4253761907229464, + 0.3917040649639126, + 0.42819082151605825, + 0.44362841839819656, + 0.44124774989325977, + 0.4259046836181828, + 0.38810731449732894, + 0.4218761942891247, + 0.4340266474869726, + 0.427676527928632, + 0.40772216328356964, + 0.4147600776596287, + 0.4296138130352052, + 0.4118538118751073, + 0.3895941562126031, + 0.4096700437759658, + 0.4140825826112697, + 0.3968062504646187, + 0.44437602629743145, + 0.418588476423655, + 0.3772434163841533, + 0.3802625062175436, + 0.38103686912185225, + 0.38845853112745604, + 0.4225958090725965, + 0.4198445865794363, + 0.39857328745884474, + null, + 0.3823873554951641, + 0.36638056398300606, + 0.3792221633177435, + 0.3652225276205327, + 0.4108069971235039, + 0.40196556051812293, + 0.3729199717580599, + null, + null, + 0.39152507876646414, + 0.392727787179276, + null, + 0.3854325464353093, + 0.3475193797797695, + 0.3872096032023553, + 0.3647076723017431, + 0.36989925613636643, + 0.36352258270544313, + 0.4336547413085888, + null, + 0.3958693874266348, + 0.39324439251865956, + null, + 0.38790833743445224, + null, + 0.38038495485418916, + 0.3867505030875236, + 0.38645195355138867, + 0.38038742595503655, + null, + 0.34011602533183416, + 0.38040300283700096, + 0.35764233819373736, + null, + 0.33689908019298886, + null, + 0.36639146945100903, + null, + 0.36827560159181566, + 0.37416221442962494, + null, + 0.34848178370074345, + 0.41334996538827645, + null, + null, + 0.38838069374812023, + 0.364823071797308, + 0.3505601088428192, + 0.3468676132411679, + null, + 0.36169796713336494, + null, + 0.3867217467128588, + 0.3654133479815331, + null, + 0.33906324612542965, + 0.3510245360756762, + 0.36595097074931204, + null, + 0.3531794376285649, + 0.34919241516301236, + null, + 0.4178125712621421, + 0.4038478346656535, + 0.41875091303047207, + null, + null, + null, + null, + 0.41647631825663844, + null, + null, + 0.3914582143581269, + 0.37389550186115283, + 0.3921517009179087, + null, + 0.42530125352854137, + 0.40084533881757967, + null, + null, + 0.3940098331083636, + 0.37574411922252043, + 0.3817174554085451, + 0.3542618954229835, + 0.3657973615668031, + 0.3845604125997211, + 0.36396151112555863, + 0.3649599718419484, + 0.356703842453889, + 0.37633498936812837, + 0.3508038430547034, + 0.3806513571961797, + null, + 0.38993636216870226, + 0.36824730938454714, + 0.39187764839017336, + 0.38717649185501407, + 0.39008392884351795, + 0.3980247949557819, + 0.39288662172522465, + 0.3945742600035983, + 0.40131956436102745, + 0.38556928146803965, + 0.3672380500285164, + 0.3618986799516686, + null, + 0.3513022582685628, + 0.4069032847375136, + 0.40180566896808173, + null, + 0.35997531891177714, + 0.3733666067891329, + null, + 0.3747371370962934, + null, + 0.3704589318589964, + 0.3667032501035083, + 0.3832454049007504, + 0.3916435523414453, + 0.3992069832657759, + 0.3739650912140343, + 0.36158933248143516, + null, + null, + 0.39338849657863484, + 0.42142465836444415, + null, + 0.37708707745373343, + null, + 0.375609500238351, + 0.3668101603103933, + 0.36493152241887233, + 0.41456282867533994, + null, + null, + null, + null, + 0.38342655692406974, + null, + 0.38548017808104884, + 0.4034538302278593, + null, + null, + 0.37722136666074635, + 0.36001639550440573, + 0.3589903368034643, + null, + 0.36039306348909256, + 0.3665576657870841, + 0.3626548793084804, + null, + 0.38973378171621337, + null, + null, + 0.36648089657456995, + null, + 0.34672154628536006, + null, + null, + 0.3750209295828608, + 0.40779406941558494, + 0.40932405976954106, + 0.4065697759861481, + 0.41397069999237895, + null, + 0.3878642218097058 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077, + 0.652321630804077 + ] + }, + "junctions": { + "mean_pearson": 0.28730207196171553, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2665980442498605, + 0.34068707886385435, + 0.3821131240378525, + 0.33240110152202806, + 0.36372399497506014, + 0.2939191110141062, + 0.10256231347112027, + 0.2342663692202889, + 0.25468549747516567, + 0.21292093490288122, + 0.38908493512713727, + 0.16320229335516578, + 0.21273293036372018, + 0.18226634596453478, + 0.19375905030629892, + 0.36648617635976377, + 0.36562584623930827, + 0.1938812542768831, + 0.19497085777280837, + 0.31832261074539936, + 0.20771898526074933, + 0.23419335883014414, + 0.2902207712599234, + 0.3137728903249585, + 0.21895547334357401, + 0.2195488489714471, + 0.3776706209497018, + 0.3905989604033798, + 0.19899847932796877, + 0.2901718679057803, + 0.36995960881570317, + 0.4273805285107128, + 0.37029661174360046, + 0.3382154980364726, + 0.36354943449289284, + 0.35932696674881637, + 0.31173564436831447, + 0.31415759793271114, + 0.29460824555815623, + 0.18033091344628918, + 0.40292932297521933, + 0.25994743519054697, + 0.25085766302548185, + 0.35874329637392316, + 0.2928860668749966, + 0.36133301581288013, + 0.18842385791157634, + 0.37801190953102204, + 0.21134346966952666, + 0.4105440367890166, + 0.2830026986740668, + 0.21194278859167878, + 0.17393542989650948, + 0.23318470482558465, + 0.33465673463434126, + 0.24358696417079856, + 0.1727483200541059, + 0.3576974284085981, + 0.2597455799343924, + 0.2209211384488072, + 0.17502696476744906, + 0.24369858697670163, + 0.1953352926307307, + 0.12689705866358075, + 0.17151135247964927, + 0.2552061902387451, + 0.4174622396437886, + 0.3758376799598066, + 0.2677452985998697, + 0.21464719513469355, + 0.2265822347868716, + 0.22598619755785138, + 0.20426652912827697, + 0.19543502318635478, + 0.37155122065938156, + 0.31176737974723945, + 0.37659203933891766, + 0.31244516672837497, + 0.12895957769567393, + 0.2952716800294283, + 0.3241216755949203, + 0.3608796075719692, + 0.148603988627282, + 0.24764366307664776, + 0.3561266936745582, + 0.2091114014687628, + 0.19351894146674223, + 0.3126320408948791, + 0.18427567583500484, + 0.39181585398616847, + 0.2985943475501639, + 0.4151119220930804, + 0.19578274528614453, + 0.2545769122792625, + 0.245052933148838, + 0.4143038858128491, + 0.3712688053716359, + 0.21090607082077623, + 0.21082252305402685, + 0.19729545117926253, + 0.22726961831119738, + 0.3708537638862618, + 0.34564035519261777, + 0.12530578422345642, + 0.12381219456276842, + 0.15567784579185046, + 0.31378034109353675, + 0.3563188187094473, + 0.1536016705203741, + 0.2112356905387709, + 0.35216316049799024, + 0.26406518748056196, + 0.3925588487251669, + 0.3320632065939196, + 0.3158872701425496, + 0.3005708835958289, + 0.33512600559507816, + 0.31149937502880704, + 0.3154645665184124, + 0.4079325662662902, + 0.36542730286873704, + 0.25632753673182945, + 0.32960179909218035, + 0.14444520239217495, + 0.3340114336695799, + null, + 0.32871281357897675, + 0.17153899149766075, + 0.1154733341923646, + 0.2231706605726823, + 0.32390426295987046, + 0.2764910796163618, + 0.25124691810879374, + 0.2619725858640493, + 0.29576465786994655, + 0.2947090706780976, + null, + 0.20723650140295954, + 0.28479291826093106, + 0.2222664611396606, + 0.26495915266084674, + 0.2617025777659124, + 0.26809450951990826, + 0.2540642718284369, + 0.1983621559336853, + 0.1460689011175637, + 0.20340964967922615, + 0.4187951807973346, + 0.24817797485123477, + 0.2986286563712002, + 0.1900952322883112, + 0.1854292586454286, + 0.34408443581708803, + 0.2591399113481738, + 0.28773589139751987, + 0.3431789817257652, + 0.31069407878000627, + 0.2764236308188328, + 0.16710708725872062, + 0.12940126203416144, + 0.22887864968833974, + 0.15938377201903436, + 0.17952660115159708, + 0.2002822235507584, + 0.19808103898541188, + 0.28952764203067793, + 0.23801218273423927, + 0.2578933378959106, + 0.3120624252220208, + 0.2687807891363912, + 0.3476789325269097, + 0.24668219824483997, + 0.37710095015012224, + 0.2520662472990974, + 0.312964073772974, + 0.41231928058084755, + 0.21788901051660878, + 0.24939662739599974, + 0.12599683074364237, + 0.18464346404489307, + 0.3841808446624781, + 0.29030678039720076, + 0.2995438007066852, + 0.17939031607873465, + 0.24179051185190334, + 0.2930768835359673, + 0.17822499276390788, + 0.2244347160033565, + 0.278361993889009, + 0.28954394930657623, + 0.23713122363178787, + 0.2580340176882157, + 0.2731345130777016, + 0.3575211780390219, + 0.3416962139435254, + 0.35750539110363516, + 0.20715421329635472, + 0.35686815672970396, + 0.19362451773490155, + 0.18799223972641935, + null, + 0.1575408947164885, + 0.41475166594638113, + 0.3851219532405604, + 0.24604050944999864, + 0.3482036113620076, + 0.27170745929973517, + 0.25570809846045556, + null, + null, + 0.1780680467755505, + 0.3057736139282136, + null, + 0.3704281115225499, + 0.3217458810449532, + 0.2081779849684391, + 0.23233962905595182, + 0.4022961695134643, + 0.3374873057804688, + 0.0578017225012016, + null, + 0.2602335679807441, + 0.28568897057912657, + null, + 0.2389792847847334, + null, + 0.29781043364145093, + 0.3094653248535126, + 0.2626602584759255, + 0.2892000729736897, + null, + 0.16620187386356036, + 0.19090967333664136, + 0.16603781063457815, + null, + 0.29226839513125213, + null, + 0.30471342523304945, + null, + 0.19967363030009777, + 0.3534576672821739, + null, + 0.26285440517616093, + 0.19623029990509874, + null, + null, + 0.2061580630351126, + 0.2625559203082271, + 0.17522004905600394, + 0.2688531082378312, + null, + 0.30881861588973175, + null, + 0.3340089690840448, + 0.26204265813707533, + null, + 0.2873208801560381, + 0.3205956068997277, + 0.28114448817835547, + null, + 0.37080457422733387, + 0.19137571326504482, + null, + 0.28261205622886126, + 0.32862833649550144, + 0.21820507754154916, + null, + null, + null, + null, + 0.1826969163122049, + null, + null, + 0.2069620052808391, + 0.17660314852450584, + 0.2755521760468594, + null, + 0.3149380246678198, + 0.21111698870645337, + null, + null, + 0.30943276786951096, + 0.2888064647649539, + 0.26963664333321785, + 0.36638495760221745, + 0.3033716152001214, + 0.3661479922205464, + 0.31444772297975765, + 0.24569997208777292, + 0.2915389104789456, + 0.1343996876889881, + 0.3097809747500719, + 0.3539187356779667, + null, + 0.49641339798086226, + 0.31492220140217314, + 0.21061545658520311, + 0.25132314914043785, + 0.4408828589262406, + 0.20300752539775616, + 0.15489345941764485, + 0.31115047389558614, + 0.3927879647338218, + 0.15040047234787704, + 0.3284975028035623, + 0.289247120198511, + null, + 0.21702478505250725, + 0.1837466888975376, + 0.17671950741902304, + null, + 0.351218541070434, + 0.14676822581361346, + null, + 0.29200692626705566, + null, + 0.18140455223544272, + 0.32442012058932557, + 0.20265759240444708, + 0.22768048760330303, + 0.19485147838158104, + 0.2671115206314757, + 0.32526740164869017, + null, + null, + 0.23140176320149422, + 0.24385384180839567, + null, + 0.27212205475254475, + null, + 0.2730186582735051, + 0.296220948200394, + 0.26274474954226046, + 0.25929488559472685, + null, + null, + null, + null, + 0.1660468194564466, + null, + 0.20766425737187477, + 0.27837048801596287, + null, + null, + 0.2352094581633351, + 0.2932207323190851, + 0.34935982531164783, + null, + 0.2953849505772673, + 0.3579391944345852, + 0.17919170917183735, + null, + 0.24204667689747525, + null, + null, + 0.2430233603253886, + null, + 0.15980949769765537, + null, + null, + 0.2765393521391135, + 0.15768382290473365, + 0.3556514248329953, + 0.2858829191143867, + 0.2794524195752934, + null, + 0.21022799728449937, + 0.28158538262741445, + 0.2235674023791936, + 0.457006322938668, + 0.19241166346507768, + 0.4608887146723263, + 0.31649025269490316, + 0.15946528476058866, + 0.17770006943371475, + 0.35971039287890627, + 0.3361356662196308, + 0.20792051887224655, + 0.3824403212825274, + 0.3022362162246344, + 0.2128745961816383, + 0.251923604192159, + 0.24426876211610093, + 0.2437121515487553, + 0.3408263773074796, + 0.351011851808554, + 0.3403166025324261, + 0.37412723942975795, + 0.4390228734122012, + 0.20934262302153028, + 0.4684319755976841, + 0.25630468003464457, + 0.27380377751927815, + 0.3761531464321588, + 0.3901891264975901, + 0.34222859266503597, + 0.12987502579621366, + 0.24838025418701282, + 0.44340211084978787, + 0.2642115826340411, + 0.39626078189778746, + 0.21846509492138955, + 0.3185436212540015, + 0.16092298588452933, + 0.2788502410346191, + 0.3578998613532574, + 0.33258577149632884, + 0.3690793016414916, + 0.11717600652892875, + 0.14429087596670684, + 0.10675453565227826, + 0.21719074621169243, + 0.25907879470571465, + 0.24459920375937816, + 0.2746334014737521, + 0.20280888331172425, + 0.25182570501252316, + 0.2340034759217676, + 0.2395572240441139, + 0.17351919048436673, + 0.17588099697903525, + 0.09889062183274638, + 0.2413423914402218, + 0.26332912879495607, + 0.29812633168694597, + 0.23573131809637576, + 0.26318770022504717, + 0.4387297563818994, + 0.325130562614356, + 0.37606756336822245, + 0.2566626480584447, + 0.19603701265458637, + 0.25226350175930873, + 0.29596982810991956, + 0.14050764477756517, + 0.10325414761350112, + 0.23185103924913522, + 0.37202338740559027, + 0.28816319834934795, + 0.29189151591552476, + 0.3345253403937387, + 0.23645024867182063, + 0.2811391516933547, + 0.3631428298415666, + 0.30594232536221544, + 0.3399219823646086, + 0.40389738157397814, + 0.24873690389104097, + 0.298715820373561, + 0.4003888082732214, + 0.24644459705635074, + 0.2867895572205362, + 0.22531729994345895, + 0.4015449867466742, + 0.37594166931023726, + 0.27191540445977125, + 0.34116424297195347, + 0.3818452499795697, + 0.27909240409331054, + 0.35823267680426624, + 0.1904725672698323, + 0.1703603101665049, + 0.20244260005346085, + 0.26790020713279966, + 0.2682492048756823, + 0.38859559637640706, + 0.40374234877227055, + 0.18992361268842056, + 0.2670930547550399, + 0.25954330660094704, + 0.29124683858812084, + 0.3417971016718244, + 0.3261526896619893, + 0.21291483221738441, + 0.2991003721712464, + 0.28691882906679933, + 0.31385312157266043, + 0.1332597877085922, + 0.35962936477968516, + 0.2342612107797588, + 0.2950436141542155, + 0.3240533034439094, + 0.1774925068772185, + 0.3560305622052914, + 0.16087988200468398, + 0.2176860577124854, + 0.1483171600778371, + 0.40416439997277687, + 0.2447967967214106, + 0.1630757547896433, + 0.35194627658221944, + 0.39962779724391745, + null, + 0.36428355926088263, + 0.4536878701090341, + 0.25260655494430634, + 0.3763686617459942, + 0.33362249516303083, + 0.13791439619324336, + 0.19830393794152645, + 0.402160924921325, + 0.31430231288599125, + 0.22312443198106777, + null, + 0.23503072911319778, + 0.19323550435143155, + 0.19029960395557521, + 0.23485107381463488, + 0.22820384844741728, + 0.1905915996039282, + 0.19145257218598838, + 0.4021892012478252, + 0.0866579756699162, + 0.21291560752140698, + 0.3122036750544187, + 0.11596560152857521, + 0.1974285929115107, + 0.312160984077855, + 0.40227158718107664, + 0.21420039862387183, + 0.49878075002363037, + 0.47453002272802347, + 0.46980162304658607, + 0.3849361569293733, + 0.29828272413904267, + 0.25222886091121116, + 0.15845366330741079, + 0.23091158360089406, + 0.20735798604330558, + 0.3447045172487265, + 0.44136085317685525, + 0.28591846788335096, + 0.24179722705619583, + 0.3317459398416249, + 0.2654410255139518, + 0.36853197014510053, + 0.25549252794556343, + 0.24207947117668782, + 0.26564539915865537, + 0.231865129951112, + 0.13353134365165778, + 0.22241708143858716, + 0.23499226203559725, + 0.38584779395702307, + 0.2163385500710034, + 0.20404769510659154, + 0.23362414357149314, + 0.13119659600455114, + 0.16606439692070382, + 0.5076207686326115, + 0.40896571781339247, + 0.2921607738163306, + 0.4812270504425864, + 0.3772388126518459, + 0.21267007816540826, + 0.41449646580824995, + 0.3322342097261504, + 0.5005844605219155, + 0.2527176649242696, + 0.233217785676201, + 0.2374958906940056, + 0.2545323810001678, + 0.24372382256554534, + 0.23680782983990753, + 0.2746673151385342, + 0.3520699189579317, + 0.5156252200474738, + null, + 0.41799911902860637, + 0.36416083199542126, + 0.4891526412055102, + 0.4398551034224803, + 0.41824665953372325, + 0.24753962482899358, + 0.34471327528822576, + null, + null, + 0.37228260183822215, + 0.402681544806616, + null, + 0.15619376926665116, + 0.35804718321307716, + 0.09878787170962232, + 0.26032357629313796, + 0.22480426025041483, + 0.2412327465037967, + 0.30233659739450336, + null, + 0.3767433030198332, + 0.37006811136943996, + null, + 0.3402255131322309, + null, + 0.48563078440538526, + 0.2875643511887843, + 0.2489104784331828, + 0.2712848347961047, + null, + 0.4255906206253179, + 0.21585748855341674, + 0.367365338689496, + null, + 0.37368869978911695, + null, + 0.39836103180380866, + null, + 0.29446453968644287, + 0.3133090983082097, + null, + 0.41342917266427576, + 0.3676220939721567, + null, + null, + 0.379608518879261, + 0.2996237887433423, + 0.14823356147104802, + 0.319841100370895, + null, + 0.408626952763674, + null, + 0.25435036950959566, + 0.3717336049043922, + null, + 0.2006799214534338, + 0.4095175856566227, + 0.2887116873620019, + null, + 0.2721226569338983, + 0.4670854610310641, + null, + 0.3291059904553073, + 0.39999621700317656, + 0.3842526434049694, + null, + null, + null, + null, + 0.3323110789402276, + null, + null, + 0.47076062036712224, + 0.46195782445032374, + 0.40563668252618373, + null, + 0.26130164872362327, + 0.3744632021657229, + null, + null, + 0.3193902084112763, + 0.08159752910151158, + 0.3469162415881775, + 0.22442099044763086, + 0.3003040753214369, + 0.3577053005928684, + 0.11528589688975525, + 0.22350496470044562, + 0.18200121603373656, + 0.3014552382994906, + 0.42354657746172863, + 0.22844411421005015, + null, + 0.1749799811821536, + 0.4304200059593712, + 0.3330875152801128, + 0.3915246838732168, + 0.2036157652870963, + 0.3629009060063294, + 0.37339900887376026, + 0.26054077317762786, + 0.2905600147000313, + 0.360437505424274, + 0.3795721457847274, + 0.3089643950955283, + null, + 0.35573234561095335, + 0.4072430567992227, + 0.3098905343473311, + null, + 0.4109221627520168, + 0.23289408841694947, + null, + 0.3531340538441746, + null, + 0.4110297739442294, + 0.4750230029825383, + 0.27812472913491204, + 0.42091449897353456, + 0.40195087731647494, + 0.4566032286196652, + 0.4782667336955374, + null, + null, + 0.3676260011107296, + 0.4973679498273914, + null, + 0.2896612820658603, + null, + 0.2807409653150641, + 0.42412878462844755, + 0.42973263041342824, + 0.3743905569281724, + null, + null, + null, + null, + 0.2760663952556666, + null, + 0.36886386200567434, + 0.5025201812869421, + null, + null, + 0.4291399032569786, + 0.16756154825815422, + 0.34350513244610864, + null, + 0.3536932463105654, + 0.33267666198600626, + 0.4038823388870099, + null, + 0.3405901909452432, + null, + null, + 0.37093210446152075, + null, + 0.3080531186233483, + null, + null, + 0.40712143440536364, + 0.45110128522836745, + 0.5147499464328104, + 0.49674208519358176, + 0.3391830320129355, + null, + 0.4169534554863921 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.8940397350993378, + 0.9046052631578947, + 0.8896797153024911, + 0.8670886075949367, + 0.9172185430463576, + 0.8915254237288136, + 0.8661971830985915, + 0.8819444444444444, + 0.8942307692307693, + 0.8829787234042553, + 0.8821548821548821, + 0.8745762711864407, + 0.8798701298701299, + 0.90625, + 0.8966666666666666, + 0.881578947368421, + 0.8706293706293706, + 0.8782051282051282, + 0.9044585987261147, + 0.8856209150326797, + 0.8884892086330936, + 0.9064516129032258, + 0.8733766233766234, + 0.8773234200743495, + 0.8683274021352313, + 0.9013157894736842, + 0.9102564102564102, + 0.8983050847457628, + 0.902027027027027, + 0.8689655172413793, + 0.8458904109589042, + 0.8964401294498382, + 0.8694029850746269, + 0.8966565349544073, + 0.8681318681318682, + 0.889589905362776, + 0.8587786259541985, + 0.8934169278996865, + 0.8706293706293706, + 0.9093959731543624, + 0.8976109215017065, + 0.8969072164948454, + 0.8740458015267175, + 0.8592592592592593, + 0.8852459016393442, + 0.8817567567567568, + 0.8819188191881919, + 0.8458781362007168, + 0.8404255319148937, + 0.8979591836734694, + 0.8686131386861314, + 0.8860294117647058, + 0.8712121212121212, + 0.8843537414965986, + 0.8544776119402985, + 0.8877551020408163, + 0.8685121107266436, + 0.9006622516556292, + 0.8767123287671232, + 0.8697183098591549, + 0.8827586206896552, + 0.8918032786885246, + 0.8874172185430463, + 0.8805460750853242, + 0.8791946308724832, + 0.9082278481012658, + 0.8918918918918919, + 0.8825503355704698, + 0.8771331058020477, + 0.8954248366013072, + 0.8896797153024911, + 0.8870967741935484, + 0.8956228956228957, + 0.8973509933774835, + 0.8888888888888888, + 0.8915254237288136, + 0.8778135048231511, + 0.8963210702341137, + 0.9, + 0.9081632653061225, + 0.8807947019867549, + 0.9082278481012658, + 0.9143835616438356, + 0.883495145631068, + 0.8912280701754386, + 0.9071207430340558, + 0.9041095890410958, + 0.9084967320261438, + 0.9035714285714286, + 0.9082278481012658, + 0.89198606271777, + 0.8900343642611683, + 0.8982456140350877, + 0.8877551020408163, + 0.875, + 0.8888888888888888, + 0.8983050847457628, + 0.8927335640138409, + 0.9006622516556292, + 0.9082278481012658, + 0.8666666666666667, + 0.8852459016393442, + 0.8825622775800712, + 0.8977635782747604, + 0.89419795221843, + 0.9033333333333333, + 0.8576779026217228, + 0.8996655518394648, + 0.9069767441860465, + 0.8885135135135135, + 0.8964285714285715, + 0.8691275167785235, + 0.8863636363636364, + 0.8779661016949153, + 0.891156462585034, + 0.8885135135135135, + 0.8862068965517241, + 0.875, + 0.8736462093862816, + 0.8873720136518771, + 0.8821428571428571, + 0.8949152542372881, + 0.8676470588235294, + 0.9078947368421053, + 0.854251012145749, + 0.9919354838709677, + 0.9134948096885813, + 0.8885017421602788, + 0.9053030303030303, + 0.9029850746268657, + 0.8804347826086957, + 0.903448275862069, + 0.8989547038327527, + 0.8872180451127819, + 0.8842105263157894, + 0.8795986622073578, + 1.0, + 0.8915254237288136, + 0.8731343283582089, + 0.9111842105263158, + 0.8832116788321168, + 0.8656716417910447, + 0.9007352941176471, + 0.8862068965517241, + 0.887719298245614, + 0.8587786259541985, + 0.8582677165354331, + 0.8888888888888888, + 0.8382352941176471, + 0.901060070671378, + 0.8481481481481481, + 0.8971631205673759, + 0.8827838827838828, + 0.8928571428571429, + 0.9, + 0.872852233676976, + 0.8787878787878788, + 0.9061371841155235, + 0.8904593639575972, + 0.8795620437956204, + 0.8712121212121212, + 0.8819444444444444, + 0.8871473354231975, + 0.8954703832752613, + 0.8867924528301887, + 0.8920863309352518, + 0.8961937716262975, + 0.9205298013245033, + 0.8716981132075472, + 0.8701754385964913, + 0.8608058608058609, + 0.8771331058020477, + 0.8909774436090225, + 0.8938356164383562, + 0.8846153846153846, + 0.8793774319066148, + 0.894927536231884, + 0.8815331010452961, + 0.8860294117647058, + 0.8846153846153846, + 0.8815331010452961, + 0.8656716417910447, + 0.8560885608856088, + 0.8817204301075269, + 0.8802816901408451, + 0.8929889298892989, + 0.8904109589041096, + 0.8956228956228957, + 0.8817204301075269, + 0.898360655737705, + 0.8888888888888888, + 0.8823529411764706, + 0.8857142857142857, + 0.8591549295774648, + 0.8606271777003485, + 0.872791519434629, + 0.8786764705882353, + 0.8888888888888888, + 0.9052287581699346, + 0.8933333333333333, + 1.0, + 0.9046052631578947, + 0.8813559322033898, + 0.8970099667774086, + 0.9054054054054054, + 0.9035369774919614, + 0.8798586572438163, + 0.8585526315789473, + 1.0, + 1.0, + 0.9207920792079208, + 0.8961038961038961, + 0.9923076923076923, + 0.8827586206896552, + 0.8351648351648352, + 0.8503649635036497, + 0.8771331058020477, + 0.890728476821192, + 0.8689655172413793, + 0.8859060402684564, + 1.0, + 0.8975155279503105, + 0.8762886597938144, + 1.0, + 0.82421875, + 0.9929577464788732, + 0.8472727272727273, + 0.887719298245614, + 0.8732394366197183, + 0.8855218855218855, + 0.9782608695652174, + 0.8913043478260869, + 0.8802816901408451, + 0.8982456140350877, + 0.984251968503937, + 0.8451178451178452, + 0.9923076923076923, + 0.8863636363636364, + 0.9915254237288136, + 0.8833333333333333, + 0.8859060402684564, + 0.9854014598540146, + 0.8661710037174721, + 0.8786885245901639, + 1.0, + 1.0, + 0.8900709219858156, + 0.8986928104575164, + 0.8856209150326797, + 0.8720538720538721, + 1.0, + 0.872791519434629, + 0.9767441860465116, + 0.8636363636363636, + 0.8865979381443299, + 0.9930555555555556, + 0.9, + 0.8681672025723473, + 0.8836363636363637, + 0.9855072463768116, + 0.8642857142857143, + 0.8795986622073578, + 1.0, + 0.8918032786885246, + 0.8701754385964913, + 0.8813559322033898, + 0.9925373134328358, + 1.0, + 1.0, + 1.0, + 0.8704318936877077, + 1.0, + 1.0, + 0.8877551020408163, + 0.8723404255319149, + 0.8846153846153846, + 1.0, + 0.8737201365187713, + 0.8835616438356164, + 1.0, + 0.9923076923076923, + 0.89198606271777, + 0.8623188405797102, + 0.8729096989966555, + 0.8851351351351351, + 0.8873720136518771, + 0.8673469387755102, + 0.8805460750853242, + 0.8858131487889274, + 0.8677966101694915, + 0.8647686832740213, + 0.8853503184713376, + 0.8856088560885609, + 1.0, + 0.8842105263157894, + 0.8783783783783784, + 0.9, + 0.8888888888888888, + 0.8809523809523809, + 0.9027777777777778, + 0.887719298245614, + 0.8620689655172413, + 0.8819672131147541, + 0.875, + 0.8695652173913043, + 0.8737201365187713, + 0.991869918699187, + 0.8906752411575563, + 0.8817567567567568, + 0.8910256410256411, + 1.0, + 0.8753993610223643, + 0.8709677419354839, + 1.0, + 0.8625429553264605, + 1.0, + 0.8910256410256411, + 0.8784722222222222, + 0.8910891089108911, + 0.8623188405797102, + 0.8945578231292517, + 0.863013698630137, + 0.8636363636363636, + 0.9929078014184397, + 1.0, + 0.8758169934640523, + 0.8707482993197279, + 0.9776119402985075, + 0.8628158844765343, + 0.9923076923076923, + 0.8870431893687708, + 0.8825503355704698, + 0.8581818181818182, + 0.8837209302325582, + 0.9928571428571429, + 1.0, + 1.0, + 1.0, + 0.8698630136986302, + 1.0, + 0.8858131487889274, + 0.9015873015873016, + 1.0, + 1.0, + 0.8908450704225352, + 0.8737541528239202, + 0.8777777777777778, + 0.992, + 0.886986301369863, + 0.8922558922558923, + 0.8546099290780141, + 0.9838709677419355, + 0.9044585987261147, + 1.0, + 1.0, + 0.8976109215017065, + 1.0, + 0.8996763754045307, + 1.0, + 1.0, + 0.8687258687258688, + 0.8863636363636364, + 0.8865979381443299, + 0.8694158075601375, + 0.8802816901408451, + 1.0, + 0.8874172185430463, + 0.849624060150376, + 0.856140350877193, + 0.8442028985507246, + 0.8464285714285714, + 0.9113924050632911, + 0.8612099644128114, + 0.8397212543554007, + 0.8264150943396227, + 0.856655290102389, + 0.8561151079136691, + 0.8512110726643599, + 0.8447653429602888, + 0.8527397260273972, + 0.8595890410958904, + 0.8475836431226765, + 0.8351648351648352, + 0.8380281690140845, + 0.8296296296296296, + 0.8535714285714285, + 0.8771331058020477, + 0.8566433566433567, + 0.8551236749116607, + 0.8862876254180602, + 0.8609022556390977, + 0.85, + 0.8727272727272727, + 0.8675496688741722, + 0.8815331010452961, + 0.8566308243727598, + 0.8300395256916996, + 0.8666666666666667, + 0.8713235294117647, + 0.8467741935483871, + 0.8865248226950354, + 0.8608058608058609, + 0.8687943262411347, + 0.8406374501992032, + 0.8368794326241135, + 0.8566433566433567, + 0.8453237410071942, + 0.8204225352112676, + 0.864963503649635, + 0.8621908127208481, + 0.8470588235294118, + 0.845360824742268, + 0.8597122302158273, + 0.8233082706766918, + 0.8664122137404581, + 0.8452830188679246, + 0.8461538461538461, + 0.8433734939759037, + 0.8587360594795539, + 0.8409090909090909, + 0.8612099644128114, + 0.8308270676691729, + 0.8513011152416357, + 0.8741007194244604, + 0.8602150537634409, + 0.8220640569395018, + 0.8369565217391305, + 0.8501742160278746, + 0.8689138576779026, + 0.8873720136518771, + 0.8644688644688645, + 0.851985559566787, + 0.8639705882352942, + 0.8678571428571429, + 0.803088803088803, + 0.8517110266159695, + 0.8315412186379928, + 0.8941605839416058, + 0.855072463768116, + 0.864406779661017, + 0.8530465949820788, + 0.8387096774193549, + 0.8442028985507246, + 0.8344827586206897, + 0.8480565371024735, + 0.8489208633093526, + 0.8415094339622642, + 0.8477508650519031, + 0.8493150684931506, + 0.8517110266159695, + 0.8592592592592593, + 0.8475177304964538, + 0.8529411764705882, + 0.851145038167939, + 0.8645833333333334, + 0.8248175182481752, + 0.8315018315018315, + 0.8466898954703833, + 0.8220640569395018, + 0.8517241379310345, + 0.8401486988847584, + 0.8210526315789474, + 0.8387096774193549, + 0.8408304498269896, + 0.8587360594795539, + 0.8527397260273972, + 0.8387096774193549, + 0.8352490421455939, + 0.8284671532846716, + 0.8586956521739131, + 0.8714285714285714, + 0.8514492753623188, + 0.8611111111111112, + 0.8452830188679246, + 0.8629629629629629, + 0.856140350877193, + 0.8434163701067615, + 0.8415492957746479, + 0.8539325842696629, + 0.8509933774834437, + 0.8494623655913979, + 0.8637992831541219, + 0.85, + 0.8445229681978799, + 0.823076923076923, + 0.8221343873517787, + 0.8228782287822878, + 0.8631578947368421, + 0.8387096774193549, + 0.8095238095238095, + 0.8798586572438163, + 0.8401486988847584, + 1.0, + 0.8581560283687943, + 0.8428571428571429, + 0.8652482269503546, + 0.8181818181818182, + 0.851985559566787, + 0.8706293706293706, + 0.8480565371024735, + 0.8609022556390977, + 0.8571428571428571, + 0.8282442748091603, + 0.9906542056074766, + 0.8885017421602788, + 0.8198529411764706, + 0.8846153846153846, + 0.8885017421602788, + 0.8513011152416357, + 0.8333333333333334, + 0.8461538461538461, + 0.851063829787234, + 0.8327137546468402, + 0.8283582089552238, + 0.8202247191011236, + 0.8231046931407943, + 0.8168498168498168, + 0.7884615384615384, + 0.8302583025830258, + 0.8458781362007168, + 0.8708487084870848, + 0.8528301886792453, + 0.8655172413793103, + 0.8717948717948718, + 0.868421052631579, + 0.8581081081081081, + 0.825925925925926, + 0.822463768115942, + 0.8278388278388278, + 0.8737864077669902, + 0.8601398601398601, + 0.8202247191011236, + 0.8543307086614174, + 0.8493150684931506, + 0.8853046594982079, + 0.8204081632653061, + 0.8550185873605948, + 0.8369565217391305, + 0.837037037037037, + 0.8410852713178295, + 0.8480565371024735, + 0.8560885608856088, + 0.8292682926829268, + 0.8546099290780141, + 0.8607142857142858, + 0.8636363636363636, + 0.8503649635036497, + 0.8222996515679443, + 0.8401486988847584, + 0.8382352941176471, + 0.8424908424908425, + 0.8767605633802817, + 0.8518518518518519, + 0.8550185873605948, + 0.8525179856115108, + 0.8345070422535211, + 0.8571428571428571, + 0.8487972508591065, + 0.8523985239852399, + 0.872791519434629, + 0.8523985239852399, + 0.8315412186379928, + 0.841726618705036, + 0.8415492957746479, + 0.8192307692307692, + 0.8870967741935484, + 0.8737201365187713, + 1.0, + 0.8865248226950354, + 0.8720538720538721, + 0.8171641791044776, + 0.8498168498168498, + 0.8904109589041096, + 0.8639455782312925, + 0.8409893992932862, + 0.9920634920634921, + 0.9916666666666667, + 0.9090909090909091, + 0.8925081433224755, + 0.992, + 0.856140350877193, + 0.8373702422145328, + 0.8307692307692308, + 0.8532423208191127, + 0.8571428571428571, + 0.8434163701067615, + 0.8621908127208481, + 0.9921875, + 0.8709677419354839, + 0.8481481481481481, + 1.0, + 0.8246268656716418, + 0.992, + 0.8203389830508474, + 0.8718861209964412, + 0.8648648648648649, + 0.8526315789473684, + 0.9925925925925926, + 0.8534798534798534, + 0.8526315789473684, + 0.8382352941176471, + 0.9922480620155039, + 0.8362989323843416, + 0.981651376146789, + 0.8515901060070671, + 0.9841269841269841, + 0.8314606741573034, + 0.850909090909091, + 1.0, + 0.8363636363636363, + 0.8458781362007168, + 0.9851851851851852, + 0.968, + 0.8793103448275862, + 0.8657243816254417, + 0.868421052631579, + 0.8451178451178452, + 0.9915254237288136, + 0.8326848249027238, + 1.0, + 0.8475177304964538, + 0.8544776119402985, + 1.0, + 0.8191881918819188, + 0.8461538461538461, + 0.8364312267657993, + 0.9922480620155039, + 0.8478260869565217, + 0.8767605633802817, + 0.9747899159663865, + 0.863481228668942, + 0.8695652173913043, + 0.8673835125448028, + 0.9779411764705882, + 1.0, + 1.0, + 1.0, + 0.8788927335640139, + 1.0, + 0.9908256880733946, + 0.8546099290780141, + 0.8809523809523809, + 0.8716216216216216, + 0.9922480620155039, + 0.8811188811188811, + 0.8675496688741722, + 1.0, + 0.976, + 0.8754325259515571, + 0.872852233676976, + 0.850909090909091, + 0.8712871287128713, + 0.8631578947368421, + 0.8540145985401459, + 0.8125, + 0.8673469387755102, + 0.8603773584905661, + 0.8603773584905661, + 0.8779661016949153, + 0.8439716312056738, + 1.0, + 0.8645833333333334, + 0.8637992831541219, + 0.8673469387755102, + 0.8596491228070176, + 0.8671328671328671, + 0.8531468531468531, + 0.839041095890411, + 0.8461538461538461, + 0.8624161073825504, + 0.8445229681978799, + 0.8333333333333334, + 0.8327272727272728, + 0.984251968503937, + 0.8642857142857143, + 0.8785714285714286, + 0.8933333333333333, + 0.991869918699187, + 0.8615916955017301, + 0.8528428093645485, + 0.9754098360655737, + 0.8557377049180328, + 1.0, + 0.8803986710963455, + 0.8647686832740213, + 0.875, + 0.8274509803921568, + 0.8561643835616438, + 0.8470149253731343, + 0.8212927756653993, + 1.0, + 0.991869918699187, + 0.8591549295774648, + 0.8279569892473119, + 1.0, + 0.8505338078291815, + 1.0, + 0.8430034129692833, + 0.8477508650519031, + 0.8566433566433567, + 0.8529411764705882, + 0.9781021897810219, + 1.0, + 1.0, + 1.0, + 0.8450704225352113, + 1.0, + 0.8423076923076923, + 0.8787878787878788, + 1.0, + 0.9769230769230769, + 0.8637992831541219, + 0.8666666666666667, + 0.8428571428571429, + 1.0, + 0.8422939068100358, + 0.867595818815331, + 0.8668941979522184, + 0.9813084112149533, + 0.8896103896103896, + 1.0, + 1.0, + 0.8909090909090909, + 0.9916666666666667, + 0.8528301886792453, + 0.9908256880733946, + 1.0, + 0.8523985239852399, + 0.8546712802768166, + 0.875, + 0.8636363636363636, + 0.8440677966101695, + 1.0, + 0.8542372881355932 + ] + }, + "contact_maps": { + "mean_pearson": 0.2705776457604379, + "valid_tracks": 28, + "per_track_pearson": [ + 0.2565871853463348, + 0.26475013520147417, + 0.32355318542019207, + 0.22031812285701244, + 0.1765931685872806, + 0.22520017346747073, + 0.2786013983012811, + 0.2173355156675883, + 0.358238746918938, + 0.3151970322071808, + 0.21633427276216197, + 0.2806867402714056, + 0.3098554056766012, + 0.19558727736396359, + 0.25555666956856987, + 0.40077209402598113, + 0.07996429766819213, + 0.30247366660524944, + 0.2745366451070634, + 0.29553767708807394, + 0.23164770506556373, + 0.2429706254059481, + 0.23876520375562726, + 0.313104288816912, + 0.3213625983857425, + 0.3299080756519726, + 0.29664865128586604, + 0.35408752281261324 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.3898467116858455, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..3a5de1f6c47a2374ac8573663279644d5f3665f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S018/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5538157472017403, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5865355151069438, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.2303778898517866, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7565102879241177, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3773237835414827, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.21456757102599244, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.29875151449827153, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.20056296822880718, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.1908762737131009, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22817931310143777, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5850371065043171, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.2131857876397151, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.28050545853335335, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/README.md new file mode 100644 index 0000000000000000000000000000000000000000..c89a9ea0922bd77f6caa53453addb0fa3a4a4183 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/README.md @@ -0,0 +1,103 @@ +# SeqFunc scale13-small after 1 complete epochs with transcript-direction junction decoding with expanded usage-site selection + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **41,020 training examples in 10,255 optimizer updates**, 1 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +176.9 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The only change in usage34 is the sparse usage-position selection rule. It returns the 4,096 +central-window bases with the largest maximum raw splice logit across the four site types, +sorted by coordinate. Conditional usage values still come from the same 128-bp pooled head. +The previous rule reused the central subset of 512 full-window candidates per splice type. +All weights, junction candidates, other track computations and perturbation inference are +unchanged. Both entry points use this one frozen model; no coordinates or targets are supplied +at inference. This rule was selected for a full validation trial after a fixed 256-training-window +diagnostic (seed 3217), with no fitted thresholds or development-label coefficients. + +The earlier 4,000-example model did not benefit from the same larger budget; this experiment +revisits the fixed rule after a complete training epoch. Greater site recall can also expose +incorrect conditional predictions, so no validation improvement is assumed. The known-position +training diagnostic is never used at inference and is not a performance upper bound. Training +provenance and compute are inherited unchanged; this candidate adds no optimizer updates. +Native/recovery equivalence and exact agreement with the parent's unaffected outputs precede +the full 200-window validation. Perturbation outputs may be reused only after all relevant code, +weights and runtime hashes match, then every development variant metric is rescored. + +## Measured validation results + +Family-mean Pearson: **0.394309** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.539431 | +| dnase | 0.630721 | +| procap | 0.492437 | +| cage | 0.256694 | +| rna_seq | 0.517268 | +| chip_tf | 0.415247 | +| chip_histone | 0.438092 | +| splice_sites | 0.047388 | +| splice_usage | 0.442241 | +| junctions | 0.287302 | +| contact_maps | 0.270578 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5538 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5865 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.2304 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7565 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3773 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2146 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2988 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.2006 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1909 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2282 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5850 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2132 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2805 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..27a22789771d40ee686665a36ffba4b14f310deb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/package_manifest.json @@ -0,0 +1,51 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "files": { + "README.md": "de867cc145bfecd4650411d8144e436a87421d157c591bbd931f84a6a4bcba76", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "decoding_policy.json": "536c326d6752057fa48b9ad38c7120de5c9e6362068a233d999166f09f9dbc75", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "0595a0feea648c3d64e4428515828b4eb404a55ed1e2dfb27e2ff73d126e03c8", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "68c3a544a9703c39999bde688d39dd2d020a76b6c7cae72f25d1d65720a6569b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/make_usage_selection_candidate.py": "f687c04439ca3a68b6b2fd135078efe92550b3ffc0c8ef4cbf1c1a0698a7b074", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/usage34-study.json": "51aa61605277424e2482669a6aefb289c4b581d0c6fdf3aef995e51161806e97", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "53c826d3593df4090e6832f77df2d21043c67e7f7fb22a5a2f51ef8dcbefc865", + "usage_selection_policy.json": "51aa61605277424e2482669a6aefb289c4b581d0c6fdf3aef995e51161806e97", + "validation_tracks.json": "73b7045b3f64d968460187a526ecbff99c0e8542d049e9f668c8fd3229b502d2", + "validation_variants.json": "0b3e36ff456063327edeaff045da9126c9b1f1f0b26d323ed76c0fa6696ec5b4" + }, + "usage_selection_policy": "central4096_max_raw_logit" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/usage34-study.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/usage34-study.json new file mode 100644 index 0000000000000000000000000000000000000000..83c0d49a0311e20ebec0fab260a9f20f0b133a5d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/training_history/usage34-study.json @@ -0,0 +1,55 @@ +{ + "created_utc": "2026-10-25T02:26:23.204684+00:00", + "candidate": "usage34", + "parent_package": "/workspace/research/experiments/scale13-small-epoch01/package", + "parent_evaluation": "/workspace/research/experiments/scale13-small-epoch01", + "parent_native_check": "/workspace/research/studies/scale13/native-small-epoch01", + "parent_weights_sha256": "0595a0feea648c3d64e4428515828b4eb404a55ed1e2dfb27e2ff73d126e03c8", + "parent_predict_sha256": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "output": "/workspace/research/experiments/usage34", + "policy": "central4096_max_raw_logit", + "change": "Only usage position selection: full4096central positions ranked by max raw splice logit; conditional head, other outputs and both-entrypoint weights unchanged. No added training.", + "training_diagnostic": "/workspace/research/analysis/scale13-small-usage-discovery-train256.json", + "training_diagnostic_sha256": "3902f5ca897c06aef68bca94abbf1d4b6d34b423777bb3bf97cbbd4e5dd16865", + "training_evidence": { + "training_windows": 256, + "seed": 3217, + "policy_results": { + "current_global512_per_type": { + "mean_pearson": 0.4218272982468924, + "position_recall": 0.35608867675689254, + "positive_observation_recall": 0.4792380748608904, + "positive_asinh_target_mass_recall": 0.5198967478509209 + }, + "central4096_max_raw_logit": { + "mean_pearson": 0.4882935722456697, + "position_recall": 0.8279218740187151, + "positive_observation_recall": 0.9061336794240705, + "positive_asinh_target_mass_recall": 0.9335667731077516 + }, + "known_positions_oracle": { + "mean_pearson": 0.4640852573367183, + "position_recall": 1.0, + "positive_observation_recall": 1.0, + "positive_asinh_target_mass_recall": 1.0 + } + }, + "caveat": "Training windows only, no validation claim. Known-position readout bypasses discovery but is not an upper bound: it can expose false-positive values at negative queried positions, so its Pearson can be lower than a sparse policy." + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "source_sha256": { + "make_usage_selection_candidate.py": "f687c04439ca3a68b6b2fd135078efe92550b3ffc0c8ef4cbf1c1a0698a7b074", + "run_usage_selection_study.py": "298acc5632d8a7e48a20758608bf700ee3ca1ebaa20d4cab641f5020cd6c0323", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d" + }, + "full_validation_required": { + "tracks": 200, + "variant_queries": 2697, + "variant_metrics": 13 + }, + "variant_reuse_policy": "Only byte-identical weights, inference code and runtime allow reuse of all parent predictions; run score_variants.py again. No prediction cache is packaged.", + "published_choice": "Promote only a fully evaluated reward improvement; report all variant anchors.", + "existing_selection_pools_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..3a5de1f6c47a2374ac8573663279644d5f3665f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S019/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5538157472017403, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5865355151069438, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.2303778898517866, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7565102879241177, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3773237835414827, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.21456757102599244, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.29875151449827153, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.20056296822880718, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.1908762737131009, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22817931310143777, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5850371065043171, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.2131857876397151, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.28050545853335335, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..6fbb36253ad0b167bb556f6cb97159c8042ea6f4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch02/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..155ef4138af508a739d6af8d7cd283308fb7b4a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/dataset.py @@ -0,0 +1,293 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S020/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/README.md new file mode 100644 index 0000000000000000000000000000000000000000..d7990b7cb00ae75958290e9462560b16b34a4840 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/README.md @@ -0,0 +1,85 @@ +# SeqFunc scale13-large after 2 complete epochs with transcript-direction junction decoding + +One model with **215,626,632 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 16 transformer +blocks of width 1024 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-64 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. At the saved small96k/large78k boundaries, an optional training-only byte cache was enabled after exact256example and integrated16example tensor/ID checks. Source compressed bytes are verified on every cache hit, with read-only source fallback. Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. The cache is disposable and is disabled for validation/public inference. 215,626,632 parameters; local width 64, trunk width 1024, 16 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **82,040 training examples in 20,510 optimizer updates**, 2 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 8 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +406.8 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +## Measured validation results + +Family-mean Pearson: **0.423123** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.550818 | +| dnase | 0.643804 | +| procap | 0.515571 | +| cage | 0.294635 | +| rna_seq | 0.563978 | +| chip_tf | 0.438595 | +| chip_histone | 0.465365 | +| splice_sites | 0.052671 | +| splice_usage | 0.492826 | +| junctions | 0.344249 | +| contact_maps | 0.291845 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4376 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5624 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0747 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7788 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.2841 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2319 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2426 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.2212 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1753 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2234 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5278 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1948 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2585 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..f7bff0af8269a526030c1976fe4492dcb5523316 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-large-epoch02/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..deac57723b460edad7df2cb4d41be9afc254acec --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/package_manifest.json @@ -0,0 +1,57 @@ +{ + "parameters": 215626632, + "source_checkpoint_sha256": "b8df049ea8b87305b46c68f7481624c6740b16c5dffe4de60b7751693abccdcd", + "files": { + "README.md": "cf615ac54aeacb6672cf9af04712b96fead0b8c7b0dc45e1f2639ba3d8a049a5", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "6e7c605f2683881b9f59571f042cb7df93a3a0f224d8f2cb4e37be0a123cdb27", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "4616441df1c88ea6a9f911a4226fea581c2c610e90ba583a34705e961b06d735", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "1eb6eafaaec3f950708cabd7dc43a3ed22e67be602d7c036fe0fbbe0fbbff802", + "training_assets_manifest.json": "7e4a33e58edeb9c25fb59ec5b1f3b65ff2b1ca83abb9c3fc0770ec4f57af06c9", + "decoding_policy.json": "57c706f91fe86109b5afd7222bc90f881199bdb03b7bb75b8fcc41e6aa42d5cb", + "validation_tracks.json": "eeec21adce86c14356ed29a5bc0e76e84f847bceba41c0faefa8e3a8d6793d4d", + "validation_variants.json": "f806730a18850e289cb95020db0ac6f69d4203c8ab53c2ef31845fd798a36f28" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..dc0f0b20833fc99253b7dcf6023fd5e7a61b1128 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/predict.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8e314c842cca039da1d2be8da7fff6a146178651 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_assets_manifest.json @@ -0,0 +1,21 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3de85396b2b503e0e87c323a7d09b18f3cf7cec1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 96 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 96 +gradient_checkpointing: true +head_hidden_size: 768 +hidden_dim: 1024 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 64 +mlp_expansion: 4.0 +num_attention_heads: 16 +num_downsampling_blocks: 7 +num_sequence_blocks: 16 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..957817db1b07b99c360c7874bbdc5cb259862957 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/training_provenance.json @@ -0,0 +1,119 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 96, + "channel_growth": 96, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 1024, + "num_sequence_blocks": 16, + "num_attention_heads": 16, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 768, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 64, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 2, + "offset": 0, + "global_step": 82040, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-large", + "config": "/workspace/research/studies/scale13/scale13-large.yaml", + "resume": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 8, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 24410.930656648998, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 20510, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..4672e51d441ded5054e72e5659eb6c848c400d46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5508175701864864, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6725485809637025, + 0.6938591542429478, + 0.4396543301492332, + 0.6039888039408337, + 0.45330679908464655, + 0.452304678330736, + 0.46459774408719307, + 0.5455054635476084, + 0.5454905760035799, + 0.6031798328867224, + 0.5497800554527652, + 0.5282979798309383, + 0.37831950866760417, + 0.5039839851629664, + 0.649949762566685, + 0.47456100641919724, + 0.5265875037732498, + 0.49306201452539855, + 0.5268935331888892, + 0.5092282729058893, + 0.4795635688956239, + 0.44802006201555883, + 0.5189514796144958, + 0.4972057921476662, + 0.49248547705699564, + 0.47472929609985204, + 0.4524106931994684, + 0.4730757434141925, + 0.48105430832058843, + 0.5437407018189762, + 0.455912311875739, + 0.49175693270244925, + 0.44458971951473286, + 0.37572901260302555, + 0.3417645719204831, + 0.5507944029347663, + 0.4927234345071916, + 0.5944859327597268, + 0.47886205266571263, + 0.4678620303673778, + 0.5518581956384215, + 0.5539197166716443, + 0.4685211142704373, + 0.4489554665823478, + 0.5241181515629435, + 0.5327966353230609, + 0.41267426668313006, + 0.4623373496855866, + 0.4556228957770416, + 0.5007060794343151, + 0.469947729485487, + 0.4375901417450085, + 0.5292260814479574, + 0.5677391365288225, + 0.42979421297380915, + 0.6683274276292801, + 0.7779504372585914, + 0.5289403472864307, + 0.5770787165937096, + 0.5923154798429114, + 0.7442378586819066, + 0.46690640961770763, + 0.6811677475707104, + 0.63313164159428, + 0.614829978865024, + 0.4792106399453634, + 0.5442732762544809, + 0.6028126609113846, + 0.7580319070510647, + 0.6834292701101305, + 0.5023403863855894, + 0.4505547984240999, + 0.4291493835876423, + 0.5101470730570232, + 0.4784041326295036, + 0.4610422616182205, + 0.47164738024141223, + 0.5079732092787538, + 0.5480386292065261, + 0.46438077290448576, + 0.542198535163472, + 0.5682076568200858, + 0.4834315241144592, + 0.45795149925908235, + 0.4919227122897909, + 0.48122847717588857, + 0.4646365244831621, + 0.4534974960383764, + 0.42708882045862817, + 0.475205842959134, + 0.4798875985223147, + 0.4775738911966216, + 0.4615243209978485, + 0.476483114715519, + 0.5389981253307621, + 0.3633342580599288, + 0.41847351485898915, + 0.4139786179814907, + 0.46505134169631707, + 0.481713977391384, + 0.4319368664203194, + 0.4365122130933752, + 0.47789975399265366, + 0.4405369174248347, + 0.5146722056636625, + 0.5204418297827604, + 0.5104743567276164, + 0.49013544478093385, + 0.5140521942166637, + 0.5153289140362628, + 0.5733205133655617, + 0.497821983140426, + 0.49113533137401866, + 0.42009928449991163, + 0.4450349783171224, + 0.41153295549879587, + 0.4507421756933634, + 0.45831199858569405, + 0.4639168561706741, + 0.4478058177732037, + 0.6290371652720846, + 0.645195255096194, + 0.7121996046243245, + 0.6933274711302067, + 0.6847545386631672, + 0.5531888255413654, + 0.828826424248559, + 0.6987468662510653, + 0.7306534393112811, + 0.6612472096474065, + 0.70737121165242, + 0.5452569805537498, + 0.5707865958244974, + 0.5454557738569068, + 0.6919211546619944, + 0.7142451886472941, + 0.675159945163663, + 0.522003917636093, + 0.4081576727916938, + 0.7843307712200485, + 0.7560222802007661, + 0.7741247347816286, + 0.7660650516280781, + 0.5566388343921295, + 0.8236136813418772, + 0.7368255928736851, + 0.704460969867264, + 0.7032042943449196, + 0.8075243710880253, + 0.7771834593836172, + 0.5914181510095843, + 0.6031132799059722, + 0.6796074411252374, + 0.7133656374324564, + 0.7305337324101011, + 0.5799446024731287, + 0.7151233618571644, + 0.5785365743757914, + 0.579482374197702, + 0.6556087610875126, + 0.5659040107566451, + 0.7372534575159956, + 0.7494660696502671, + 0.6274823463252195, + 0.5703961333453694, + 0.5650955778018198, + 0.7096268775844697 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.643804329650295, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5474249866739545, + 0.9194275166055362, + 0.77999304251123, + 0.5076984327567804, + 0.5133838004921847, + 0.5248901613257457, + 0.9303493009999333, + 0.7091699708832515, + 0.4454941308655394, + 0.34461535977627233, + 0.8764116208763516, + 0.7015749213135025, + 0.4249461597889054, + 0.5077559127430502, + 0.49225895666306935, + 0.8044401459244839, + 0.7908071997430431, + 0.8486705867116632, + 0.8624990385865205, + 0.8906706974196699, + 0.6899645218505969, + 0.6073716112953184, + 0.8972995470725343, + 0.5233752767049965, + 0.5367619261589822, + 0.7778098049320084, + 0.460448636934175, + 0.6720304639584671, + 0.5263417866555391, + 0.5423503746539875, + 0.6260126600449674, + 0.46463721350020826, + 0.8407198271560635, + 0.6275074103212143, + 0.8788374902194741, + 0.8864605790793495, + 0.4476131062412663, + 0.8271128205840783, + 0.8731049131186138, + 0.8591534514853497, + 0.8269624501438579, + 0.6698516816176047, + 0.704137813793203, + 0.6320768081690851, + 0.5270616147329147, + 0.6384172457991911, + 0.7324704634644159, + 0.49464846718237393, + 0.5890447135578997, + 0.4038714022569794, + 0.38144581260878285, + 0.5607662139949364, + 0.49423919382341414, + 0.8236235569474729, + 0.4382347196053672, + 0.4014799756535367, + 0.45948257075165283, + 0.5078505419683421, + 0.47843328856725864, + 0.5196252164103747, + 0.44776375995658046, + 0.6070675817308698, + 0.493086365697982, + 0.4836089022198637, + 0.5116835847307487, + 0.5185658693938943, + 0.5510922703360066, + 0.475915535531656, + 0.3799997417339941, + 0.42607130926780684, + 0.5057857367356448, + 0.6416476259692164, + 0.5033181328996527, + 0.6267620058396965, + 0.36112358622747226, + 0.8061425595682058, + 0.8898464160214575, + 0.6546042910338021, + 0.4422959788215162, + 0.45738138842634035, + 0.5220976315828557, + 0.5059456266973075, + 0.46594266885096375, + 0.4807584016959385, + 0.569815988985323, + 0.4972373574387906, + 0.6128870803498077, + 0.5027808046038538, + 0.6382206466494165, + 0.4938779079652002, + 0.6324494133639111, + 0.41140160201302345, + 0.7253642198870154, + 0.47132631126873525, + 0.7182214431951538, + 0.503431987639337, + 0.7443371490998849, + 0.712581162862771, + 0.3749612477529617, + 0.38113100076027473, + 0.4743121030704761, + 0.3765390809177277, + 0.3594311954007516, + 0.7236080581321299, + 0.6908379729674006, + 0.41076260401887144, + 0.47591810618554997, + 0.5912402633496303, + 0.8449826606382923, + 0.5558399433688673, + 0.5515719209250285, + 0.5752390577963603, + 0.5046637632799514, + 0.440482238595063, + 0.7792078199027416, + 0.5687769603523108, + 0.4932604864395861, + 0.5524241264269137, + 0.6470137944414381, + 0.45515226526503594, + 0.527121313873056, + 0.5020937596533508, + 0.5554822256161508, + 0.5434723783702744, + 0.613763800742699, + 0.555196915060212, + 0.6399963085903619, + 0.6685268039733544, + 0.700051214819576, + 0.6237441147383521, + 0.46839202657700346, + 0.6094442282499823, + 0.6396730477417187, + 0.7896056401630128, + 0.4490618303370531, + 0.3355479984702816, + 0.5980491771234436, + 0.4800715549133919, + 0.6481210268944543, + 0.611417905526287, + 0.6207133774561985, + 0.5662485782112202, + 0.48827394765015947, + 0.7777265262788471, + 0.6196785606955951, + 0.7592509560265689, + 0.401263275353111, + 0.3798545321338966, + 0.4036717196582831, + 0.3649359608461915, + 0.8900881369794063, + 0.3291889688530219, + 0.891966669991504, + 0.7514930491381642, + 0.46829496180038055, + 0.36099220866018267, + 0.3671717642647664, + 0.3773786407508581, + 0.6518412120099063, + 0.7490001285278839, + 0.7427063038552006, + 0.6689929316123489, + 0.5713785067634661, + 0.4826050352813748, + 0.6870183621988553, + 0.6228647148812925, + 0.5156924320481826, + 0.6407281132859088, + 0.4632440734526387, + 0.49804104914800357, + 0.4167705393697866, + 0.8275630314173016, + 0.471515062392103, + 0.47180494318584215, + 0.6529029195391376, + 0.7004007485329836, + 0.521641928010101, + 0.3570912184757382, + 0.8731873774527373, + 0.38044456875992677, + 0.7701579012988385, + 0.5025271548347597, + 0.5997044313534103, + 0.7302521354663768, + 0.45766538847998367, + 0.5650310378332559, + 0.6044106552615609, + 0.8281825423166699, + 0.4206426805914778, + 0.6133834344139231, + 0.5669258506205567, + 0.70186681077153, + 0.8920017621075308, + 0.44377574709073736, + 0.600546987630373, + 0.49759296383679835, + 0.6334313078785493, + 0.7171874715767196, + 0.5693364280000316, + 0.4627594071424868, + 0.8669177635983671, + 0.8761427205976581, + 0.7482824454503542, + 0.8300002987505932, + 0.8409811093488848, + 0.8307089561512611, + 0.8462191440571271, + 0.37204129254953683, + 0.3929603210656972, + 0.5173903876510395, + 0.47084273810208216, + 0.46344906730664825, + 0.6086509552769429, + 0.858780945197092, + 0.8434036571958079, + 0.6826439513773335, + 0.8265311405890883, + 0.8231934872505048, + 0.8535619776164978, + 0.8994779982044326, + 0.5358220659323929, + 0.6128659977863298, + 0.6847352678494674, + 0.6408165643952635, + 0.8216687877357015, + 0.5851263227002256, + 0.9077881380575591, + 0.8808125618613353, + 0.8655301582008292, + 0.8979882615764463, + 0.8893651288655023, + 0.8732510570574005, + 0.8307164863568554, + 0.8548778359142626, + 0.9034134882733046, + 0.45783381966530823, + 0.9206947692101566, + 0.8733787515412976, + 0.8180692975005657, + 0.6100649189694595, + 0.868016010224041, + 0.5299105526068414, + 0.7476196645994065, + 0.8685810692258167, + 0.5991870993836372, + 0.5899868789650806, + 0.7010059937612251, + 0.5284914486683451, + 0.5538105899205048, + 0.7306291054355761, + 0.8835113744085366, + 0.7504683794653937, + 0.8155075641752006, + 0.633536714307484, + 0.8599723483874581, + 0.8038350751380426, + 0.8364095769187923, + 0.9004402983554816, + 0.8262841931284529, + 0.8919836848607038, + 0.8652275535675238, + 0.5430741513763093, + 0.8690534641892879, + 0.5806472286848392, + 0.549970886431142, + 0.870633142112893, + 0.8974524425727194, + 0.8736545377102786, + 0.6093499112180651, + 0.6397668122449021, + 0.6644924749787725, + 0.7939341364923892, + 0.8905345743401103, + 0.714983493756341, + 0.9046408147892443, + 0.9122346852467912, + 0.6242982879563955, + 0.34083603929984707, + 0.5316671921694531, + 0.8936377393022572, + 0.4589248421488829, + 0.6271847750392938, + 0.8806373316186062, + 0.8250968985077652, + 0.6391205906656948, + 0.8602487422239097, + 0.8900739860569505, + 0.8835256014069792, + 0.8724654269831086, + 0.8622009546998302, + 0.8868762088781035, + 0.8549605099777433, + 0.9226496571192325, + 0.7789448296390702, + 0.8315147840564411, + 0.9017688301132658, + 0.5681150145422939, + 0.5764848131074369, + 0.570926248991945, + 0.5458715330584502, + 0.9019670160310148, + 0.7976243707253913, + 0.8439029285778982, + 0.7948925255785256, + 0.765306659585681 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5155707474521652, + "valid_tracks": 12, + "per_track_pearson": [ + 0.47975714048888884, + 0.5306046125501384, + 0.49605576444273386, + 0.5178803886823833, + 0.5236110597345308, + 0.5494104711212304, + 0.47004563790648557, + 0.5262799202448234, + 0.4899116611000762, + 0.5238784065555439, + 0.5080753107608997, + 0.5713385958382486 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.2946354723906322, + "valid_tracks": 546, + "per_track_pearson": [ + 0.28672678505704224, + 0.3296707966425556, + 0.27985559276504895, + 0.18730731230484926, + 0.10216278265047439, + 0.2895416505177462, + 0.19092491385759586, + 0.19116602461422857, + 0.3052876876908939, + 0.279092906706615, + 0.31536352707390647, + 0.18977053174802566, + 0.20021530411782865, + 0.16567739896026454, + 0.1634796983902518, + 0.21781375308330347, + 0.2440969037356291, + 0.3271015262795436, + 0.33178307572123594, + 0.3139010418961939, + 0.2548331020451181, + 0.2617407860683316, + 0.3845272417226645, + 0.21687904110684433, + 0.2463249332252453, + 0.3168511443763484, + 0.28490812273947047, + 0.3083605236530734, + 0.31620765318502997, + 0.3158366982704072, + 0.21548326766021456, + 0.3512390222874708, + 0.23291829185474833, + 0.20007358989679078, + 0.35069305684579577, + 0.3001869072725927, + 0.356318825912892, + 0.38095098198195537, + 0.35519748793449785, + 0.3496825962759855, + 0.3918114994312157, + 0.32665332350650805, + 0.30453760447103106, + 0.28176602437655546, + 0.19474549882638023, + 0.26524147118673863, + 0.28007608125999783, + 0.31892398586918225, + 0.18223670150041973, + 0.23062851928269726, + 0.3385291417277975, + 0.32758664183533787, + 0.33346474607703885, + 0.24407407803252154, + 0.38098027921564326, + 0.37992906531735476, + 0.2227480775198433, + 0.3856131621397614, + 0.26907643162486666, + 0.3778360501572858, + 0.3540701060921155, + 0.242161396494465, + 0.3080517814550338, + 0.25165564037847366, + 0.3242813965330848, + 0.3697637540294675, + 0.3639154934625164, + 0.23906209820074825, + 0.17423978383145378, + 0.14944094428575633, + 0.2856069139724508, + 0.33100945975657015, + 0.2930177760153357, + 0.2878547974687151, + 0.3247075357936338, + 0.2714336098747711, + 0.3219003769351255, + 0.303728356630465, + 0.34090358988695774, + 0.33912901657747974, + 0.2952534437473288, + 0.17485173463398984, + 0.30688162579065753, + 0.39599231753258635, + 0.17411737694800555, + 0.30970159501284833, + 0.27524290301484533, + 0.3805938237978836, + 0.24146506612193042, + 0.32030887997054547, + 0.2563711678890319, + 0.2978790985641202, + 0.3047947956908475, + 0.15758569640341702, + 0.3286921858890318, + 0.2689415591125045, + 0.3178933691783753, + 0.31305306727979626, + 0.28949368807391884, + 0.18597145878792024, + 0.29650042792714904, + 0.3029552752634316, + 0.3284504730984128, + 0.3870523414008909, + 0.3505028680140998, + 0.33992842736159096, + 0.24779740270956943, + 0.3259589126121184, + 0.2608387968586332, + 0.26351630314762786, + 0.2748173196912187, + 0.2746469211998569, + 0.29778065221447264, + 0.3000378550513345, + 0.24948523291050612, + 0.3253285846374477, + 0.29711859479266006, + 0.3097314977538221, + 0.2912466144443527, + 0.3326627441525248, + 0.3347004781103221, + 0.3215129936118818, + 0.21760479497408095, + 0.25714559917498825, + 0.27630676433460793, + 0.3197434389087033, + 0.2990695963038115, + 0.32211342262576925, + 0.3957568965657054, + 0.3471479813983735, + 0.34045041342593435, + 0.2525271940208397, + 0.33880890941161423, + 0.3830465528435591, + 0.39464892486406045, + 0.29427818890051904, + 0.36200044296418754, + 0.29431940141358626, + 0.16819743429630848, + 0.342959860067821, + 0.35710077006864405, + 0.29382595916762494, + 0.33362728227365107, + 0.2624785203196354, + 0.2895721074261562, + 0.30339716627505464, + 0.30959448721280863, + 0.40069411785269965, + 0.2808996115296786, + 0.28197219428783155, + 0.37135186750813787, + 0.2757138577643582, + 0.32076587764274916, + 0.32513366298064217, + 0.30446330676093825, + 0.2957439444494071, + 0.3789509767240082, + 0.3971917154916838, + 0.27058177840741093, + 0.26124159909244926, + 0.22376421067969068, + 0.27612527505020884, + 0.28393955787113867, + 0.31453295862888986, + 0.2898662827330601, + 0.2556524715704, + 0.29118416480940335, + 0.23023085872803462, + 0.2028839461811824, + 0.3541209120900278, + 0.30184345433249915, + 0.24671336839786803, + 0.3028560294528493, + 0.40719679864637737, + 0.30422785829799315, + 0.1962349266078011, + 0.2061232384601952, + 0.2753740736065976, + 0.34141788028381204, + 0.34954336750376036, + 0.347518306541712, + 0.35773265047315916, + 0.20839337808860836, + 0.33904878331690663, + 0.29318527898919094, + 0.24112147263763933, + 0.2874605567636265, + 0.21057253413983937, + 0.39177949846471055, + 0.3128945410196215, + 0.2673413508862475, + 0.2006351475535074, + 0.19825142906070958, + 0.2794200381414733, + 0.30292962427791664, + 0.25753176231396807, + 0.29750943266150887, + 0.3025611839287319, + 0.22953258736506, + 0.299241670334327, + 0.1575792559320236, + 0.15490290910456683, + 0.14936718175462743, + 0.14296493854870576, + 0.18005205560281468, + 0.28307915855924104, + 0.28507837613569975, + 0.24934380766085842, + 0.1466400418629528, + 0.26577854319647265, + 0.3448027236757326, + 0.3380067151869226, + 0.3554625538642077, + 0.3253055613897369, + 0.30942593892820447, + 0.2660502726370398, + 0.3495629008863555, + 0.360007174827077, + 0.31577512694276405, + 0.3554016434692363, + 0.3234387525834511, + 0.25731126973816737, + 0.3255457083854542, + 0.31629021740031393, + 0.37900837803982573, + 0.36821659751597263, + 0.3158159915337297, + 0.3103890157184412, + 0.3371619495641452, + 0.3679593624354596, + 0.22962476217805447, + 0.23227266566659813, + 0.37429929420941727, + 0.2948637368353587, + 0.26197965044565796, + 0.28087382571851344, + 0.37750680383003127, + 0.2375517112214716, + 0.32858240088286395, + 0.32243529949257765, + 0.15781034527006602, + 0.23938172352854611, + 0.20505309771480787, + 0.3335220355743685, + 0.27709439492593446, + 0.32123824471153195, + 0.13671933949708648, + 0.36071409699507684, + 0.30568504266824553, + 0.3491112523997767, + 0.3269988125536662, + 0.23361413120780877, + 0.39447069670731627, + 0.18533323629618806, + 0.33767147567513184, + 0.278087894791067, + 0.38655067977051083, + 0.34711211794488833, + 0.36142144543935856, + 0.40583881981029685, + 0.28222961989224926, + 0.3169110452362138, + 0.160605430477822, + 0.3268266974739909, + 0.35267029792600335, + 0.24273435724669593, + 0.16917676315830882, + 0.16769584721343997, + 0.1976719343309492, + 0.29551845938498733, + 0.3647963924975467, + 0.19129516632273938, + 0.15511047659239016, + 0.2910130792656012, + 0.356874268894673, + 0.30141056209636613, + 0.21335059419378763, + 0.0985474078058261, + 0.31266102226776, + 0.2103407007901355, + 0.23229859101050357, + 0.30990832601793555, + 0.2801986603723117, + 0.32542278039973793, + 0.1943288069968905, + 0.21207835245319137, + 0.1961175122124233, + 0.18397839460623952, + 0.19999923006362366, + 0.2558124668165382, + 0.3340157529802396, + 0.3347852362048687, + 0.32019803174005484, + 0.26042469940305485, + 0.26682526032092313, + 0.38576147707779695, + 0.2109530484131725, + 0.2519885151228601, + 0.34174478756517895, + 0.2800597001809609, + 0.3084430696029626, + 0.331902479684685, + 0.32549509572601115, + 0.20796727243359092, + 0.3534762762736256, + 0.21888777055479466, + 0.19679900449992882, + 0.35066202962202203, + 0.32293561419998346, + 0.3859789164125656, + 0.3916498125326388, + 0.4193943686215211, + 0.3630408018744926, + 0.39702480529608725, + 0.32444090777669504, + 0.3183272851952026, + 0.3026555930639774, + 0.23920960837826163, + 0.2831946168336349, + 0.32086397378209264, + 0.3702965347346769, + 0.18242176972835447, + 0.23575709590290148, + 0.32489951658407823, + 0.3399345932521189, + 0.33221956334921837, + 0.24443848227778586, + 0.39012550622192016, + 0.3756167957456893, + 0.24157032867844985, + 0.3897560732068505, + 0.2658178607915555, + 0.39015092685213115, + 0.3556076247090235, + 0.2412408959227729, + 0.312201710020247, + 0.26758969961667595, + 0.34561062051171254, + 0.3686416062564716, + 0.35719922167679596, + 0.23121365162242852, + 0.18754825484179938, + 0.15527542633797992, + 0.2943153027183014, + 0.34794883606872196, + 0.30510759346441385, + 0.29273761145540167, + 0.3396205638878218, + 0.27472816234735886, + 0.3103545407657176, + 0.2995613185608683, + 0.34555343086458595, + 0.3505093337129622, + 0.3076930883206855, + 0.1862944884785014, + 0.3204827694765823, + 0.4086279948621076, + 0.18678366519946069, + 0.3327876030419585, + 0.3312861420495395, + 0.3978028764576378, + 0.23964430863655958, + 0.35277214199777446, + 0.26273084531044716, + 0.3116407467727909, + 0.31976291990661027, + 0.16830287024116555, + 0.32873084739319425, + 0.2694565984700752, + 0.3150541703098627, + 0.3380752682527882, + 0.2923280108059308, + 0.18687857528151094, + 0.3105736109656452, + 0.3211940249891973, + 0.3380492503504177, + 0.38860425220562644, + 0.35110874121613367, + 0.3534855421662178, + 0.25090873128449714, + 0.32015139791593095, + 0.26207881567055036, + 0.268517264120695, + 0.2753974272572599, + 0.278190708196841, + 0.310590649164248, + 0.3165917412065552, + 0.24981559962209546, + 0.3286916153137114, + 0.2963802990539575, + 0.3150721088266518, + 0.2924008544706517, + 0.3368183355878893, + 0.33576552794099546, + 0.32949714409336206, + 0.23407169145706286, + 0.274521865170487, + 0.2972762468250128, + 0.3410821430760967, + 0.30143342231802106, + 0.3454164567064265, + 0.40635462282989626, + 0.3830842717048503, + 0.34005432808145875, + 0.24892582889805615, + 0.3195722105173142, + 0.36594524657863864, + 0.4001950490023849, + 0.3049344738511856, + 0.3752087909691249, + 0.3056367175501965, + 0.19724491794944257, + 0.35263521181991525, + 0.3446006537381568, + 0.29838108700313054, + 0.35707676901529695, + 0.2611538402644633, + 0.28881342063486576, + 0.30407483102816374, + 0.30235863092417364, + 0.40209248770452344, + 0.33296954222963276, + 0.2854141038428617, + 0.37313939218463854, + 0.2835747687893561, + 0.3335891658916311, + 0.3301218222208315, + 0.3097987029434231, + 0.3176900500477562, + 0.3736073660668456, + 0.40843330066047856, + 0.27012831424138745, + 0.29236483412475534, + 0.226710893298352, + 0.26118266357306263, + 0.3072142014624737, + 0.3538626075627674, + 0.3232506118469962, + 0.25444943161403266, + 0.2985360570849813, + 0.27002845525341934, + 0.1888216280513926, + 0.39209031920305465, + 0.3140922613227027, + 0.27442622367402575, + 0.289720677965116, + 0.4147680611910818, + 0.31265011392261666, + 0.18196087076843478, + 0.22067146058070702, + 0.305162685276622, + 0.314674762788886, + 0.3701467184851825, + 0.3594367251523208, + 0.3689544309120413, + 0.2044085311013319, + 0.35681506956310893, + 0.33140506967188665, + 0.28013064297224716, + 0.30578795037823725, + 0.23405067619493491, + 0.38748490631388577, + 0.32083502740462194, + 0.28440842158532514, + 0.24243016501660408, + 0.17473884843601067, + 0.2326167970874224, + 0.3427120973208561, + 0.28972852676081084, + 0.32997287181938895, + 0.3300831555017158, + 0.2512258288329531, + 0.2848111350770402, + 0.14211390569322302, + 0.1194183842150956, + 0.12727106567137486, + 0.12152438899269612, + 0.1431860604353501, + 0.31244981895407925, + 0.2547872867724474, + 0.25221586903796833, + 0.1411812990411093, + 0.29255477667572893, + 0.36840217755926585, + 0.40049266996048444, + 0.4005858675443381, + 0.36183219450845144, + 0.35394231685469235, + 0.2859481435124677, + 0.37440990030975774, + 0.3746769166684995, + 0.3260920373333018, + 0.37449970242476893, + 0.3488068231493762, + 0.30844080903743737, + 0.36239736470332257, + 0.33407223756422305, + 0.4310013336129673, + 0.37963935740052834, + 0.34543879755069884, + 0.3251437088238416, + 0.32199960142151174, + 0.36282881627608765, + 0.19667271103656356, + 0.18930610805535314, + 0.4042340712202176, + 0.30694825432370215, + 0.2901362270406724, + 0.2975393058518045, + 0.40102477034772194, + 0.2846354799913105, + 0.3355511179097753, + 0.3250875120118377, + 0.1465948208463563, + 0.22349383250468466, + 0.19505719807670527, + 0.3539497742794725, + 0.2715700117883773, + 0.3348828645173264, + 0.14816215846912545, + 0.37084172956215655, + 0.3228403554963279, + 0.36830142996190335, + 0.3584587711569715, + 0.25955062746344354, + 0.44681327612114896, + 0.17779252402076384, + 0.3429707762892309, + 0.25563467438843834, + 0.3897239754878144, + 0.36817866251708664, + 0.40013258301083915, + 0.4098278446176172, + 0.30888381160402395, + 0.3209593665726041, + 0.14781127363265573, + 0.32456137489704623, + 0.36388377251554216, + 0.2892202918306872, + 0.15549143139951888, + 0.15541430116534444, + 0.18756730307976224, + 0.31528711218557726, + 0.37332723124645906, + 0.17422655966473183, + 0.13409086634020173 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.5639778974831469, + "valid_tracks": 613, + "per_track_pearson": [ + 0.5109350949448141, + 0.6505519694331835, + 0.5720389642703545, + 0.6107004605058506, + 0.6333071294585778, + 0.37968820496884725, + 0.4717709701092413, + 0.6128563891136971, + 0.59418152316386, + 0.6221296120105504, + 0.5579762140383758, + 0.5499603057513307, + 0.47700493228393087, + 0.5919556060213476, + 0.546477970921886, + 0.5728438434507763, + 0.6528477190685666, + 0.5634389860353276, + 0.5894144744765929, + 0.5514770892399534, + 0.5108785413547278, + 0.6470949368575937, + 0.5052723025475948, + 0.4817005103749829, + 0.4400091976388939, + 0.651933061926303, + 0.5357203066756376, + 0.591386839690261, + 0.5794722374459386, + 0.5967103211157847, + 0.5457302715608469, + 0.59049526917845, + 0.5840072356425255, + 0.5289255256336655, + 0.6198148905588682, + 0.5135104295521744, + 0.6060383011347538, + 0.6055028254513349, + 0.557310009186677, + 0.6010693497801058, + 0.6269568737273177, + 0.5598326470159002, + 0.6072734171198007, + 0.632511469300004, + 0.5312621149903087, + 0.45103751521693897, + 0.5146901653996323, + 0.6096235951975927, + 0.5561415886519864, + 0.6262260933713932, + 0.6173163614544455, + 0.6038756153832391, + 0.441522676074545, + 0.5395263016927445, + 0.4254606177607043, + 0.569587477976146, + 0.53850252185854, + 0.6272682650079564, + 0.6684386931207268, + 0.6906366161778639, + 0.6911438506940315, + 0.6302746351555006, + 0.6638229712373475, + 0.46762064405374465, + 0.6137968003168655, + 0.6133912010796452, + 0.6440046252363129, + 0.6688975520438251, + 0.6322251212629324, + 0.6849870662588865, + 0.6261935500005283, + 0.6363667387777792, + 0.5886153836191108, + 0.6136853951272032, + 0.6008482177407162, + 0.6055964502450868, + 0.6318036947207224, + 0.6474822504240706, + 0.5650427067151021, + 0.662784606207154, + 0.6157413414251764, + 0.4970659460321981, + 0.6688929029132468, + 0.6409104222727182, + 0.5645762941365156, + 0.6066423203784791, + 0.6499196530454774, + 0.5187392618681683, + 0.6092897494326, + 0.539161273363611, + 0.4912547905699714, + 0.6536091817798845, + 0.6546962345133877, + 0.6362144511775906, + 0.629491500035382, + 0.6554082159314923, + 0.6679785358505967, + 0.6406937299943376, + 0.5628087473066707, + 0.49623721705183266, + 0.448745473661707, + 0.45512982481690256, + 0.5176686067445738, + 0.6299838558798614, + 0.49982393971773265, + 0.521899702716382, + 0.6470552005994374, + 0.5639503086034665, + 0.6214757906336342, + 0.5428108313781329, + 0.5961915972425612, + 0.5939659567021159, + 0.5488770288500029, + 0.5640370571665855, + 0.6271669366781483, + 0.6030392884634311, + 0.6167492128444904, + 0.5560012705363737, + 0.5376119681905502, + 0.5898340683263381, + 0.6347845510839465, + 0.5481342838554042, + 0.5225527495487032, + 0.5779772702215702, + 0.663152901664529, + 0.4280837302644713, + 0.471376221018793, + 0.4350231293375722, + 0.4586260613828552, + 0.5192341566054599, + 0.6265867708230283, + 0.5115234457624858, + 0.5735611296946072, + 0.5725730207944766, + 0.5630354106562792, + 0.6185438266150542, + 0.6021314141283831, + 0.6311633662848798, + 0.5324135544613332, + 0.6084178789412088, + 0.5054797173061183, + 0.5660453413435032, + 0.6353967649337249, + 0.5377052090209015, + 0.44943626260713276, + 0.5093520722285875, + 0.5191271969527811, + 0.5436488779949205, + 0.508064658220323, + 0.5301984565470281, + 0.5122544120508421, + 0.526153051943528, + 0.6003736324496478, + 0.4689097159650308, + 0.5311358511575656, + 0.49013673966242866, + 0.5694343364665387, + 0.519009766090655, + 0.574887290716232, + 0.5762354186566829, + 0.6181677631398151, + 0.5833377748965779, + 0.5803334523328019, + 0.48327855356140115, + 0.48148707645184075, + 0.48199279620406216, + 0.4876894802034886, + 0.4837135637910851, + 0.54061322781481, + 0.5334714319628826, + 0.6290096315139555, + 0.5431596097367806, + 0.6393492297141942, + 0.5368677995219038, + 0.48354178147975796, + 0.519467210747405, + 0.5045195438913519, + 0.6218437796943843, + 0.432654094317517, + 0.44585674233006245, + 0.4259439666498099, + 0.5945167576389361, + 0.4530040873323442, + 0.5919203611778091, + 0.40114688942752624, + 0.5306016027414678, + 0.5964156794059216, + 0.38196233119331646, + 0.5110738148563791, + 0.5103846678918049, + 0.608008511467767, + 0.5656768640622456, + 0.53974760775828, + 0.502845041476574, + 0.5983024175802745, + 0.5251053980576514, + 0.5446044437012062, + 0.3195995440549957, + 0.5553284386745309, + 0.5592632424614484, + 0.4910503515573146, + 0.6115076091737451, + 0.6046686529258721, + 0.5790230775287607, + 0.6031856083257425, + 0.6068577149228844, + 0.5927979624629498, + 0.3041071537609384, + 0.5859076228563, + 0.5783579629027221, + 0.5146830333037025, + 0.5022516796121799, + 0.5607026877307791, + 0.5643771371986365, + 0.5209141875384042, + 0.6131292690182707, + 0.5477868569377146, + 0.48655282336388556, + 0.5878907620843543, + 0.6391502337334702, + 0.6077296214606019, + 0.6560547877245333, + 0.5673619903022489, + 0.566639834266163, + 0.5579473104385053, + 0.558823898715758, + 0.5522475935397892, + 0.5632588385241285, + 0.4473905519454379, + 0.6442217373762062, + 0.5250863368517937, + 0.5767686799517163, + 0.6192143058371037, + 0.42792615441577275, + 0.6614009705114715, + 0.40191669790991535, + 0.3636558670088556, + 0.6035501742016898, + 0.62534437706704, + 0.5963892650143984, + 0.6079723001786321, + 0.6378412621021939, + 0.5944083669123583, + 0.6210393298950964, + 0.31181852762117296, + 0.5349249800717636, + 0.5556194287655762, + 0.6107025041918575, + 0.5293599696462975, + 0.5344894439593073, + 0.482974544518433, + 0.5428366804021575, + 0.5222020911685199, + 0.5228090190763336, + 0.5790523567473884, + 0.5587705276349582, + 0.46611993085048203, + 0.6193908645405543, + 0.5583667804130027, + 0.6152636679299925, + 0.48355682792113525, + 0.5081671372392405, + 0.6421159631720749, + 0.6461863565425325, + 0.5694464359972164, + 0.6045243308835578, + 0.6479952469750798, + 0.4865359723147223, + 0.5895828865259186, + 0.5867819604045333, + 0.4065735678962702, + 0.5321476649438664, + 0.6936155984180529, + 0.6857848633717882, + 0.6818232425163461, + 0.6596306546157912, + 0.3739308966971314, + 0.4605338469381476, + 0.6557217984665763, + 0.6164211915269838, + 0.6409646945243762, + 0.5650070808581173, + 0.5124900060048335, + 0.5146030326845895, + 0.5912813291564185, + 0.5403996460818775, + 0.5596102516744759, + 0.6334659314117145, + 0.637789058205006, + 0.6846733713093864, + 0.5587390714163198, + 0.5328207390466462, + 0.6686924787509063, + 0.5328837586782985, + 0.4441748630738951, + 0.4191537745712638, + 0.6582923191335873, + 0.6934869102156591, + 0.6988029967315923, + 0.6169174199658589, + 0.6324384676036381, + 0.5388855762283217, + 0.6169421854916287, + 0.5260336031209841, + 0.6076394242909124, + 0.6418843073118963, + 0.6168335021367797, + 0.6743136827921259, + 0.6855355864816426, + 0.636671895456356, + 0.666751661009596, + 0.6444517227161871, + 0.6119264366318246, + 0.6396706594118543, + 0.5893957146999071, + 0.5069673408123251, + 0.5217372240027355, + 0.5576296131353233, + 0.6541941026015572, + 0.5649191706265627, + 0.6223521590462906, + 0.6709587071810021, + 0.6375767283412825, + 0.42690402036201713, + 0.5281042884411734, + 0.3705007103097171, + 0.5895022378353146, + 0.5622476926117861, + 0.6337208263122606, + 0.6787960077743725, + 0.6405081665286594, + 0.6652408972401115, + 0.6952554832519426, + 0.6982843004302988, + 0.47605787432955204, + 0.6798407274020376, + 0.6711542401286891, + 0.6842228538245475, + 0.7238904702968861, + 0.6950686296481352, + 0.7246547011290757, + 0.6512897581566301, + 0.673549885207944, + 0.6130792609177951, + 0.679661501893026, + 0.6498676111774117, + 0.657723679133599, + 0.6904866205869684, + 0.7001859587619489, + 0.5192805336912084, + 0.638228981770172, + 0.6034350187202938, + 0.5371990479223571, + 0.6971729946996222, + 0.6684997267350539, + 0.620807285089072, + 0.6263015332181204, + 0.6146038518191206, + 0.5151947139756563, + 0.6327738359644488, + 0.5370419649326573, + 0.5526984298669446, + 0.6773302989789168, + 0.6283480855887639, + 0.6245175906394425, + 0.5898532789374061, + 0.6269429323220121, + 0.6412065723905677, + 0.6144649423366508, + 0.5657098850192716, + 0.5455703394705469, + 0.43827572364746714, + 0.4258757802040381, + 0.47743848980149484, + 0.6887189834926071, + 0.48509903916843183, + 0.5057026324904568, + 0.688551212254764, + 0.5669119452506018, + 0.6068551946015843, + 0.54730689963218, + 0.6115125709330408, + 0.567386158013486, + 0.534558082462133, + 0.6310853356331697, + 0.7055071089484518, + 0.6236492641369835, + 0.6384247311468214, + 0.5625374656603876, + 0.5233887330589182, + 0.5795078668997452, + 0.687947822536973, + 0.5431626019729278, + 0.5522306304894536, + 0.5784827788496087, + 0.6710808797162812, + 0.39426903805566094, + 0.43968456734829575, + 0.43527100715861566, + 0.42283840590746247, + 0.5659741972007184, + 0.6643935436876063, + 0.5049527738563094, + 0.5788474212922241, + 0.6937762960918363, + 0.6187651155577237, + 0.6467868111767172, + 0.5923149315174934, + 0.6691085094503889, + 0.5394445008089579, + 0.6136293628218693, + 0.49750698565617263, + 0.5789494349347426, + 0.6440916508427363, + 0.5454671129068007, + 0.45596060633689756, + 0.4978165973068207, + 0.5723093743814743, + 0.6057937166200085, + 0.6456194840465254, + 0.5096676829620479, + 0.5166839367766024, + 0.5507961852596215, + 0.633926636140593, + 0.47139335536508026, + 0.5149373257325146, + 0.5319752583368457, + 0.6646044890782936, + 0.4994019045009679, + 0.6266155053368048, + 0.5861937026443053, + 0.6260803025698878, + 0.5993644733840032, + 0.6177763216580359, + 0.39316672671008535, + 0.4071021086017822, + 0.4024360338460831, + 0.40090962822440523, + 0.47463204390841846, + 0.5726187444908414, + 0.563270896779512, + 0.667916751566568, + 0.5956448940370013, + 0.673797977968841, + 0.5453391998457043, + 0.5492275388782475, + 0.544992291321001, + 0.5016939764430659, + 0.6699598995330686, + 0.39201757279274346, + 0.4952753266579273, + 0.4137071498941898, + 0.6576818630498081, + 0.4384138818795096, + 0.5585212735877094, + 0.40273234106024425, + 0.557189802637339, + 0.6209270892351835, + 0.43612734691005356, + 0.5362377990993848, + 0.48059039309071055, + 0.5946238937899098, + 0.6469284321423465, + 0.5386347585441045, + 0.5108343219189477, + 0.6169433333457317, + 0.6076121632057562, + 0.6048839429167775, + 0.306203157615982, + 0.6301361001717429, + 0.6272283531762857, + 0.4759665739730286, + 0.6752254605409749, + 0.6118863720954933, + 0.6286938856557494, + 0.6287980098165795, + 0.6024571462419198, + 0.5773015139098024, + 0.28189718707410677, + 0.6584224134887063, + 0.5408551206679969, + 0.6470135828492088, + 0.6264395782227541, + 0.6957519896210572, + 0.6474933285919303, + 0.5608196434741227, + 0.6563787036803413, + 0.6749031440034479, + 0.553084189859511, + 0.7004120881568638, + 0.6121574958971457, + 0.5686543537379007, + 0.6663674955429034, + 0.5423902805634657, + 0.5145595285895165, + 0.5047172606011632, + 0.5084615127463674, + 0.4635688732596927, + 0.5322056952858537, + 0.3756309194083391, + 0.6645905249259009, + 0.466729284387818, + 0.573999265187561, + 0.7045715514969332, + 0.41757771507329056, + 0.6857148462646389, + 0.3756534116592473, + 0.38049022599833343, + 0.6349394474917814, + 0.6051078715465865, + 0.5515305490550069, + 0.5642586685233527, + 0.6690020011032819, + 0.6563944773442512, + 0.6869897538273885, + 0.33211185475376925, + 0.5775096973774665, + 0.5642759353918068, + 0.6445546876923578, + 0.5744776228181354, + 0.5570660948892113, + 0.5422518216152382, + 0.5473593891139833, + 0.5686845331075191, + 0.5620547143329052, + 0.6225375983116563, + 0.5258436539954365, + 0.4186685696635757, + 0.6432328030936583, + 0.5866055001065874, + 0.6728057760027902, + 0.43219872596659314, + 0.40905590805904773, + 0.5905141705571294, + 0.6068121582678104, + 0.6049106900849067, + 0.6352373690248341, + 0.6860236467923896, + 0.43413483179153994, + 0.6336019312800197, + 0.6031715442989285, + 0.4097501585230974, + 0.5986323759626123, + 0.30409091617670403, + 0.48729712489659677, + 0.5524925104984724, + 0.47781584170024055, + 0.5486857091321189, + 0.4781967757768241, + 0.5908011186424393, + 0.6079426739413748, + 0.4934895725291597, + 0.4746517965785874, + 0.4765919356323327, + 0.4566962180526305, + 0.46142927939771955, + 0.44907773454125316, + 0.46403208338641866, + 0.4474916188168736, + 0.4291161419416344, + 0.39156689549375995, + 0.5802943413922134, + 0.496936838668669, + 0.5152531603419279, + 0.5033204487958279, + 0.5164238322014052, + 0.5118111877114434, + null, + 0.5490953901142889, + 0.5677590074793227, + null, + 0.4424830417067812, + 0.5915085182056551, + 0.45225730164570843, + 0.599259360606041, + 0.6282335266589536, + 0.59132601511416, + 0.5793817680675821, + 0.5777101307060013, + 0.5849790852894784, + 0.5547385320175041, + 0.5141324710515389, + 0.5135563614347081, + 0.46740802271974324, + 0.5639315557606528, + 0.45440272248345376, + 0.5396863792387354, + 0.515820274277287, + null, + 0.5536941398363125, + null, + null, + 0.527730194293968, + null, + 0.5575189283789851, + 0.5121728774883666, + null, + 0.571437604357735, + null, + null, + null, + null, + null, + null, + null, + 0.5536158742137897, + null, + null, + null, + null, + 0.46699067777283854, + null, + 0.5407518767954521, + null, + 0.5562593450103256, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.5656817892747467, + 0.434739282455659, + 0.5367486126942982, + 0.5589966439930935, + 0.49106541160685546, + 0.5877463494030846, + null, + 0.5785801654110883, + null, + 0.540326337799261, + null, + null, + 0.5767352639579582, + 0.5905156207084636, + 0.507585320677559, + 0.49415487173367634, + null, + null, + 0.5290687872583788, + null, + null, + 0.5461154663661217, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.5475887241069646, + 0.5290463583954681, + 0.5560177580248309, + 0.508641177828032, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.43859545145212453, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.23105878841519972, + 0.3292245244534398, + 0.3055731104442136, + 0.45154189870362915, + 0.38196289640968806, + 0.289570783922568, + 0.37550067126895365, + 0.42839325922285254, + 0.35397225792340503, + 0.3646575658846223, + 0.4029048337368651, + 0.3938961627447437, + 0.29154439497188245, + 0.42393153912259146, + 0.36374595362455586, + 0.36658176172075435, + 0.4070513446799974, + 0.33863071359859687, + 0.48516464123418607, + 0.373315444094235, + 0.3719336096601082, + 0.33882550372186526, + 0.31565526700543034, + 0.4974214432306272, + 0.34222892536519667, + 0.37630144907625696, + 0.3022002140937579, + 0.3721610845947479, + 0.3957475902246244, + 0.4266848123946114, + 0.40904168706106725, + 0.40310948713773564, + 0.48822997142078006, + 0.2861514788051406, + 0.6478700349020475, + 0.2871454542329867, + 0.2375721841449516, + 0.18191973528666225, + 0.32411928418638625, + 0.22573419474124623, + 0.27739864424405236, + 0.4624681545191342, + 0.3076741828977772, + 0.39887366293831705, + 0.29571637926851313, + 0.38779011563611154, + 0.295264523761715, + 0.2262484543412229, + 0.4663226896887725, + 0.43409921580849686, + 0.38489515220688325, + 0.367007421943556, + 0.3891892354157154, + 0.4165369311959297, + 0.41672712263725725, + 0.39159570842026165, + 0.3120125223440161, + 0.437920805482247, + 0.4509731294246166, + 0.44975780434780344, + 0.2989218582350674, + 0.2100108218645727, + 0.3957452433273108, + 0.4118934273235607, + 0.27182646037647196, + 0.34239397805662364, + 0.34518029039679976, + 0.45954584446204316, + 0.4720100312449903, + 0.423932521197424, + 0.3598763404275669, + 0.703218349679536, + 0.3462004389753675, + 0.36737952561471604, + 0.36033089920988703, + 0.42895053703127833, + 0.29816705671005916, + 0.6380805418255657, + 0.4016806191408404, + 0.4127427299467652, + 0.39694616560241125, + 0.4046810801364091, + 0.4734237621629732, + 0.40634788906320796, + 0.34031604388719766, + 0.30688214490830545, + 0.44480824628882104, + 0.4108212192121153, + 0.8005207555815019, + 0.4796101291535059, + 0.36965507712740203, + 0.3397549500949711, + 0.4035019120745471, + 0.3575035039785599, + 0.40312366247997555, + 0.5155235523061774, + 0.33745779055837966, + 0.3519760156130564, + 0.3640346048202948, + 0.30908217022355877, + 0.2476564611799725, + 0.37106638532659, + 0.33453926676688284, + 0.3243006363535454, + 0.3707637957354949, + 0.24450902282168233, + 0.30960064226586553, + 0.3210883444213977, + 0.38059626234221017, + 0.38593920052041775, + 0.2557604822556715, + 0.34570278268103527, + 0.32884177999383857, + 0.3646408229831533, + 0.3781824980200107, + 0.43860591949726246, + 0.38702366492980145, + 0.3970903159661532, + 0.37194797020383685, + 0.42256728699581564, + 0.3722677118906251, + 0.4217729457531058, + 0.3906517035280549, + 0.31519928409231596, + 0.3820826860836752, + 0.33534468268017115, + 0.3609240699278178, + 0.34287973076816897, + 0.36699136145814415, + 0.5286975822790672, + 0.41017032083202043, + 0.3597661639448057, + 0.4117786642742426, + 0.6150058373883449, + 0.3417341403869183, + 0.3891754208281937, + 0.38888110396994097, + 0.31320940443970957, + 0.34887871559890743, + 0.33863730435372236, + 0.3735458740480194, + 0.42985300413495003, + 0.5193274822597265, + 0.35881770742281804, + 0.39763361698552, + 0.48650082434088693, + 0.43614601446212814, + 0.29157989854696703, + 0.4130914908263941, + 0.45019728319331165, + 0.42193662964153905, + 0.3514568624143643, + 0.4280092188767209, + 0.473168062828687, + 0.4143605711607482, + 0.41005652096367634, + 0.3236283712629251, + 0.37150669728891583, + 0.38269174836718656, + 0.43073726586732414, + 0.41381441652682643, + 0.37232951838891004, + 0.3967132880279509, + 0.400252834286373, + 0.40363722907081695, + 0.3774849912012924, + 0.3675559779435796, + 0.4638859414102041, + 0.3080218336861559, + 0.40705296716625966, + 0.39872543956508105, + 0.3536837297822314, + 0.4823444924632065, + 0.3587284587815373, + 0.33621255728067856, + 0.42095657716242607, + 0.29049035428393744, + 0.43349870892392, + 0.442183670849033, + 0.3444546981559732, + 0.4154265694788857, + 0.39589173791326465, + 0.32986503470670686, + 0.39153359876029964, + 0.3437037065407244, + 0.4080461742432862, + 0.4253670978792581, + 0.3154679963522417, + 0.3434469489336121, + 0.3974050683457908, + 0.34236855955044193, + 0.352567141248151, + 0.41713862158335324, + 0.47865676236402505, + 0.34544034979207955, + 0.3488068470174136, + 0.36085006941078307, + 0.35928069183216876, + 0.2816356328179783, + 0.44899367079589436, + 0.37336417513401315, + 0.34463031361373947, + 0.3829660381916616, + 0.3348142344383743, + 0.4421037292640323, + 0.37031496399532354, + 0.4411170421539948, + 0.3700423603806786, + 0.32884900636572717, + 0.48857506089589353, + 0.35976357650365864, + 0.4178264119417101, + 0.3637006999659994, + 0.3980843707685886, + 0.8412639779223005, + 0.43406740430378155, + 0.3967193666786414, + 0.44444813308635767, + 0.45090231032832945, + 0.5442672774437143, + 0.6075574258780226, + 0.6194502193605942, + 0.4078245453815131, + 0.3806801161353683, + 0.47790756271713936, + 0.7280381981813757, + 0.3505812048546432, + 0.37387119472257685, + 0.5412194303497737, + 0.5127404289318697, + 0.5509807105470644, + 0.4340975726417999, + 0.3216793729133323, + 0.38402295409315934, + 0.5931441819691065, + 0.4294108714334172, + 0.346067307697504, + 0.6086266041280208, + 0.5460014027500604, + 0.4431024343930996, + 0.48632176225277807, + 0.7045715698944098, + 0.4709188841142148, + 0.49485059553219335, + 0.43029576956801774, + 0.7595112721351249, + 0.3209968241472183, + 0.6366149499990422, + 0.389590887403622, + 0.5423360990331656, + 0.4312063537177079, + 0.3878583827937086, + 0.6990108355108863, + 0.5036308558686219, + 0.32683223337413764, + 0.3709902775063252, + 0.643316472283122, + 0.373541708370313, + 0.43469118619671215, + 0.4642497651038166, + 0.36483857816591214, + 0.46450071060977716, + 0.4637095960015957, + 0.4120621229660001, + 0.38713894576587987, + 0.4703235312017183, + 0.34821131407903605, + 0.33486803989438074, + 0.6248228490353969, + 0.3769097787772535, + 0.44813107540263525, + 0.5563909474002, + 0.3737742969451316, + 0.6558562694126259, + 0.3354052893896984, + 0.34516835793255346, + 0.5529070598956581, + 0.4826126160619135, + 0.5780545425002654, + 0.724686204552247, + 0.6621909621020005, + 0.3968400031031409, + 0.4651451293522568, + 0.5513394045101014, + 0.34332826997648525, + 0.3316717614548791, + 0.5617005622540344, + 0.4993466983500576, + 0.7213679983465885, + 0.4761671371318698, + 0.3609686915809383, + 0.44401070433220874, + 0.3630703992026434, + 0.46589650679211525, + 0.7010341047059662, + 0.6208951088337725, + 0.5003017784191551, + 0.5233313110840369, + 0.4855954587105356, + 0.5093161969871658, + 0.7372317623835039, + 0.3077806122104056, + 0.44243414623540767, + 0.30494562995190233, + 0.5996625964368156, + 0.3679617705178451, + 0.7321078293621671, + 0.4088056655391343, + 0.42571845863458, + 0.6892730459844753, + 0.5077298443668745, + 0.3208468813848651, + 0.5621909727993146, + 0.39314151033452227, + 0.5086831580213621, + 0.49797759748767223, + 0.34245710445458016, + 0.5018767823923213, + 0.38890842721870644, + 0.4499628567564036, + 0.38736506458435077, + 0.34122719316757827, + 0.40074002275876897, + 0.5148392879292765, + 0.5755505404458712, + 0.6605704447600156, + 0.4119570995244423, + 0.3032843151848139, + 0.44374349533546825, + 0.45455015366188983, + 0.42787414426647635, + 0.4719197345897773, + 0.46353023530419674, + 0.3724247911077679, + 0.3321425391166289, + 0.5057189636942887, + 0.760732314976042, + 0.5225242661006166, + 0.5962880993140145, + 0.5875189666660527, + 0.7634279406744116, + 0.5444927414695916, + 0.3443703710575912, + 0.580873159619472, + 0.3758915253356853, + 0.5284158603564278, + 0.4560801867615683, + 0.3803802273938689, + 0.4899130645010388, + 0.7763498093027531, + 0.2463321773377713, + 0.4611221606553122, + 0.46479657067962044, + 0.618160670846377, + 0.38423448906038005, + 0.3154000356157817, + 0.41979911825997956, + 0.719760441624173, + 0.3858189579675735, + 0.4310238147324101, + 0.5026842077014049, + 0.45373112203400545, + 0.42295059533041973, + 0.4858993005489552, + 0.6085157330373265, + 0.45868660840390835, + 0.4037208745386321, + 0.5368802860769866, + 0.4464277834693783, + 0.565630342841787, + 0.528026281703482, + 0.5129625904810003, + 0.5379154987274615, + 0.32768492595268645, + 0.34675505870251194, + 0.5294633185625792, + 0.31805619941474406, + 0.6950809533530402, + 0.3138027031071948, + 0.5591995142403059, + 0.5572085320497853, + 0.6818889376497425, + 0.30098830423668804, + 0.35583125279669603, + 0.32841210632033985, + 0.37094866699162976, + 0.3211169497407397, + 0.30239396110011174, + 0.6159019909591854, + 0.37533522596071767, + 0.6769869139768412, + 0.5785414098987702, + 0.4418490499154866, + 0.5710059825814373, + 0.3761154746323818, + 0.5153807229018383, + 0.5112125080330243, + 0.47232198664747393, + 0.37638865153608964, + 0.4606716741081868, + 0.48382176321699916, + 0.5232984444488368, + 0.3875033136480341, + 0.3400612719432382, + 0.614715412386635, + 0.46731104150284974, + 0.6539947201021332, + 0.6260978448600065, + 0.5127894849379369, + 0.326294062616826, + 0.35048660892926264, + 0.35314873127736146, + 0.460190265547884, + 0.7433872757051415, + 0.4683454808506625, + 0.37401092466741226, + 0.7370428482069397, + 0.4261661739855277, + 0.6169800838858104, + 0.5290081448312807, + 0.41851064461572474, + 0.4810908609936368, + 0.4414036933724028, + 0.45587990073319296, + 0.5768712763105484, + 0.5327303395225078, + 0.5226433671634974, + 0.42563571838681696, + 0.4534993266234308, + 0.46750270728552, + 0.46726616931451753, + 0.7631465704612735, + 0.4697980440973136, + 0.4769736517489125, + 0.4242230032917939, + 0.3006683989840612, + 0.6623073725675662, + 0.4096281052898145, + 0.33552356870975336, + 0.5104026675630601, + 0.4268437893149159, + 0.32657162375237103, + 0.48311361152995663, + 0.4240789862181731, + 0.7724713152562618, + 0.3117544496751837, + 0.3857995005759684, + 0.3409501359853112, + 0.4260089506876892, + 0.37999381090472234, + 0.30873931920244124, + 0.39092462496449404, + 0.2769772546921074, + 0.4069688100360812, + 0.7337164283729305, + 0.3085284901038793, + 0.4009005606294847, + 0.49391286385686023, + 0.40853211280429397, + 0.3942216711060716, + 0.4715282619370363, + 0.35384363342063985, + 0.5096492905361321, + 0.507781950262926, + 0.25966283538655577, + 0.6269672867764277, + 0.6780862153449053, + 0.49806435397586835, + 0.47284047796246625, + 0.5351568146157308, + 0.41993067157676944, + 0.4451172688595311, + 0.5521955901513178, + 0.7755783640337112, + 0.3810040405167697, + 0.6029682065832525, + 0.3273353014588056, + 0.4526461135248951, + 0.46576921851228226, + 0.465018739101331, + 0.3893842989539477, + 0.3210973341058682, + 0.33163169881771837, + 0.3934371377563478, + 0.5647709533831227, + 0.45070701511362743, + 0.32466149552661405, + 0.3988717447477906, + 0.45473562527493605, + 0.29771771912964773, + 0.5359901006993341, + 0.465771833344879, + 0.3900688323553116, + 0.4353105771472244, + 0.3163644543025986, + 0.4169682535894346, + 0.48331153894395923, + 0.5783237605717417, + 0.42825844260054485, + 0.7763750830399956, + 0.347831857411961, + 0.5509997793074034, + 0.7368599375883711, + 0.6296894782702853, + 0.3087338435644968, + 0.48734260884245495, + 0.41372478076205105, + 0.34061770830139126, + 0.35477190609954995, + 0.36915463551435834, + 0.44732266636942003, + 0.492485646669664, + 0.3635847361595533, + 0.607537236578061, + 0.7584929846822877, + 0.41882947895739037, + 0.41229020830684754, + 0.422252836888014, + 0.4811310458385787, + 0.5008110498462961, + 0.4834968843034002, + 0.49168747255516165, + 0.3440931625137734, + 0.6491106790874448, + 0.4761072198483759, + 0.5431553312310565, + 0.7047717396568276, + 0.31669780056618607, + 0.5859472455223629, + 0.377992213485733, + 0.44077938541564077, + 0.3457721438278905, + 0.5081387901856623, + 0.48316553860726325, + 0.32354465817704237, + 0.41067451767988356, + 0.4088162262187123, + 0.5354969872744644, + 0.5835185412028852, + 0.4398301383632991, + 0.4453872666047085, + 0.402567590782507, + 0.7341364320946439, + 0.4456841028660847, + 0.5050985352638616, + 0.453002374558106, + 0.4564247480959181, + 0.5990281328771219, + 0.3356973337014439, + 0.5635263118027203, + 0.36396804769597757, + 0.5347422153606304, + 0.41009415250284204, + 0.33383793973697656, + 0.34634385622494324, + 0.4558716813350233, + 0.35745213219245286, + 0.4988992286002121, + 0.7250064764883911, + 0.7198171834150356, + 0.3886132495358546, + 0.3390439699693902, + 0.6958991325306269, + 0.44282758654227794, + 0.38088176273980356, + 0.5058759728583695, + 0.35278193974423444, + 0.5894826391129837, + 0.4267319306088831, + 0.5580138651334768, + 0.5708909996012059, + 0.41352730493246725, + 0.4991237447756671, + 0.6480646622767038, + 0.5102846126410702, + 0.34912943648781647, + 0.4059785788818007, + 0.2954861832678346, + 0.5295058467280109, + 0.34152241400139305, + 0.5075423681866637, + 0.4919570610804307, + 0.3719778707377333, + 0.717492725616639, + 0.32366950278191137, + 0.7235026353291111, + 0.42168596775687833, + 0.5291997623452801, + 0.7407873041298543, + 0.4633218096845937, + 0.5466298144822519, + 0.38448785442959, + 0.6972578196117194, + 0.393642361200517, + 0.31583419935783547, + 0.5450361947455302, + 0.5129242592466765, + 0.6327667508521698, + 0.5445608607155513, + 0.3802026730292499, + 0.47337122473077603, + 0.29652763466723064, + 0.43428134107529903, + 0.5623302677794667, + 0.5034279671213476, + 0.34469391188367676, + 0.5313997679813502, + 0.5787263825233959, + 0.5672672809535704, + 0.43072342703997324, + 0.3714342828049895, + 0.5015162571332732, + 0.4518480902721717, + 0.550130205965775, + 0.46416779183253276, + 0.3500714726076359, + 0.3965180995555998, + 0.49989488804571325, + 0.38729009628285216, + 0.7295616669500878, + 0.5104405537123121, + 0.44622969961273856, + 0.48835690696512235, + 0.389132812902414, + 0.43556472728556206, + 0.534804836711378, + 0.5455993826602465, + 0.7070345795451188, + 0.5293818980612477, + 0.5710108048249202, + 0.5171281940566134, + 0.48004415716985155, + 0.5702884918817713, + 0.39151316660920255, + 0.6261315102933558, + 0.5894924993865682, + 0.7210399262827115, + 0.5091526548484591, + 0.5388090193537689, + 0.5329781785905621, + 0.41052974224066474, + 0.5323612303804729, + 0.33118078516395666, + 0.675708114996505, + 0.474665442946885, + 0.7312404869936944, + 0.7530143556768779, + 0.45465831792087263, + 0.44013440108615076, + 0.4945636800263113, + 0.34754354304640767, + 0.48651453818863033, + 0.7529194462565281, + 0.48299892190367816, + 0.5045280933377408, + 0.5787605911109587, + 0.5080863269776177, + 0.343516551358324, + 0.6048202258282713, + 0.42460275722531643, + 0.3786170630994553, + 0.7291831426837102, + 0.4032783484823783, + 0.4240635120636742, + 0.38776502092155485, + 0.47805532015472435, + 0.5537144946454857, + 0.618642556742368, + 0.6740453758967159, + 0.6165558751006094, + 0.5838217785285321, + 0.4476614513237667, + 0.6034381092299794, + 0.5515734545653735, + 0.7761570192567446, + 0.7335570795117405, + 0.3397594363492785, + 0.3662264768606365, + 0.3412181258438909, + 0.5064594826711384, + 0.40618090235069587, + 0.7819321427484063, + 0.3687250634573978, + 0.41115554226069, + 0.5615087696534656, + 0.4759954178585042, + 0.54718799862932, + 0.6180307668187076, + 0.4555327723701162, + 0.6160780486633348, + 0.37160284085837686, + 0.6762164584851648, + 0.7758951414607679, + 0.7615572259696404, + 0.5831957524041113, + 0.42451409800077594, + 0.5500786367003427, + 0.7357142836365714, + 0.6846269089073347, + 0.619054588286125, + 0.4746210997796064, + 0.45462606174158676, + 0.48712877768280766, + 0.7323239174698702, + 0.4147545938959419, + 0.4239997954354564, + 0.3153541705422162, + 0.5923517160290585, + 0.3076589698023822, + 0.3966007429424123, + 0.670443638107442, + 0.5360411439653409, + 0.6808450993125099, + 0.5277893966654464, + 0.5133167075517647, + 0.6624245828794769, + 0.6273914316261735, + 0.5603477136314059, + 0.37543622390911446, + 0.6390240852032251, + 0.5145139987005924, + 0.39662005735519507, + 0.3471891695779746, + 0.4793457301919996, + 0.6354740488215074, + 0.5063065191743219, + 0.6071315039903835, + 0.566231750046268, + 0.420229021124035, + 0.3984317456369068, + 0.6824095200696145, + 0.49508354430813833, + 0.5988130446794662, + 0.7203826965798416, + 0.4116530961726458, + 0.36927593460951663, + 0.36681412498087357, + 0.3587772695236903, + 0.7092655434478716, + 0.7308743547615771, + 0.5609594396223183, + 0.41573565116939876, + 0.4536553688453035, + 0.37816418459566864, + 0.7756700190648879, + 0.3426013806937034, + 0.4757074358137515, + 0.3308569686678535, + 0.5013280893960823, + 0.36589905600225814, + 0.5596113739500858, + 0.47279235274663584, + 0.3412586300923215, + 0.6906102067277307, + 0.5016702697966504, + 0.4097697551750534, + 0.339474470869585, + 0.35086455107535364, + 0.3991701976585649, + 0.409838609770359, + 0.3641608858205321, + 0.3616651955093879, + 0.39356812403735786, + 0.385022258379619, + 0.40405801438557304, + 0.33247113321863386, + 0.4317376586368492, + 0.204570598092883, + 0.26908772539874853, + 0.3336399950733864, + 0.38479346118549, + 0.406107283696992, + 0.4202427171676007, + 0.3312925746608253, + 0.35154847596350036, + 0.3348419140557661, + 0.650077346573965, + 0.31993721379694173, + 0.3161535082970124, + 0.7690029832190454, + 0.3574056075829124, + 0.40798206983190655, + 0.336521653154848, + 0.4399079158531748, + 0.2861108073272883, + 0.3934680954154604, + 0.3490607473369076, + 0.4524677964530166, + 0.2650288996255517, + 0.4077604883616233, + 0.39556216274492745, + 0.4939544883337654, + 0.38064334415989265, + 0.46675726081313684, + 0.30255221850702846, + 0.42943822791141967, + 0.3806987373554409, + 0.3460293034903091, + 0.46615076150079476, + 0.3562627074802079, + 0.46389062429207023, + 0.377635457034115, + 0.36121656026475807, + 0.8049040020589004, + 0.300289229868319, + 0.4329808078488589, + 0.36126624147187986, + 0.39905645214877544, + 0.3194001218323511, + 0.378757082144552, + 0.44685237093562835, + 0.5559554752019765, + 0.33915742143139865, + 0.37916614613032257, + 0.28434307183070445, + 0.40431385394137487, + 0.4235566028931148, + 0.4141181333948507, + 0.453200542960377, + 0.35900671699649706, + 0.35463326905252945, + 0.45153064979728236, + 0.40315686482947255, + 0.43604719446407886, + 0.30171392325558044, + 0.3089571394154474, + 0.4659544265110956, + 0.445326516425831, + 0.39905147648941675, + 0.35386914729254065, + 0.3619204821686515, + 0.8268794122468643, + 0.3139174911250531, + 0.32712580222493953, + 0.3461928283083299, + 0.3206429828062955, + 0.32390837896899977, + 0.34311199890907224, + 0.30496278512812003, + 0.3483260795030572, + 0.3431769745237849, + 0.45858142933505, + 0.4197776537574333, + 0.4337465362163078, + 0.47463973298602247, + 0.439529454649476, + 0.3836347056698372, + 0.5538022803254916, + 0.4269889824610411, + 0.44016725167463605, + 0.4920453311716527, + 0.4451411283795456, + 0.3848817772208914, + 0.3218976678297823, + 0.3982085546664423, + 0.49822129041524904, + 0.49196764568936086, + 0.3487707514969559, + 0.37276584885161335, + 0.4027366210132903, + 0.44985208660049264, + 0.35597270950667204, + 0.39192543905096777, + 0.4760735461078504, + 0.5162942143568037, + 0.3419947059997619, + 0.42962755284719617, + 0.35004104122348867, + 0.3508759336092102, + 0.35265941342108803, + 0.7846775953637135, + 0.41317884329153703, + 0.4323509769695818, + 0.34653320294784257, + 0.3728321417202593, + 0.385078994520186, + 0.6958207875558015, + 0.42570797455080295, + 0.41000623395039454, + 0.44228274210618534, + 0.40975512349679766, + 0.4252736921500838, + 0.44162170504264425, + 0.4863745303760163, + 0.2944600728577789, + 0.3592717361520659, + 0.44761122909765216, + 0.4328503698538787, + 0.4036624334548859, + 0.3956118337240331, + 0.3935073761606069, + 0.33236646709169343, + 0.39077705782331995, + 0.39210863504653476, + 0.38397397933039057, + 0.612256649500866, + 0.3631962256609947, + 0.41699025028802805, + 0.33628970064646285, + 0.5636452363639678, + 0.3937774719868224, + 0.3666291129133509, + 0.42966859882522956, + 0.41839260483665813, + 0.43357270457031893, + 0.35674361343071315, + 0.5385406583008704, + 0.45357504158595346, + 0.3488556667758444, + 0.42465268152671265, + 0.43720187727851306, + 0.4603313726772673, + 0.3868612548542732, + 0.4662473401954372, + 0.3456014671915862, + 0.6523607781156416, + 0.3871320867276063, + 0.45633192645719933, + 0.4553106064513014, + 0.3967357130228331, + 0.40039216918853393, + 0.41923420784585025, + 0.33404090349578636, + 0.39312472585934205, + 0.591767446129981, + 0.3860369987497196, + 0.3676776262334549, + 0.510555998431506, + 0.4198194094545908, + 0.4067207948333864, + 0.6089014422482794, + 0.3937136466990174, + 0.4535280735657499, + 0.4063943723963365, + 0.38713226911471815, + 0.4795172808901258, + 0.3837453068368155, + 0.47557690950038367, + 0.406459610598692, + 0.46557636589223206, + 0.3649135847654316, + 0.4595407422961256, + 0.4188700103307459, + 0.35052879224783506, + 0.328384525177203, + 0.37138195929657897, + 0.42690411298568937, + 0.540498510016305, + 0.4531575943482299, + 0.44791003327965834, + 0.46807356623809454, + 0.38192874599888277, + 0.37338524331526024, + 0.4622593683989956, + 0.364782378845898, + 0.4017261067464438, + 0.47587608261156283, + 0.38488285337179545, + 0.4615381241250186, + 0.5027088928718352, + 0.38855982352762874, + 0.39484727715383566, + 0.51223841095554, + 0.40499102994831926, + 0.430131763411656, + 0.3378728278653394, + 0.47775908960961055, + 0.6014193833341357, + 0.7891938343347463, + 0.4204295965165994, + 0.3975980436694703, + 0.4208789575706433, + 0.37724615714889304, + 0.35563398980725197, + 0.37607170016955754, + 0.44274680067078265, + 0.43131309439835464, + 0.4049486959744955, + 0.44057186862673015, + 0.45742508832384615, + 0.47274928528187726, + 0.3600948785465294, + 0.35240091587214417, + 0.44425636209504077, + 0.4013785397894257, + 0.384839971656494, + 0.4046125241277243, + 0.3936091808062496, + 0.40097132625295173, + 0.30273549228425656, + 0.45403045389567215, + 0.43669706189442103, + 0.3896327797686529, + 0.3941987478209783, + 0.31610682346845653, + 0.4205667034810593, + 0.29613704044638967, + 0.5138426998961475, + 0.39520922775061884, + 0.34530191989779235, + 0.4419008577078371, + 0.4168511950417651, + 0.3400612571555123, + 0.5946198114588989, + 0.49611321804085223, + 0.34396156760831115, + 0.4792944606084569, + 0.40736847162254947, + 0.4216244452918404, + 0.4269904463767909, + 0.31028213978498836, + 0.37927507193669396, + 0.4799541781756389, + 0.37014763244575727, + 0.4374169194350487, + 0.41875091920597357, + 0.45984508050119594, + 0.4779800436568994, + 0.39310569353318703, + 0.5076391209677977, + 0.4689608300228714, + 0.3941241189136502, + 0.34264143356708143, + 0.42029556019302317, + 0.3557185400795462, + 0.3992799248867524, + 0.4204558558171587, + 0.4832467669306576, + 0.5154921031968857, + 0.43260958969622654, + 0.4117005773295438, + 0.4482973720600095, + 0.4377676695110751, + 0.4365664024802837, + 0.5242996570042064, + 0.39626907575700954, + 0.32880089481119523, + 0.4895829535141042, + 0.4035500697977756, + 0.44042809498314306, + 0.42789247052050844, + 0.4126652763393124, + 0.4936660430308211, + 0.36338633796988506, + 0.41969130221948925, + 0.3496872019797372, + 0.3550227578537683, + 0.35055876416736637, + 0.40714380445071013, + 0.34726056970397245, + 0.4227686591817648, + 0.5242039528249433, + 0.4174927095492887, + 0.4840806256951661, + 0.77243539409329, + 0.5305338581757941, + 0.43568743496331325, + 0.40884375829200315, + 0.45965285127302746, + 0.4379863392906032, + 0.3391556152927501, + 0.5649672326407128, + 0.458590306649348, + 0.41815615524218747, + 0.4589662340005546, + 0.3879632454240037, + 0.4883131455290965, + 0.5079538869759751, + 0.39270489889391547, + 0.5322518719164152, + 0.4792543031619419, + 0.36794086378987834, + 0.4477304841420888, + 0.4930750722909192, + 0.40035733338891516, + 0.4219878240313737, + 0.34187015961604683, + 0.5624654710349901, + 0.5023349697055318, + 0.8160363015939442, + 0.5128027399073465, + 0.43196034764808283, + 0.2831728852479007, + 0.46862505840863217, + 0.43159237935836603, + 0.6022056682627197, + 0.33963356096556824, + 0.5175259278840298, + 0.3310517798221015, + 0.3670581661627791, + 0.36084106462958976, + 0.35105762464862755, + 0.42564963790115345, + 0.3628645474840335, + 0.41540192122393454, + 0.4483103525504689, + 0.39668630879361094, + 0.3574368576523476, + 0.3564367516681435, + 0.40929011040843455, + 0.3440336772506802, + 0.5057725851536065, + 0.41747616400290116, + 0.48261711437207294, + 0.4095229333901543, + 0.37383562412543453, + 0.5554581614161722, + 0.45515074880934153, + 0.3631139228875331, + 0.5840626986190965, + 0.3120068471053061, + 0.4329398875109493, + 0.3607068715484339, + 0.3700656798478341, + 0.330279245952625, + 0.372373154047477, + 0.47104822328713375, + 0.35059348568633075, + 0.686947088551275, + 0.44169058610090534, + 0.3629858171576956, + 0.38848353471609803, + 0.36367970975811803, + 0.5756116678025303, + 0.32413642560494793, + 0.3689157929860581, + 0.3791705610144318, + 0.41991984226376766, + 0.3530305593896813, + 0.4074744973909595, + 0.30923744193276115, + 0.4296386216616736, + 0.762854822983033, + 0.4550353746940792, + 0.37646946753716704, + 0.3899529679610284, + 0.4135718651695926, + 0.3281131860279047, + 0.36763470821598276, + 0.42890285581062515, + 0.7472331998432657, + 0.5444090678716723, + 0.43102641391465485, + 0.38290532990749854, + 0.351571023386887, + 0.39744710325204824, + 0.32297835875791664, + 0.4238855074876803, + 0.4953172901426716, + 0.31560042546701844, + 0.37528617235986655, + 0.3703918010327989, + 0.459785590425184, + 0.35720364454008724, + 0.5365357495215716, + 0.4732488159054042, + 0.29785276305233827, + 0.40879707360853096, + 0.4129515898274147, + 0.5156359762107321, + 0.5135955138834492, + 0.2406391787473332, + 0.71434144993378, + 0.707105396041166, + 0.2032616888638975, + 0.3045212155484218, + 0.3763242706914816, + 0.4390703465122727, + 0.29105381323594437, + 0.3795906555561934, + 0.4549968269129039, + 0.46469126553607903, + 0.46417176353598577, + 0.3604387923885252, + 0.37084088560868395, + 0.3451088249056871, + 0.3163836179576507, + 0.4236929623536362, + 0.3836102508361445, + 0.37113373283469137, + 0.48410642022899697, + 0.40470773710498303, + 0.5304322323833495, + 0.43261925664593925, + 0.5984568356343926, + 0.42625456224570785, + 0.4256131998843725, + 0.4770487048014537, + 0.43885353839273944, + 0.3271120558947302, + 0.6146050746335834, + 0.6106329200280123, + 0.4338850555054725, + 0.36340825836748275, + 0.3767613730262285, + 0.5090062920761665, + 0.41161746337440225, + 0.4439238281186896, + 0.43441618055820624, + 0.46734609659719106, + 0.421437992089824, + 0.3626093011178111, + 0.377365473775902, + 0.5305318101757268, + 0.47322976743614814, + 0.4516464771186925, + 0.4718100807449482, + 0.3764323854573619, + 0.2590016685521763, + 0.31279606187442177, + 0.335789668081388, + 0.5146089896735392, + 0.4491206212412893, + 0.3178118192853128, + 0.451056134211367, + 0.35859561855971733, + 0.33631017170324135, + 0.35760931276739966, + 0.3847935861670136, + 0.34690529450036495, + 0.42824027626272376, + 0.613845733954318, + 0.3542399525339887, + 0.4302203286040769, + 0.5658942252677931, + 0.3911938903327852, + 0.5057237209694032, + 0.5787958788165182, + 0.44160732493882615, + 0.4656366094273104, + 0.4406507476426265, + 0.4036241834174605, + 0.3069554597641518, + 0.3563597421823873, + 0.4740074410998076, + 0.499958516038921, + 0.46454048751966526, + 0.3403941603538676, + 0.45450794106646053, + 0.4603008578861563, + 0.49216303375477277, + 0.40245066882804464, + 0.4602943671338124, + 0.3793435181392111, + 0.46178876064570534, + 0.43069403267631523, + 0.4082396339292669, + 0.34667210471995114, + 0.4454532934705412, + 0.4616883638280373, + 0.29339789777354247, + 0.36053250138339527, + 0.4368825543465767, + 0.38560487872844573, + 0.4107266525693797, + 0.4836530211259693, + 0.49703330207323976, + 0.42854521089917214, + 0.4571777181213311, + 0.35796121114504587, + 0.44266188868200884, + 0.41221312694883716, + 0.38803256441298756, + 0.3975126193397413, + 0.4358876637669973, + 0.39758179928131027, + 0.46266208335098324, + 0.4740722169215661, + 0.7162821510069028, + 0.5386556994114331, + 0.4954266646747137, + 0.5347857196721336, + 0.5063309083502581, + 0.49054658456709765, + 0.5309078607941461, + 0.3537304448735456, + 0.3951535137758176, + 0.4606716405654503, + 0.3434131218950434, + 0.4447512732491447, + 0.49032573816396985, + 0.43223268209317256, + 0.44085059822247524, + 0.4560167100532296, + 0.3192358969217856, + 0.5163484943171078, + 0.48400567596448757, + 0.37956083595082524, + 0.4278776172705805, + 0.3811633257896073, + 0.3532877619056038, + 0.44021444262847204, + 0.437250239941931, + 0.4189176784381858, + 0.45316666194915856, + 0.3884905865049757, + 0.4700315631345628, + 0.367950275532766, + 0.3893136675424444, + 0.42690736175440663, + 0.49459588263099546, + 0.4872484774128724, + 0.46492315002312806, + 0.4015553869717223, + 0.34838351551299535, + 0.3513737081705461, + 0.4359000102909796, + 0.4126936788042563, + 0.448674716854449, + 0.4683838431504323, + 0.4252736071034312, + 0.4019462214099272, + 0.35485085499804037, + 0.4080656148179307, + 0.5184261403935678, + 0.39829754266472867, + 0.2061619772839879, + 0.34742729837582026, + 0.44221871609496605, + 0.30584317327143173, + 0.2965028450929271, + 0.45306882283243916, + 0.4511536141634578, + 0.32795703064651827, + 0.3086168252751979, + 0.39159981033118246, + 0.38178340469742866, + 0.49831810341641974, + 0.29290027467479646, + 0.2976206513503663, + 0.3115967425491192, + 0.33175117011888233, + 0.25963216038801235, + 0.31038392760365285, + 0.29767269313923006, + 0.3175684060138517, + 0.31821522482111697, + 0.42353367382482593, + 0.343273461408068, + 0.38393442546382267, + 0.38796705521265284, + 0.4789231118391553, + 0.3356079316893229, + 0.3952855705751021, + 0.3874345699025666, + 0.3972303504025818, + 0.3906000943503025, + 0.40265596654150393, + 0.38928498277238116, + 0.6031848936841708, + 0.3105825782887767, + 0.4251995353994879, + 0.5381401492007849, + 0.46800847189774314, + 0.362090216955172, + 0.44169813049143186, + 0.3976090424155557, + 0.4344831081450145, + 0.4566074061201406, + 0.357464934816835, + 0.5630808886651287, + 0.4136806021664694, + 0.40828246295213627, + 0.34838704613493937, + 0.42185645906474867, + 0.37551797974736306, + 0.43826399583157316, + 0.4611914257169526, + 0.4298746955611565, + 0.45633039445542173, + 0.5651733042628638, + 0.4225062902509193, + 0.3911191494819418, + 0.5012389608538856, + 0.5715919184762096, + 0.5062970768393011, + 0.43466434708741, + 0.5327552785389809, + 0.5290042960336983, + 0.37701903269284737, + 0.38026957504317355, + 0.42356114072279044, + 0.3373873172192625, + 0.6079977397708943, + 0.48164373285713574, + 0.4076930860494222, + 0.34876895277374276, + 0.4562662260786417, + 0.3956901774907631, + 0.4926750150316796, + 0.4316463230139378, + 0.39791616080411, + 0.514928765310536, + 0.3127991728658333, + 0.3979963147757816, + 0.5125169781381786, + 0.33529129776482647, + 0.40842934936580094, + 0.31268282883646425, + 0.35775995718711506, + 0.37501142393004533, + 0.3874734751765199, + 0.4475783264678856, + 0.3371259280228998, + 0.3126255579915808, + 0.4208760459978723, + 0.3813798839281242, + 0.3619527035327423, + 0.4363718005637848, + 0.4592792337968443, + 0.4135288106634641, + 0.5116529611213417, + 0.5546594324769333, + 0.44427678551359734, + 0.3427752434899648, + 0.47105512330734145, + 0.42712269824137156, + 0.38227248674031433, + 0.41464962225004726, + 0.316335634957839, + 0.3136716900944477, + 0.30227946835180775, + 0.3548373989997008, + 0.42272637897502435, + 0.6432970398290381, + 0.4263753463524418, + 0.40647383203902815, + 0.2944449033647195, + 0.3505587144342734, + 0.3131487463068518, + 0.4015191961704208, + 0.36239474278005274, + 0.38175276759623233, + 0.34519729266625854, + 0.4117589813227876, + 0.3724755212081406, + 0.4182554852813333, + 0.39583752427813657, + 0.3563856144600165, + 0.4875895858897822, + 0.48516436798110374, + 0.3667635834677421, + 0.4123624648207437, + 0.4156194965698389, + 0.3756567148682933, + 0.34483220519644214, + 0.5027491859222271, + 0.4397232054507549, + 0.5726033095461994, + 0.43837777169081876, + 0.4284765419341445, + 0.4383863112186043, + 0.441329280741601, + 0.3412629170855072, + 0.5690720657428617, + 0.3918269417204143, + 0.22947698177243286, + 0.4436763730753429, + 0.5566054832039302, + 0.35496640410323615, + 0.4509570075202907, + 0.7499131715728199, + 0.42981208773390944, + 0.5012561877035585, + 0.3069371587445485, + 0.46560789434932637, + 0.30452314068351394, + 0.30116783965551136, + 0.3178261598211878, + 0.2717202663550177, + 0.4280016780804979, + 0.22809091886782215, + 0.3688940912552988, + 0.42099749616149834, + 0.42165556837687357, + 0.382269146234616, + 0.38165040186459165, + 0.3247478449361774, + 0.4184743166688194, + 0.5015618752019099, + 0.5225889999068339, + 0.32788770477985685, + 0.4345151027709203, + 0.3906809083600358, + 0.7197354126654849, + 0.37682945834456893, + 0.5465115071487603, + 0.7298046771968542, + 0.7071810342229974, + 0.4493166973794163, + 0.3566834669247113, + 0.3173902686338189, + 0.4360320516385738, + 0.4754947275295097, + 0.4784658743747778, + 0.5524037903455186, + 0.4049281810732774, + 0.5362377172879256, + 0.4061928642649536, + 0.3665008550631812, + 0.6906528313622509, + 0.6728253113532849, + 0.4519036155674899, + 0.3961651238167257, + 0.4432872145953565, + 0.40000164794198845, + 0.433289015879726, + 0.6340819309756995, + 0.6513558510277454, + 0.40702149354842343, + 0.41955600903317575, + 0.42141188198952506, + 0.3509464676017726, + 0.4726501466773215, + 0.3240319837059135, + 0.46464815778849305, + 0.47270748298075815, + 0.34062335141304756, + 0.42862007038873995, + 0.39553857540527554, + 0.33700024614951124, + 0.5101453769019325, + 0.4254059571791166, + 0.3901140692725583, + 0.35977366735693994, + 0.46553921716234764, + 0.4092171902248055, + 0.33099609665526597, + 0.33851813530995745, + 0.28432011800985524, + 0.2375688555948624, + 0.2783671661288261, + 0.4443840080765731, + 0.486697842937683, + 0.3430715174376323, + 0.3172430294880029, + 0.34684765414599344, + 0.4583111113561696, + 0.4175333703453821, + 0.48289059275284857, + 0.3546094368958009, + 0.4552828742627919, + 0.5079016355686471, + 0.36733712736850954, + 0.5020787537533387, + 0.45420356247810045, + 0.41919653717591576, + 0.41022047204774786, + 0.34571630817566645, + 0.22861530551159762, + 0.2769256158844699, + 0.2609207481664926, + 0.5036292042099708, + 0.4286619180110646, + 0.421211246339062, + 0.45504715457364553, + 0.4794290446317696, + 0.3255539819951414, + 0.412362110511821, + 0.3555867908912568, + 0.711675898374346, + 0.42254970017469895, + 0.4393561988634226, + 0.34527305698834554, + 0.43047210885115683, + 0.46600773370503806, + 0.41498429736188985, + 0.38629723151141193, + 0.45473365630159723, + 0.42711169823290646, + 0.5308621914770177, + 0.40165861250479046, + 0.4676393592983697, + 0.41931140618389434, + 0.41236547020930386, + 0.3924532269443431, + 0.45246736230357426, + 0.41788447445223437, + 0.46713754887804565, + 0.32548900194350366, + 0.31675371103652816, + 0.31817168759644254, + 0.4023831389406821, + 0.43740489962306967, + 0.39191080903060865, + 0.41918725402929, + 0.46074114353903867, + 0.4146052911797776, + 0.4214679491322145, + 0.4228396879543003, + 0.26575960647712243, + 0.4100335532499568, + 0.31000104911544196, + 0.42248023741927143, + 0.22254787600169895, + 0.6599408255921074, + 0.3096624256976025, + 0.4301246785831639, + 0.5485651841890012, + 0.46030369019059336, + 0.26998515423140146, + 0.4151509327038289, + 0.3341809885186651, + 0.7691252491026667, + 0.27350358119512336, + 0.5024527514110727, + 0.4947124319954889, + 0.5070217049936252, + 0.6294165502572433, + 0.40653197153172516, + 0.565392846571479, + 0.4416375096728487, + 0.37450823748765844, + 0.32823507393919693, + 0.43857315565981164, + 0.3027115823562234, + 0.7089774233854579 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.4653647077415584, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.5453676294968282, + 0.7494895783795542, + 0.725773924708241, + 0.8096836041785997, + 0.723016883267764, + 0.6463329035807432, + 0.7046964286768287, + 0.6506281865909156, + 0.42542837459409055, + 0.696879215439332, + 0.34422463342733733, + 0.4854658037749917, + 0.4061871632587162, + 0.36569098246357706, + 0.40343738249050726, + 0.3355450659107699, + 0.3833201383064721, + 0.5551117197728348, + 0.5598266788779727, + 0.44895864590827816, + 0.34410853721427986, + 0.6780284580073752, + 0.45941828828823217, + 0.60116126234162, + 0.37583867119130926, + 0.3316841728911472, + 0.3882525823033881, + 0.4819061468871633, + 0.6056315968784686, + 0.4879941519019792, + 0.32585500356135166, + 0.4079624356053967, + 0.41450051411341293, + 0.3675810402941761, + 0.472366678507032, + 0.36430051426736076, + 0.46881100636799056, + 0.5266224240395301, + 0.581271824025781, + 0.6232434375970818, + 0.7832304331192586, + 0.7752583548729465, + 0.7674454658891289, + 0.6475985335541279, + 0.4640264100804721, + 0.5325668992084212, + 0.7729479963715705, + 0.4166805897215926, + 0.3798094254319286, + 0.528805945066557, + 0.42235762728737486, + 0.7718361419934107, + 0.7885541888326685, + 0.4727359086365452, + 0.4647835859557916, + 0.3952865418340139, + 0.3740741549414876, + 0.4313723742592237, + 0.41546320391873814, + 0.5230481510619868, + 0.47662577515758314, + 0.4439262312077879, + 0.45968730460550816, + 0.506100799178976, + 0.578814059681018, + 0.41469361235834457, + 0.40492646534110216, + 0.5713071520113301, + 0.41287849483730654, + 0.4626434327217378, + 0.47593412945290087, + 0.47668749024575685, + 0.4561510405122208, + 0.5094868145783764, + 0.4206574457059199, + 0.5168314222101696, + 0.33228854532769114, + 0.37797369830372535, + 0.4075105851838584, + 0.35121062431036226, + 0.44316684256368516, + 0.5588079377840636, + 0.3960736599979363, + 0.40856823005292137, + 0.39528488393986877, + 0.4780053720276599, + 0.4524264150055181, + 0.4362548893972261, + 0.3637403525131652, + 0.40050884764584455, + 0.35378494789996484, + 0.35891832240179683, + 0.5055920372897923, + 0.5482745888422782, + 0.4369028150810878, + 0.5703175922057768, + 0.38601642154254134, + 0.5449535780582232, + 0.42791136605751473, + 0.46136978264020406, + 0.4556439668890554, + 0.5237351718520179, + 0.32780023971042827, + 0.4834892361711834, + 0.35084825327148056, + 0.37053834753720893, + 0.4810022462498671, + 0.53680658729004, + 0.3620264043357905, + 0.8031082563494553, + 0.7970068063865403, + 0.8024581139651014, + 0.8217232729230939, + 0.577741507300194, + 0.6975971109475156, + 0.7114478080728903, + 0.7496502262262881, + 0.6944722831450229, + 0.5831778374231565, + 0.6527140846828744, + 0.5690249919532513, + 0.7670192588835415, + 0.6580280089184383, + 0.4316509459933102, + 0.34761731359280434, + 0.3218923700753977, + 0.3542126555866892, + 0.552195073744926, + 0.43735697762104014, + 0.3082660620476764, + 0.47034713570542613, + 0.437288955293049, + 0.4895480645665175, + 0.3845296730343606, + 0.4188067303692059, + 0.4735393173238989, + 0.46813839747327546, + 0.3710796317835823, + 0.4027461462539568, + 0.4014105115458156, + 0.3740422258811143, + 0.33007298905244653, + 0.4406148486594498, + 0.6060680905069671, + 0.32034800859453244, + 0.5632373210935672, + 0.3618870373230199, + 0.49559522647955934, + 0.44486311290826763, + 0.574359333550108, + 0.33024541327151685, + 0.503603731165702, + 0.33879370718665974, + 0.3697958606147153, + 0.4167287623597447, + 0.36421543387251487, + 0.28696777282492797, + 0.21341031731644178, + 0.37847779295883244, + 0.37757899713206267, + 0.5006895693218324, + 0.46475988052372014, + 0.5440121842085577, + 0.47343632931633395, + 0.39193208880397995, + 0.4748395293764142, + 0.507178058302159, + 0.4978379193930207, + 0.41254504236011547, + 0.4552758263750948, + 0.41670484674188535, + 0.3264721652483921, + 0.49709569735119763, + 0.46280457966575655, + 0.29109923204952676, + 0.3210109348944474, + 0.5024974686234972, + 0.5534570821243318, + 0.5693617307909079, + 0.4732611667720946, + 0.3420058527781028, + 0.5399455866279583, + 0.4939258731273274, + 0.5039164460163614, + 0.32061014240508195, + 0.539549919230663, + 0.46268078573436755, + 0.5043154079829949, + 0.27116673277411746, + 0.439351357715601, + 0.3465096488351476, + 0.28475426378915125, + 0.5212013161453318, + 0.44470166365098673, + 0.537235358223714, + 0.39635583503880406, + 0.5876147450874809, + 0.36163483005342295, + 0.4635442888483765, + 0.4165218276111386, + 0.3507005127589989, + 0.2868270290536468, + 0.38689493851492013, + 0.3822735062721963, + 0.6248878523883026, + 0.3687054099673084, + 0.30537101876537126, + 0.46485883972477793, + 0.5493561543789367, + 0.41336909967340085, + 0.46229342604171747, + 0.4185916049768282, + 0.34617412231182665, + 0.3940415846887361, + 0.38817932283583323, + 0.5117641058297667, + 0.5835338245057758, + 0.5847035040235324, + 0.5480284169215018, + 0.35414284939762797, + 0.5153342865762611, + 0.4617894606674658, + 0.29594389647037866, + 0.3501363731462895, + 0.3599713291966064, + 0.2340355100840614, + 0.5465667170193836, + 0.34553460243794937, + 0.3243805041301867, + 0.5341201635479446, + 0.30263747346965575, + 0.2739916620814827, + 0.5288271509173603, + 0.4162171707671609, + 0.3312007851731089, + 0.59990532420719, + 0.5548504794280056, + 0.3177485294245803, + 0.3111430284048086, + 0.4864474308352433, + 0.33890918365062617, + 0.45717157834111577, + 0.32966975513820723, + 0.4951358556746582, + 0.4599546327235088, + 0.4031226255103287, + 0.3678440042867347, + 0.34606683255596044, + 0.4659528916368232, + 0.36051113031738324, + 0.4388913106998212, + 0.3457654578382026, + 0.41233616931832795, + 0.37576004148608416, + 0.37591425408649337, + 0.4440287827518788, + 0.4170042390700565, + 0.38350297777392833, + 0.4323499749197918, + 0.4465528401080879, + 0.33723918170088724, + 0.5285380329285561, + 0.3416740293036664, + 0.36200310558761467, + 0.4036111459893793, + 0.331588494722733, + 0.3686439276170089, + 0.5208303545158667, + 0.287060267025454, + 0.3593005592054671, + 0.3524411949148046, + 0.3610519871414875, + 0.3724557122687258, + 0.41954874425966127, + 0.38813256034465426, + 0.4872514721514514, + 0.3096556890874339, + 0.45040225235426234, + 0.5090471640069595, + 0.3237141007511757, + 0.32145125071575303, + 0.5958948895608567, + 0.4875455986389976, + 0.29554608094680446, + 0.523241477976124, + 0.4626562497031546, + 0.43922854149028484, + 0.42653482733327874, + 0.48719203187470905, + 0.4779881769587394, + 0.5018467471672932, + 0.4961931114823375, + 0.500255618986955, + 0.4181460840246447, + 0.43911969893030184, + 0.5645325239409048, + 0.47818825500369627, + 0.48032816649104787, + 0.6540087656222547, + 0.3368431229200373, + 0.5992229768992255, + 0.6544536909979184, + 0.719771937662915, + 0.6902730391958243, + 0.5397078310346561, + 0.5678761875965528, + 0.40105345731849557, + 0.6888377619611978, + 0.4941488408016359, + 0.501114451459364, + 0.32322198879758796, + 0.5240118598315352, + 0.4388524094348532, + 0.41312298559161464, + 0.5216194355681766, + 0.3597962732361936, + 0.43193944469782697, + 0.3360611331891501, + 0.4609886602597447, + 0.3851611003723526, + 0.36263074934066886, + 0.40336624182506076, + 0.2876101115846781, + 0.3573074928620724, + 0.392303281068778, + 0.40713740999494835, + 0.38631585032026905, + 0.3957474193710015, + 0.5934903397192373, + 0.48271028839868124, + 0.44633457332582765, + 0.489276788925392, + 0.4389748384206127, + 0.3443918484914877, + 0.31997747113770375, + 0.5580004277322143, + 0.3954408543770507, + 0.350659020507429, + 0.4493728402459922, + 0.5481268595615492, + 0.586634046832113, + 0.44230471319451814, + 0.36291513619564186, + 0.4866776566051479, + 0.37764250309218755, + 0.35163797190423357, + 0.4993094083740595, + 0.34658963299572143, + 0.8160715442621794, + 0.5420960695645711, + 0.4435230164969937, + 0.4449391294216766, + 0.4733703676121183, + 0.46885834835080387, + 0.43116682456945105, + 0.544267900984258, + 0.41648641934166053, + 0.5618937639818282, + 0.46768689254735957, + 0.5253162726647403, + 0.5768697254261231, + 0.49056190356509255, + 0.4999409180173502, + 0.4111113788704751, + 0.43418519127543287, + 0.6163357446291083, + 0.5935321689206263, + 0.4448553926472615, + 0.6082760725905768, + 0.4669377639933447, + 0.5719232384074644, + 0.46944031173569817, + 0.5068542397237589, + 0.45393338218582435, + 0.5409708275925005, + 0.5541906809597523, + 0.5948187444056149, + 0.525319073015206, + 0.5219672695502813, + 0.519127359676471, + 0.4890288449379207, + 0.40303642456475314, + 0.40657898327531183, + 0.32971672813181274, + 0.3913975864093166, + 0.3295254242343199, + 0.3517007166658443, + 0.36897028252591463, + 0.5062454645514003, + 0.359060455947927, + 0.3521968378069201, + 0.40292083539324663, + 0.5506386732284761, + 0.6236735255629023, + 0.5804411050638796, + 0.5629273759180676, + 0.4299767073898752, + 0.5781755912099799, + 0.46169055813129367, + 0.64372020519753, + 0.48228173739412294, + 0.5617445760919856, + 0.4430165793118243, + 0.5449110920487881, + 0.5671570404497039, + 0.4791596483031407, + 0.479220386138489, + 0.5213861058725616, + 0.5735112999638151, + 0.5334517012856294, + 0.379699485902065, + 0.40262484631195444, + 0.3478702662865202, + 0.3765247530625823, + 0.5251004278240086, + 0.5600955019460033, + 0.3301514871051546, + 0.5989043455723978, + 0.5765205013762282, + 0.47821439302825325, + 0.40977262204690623, + 0.4723725901776334, + 0.5608797547322636, + 0.5123885483331988, + 0.46695053769318734, + 0.8453232743565463, + 0.44804151195903463, + 0.5352907104797018, + 0.49328859401396086, + 0.5439277797926537, + 0.4684850340595499, + 0.4054189196299017, + 0.583689343695533, + 0.4626139382603754, + 0.4211234802969318, + 0.451895407394645, + 0.4501028582112918, + 0.45828433579981664, + 0.42129745352290837, + 0.4567832724376454, + 0.44745615461278393, + 0.4806293097232784, + 0.33180668338421393, + 0.4935349497984668, + 0.3386487943986121, + 0.395195766822891, + 0.30278698105577506, + 0.4961204323543068, + 0.3816937490526689, + 0.3373073866864711, + 0.5292065987680225, + 0.4161568465175909, + 0.2938822452103734, + 0.5333055225367946, + 0.48258353723186154, + 0.46355770401999463, + 0.5168857810011964, + 0.4120007227241628, + 0.41799002178590855, + 0.6140250421456563, + 0.42718681641335293, + 0.5485431390635842, + 0.4059623318687917, + 0.3214728601471469, + 0.29564708231322884, + 0.404585174600986, + 0.48769061070138153, + 0.5672286722783775, + 0.5485620588636998, + 0.4729807693271748, + 0.5273887938244959, + 0.4728255390059217, + 0.4935802079390641, + 0.4277593288042355, + 0.416806817620943, + 0.4225529332532841, + 0.3935312995373847, + 0.4334989697818617, + 0.3906399165030309, + 0.4207308216760718, + 0.5007939074754435, + 0.4373548678233343, + 0.43903933707569814, + 0.369281118210349, + 0.3820569090539373, + 0.48890090505693135, + 0.533185266030605, + 0.5199521429139096, + 0.5386902070666264, + 0.5599911532855454, + 0.588289507678922, + 0.6577541091240439, + 0.6405526094820861, + 0.46858950478061484, + 0.5189949744207542, + 0.4823101097255837, + 0.5263813331485796, + 0.41145530271773717, + 0.49349360948023946, + 0.4643092899199761, + 0.565646645101346, + 0.38673954525054033, + 0.5570954906597677, + 0.6430332212734593, + 0.6116252797687366, + 0.5685164759488253, + 0.6358052919804724, + 0.537452756231259, + 0.47177359855621115, + 0.3887211637455938, + 0.4312987060906399, + 0.3381922007417521, + 0.37746483842747774, + 0.5432918653300192, + 0.5541569688859661, + 0.44305621437342024, + 0.3677183190561886, + 0.39177993861244265, + 0.4735479745447111, + 0.45935228222429136, + 0.4022547736921848, + 0.512341460051355, + 0.49981858769497645, + 0.3821377967558804, + 0.43402849621875567, + 0.3804106565973577, + 0.45593595644981894, + 0.46648462718126316, + 0.356437732434373, + 0.3976967896952158, + 0.44242309261019597, + 0.3807100710786751, + 0.3742183679515046, + 0.4036546449776735, + 0.37026532704792026, + 0.3934349008057993, + 0.4069850185793726, + 0.5144145396736998, + 0.3978520002376928, + 0.3068057913527224, + 0.5616193652772495, + 0.4034429858336139, + 0.5358290997710657, + 0.5215069337879722, + 0.3807739445470025, + 0.5024840506919533, + 0.5717600970103026, + 0.49161834314442254, + 0.6293571499575616, + 0.43587316822736355, + 0.5090071762444973, + 0.4745196970606215, + 0.4664627775844274, + 0.5206454686623904, + 0.49915472176573533, + 0.37349131030426735, + 0.46212179253609653, + 0.508240666315196, + 0.42424518256420085, + 0.3291758223344551, + 0.4557747686717492, + 0.49050928757523693, + 0.45694416206959393, + 0.6090773169950032, + 0.523761265874536, + 0.5320091818550357, + 0.5365321303825048, + 0.6029943328904303, + 0.5342989400703837, + 0.5824584358329705, + 0.588416118765302, + 0.583223261698504, + 0.397177101136994, + 0.42873485349678314, + 0.38228427617884625, + 0.33651091443131964, + 0.31476163286409164, + 0.36691934275184757, + 0.39969579351145207, + 0.3547141853806262, + 0.30983152989887003, + 0.48764736363778893, + 0.30072161713990025, + 0.45196429828592555, + 0.3547636007002967, + 0.5291344284207882, + 0.5674384750768263, + 0.503972500452279, + 0.3398052861220414, + 0.4026811601795904, + 0.5741350001125713, + 0.3606590314001433, + 0.5035056773798878, + 0.3304742533024535, + 0.8183134677220493, + 0.5268307503641558, + 0.42367428636931787, + 0.5883578003051315, + 0.5387506971081532, + 0.4836931428573587, + 0.4302442904188318, + 0.4896272095156318, + 0.5524293801054049, + 0.44716496499581904, + 0.5598143173684992, + 0.5282282851720879, + 0.31024381890280245, + 0.494933317157429, + 0.31689296083388696, + 0.37994583429066003, + 0.3300665871553948, + 0.5181942331783564, + 0.43214418816233147, + 0.38695249304109053, + 0.5213660358677433, + 0.5461601333160487, + 0.3895340978826668, + 0.48743589534135356, + 0.44262734763955913, + 0.4675362688045061, + 0.49643934221273056, + 0.4769256663078026, + 0.3615328144710955, + 0.358809782719582, + 0.5628830722778781, + 0.43949166870822665, + 0.38935825928814566, + 0.5147870021730431, + 0.41232802705136223, + 0.4301996296957186, + 0.33135915371955393, + 0.3706952692704286, + 0.4164029141688515, + 0.4712414963625929, + 0.4578632860416735, + 0.3514913675955717, + 0.4678250820953548, + 0.28927777929714976, + 0.4859290262648199, + 0.4522267630227519, + 0.48211873713226133, + 0.47834176567150405, + 0.40607682237560455, + 0.46942168311505283, + 0.35890051076114154, + 0.4883280521944359, + 0.4065769041927033, + 0.4519805275198267, + 0.44822869525693143, + 0.36303703213314875, + 0.4910860082070442, + 0.2655594243109079, + 0.6623111315671313, + 0.6648132969856331, + 0.5630164253969031, + 0.5550430561011623, + 0.598494426633715, + 0.6007339978045163, + 0.2897495384979525, + 0.3906573266198168, + 0.36148392509687355, + 0.3902495485848278, + 0.3446098636197327, + 0.34488613371467597, + 0.3992338246525668, + 0.46308671722582995, + 0.43596887646883425, + 0.35849221377421253, + 0.37717598524504886, + 0.30635043905035525, + 0.46271484857913203, + 0.253153533487864, + 0.41338239852529185, + 0.4646350417452887, + 0.4764745813013749, + 0.4865572643671513, + 0.42254632637884104, + 0.5149338306677801, + 0.2848277740172001, + 0.5095264896369885, + 0.4533386094106904, + 0.4019594294905311, + 0.39152417364554754, + 0.4258391613356953, + 0.4361548081420647, + 0.4364750927596966, + 0.4864864726808249, + 0.5042545373628243, + 0.5368723384001705, + 0.35356205628361687, + 0.39658241218902873, + 0.42209216648553627, + 0.4712152620685195, + 0.4752828364593071, + 0.4857900033195834, + 0.3442755768066618, + 0.5549439332963203, + 0.4525902574068709, + 0.4363164939492551, + 0.5603686981005924, + 0.39259629643560884, + 0.32497033670087, + 0.3118048924579526, + 0.36985772183602944, + 0.4155952242462672, + 0.38373670072553356, + 0.398608989068029, + 0.4010694331309815, + 0.5080788902668331, + 0.5817703141024914, + 0.647079276855758, + 0.5323132167883645, + 0.557752819130131, + 0.49903919616253145, + 0.554677990271027, + 0.5132967463314457, + 0.5650257187040251, + 0.37336423258300755, + 0.48114804317927856, + 0.5921598869742776, + 0.6391759613250867, + 0.21233678068197887, + 0.36512271855287787, + 0.41635592734910176, + 0.48626418574465397, + 0.47375877291534163, + 0.5147266366858986, + 0.3987687552892144, + 0.490389127336212, + 0.7370517087295991, + 0.6103642509514129, + 0.7667189098849286, + 0.7798749828716891, + 0.6473463821239168, + 0.7419765896336439, + 0.7479566767664596, + 0.38552571641258965, + 0.40493796544470156, + 0.41048769236358595, + 0.5091456801830312, + 0.35511928954116756, + 0.46465684244878275, + 0.3800131937378576, + 0.48490446988482705, + 0.40943147770262395, + 0.4093397238709239, + 0.3143436578530534, + 0.44042241770729357, + 0.4084420046516103, + 0.4611004586075633, + 0.4394775385100618, + 0.3176385194930016, + 0.36088920449697864, + 0.3460122377288458, + 0.36062165898975307, + 0.2764841462340934, + 0.3628575593115415, + 0.30978883924929473, + 0.5248173324633666, + 0.5642905080542236, + 0.3692261900450898, + 0.32586010207973104, + 0.2583274326254674, + 0.5306983894726123, + 0.6296369495371719, + 0.6394027318939988, + 0.3877810978590354, + 0.36651485849464827, + 0.2603712969060063, + 0.47908099229417955, + 0.33942994423279177, + 0.3947208169999988, + 0.4814838349536256, + 0.5248480039331996, + 0.46931494937122453, + 0.32487695594131955, + 0.46499696291169246, + 0.5156428656611327, + 0.5029387502200475, + 0.4599514697137064, + 0.5017975190296647, + 0.40551556446059617, + 0.4656094710762115, + 0.440862179859959, + 0.5077231480960115, + 0.3554963305718422, + 0.44013809968470413, + 0.615282437563788, + 0.42414208764664485, + 0.6232284603826449, + 0.33801972798651186, + 0.5466122578638936, + 0.7080905706194947, + 0.558444221946667, + 0.3997721517494, + 0.37028898950771894, + 0.34869037509165024, + 0.3242037109411427, + 0.4478233525419581, + 0.5889223118151743, + 0.4583000586783605, + 0.40815502004289694, + 0.3226117742261691, + 0.4806036279744878, + 0.5198770071459567, + 0.5627926282167857, + 0.597253197657616, + 0.4993011972135573, + 0.5020343836772678, + 0.39573915994102526, + 0.5092136542074693, + 0.306802650630075, + 0.3897869873481122, + 0.449847998885569, + 0.39072697813000384, + 0.5819715936695854, + 0.5802566415794895, + 0.609546609723151, + 0.6693677443649254, + 0.7473314914231938, + 0.595334014482192, + 0.7569285134319654, + 0.47861005064347817, + 0.4291289211188153, + 0.40287129506331504, + 0.5361335517867281, + 0.37565008925299015, + 0.5055327420225058, + 0.5169141116818265, + 0.5193818986852152, + 0.4733235721722404, + 0.443394489304, + 0.44767765670372267, + 0.4675939599726043, + 0.39011033039731036, + 0.503234520742484, + 0.4599144731867213, + 0.49638534579402616, + 0.5128903676404748, + 0.5749014638851296, + 0.5455977193041701, + 0.38561280333755926, + 0.5336426814668112, + 0.5173828263706698, + 0.346840543748155, + 0.4395962068762882, + 0.4212850477678095, + 0.41618108620479, + 0.40417264399037517, + 0.4387751220146345, + 0.4583389205163661, + 0.4293169253464218, + 0.46087906118894345, + 0.4395867377746613, + 0.31645147447631833, + 0.4709443154859267, + 0.4858357917044179, + 0.4471262255426388, + 0.5468987454844219, + 0.5726811274596034, + 0.7386705210304166, + 0.3373250302503831, + 0.3363367807402553, + 0.507110537260239, + 0.3821336178476105, + 0.2825263268412102, + 0.339060073879768, + 0.3377079646854062, + 0.6770204592859097, + 0.6209923551291843, + 0.6479118595433873, + 0.35718466809107946, + 0.38048128632341277, + 0.328096307964579, + 0.5306317420698198, + 0.44311029997428186, + 0.2634781024409029, + 0.2916754745676884, + 0.446450173994287, + 0.4234877332117686, + 0.4573369684617126, + 0.47923737368033154, + 0.44071498096654754, + 0.40397308893741474, + 0.31078545285216697, + 0.6844544584199052, + 0.5563091311892355, + 0.6889220002583323, + 0.5768949609155044, + 0.28676659120629, + 0.519847237002099, + 0.526345709601531, + 0.5265367031737441, + 0.4753203111145721, + 0.40107788801595645, + 0.5005891551895976, + 0.5381847157440514, + 0.5338080257005671, + 0.47414739669643424, + 0.37233978084660263, + 0.4469548928640381, + 0.38871143802712615, + 0.4635390578079673, + 0.6353494970840824, + 0.5228562946817845, + 0.580721303463448, + 0.4724858039829556, + 0.48116749834425926, + 0.47048907761442066, + 0.428502000833597, + 0.488957459181224, + 0.4207898521361561, + 0.2895305614925976, + 0.4220954720630746, + 0.4871891347162837, + 0.26827107769056546, + 0.7574569407910327, + 0.43736334149853046, + 0.6008574778980491, + 0.7267995540747316, + 0.67496659387161, + 0.3886032985129766, + 0.8023452366106073, + 0.49551155729283136, + 0.6678290936119377, + 0.402377036040436, + 0.5436071433229205, + 0.5406178340130727, + 0.6199445158504877, + 0.5762924510891347, + 0.6338133542656372, + 0.5672772074800926, + 0.40865235158057295, + 0.5386207748118732, + 0.4879374051183595, + 0.4201473670911081, + 0.63624489660549, + 0.32825028905084225, + 0.562161328054207, + 0.5326119518106115, + 0.627306329195094, + 0.5667206121723912, + 0.6278673512527305, + 0.5216268697565078, + 0.41103725151672954, + 0.5214321806825954, + 0.5307899325458382, + 0.5063290167232849, + 0.4090938431756072, + 0.37318733221768746, + 0.47284061791520926, + 0.5403018041000565, + 0.5570487689340105, + 0.42549849596979566, + 0.5761352995694977, + 0.5827299821575146, + 0.43196013155622764, + 0.5409864235633012, + 0.5367530285974874, + 0.5120789601618311, + 0.4584851048842968, + 0.39642536985121146, + 0.4032898480524161, + 0.47855531360866266, + 0.41997768623364645, + 0.34942309479581946, + 0.44901086311984045, + 0.36036075576280524, + 0.4685170725924159, + 0.39949336573843025, + 0.35156346588144544, + 0.3559397064108884, + 0.359337077965756, + 0.4640444536982307, + 0.4409282219041629, + 0.36877763620103265, + 0.33042755216623937, + 0.4269115922123935, + 0.5759765000833137, + 0.6417595010364949, + 0.32810011176511356, + 0.5414885553647716, + 0.5724979247407012, + 0.5615443606302664, + 0.5549431226999336, + 0.43174437607038585, + 0.49120617900266866, + 0.38116282543726715, + 0.5271796760069698, + 0.4235557649044027, + 0.6399801724340934, + 0.5132424548926744, + 0.5165921631759357, + 0.3096413147057853, + 0.34414804960780676, + 0.3997332085594262, + 0.4876506930076038, + 0.47777334280089234, + 0.5685761859481702, + 0.34827463832333677, + 0.5994097142239216, + 0.511904533441214, + 0.5392640763850985, + 0.5349279105887152, + 0.5003691252580129, + 0.42950409573377896, + 0.5396432781028463, + 0.5552443507753463, + 0.5427876278534621, + 0.5188227503129184, + 0.4459880074577443, + 0.3554624494424927, + 0.38261198235084665, + 0.38841148960460153, + 0.2560494030620823, + 0.4028439490316825, + 0.5117769712638387, + 0.43339335645418353, + 0.42328815839650974, + 0.4458318165685637, + 0.4167378487916214, + 0.5107513098796476, + 0.5407908189081483, + 0.3728271441292192, + 0.48249516778873774, + 0.5568207329537609, + 0.3915049571403467, + 0.3906055195948777, + 0.49806548964553643, + 0.46973556207028394, + 0.56301516577576, + 0.33364264079051753, + 0.5168863472685993, + 0.366252053490445, + 0.5973033776161901, + 0.5446603073706127, + 0.6052428682013319, + 0.29328968938903155, + 0.3279775794269315, + 0.3223119453955768, + 0.29503583402991274, + 0.3216911470977769, + 0.4406668529162853, + 0.4746062922087011, + 0.2903916973333193, + 0.3772317923592491, + 0.45039821008176234, + 0.45508262024486545, + 0.5271064938606649, + 0.45311442314711603, + 0.6023565597637551, + 0.2804831992847157, + 0.45721945579803863, + 0.4461545660274152, + 0.39611538883998465, + 0.42139224335927816, + 0.49804476036864587, + 0.4833467323126541, + 0.273263752813898, + 0.5196422239607853, + 0.44782093377893956, + 0.4325200881503179, + 0.560502375212506, + 0.41411167187181946, + 0.37082646149806875, + 0.4026998959746884, + 0.44055004382588575, + 0.4260845085107542, + 0.4086493081562146, + 0.43708877610336294, + 0.5572811662383375, + 0.44121109696098704, + 0.5774431751115536, + 0.48878298837216644, + 0.4760391101494821, + 0.5186557554111239, + 0.4969347086559882, + 0.4865784779965791, + 0.4672183486274173, + 0.5923712936328678, + 0.589692353881699, + 0.5464737666006704, + 0.5646529024523049, + 0.6681720482848302, + 0.46275911658468016, + 0.45669276298029715, + 0.35155434185883433, + 0.48575130481195494, + 0.5149342612653889, + 0.4854958451760178, + 0.4613961732198336, + 0.5378570809664952, + 0.3722923659914006, + 0.4805390839280646, + 0.4040771436420381, + 0.3056906694102609, + 0.46463665235972085, + 0.4463395290165245, + 0.45067962273231627, + 0.4555105120653094, + 0.35282031548181136, + 0.6071211518334554, + 0.6253445830111546, + 0.5211510014730949, + 0.5412529556288599, + 0.6157893437804906, + 0.4815970031402753, + 0.3304564198226441, + 0.43055333306312404, + 0.40033279560930546, + 0.3852858678881039, + 0.3839564889956159, + 0.41932268439198284, + 0.4718703293937692, + 0.43947422360804755, + 0.3305569143469186 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.05267068291891003, + "valid_tracks": 4, + "per_track_pearson": [ + 0.056768873086175994, + 0.04655293276906121, + 0.05590910473532467, + 0.051451821085078245 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.4928263237105423, + "valid_tracks": 626, + "per_track_pearson": [ + 0.5055009531322343, + 0.49699306292988493, + 0.4960415804866559, + 0.49645279458155045, + 0.504110166462966, + 0.5006028462796561, + 0.4334806896792585, + 0.4686399074270243, + 0.5002930616227472, + 0.500789023721765, + 0.4949074503205397, + 0.49560026245473854, + 0.47193859785722786, + 0.4768171532640446, + 0.5011625372121087, + 0.49747501689975193, + 0.4923632823468767, + 0.4919923428081603, + 0.5048954707091554, + 0.5010938919509262, + 0.5109703932259841, + 0.5149908294228777, + 0.5140835595528032, + 0.5027553414436108, + 0.4916328450202407, + 0.5020211641208969, + 0.485984767049902, + 0.49409768906332846, + 0.49132558209533583, + 0.4905646351658013, + 0.5035502232239679, + 0.5044127152329478, + 0.4768116705091907, + 0.5073320543569553, + 0.4942047773885044, + 0.5149056772287735, + 0.49413148691027825, + 0.4971042792601367, + 0.49949524120416855, + 0.5015970694688255, + 0.5071800938032842, + 0.49646973836652375, + 0.4805121409041608, + 0.4957499692641291, + 0.49662714895182986, + 0.4788596601018261, + 0.5016108612475297, + 0.5062711131410853, + 0.4974136714556532, + 0.5108945869853301, + 0.49068514575365074, + 0.4968001010992515, + 0.4966795679733759, + 0.512475446411385, + 0.5025780354411011, + 0.5057489000131004, + 0.5009125132145201, + 0.4948752946946778, + 0.49797756628254475, + 0.5005738959483365, + 0.48797604818140644, + 0.4957874609595783, + 0.5060977411857753, + 0.5003383520050655, + 0.5006419343981396, + 0.4874263097579386, + 0.49953122335910566, + 0.5028944892020231, + 0.48823700127286274, + 0.48236515039719546, + 0.5051775864997534, + 0.5043596950679561, + 0.5005261768946487, + 0.4956727741276576, + 0.5188153054380535, + 0.5238230364751099, + 0.49729545694042826, + 0.4912357951013535, + 0.49230612032126747, + 0.49431815048987227, + 0.5209873631110609, + 0.5029105890689639, + 0.5003920063009135, + 0.4963542070916296, + 0.5026973631975495, + 0.49452781108774085, + 0.4954031345604581, + 0.5017292709854331, + 0.5027008368261059, + 0.4967715594089379, + 0.4878622855451258, + 0.49109361208998925, + 0.5062144521670456, + 0.5120724669991513, + 0.5140852060534632, + 0.5190627174116791, + 0.49537069159286495, + 0.5003799197564721, + 0.4926051269221496, + 0.5037625643878447, + 0.47152471738473345, + 0.5051814716543233, + 0.51195909985224, + 0.49772383022766703, + 0.4987157969194321, + 0.5007466743599083, + 0.50017823969796, + 0.49206577038458943, + 0.5104338646075045, + 0.5073379223751437, + 0.4756711343842684, + 0.5091896249190504, + 0.514353423913593, + 0.5013227437434381, + 0.49865932913272254, + 0.5007862556659523, + 0.5025442347797922, + 0.478916643367747, + 0.48966353136306046, + 0.506667542734481, + 0.5111580044014878, + 0.4938701648300215, + 0.5015932932480325, + 0.49344640464071754, + 0.48960292297248875, + null, + 0.502614804969183, + 0.490841541098033, + 0.4845298535444401, + 0.506287418957008, + 0.4903712642120646, + 0.4831821864900417, + 0.49174673608757347, + 0.5086311708204517, + 0.5032143703396292, + 0.504367688810655, + null, + 0.5002954953094954, + 0.49818435984365933, + 0.4879418819962247, + 0.4812610910421819, + 0.5025222517608471, + 0.49080512117375674, + 0.48761624484716987, + 0.5113039721353246, + 0.4876702397981494, + 0.5038750673723258, + 0.5026992614350868, + 0.48934517550900825, + 0.5026650462788705, + 0.497823875608665, + 0.5109218615856616, + 0.48910074027132416, + 0.4940762701840969, + 0.5083598469765616, + 0.5072700680596041, + 0.5094123465904163, + 0.5063208939615734, + 0.49528864731223654, + 0.5103146168702304, + 0.5053583700068582, + 0.5092782190951964, + 0.5188306159120232, + 0.5037787688723149, + 0.49162101757602267, + 0.49924716386274337, + 0.49943601342541927, + 0.4965442604474707, + 0.47887115771566185, + 0.4614813508772625, + 0.49668508120173754, + 0.49118208973231375, + 0.48721021204130444, + 0.4907476542529516, + 0.4917988761930027, + 0.4789683941115329, + 0.5023385145032427, + 0.5078526701321194, + 0.49290188672231666, + 0.5073408374616721, + 0.4947136797532633, + 0.49572963550101357, + 0.4855879213384176, + 0.497454564785313, + 0.49110352134132723, + 0.4780553551240411, + 0.504172003589043, + 0.5004186293427657, + 0.5028707166514276, + 0.4895950989706533, + 0.5031111189326108, + 0.5063005170340376, + 0.5072030859253911, + 0.49618854059548234, + 0.49350596210506487, + 0.49429879724796427, + 0.4931925383142263, + 0.47545443568358353, + 0.5149780140052159, + 0.5129093014573186, + null, + 0.5005280649342929, + 0.5113993919254878, + 0.4927967241605592, + 0.5015868906027487, + 0.5092722635923603, + 0.5094847603616, + 0.49222982922908354, + null, + null, + 0.5221952751090022, + 0.5197175959167141, + null, + 0.4908344562379301, + 0.45588174097521966, + 0.4856201867434241, + 0.5049656161042386, + 0.5077995609922562, + 0.5032762287710718, + 0.49689244868330185, + null, + 0.5150456505825161, + 0.4950260413747104, + null, + 0.46306423095461724, + null, + 0.4768737855372149, + 0.5092133289915731, + 0.503701743278966, + 0.5091364000985942, + null, + 0.4907268757729573, + 0.4974334627853101, + 0.4990699135692397, + null, + 0.49046096175759746, + null, + 0.48705252394100745, + null, + 0.4791963826180116, + 0.4978142332883467, + null, + 0.4792606866791589, + 0.49619894931042735, + null, + null, + 0.5008750054197084, + 0.5005778130478225, + 0.5020101801230512, + 0.49642748668160064, + null, + 0.49861650560445314, + null, + 0.5076415798334466, + 0.5098862622405239, + null, + 0.43132134076983947, + 0.49375846951368024, + 0.5074138251976029, + null, + 0.497607585635319, + 0.513487633004318, + null, + 0.5059931654461999, + 0.5085997916380378, + 0.5018261381726438, + null, + null, + null, + null, + 0.5109401851390607, + null, + null, + 0.5024792757874406, + 0.5075409115044128, + 0.5048853237705264, + null, + 0.5105756804871788, + 0.5055089374401858, + null, + null, + 0.5111818810834975, + 0.5032768328512197, + 0.49497790448239515, + 0.5066051595932863, + 0.5078026480960142, + 0.5010060882416154, + 0.5019331943493277, + 0.5031595429271395, + 0.5010922193819353, + 0.49188136780383623, + 0.49394950472328336, + 0.5053118789784262, + null, + 0.5062212799999133, + 0.503023157505866, + 0.5017298829733908, + 0.49154202519361867, + 0.4775673104402993, + 0.46388635052258825, + 0.5056951839150371, + 0.489781168052466, + 0.48199368399824427, + 0.49789891176943135, + 0.5013497181973361, + 0.49851960563197817, + null, + 0.5036053796322563, + 0.5160912608253392, + 0.5064637721723385, + null, + 0.496722075422986, + 0.43511675279583534, + null, + 0.5137529150039529, + null, + 0.5008812594566373, + 0.5033432923976113, + 0.5045225120100553, + 0.4814470499631047, + 0.510937332574756, + 0.47848800269159464, + 0.4831908397910421, + null, + null, + 0.4933685815694598, + 0.4748737846767709, + null, + 0.49152072286468185, + null, + 0.49718786823005484, + 0.4957594935675948, + 0.491245126317582, + 0.49550412417461315, + null, + null, + null, + null, + 0.49662144214214615, + null, + 0.4950824936498983, + 0.5156188987989483, + null, + null, + 0.494540288908813, + 0.5136621793772731, + 0.5084926418924702, + null, + 0.4965094913163455, + 0.494264374714062, + 0.4914583028023446, + null, + 0.5073254969479952, + null, + null, + 0.5117741687511295, + null, + 0.5058519873100807, + null, + null, + 0.4908316406208329, + 0.4920662651445116, + 0.4937792974165969, + 0.4904037351525894, + 0.49254199287550665, + null, + 0.4972975008728884, + 0.5108784382947877, + 0.5263951491950454, + 0.4482788823594938, + 0.46025562632698935, + 0.48785750728627986, + 0.500161025385393, + 0.5029587785814158, + 0.5215407199434274, + 0.5191683103677398, + 0.5315727066133764, + 0.5238494373309527, + 0.4710130141912941, + 0.45317394223236873, + 0.4584175394077517, + 0.45219660735789485, + 0.44984135305883094, + 0.5197266534020588, + 0.521232785726285, + 0.4957404632497528, + 0.5033407252078437, + 0.4760457199998031, + 0.45836354855729616, + 0.44704956158967446, + 0.43341127513110694, + 0.5264094926611487, + 0.520227782084078, + 0.5256705864049921, + 0.5130092179605114, + 0.45392500560346966, + 0.44735011010129144, + 0.46787033793798066, + 0.45940933796446853, + 0.4515923218694278, + 0.480263787982495, + 0.42620577213611327, + 0.4957767410793126, + 0.42216535361053353, + 0.5124768301493802, + 0.5185922422111626, + 0.44939633948562946, + 0.4412937640772016, + 0.46241217106252513, + 0.4725075456132841, + 0.48406092112949556, + 0.4957015879067345, + 0.44054850418389746, + 0.47192505716460426, + 0.4504924791926541, + 0.48119712326215075, + 0.4627411199438502, + 0.48051940404470284, + 0.4495010913086146, + 0.49792580561526634, + 0.4859910645938626, + 0.45357812615750914, + 0.48419912425629924, + 0.5091290068587783, + 0.5079880228271829, + 0.5081546772205575, + 0.5228001666275348, + 0.5081769317237368, + 0.5086398224091196, + 0.4955424937325033, + 0.526286780162525, + 0.5199214404952242, + 0.5115349380063114, + 0.45796470410789974, + 0.5391699733196029, + 0.5273253070472754, + 0.4585758536726812, + 0.5018077077584541, + 0.5236498340990384, + 0.49457808869825065, + 0.49665780026616074, + 0.5021763243302132, + 0.490591788986027, + 0.49090560525860166, + 0.5156730180291749, + 0.5102910725218318, + 0.5166932431304989, + 0.5303177376054179, + 0.5279820420427873, + 0.49338212207003457, + 0.530133695153471, + 0.5091987602642882, + 0.5083272939604772, + 0.4960073642328381, + 0.5115313945078963, + 0.5118270221694302, + 0.527301288203368, + 0.5087566329757829, + 0.4964973137533834, + 0.5133283062966211, + 0.5246310547583818, + 0.5172058620904454, + 0.5258329753136811, + 0.5058614804213433, + 0.5270055723797519, + 0.48735971005705264, + 0.5265587828742609, + 0.5204531095734044, + 0.522431268288519, + 0.5063284075417422, + 0.5138892115585518, + 0.5268458367624805, + 0.5239478912123564, + 0.47868844494373186, + 0.5255589201541474, + 0.4892519986704585, + 0.491841762300767, + 0.5268948212486942, + 0.48236813847726934, + 0.500141198781922, + 0.5037690572564945, + 0.5015058090396194, + 0.5215609106594549, + 0.5153611232054326, + 0.5225440966705401, + 0.4650257373661712, + 0.5120290991842964, + 0.5278582909241023, + 0.5297088769988835, + 0.51482469903841, + 0.5144191626231827, + 0.5245370777373306, + null, + 0.5077067446909759, + 0.45097168680880684, + 0.5065865285330874, + 0.5205779387584856, + 0.51545708895537, + 0.5201588472484612, + 0.523461265695965, + 0.5352182332406475, + 0.5173316172619485, + 0.5272066511736107, + null, + 0.5153626135255299, + 0.5390834434221297, + 0.5219930294728748, + 0.518443005511069, + 0.5142543569686147, + 0.5356413240974812, + 0.5208289137411017, + 0.5463573931974138, + 0.538292140051457, + 0.5333330188937708, + 0.5279903376524958, + 0.5485605082029229, + 0.5371919402760165, + 0.5339725419762171, + 0.5376020993971078, + 0.527373240470085, + 0.47504937499326455, + 0.45509690494036564, + 0.4675958450229366, + 0.4605921080347825, + 0.5246571546178596, + 0.4994077023183131, + 0.5115562670403595, + 0.5106039489271236, + 0.5232554360395145, + 0.4857726714606673, + 0.45934375316886245, + 0.5157845574006722, + 0.4955236788058314, + 0.5067841779733344, + 0.5236953311724502, + 0.5117748382975222, + 0.5243156280426429, + 0.5225370435868318, + 0.5370696408347724, + 0.5349323562197733, + 0.5149781488496707, + 0.4894775785874127, + 0.5174487477328215, + 0.5154468003625395, + 0.5168669781665747, + 0.5297434173980364, + 0.4603406504773527, + 0.5288387540394204, + 0.5339336449371848, + 0.503281999401809, + 0.49411804712687957, + 0.5163691992879376, + 0.521674455755923, + 0.4977238371278123, + 0.4746115841836988, + 0.47600960313778146, + 0.5129120312126625, + 0.46926581438151266, + 0.5357805343811874, + 0.49320983523633893, + 0.48116885526087544, + 0.47303256193905646, + 0.47776986144844513, + 0.47692944785029046, + 0.5120376526281273, + 0.5001827284148881, + 0.4783304214743772, + null, + 0.47644045839024746, + 0.4571868798100845, + 0.4650570998333295, + 0.4555322449881851, + 0.4877814637177251, + 0.5063697765926244, + 0.47081488436392793, + null, + null, + 0.48064961598979855, + 0.4598836218738352, + null, + 0.48464345206319204, + 0.4231976599831279, + 0.4850443316347307, + 0.4598126963061684, + 0.46292732467268244, + 0.46592367531935663, + 0.499702204221969, + null, + 0.4903824869864021, + 0.4775304235363345, + null, + 0.46891933179761924, + null, + 0.4785399970814796, + 0.47506302234741593, + 0.4756046621215036, + 0.48296035105110124, + null, + 0.41893341803308076, + 0.4617909305874866, + 0.44243977308610605, + null, + 0.41393181295517856, + null, + 0.45575160152343436, + null, + 0.473043829555896, + 0.463814816001889, + null, + 0.4271656278058636, + 0.4985402229241657, + null, + null, + 0.5037576371082191, + 0.44394607752786014, + 0.4438915362756471, + 0.426546371777556, + null, + 0.4506449452016572, + null, + 0.46564771676446026, + 0.45407661296237106, + null, + 0.4146039157028097, + 0.438731215342233, + 0.44972757724469253, + null, + 0.4414719652413239, + 0.4514441285420849, + null, + 0.506434913606644, + 0.49677754602718543, + 0.4918404502766455, + null, + null, + null, + null, + 0.4949033728670038, + null, + null, + 0.4758786086744468, + 0.45094545516134366, + 0.49000633221022594, + null, + 0.5011989753176886, + 0.48074164526376445, + null, + null, + 0.45271942269358106, + 0.44338613708545427, + 0.4678299373351923, + 0.46075569496936325, + 0.4641665079102173, + 0.4686180818201562, + 0.44814872546173806, + 0.4664927433698014, + 0.44479020359497795, + 0.4725589987615211, + 0.44606686389734307, + 0.4651037850716848, + null, + 0.47052982691809564, + 0.4382315373828364, + 0.47672906795136566, + 0.47312750406097875, + 0.4781193662156092, + 0.47715424852641897, + 0.4583442337192692, + 0.4710056989514776, + 0.4816981191225852, + 0.4419930806528291, + 0.43151939953680096, + 0.43041685455787276, + null, + 0.4476352396202182, + 0.49967533673989445, + 0.4894124921667289, + null, + 0.4536619876723447, + 0.4607592319211442, + null, + 0.46973922207518387, + null, + 0.4640890851282806, + 0.44871503639529736, + 0.48306349170191587, + 0.4705118959716329, + 0.48622226491700454, + 0.45808932873252894, + 0.4524890812460232, + null, + null, + 0.4742440435835831, + 0.5030458195258736, + null, + 0.4675634153336523, + null, + 0.48782918447986595, + 0.45955018216242594, + 0.4314791220416161, + 0.4921682577833002, + null, + null, + null, + null, + 0.46433963715480775, + null, + 0.47592750989907673, + 0.4929845665330774, + null, + null, + 0.46759977193607816, + 0.4479338743183979, + 0.443123376489964, + null, + 0.42825013813687773, + 0.4326313955135741, + 0.4569661574420624, + null, + 0.4780321586036264, + null, + null, + 0.4500357801802411, + null, + 0.4383367486913198, + null, + null, + 0.468343577641436, + 0.49119395932832954, + 0.4907496668550386, + 0.49312108462019133, + 0.49553627010945067, + null, + 0.46228918489732906 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768, + 0.5722914307285768 + ] + }, + "junctions": { + "mean_pearson": 0.34424864485527146, + "valid_tracks": 626, + "per_track_pearson": [ + 0.5029763574165507, + 0.37082388739589645, + 0.3314855525260907, + 0.4057548188856357, + 0.40396551696697885, + 0.4730176925077274, + 0.050787674183595595, + 0.19979546530428807, + 0.46110747796118734, + 0.4036300684837117, + 0.3813341985477873, + 0.3923734123954284, + 0.24963372062377284, + 0.3021106454495029, + 0.31716879263826153, + 0.29894957181715187, + 0.3105212828983179, + 0.3227491036137914, + 0.35217592277334825, + 0.31212928611060775, + 0.3757972278355271, + 0.4963529661684754, + 0.27054756590587387, + 0.3230219985134011, + 0.37320802476669857, + 0.33373898775109234, + 0.29282756545001026, + 0.41170214908853897, + 0.3815017366380592, + 0.4028042018395275, + 0.3320181653689958, + 0.3599507156681395, + 0.43704206189019307, + 0.4327001413574128, + 0.4696258852244854, + 0.4267213738992781, + 0.4486381704231871, + 0.4477147679262089, + 0.3647892621641368, + 0.31505995785437185, + 0.31271008483394624, + 0.3383778212103373, + 0.2933504647226762, + 0.21728774854264596, + 0.37438446543782994, + 0.2909021107024816, + 0.3338481427433396, + 0.3802070104672648, + 0.2547073847903455, + 0.3898619537528743, + 0.3445277451606808, + 0.4390922462699218, + 0.2955924640064974, + 0.4238752645773749, + 0.2148646903108863, + 0.21323493304992616, + 0.30639807078713993, + 0.16657979304048914, + 0.2625899080036474, + 0.41499297910275074, + 0.21849634854945912, + 0.2727494798529781, + 0.2785661764877776, + 0.26807552290418335, + 0.3194894791547785, + 0.2701756456405704, + 0.3396901469635563, + 0.30347298963581687, + 0.3657669697567897, + 0.29436089033108853, + 0.25472536006320184, + 0.4139132805617079, + 0.2922041219428312, + 0.23039668415602868, + 0.33136611166042684, + 0.38743087530625675, + 0.3856901863969696, + 0.3281331123669043, + 0.2782795505972935, + 0.32368930337948254, + 0.3432922030804899, + 0.3312667121009351, + 0.36523540203045274, + 0.3948009542542065, + 0.3832672143509132, + 0.32593400737520384, + 0.28247588109022503, + 0.3421656111402335, + 0.26602400685111816, + 0.3209741711873125, + 0.25824692927137816, + 0.3217814521271561, + 0.2803478560708846, + 0.46738821965170757, + 0.3835114268046755, + 0.16861850577932044, + 0.34365979315012457, + 0.4098860183788587, + 0.48298068150008167, + 0.25678565973781053, + 0.39305730371884334, + 0.3306336734075644, + 0.29796550772922625, + 0.35790835706285984, + 0.29822706749805344, + 0.2795610942825794, + 0.2648630900798008, + 0.4369001521624768, + 0.22883597023913455, + 0.25096000386984674, + 0.3359664585508472, + 0.23705246670847777, + 0.3505227724163581, + 0.3118815991746065, + 0.19668102896245135, + 0.33238181960537905, + 0.27075575395319906, + 0.34190973382173884, + 0.28435867986449026, + 0.35313540233809804, + 0.31625664058116454, + 0.3750172584866176, + 0.38136017806857697, + 0.24540262495149634, + 0.22821934534397423, + null, + 0.35006677989209456, + 0.434589491681224, + 0.21939872471763627, + 0.21086873522986957, + 0.37687983964096455, + 0.4118589766084398, + 0.15997075916473735, + 0.47135806179010087, + 0.4250457216416712, + 0.3691913847300031, + null, + 0.41955106734141195, + 0.22889055877430625, + 0.3972361464708338, + 0.30226234423480475, + 0.3249054317997448, + 0.20262641011154706, + 0.2574498144157353, + 0.3688159079217239, + 0.16031566690621116, + 0.32601333351022904, + 0.26917141048049936, + 0.1626807130334249, + 0.26581075756659744, + 0.23555196442661178, + 0.37072555972188537, + 0.32665276594329185, + 0.4866155286922019, + 0.3224000246353449, + 0.48052170673549444, + 0.4332295267134404, + 0.342851113434751, + 0.363902846017331, + 0.41906524090866815, + 0.44018923685393263, + 0.35726164851587455, + 0.5362516820269843, + 0.44342207152410434, + 0.1575343194150834, + 0.44359482626079927, + 0.46994511170818615, + 0.34358459621347587, + 0.1820566413996169, + 0.16024277991692326, + 0.23277180013064494, + 0.21277833958692852, + 0.42838924002317286, + 0.20388804024822027, + 0.2851356148594354, + 0.33743124515654827, + 0.3315760710359815, + 0.2168453850653809, + 0.2625139096639995, + 0.2878486504164618, + 0.26239026471976434, + 0.18289225026219985, + 0.23899013561841526, + 0.44946476699118587, + 0.37612713289572103, + 0.3362008533358516, + 0.4221896347667029, + 0.2892710434573791, + 0.36126926409469917, + 0.4222844997737353, + 0.49899761179516333, + 0.4473867545580512, + 0.34286766949354364, + 0.3949169911372975, + 0.32532609760327375, + 0.30608037631867596, + 0.39833071705710427, + 0.24818477952716553, + 0.5383144295650958, + 0.3853901198293971, + null, + 0.35185546559394787, + 0.30368773918006, + 0.47481950675607465, + 0.3932644622234611, + 0.2183800505179983, + 0.33393388645761246, + 0.27986196600354385, + null, + null, + 0.4079300447783786, + 0.3494497440627174, + null, + 0.25259973340287856, + 0.30574528124063033, + 0.3192819551138251, + 0.35332231866000324, + 0.23482213975803817, + 0.2534054956846265, + 0.24661152129564962, + null, + 0.4358536878279073, + 0.37055296597569193, + null, + 0.2215392282748315, + null, + 0.4084717660280696, + 0.33603993858458675, + 0.36041621546456526, + 0.3287706772701672, + null, + 0.19602712471450878, + 0.21875594260111567, + 0.3233938005177916, + null, + 0.2924738995337612, + null, + 0.4173346850443165, + null, + 0.17784626070509396, + 0.28522160124247364, + null, + 0.20753253743560324, + 0.44326103725174176, + null, + null, + 0.34422740691595344, + 0.33051772188180145, + 0.2978477552256007, + 0.29926033872823277, + null, + 0.43577840802177065, + null, + 0.3280129034315887, + 0.1819485175956551, + null, + 0.3078032984298269, + 0.4432608239086646, + 0.41851154846042155, + null, + 0.30585769608502295, + 0.3219598328488824, + null, + 0.3178527013504475, + 0.3622434997151898, + 0.3346296501146654, + null, + null, + null, + null, + 0.2942297576354501, + null, + null, + 0.24077760111098434, + 0.26036594911677036, + 0.36035972170356045, + null, + 0.44960291458562923, + 0.4182050546385034, + null, + null, + 0.3440699797012067, + 0.37229500564599516, + 0.29905987139112683, + 0.2738212411728233, + 0.31906306939262774, + 0.3098771294070841, + 0.3154446063467663, + 0.3867224435739062, + 0.262128126739878, + 0.24237009445231403, + 0.3166533206349856, + 0.35233198544488803, + null, + 0.3556068980423577, + 0.4662759305955598, + 0.1154914335366013, + 0.2197618839693959, + 0.1565668630879322, + 0.3077918354936819, + 0.3736229932597095, + 0.329781959823656, + 0.42822654991246195, + 0.36614075785221706, + 0.468259164215391, + 0.3747553907512016, + null, + 0.2604248432382437, + 0.2809965255308675, + 0.2670410270537655, + null, + 0.250551604720002, + 0.2629054794652347, + null, + 0.4055083498274123, + null, + 0.40990940232383266, + 0.4070570887870525, + 0.44832917680472806, + 0.32324452645527374, + 0.31584705003552743, + 0.24927789097099373, + 0.2710626603467585, + null, + null, + 0.30043191651216916, + 0.3401928383172854, + null, + 0.32427200532453876, + null, + 0.35265336121031116, + 0.2656308312934186, + 0.33498921328984355, + 0.42423610026786296, + null, + null, + null, + null, + 0.259234634587505, + null, + 0.2990766788000501, + 0.44143229816829505, + null, + null, + 0.35957196982716205, + 0.45031887394397246, + 0.32468934945976813, + null, + 0.4260841600851864, + 0.37130805275763434, + 0.3572854122825368, + null, + 0.39151961993906753, + null, + null, + 0.35618035680544957, + null, + 0.2783073513719754, + null, + null, + 0.2703062247133687, + 0.4648473823114309, + 0.4028648250075936, + 0.42252611400261414, + 0.41934007305282195, + null, + 0.3734609402739223, + 0.4027225915293037, + 0.36934333608062925, + 0.5002594386661365, + 0.26091349977414907, + 0.40740773283021475, + 0.4086060001518342, + 0.1568960855177397, + 0.19364016229809544, + 0.4418598512352038, + 0.3296094457704511, + 0.2152526273965153, + 0.4137769079170783, + 0.3141346112302822, + 0.2571741838293234, + 0.28806073425411816, + 0.27964456371742397, + 0.3560697214740108, + 0.4077890881792769, + 0.3892935295383747, + 0.4952314087670342, + 0.4540336947391619, + 0.3893006675680478, + 0.3443440800597419, + 0.5296125311367653, + 0.3850217187067975, + 0.3869891523725536, + 0.38820927897327256, + 0.4481634268164613, + 0.4082557605373662, + 0.2168819446821564, + 0.36138109936344337, + 0.5268676591935757, + 0.32811408630821054, + 0.49187747808813936, + 0.2933324635823118, + 0.38514011163153467, + 0.20423017783284206, + 0.3373257592149033, + 0.31777033769668317, + 0.31872388456553646, + 0.3207990617429002, + 0.25242055549316905, + 0.18632585780420255, + 0.24158042209325645, + 0.2765115777310275, + 0.3611469813868141, + 0.2529487740663908, + 0.29737798299707463, + 0.24717828044670778, + 0.3309393056939781, + 0.2696636325052646, + 0.30791543153237677, + 0.2177982704314987, + 0.20026118154518976, + 0.17814915959211375, + 0.34466483372181766, + 0.35210667202647156, + 0.3347983789442066, + 0.33930787326872175, + 0.34460318502698967, + 0.4450555552102906, + 0.4129770051674854, + 0.42788606608252516, + 0.3577636796182445, + 0.22165520279182946, + 0.25462079450867775, + 0.2843724437294176, + 0.2273063538179144, + 0.190537910292144, + 0.2915680599424275, + 0.41769068427290335, + 0.30985123950087107, + 0.39582134662916724, + 0.3441133943503057, + 0.3212356453563098, + 0.2745728653847935, + 0.5103011504973178, + 0.3208443862276263, + 0.4057510583006487, + 0.4001132922482105, + 0.3363665304055505, + 0.36388517472218124, + 0.43656683992302026, + 0.33353759247323667, + 0.3294432872113302, + 0.2888634446577792, + 0.44686114903323343, + 0.43596982898320913, + 0.29266867816135667, + 0.3591888940390206, + 0.4572072042334276, + 0.34693496993405615, + 0.3166533688547743, + 0.31183156182289473, + 0.26120616807141295, + 0.2920692380483156, + 0.2971999421521435, + 0.30772170807223487, + 0.44833887933655364, + 0.41499351795490597, + 0.25197138496371724, + 0.3549604293356454, + 0.39698492981675265, + 0.3616925321546201, + 0.347876958009192, + 0.39073935527577325, + 0.3183877583193275, + 0.3507142608634738, + 0.36870711385759625, + 0.3443945342444313, + 0.31583543510239, + 0.3880041845741022, + 0.27358463176058934, + 0.4528711445357302, + 0.42393937316064945, + 0.27005440362365024, + 0.37906614575401004, + 0.4137466228356631, + 0.2863551767670922, + 0.24585937225888368, + 0.40390037033298576, + 0.31570948850475034, + 0.2640322421606738, + 0.34837692388344654, + 0.4764732136170516, + null, + 0.3757494801050344, + 0.49333665741415234, + 0.291569383922467, + 0.42770913862434345, + 0.32231917526117576, + 0.22651903303142037, + 0.29116806942706064, + 0.3654726567440532, + 0.3298005875450774, + 0.2599377505098924, + null, + 0.26690685282102955, + 0.25305490271057857, + 0.24302728864869258, + 0.18556020712494026, + 0.3149535533256333, + 0.3223727334407022, + 0.29595677809648674, + 0.39628484409550785, + 0.08516465816652395, + 0.23435201480204032, + 0.38803728042510466, + 0.2061774606121098, + 0.29485118520875203, + 0.425159699241052, + 0.4074703132862793, + 0.39442812263938154, + 0.46446592810385817, + 0.4936327209800702, + 0.5130446815801617, + 0.39824894387447357, + 0.3435066287715732, + 0.3072864167490039, + 0.2519897399202325, + 0.26285457811078056, + 0.24135700538203061, + 0.3636030785904994, + 0.3933476362934378, + 0.33889832910877266, + 0.32846922590886896, + 0.25127633292533275, + 0.3180988498676006, + 0.405795323457643, + 0.3035577173473837, + 0.3644649211340472, + 0.3416935397843478, + 0.4171622776524643, + 0.2068887807205584, + 0.3000961116543331, + 0.25932037495212085, + 0.42891775084157835, + 0.29658192006126144, + 0.3011789029090358, + 0.308586628844194, + 0.27487654270527156, + 0.2643196504990872, + 0.36393479661775807, + 0.39413256128038426, + 0.4719292459358487, + 0.4654094627843481, + 0.4244074285409743, + 0.24407718840499495, + 0.3465644395773272, + 0.3752265064937146, + 0.4312022067153957, + 0.3855270176098305, + 0.2436882077929915, + 0.24247664290836102, + 0.3046285749826262, + 0.3161487379009066, + 0.3047659666518319, + 0.3254478306949094, + 0.4613037829713755, + 0.6093982222873546, + null, + 0.48111552029898536, + 0.4300126436852335, + 0.4920519289907408, + 0.5285985504915264, + 0.603864268873932, + 0.35339245551834003, + 0.3571535491251207, + null, + null, + 0.3326406327811641, + 0.34873348976765167, + null, + 0.23680418932670386, + 0.3814073444239816, + 0.27695828872465533, + 0.3241156253698733, + 0.3609946527357062, + 0.30022710021493504, + 0.29666942097592314, + null, + 0.4576854896400251, + 0.39243146823036795, + null, + 0.34760126371048355, + null, + 0.5605155441580661, + 0.47859547245627077, + 0.4225145911406498, + 0.3072254077110412, + null, + 0.4340489753047941, + 0.3106300885334176, + 0.37528361274617356, + null, + 0.30858108618384167, + null, + 0.2705799343274551, + null, + 0.2957282360825081, + 0.3631267382081268, + null, + 0.374788939299142, + 0.4722593223247261, + null, + null, + 0.4378823399147161, + 0.2864526989856023, + 0.1743224239108085, + 0.29927257786678413, + null, + 0.43503899397319357, + null, + 0.23550638764202525, + 0.4469867948883321, + null, + 0.22263009930316588, + 0.4097613313934404, + 0.2895005218540942, + null, + 0.38048325829119184, + 0.5109555657004327, + null, + 0.2956682653294831, + 0.3780307134383597, + 0.3390777200198378, + null, + null, + null, + null, + 0.299334985516662, + null, + null, + 0.5036469941328053, + 0.46867081227195817, + 0.3457774728382619, + null, + 0.29911228676118384, + 0.3533417601032773, + null, + null, + 0.3889488765998235, + 0.13695016476299524, + 0.42491489668640503, + 0.3230873539632308, + 0.2872795523049538, + 0.3750687413218865, + 0.0920085339126971, + 0.250681429602651, + 0.20526839970122518, + 0.36956951826375795, + 0.5146768345491682, + 0.2403329902313304, + null, + 0.32892848700773863, + 0.47554132724657355, + 0.5237621698106573, + 0.4537043315889291, + 0.19007627696188678, + 0.42095811589001214, + 0.4157438013499477, + 0.4171634720442714, + 0.4159942216775851, + 0.4537152942463725, + 0.42989977941127533, + 0.3540342929617274, + null, + 0.36403099503842157, + 0.40987828614232813, + 0.3002678488259285, + null, + 0.45904722295513034, + 0.2895475926157911, + null, + 0.39930247900471316, + null, + 0.5498209732033789, + 0.48846781990257365, + 0.27817284789907226, + 0.40542755340508385, + 0.5540249142814184, + 0.4742034639366208, + 0.49086388663387265, + null, + null, + 0.5076928642146178, + 0.5668632696079504, + null, + 0.3413322265735693, + null, + 0.35956423997389375, + 0.39885745644297205, + 0.4373778665462791, + 0.506498194124974, + null, + null, + null, + null, + 0.33046389448496294, + null, + 0.4043537423400705, + 0.5984393629806792, + null, + null, + 0.45720794878032267, + 0.24426914211848105, + 0.33320744789558365, + null, + 0.31670007244206966, + 0.4052045981840863, + 0.38271507937291227, + null, + 0.35157463940261874, + null, + null, + 0.40383375439755653, + null, + 0.30497641939861314, + null, + null, + 0.42792277348658486, + 0.5455348717268961, + 0.510783025617285, + 0.5861827139104682, + 0.5080964751448901, + null, + 0.462044461377825 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.8841059602649006, + 0.8947368421052632, + 0.8932384341637011, + 0.8987341772151899, + 0.9006622516556292, + 0.8745762711864407, + 0.8838028169014085, + 0.8784722222222222, + 0.8974358974358975, + 0.875886524822695, + 0.8787878787878788, + 0.864406779661017, + 0.8896103896103896, + 0.86875, + 0.9, + 0.8980263157894737, + 0.8601398601398601, + 0.875, + 0.8949044585987261, + 0.8888888888888888, + 0.8776978417266187, + 0.9032258064516129, + 0.8863636363636364, + 0.8847583643122676, + 0.8754448398576512, + 0.9046052631578947, + 0.9102564102564102, + 0.8949152542372881, + 0.8783783783783784, + 0.8896551724137931, + 0.8767123287671232, + 0.9029126213592233, + 0.8880597014925373, + 0.9057750759878419, + 0.8791208791208791, + 0.9053627760252366, + 0.8893129770992366, + 0.8871473354231975, + 0.8566433566433567, + 0.8825503355704698, + 0.8873720136518771, + 0.9003436426116839, + 0.8816793893129771, + 0.8814814814814815, + 0.8885245901639345, + 0.8952702702702703, + 0.8856088560885609, + 0.8853046594982079, + 0.8475177304964538, + 0.8979591836734694, + 0.8868613138686131, + 0.9044117647058824, + 0.8636363636363636, + 0.9013605442176871, + 0.8656716417910447, + 0.8809523809523809, + 0.889273356401384, + 0.8675496688741722, + 0.8664383561643836, + 0.8838028169014085, + 0.8827586206896552, + 0.8819672131147541, + 0.9139072847682119, + 0.8873720136518771, + 0.87248322147651, + 0.9177215189873418, + 0.8885135135135135, + 0.87248322147651, + 0.863481228668942, + 0.8954248366013072, + 0.8896797153024911, + 0.8838709677419355, + 0.8754208754208754, + 0.9039735099337748, + 0.8817204301075269, + 0.888135593220339, + 0.9035369774919614, + 0.8762541806020067, + 0.8758620689655172, + 0.8877551020408163, + 0.8874172185430463, + 0.8892405063291139, + 0.8904109589041096, + 0.883495145631068, + 0.8947368421052632, + 0.8947368421052632, + 0.8972602739726028, + 0.8725490196078431, + 0.8821428571428571, + 0.8892405063291139, + 0.8815331010452961, + 0.8969072164948454, + 0.8842105263157894, + 0.891156462585034, + 0.875, + 0.8715277777777778, + 0.8813559322033898, + 0.8685121107266436, + 0.8807947019867549, + 0.8860759493670886, + 0.8866666666666667, + 0.9049180327868852, + 0.8861209964412812, + 0.8977635782747604, + 0.9146757679180887, + 0.8866666666666667, + 0.8951310861423221, + 0.8963210702341137, + 0.8837209302325582, + 0.9087837837837838, + 0.875, + 0.889261744966443, + 0.8961038961038961, + 0.888135593220339, + 0.9081632653061225, + 0.8885135135135135, + 0.8586206896551725, + 0.8783783783783784, + 0.8628158844765343, + 0.9010238907849829, + 0.8678571428571429, + 0.8847457627118644, + 0.8602941176470589, + 0.9013157894736842, + 0.854251012145749, + 0.9838709677419355, + 0.8823529411764706, + 0.8850174216027874, + 0.8863636363636364, + 0.8880597014925373, + 0.8695652173913043, + 0.896551724137931, + 0.8850174216027874, + 0.8947368421052632, + 0.8771929824561403, + 0.882943143812709, + 0.9838709677419355, + 0.9016949152542373, + 0.8805970149253731, + 0.9013157894736842, + 0.9051094890510949, + 0.8582089552238806, + 0.8970588235294118, + 0.8827586206896552, + 0.887719298245614, + 0.8625954198473282, + 0.8858267716535433, + 0.8611111111111112, + 0.8529411764705882, + 0.872791519434629, + 0.8703703703703703, + 0.8794326241134752, + 0.8791208791208791, + 0.9058441558441559, + 0.896551724137931, + 0.8797250859106529, + 0.8787878787878788, + 0.8989169675090253, + 0.872791519434629, + 0.8978102189781022, + 0.8674242424242424, + 0.8611111111111112, + 0.8840125391849529, + 0.8815331010452961, + 0.8528301886792453, + 0.8884892086330936, + 0.8996539792387543, + 0.9006622516556292, + 0.8754716981132076, + 0.8807017543859649, + 0.8791208791208791, + 0.8805460750853242, + 0.8909774436090225, + 0.8801369863013698, + 0.8692307692307693, + 0.8793774319066148, + 0.8804347826086957, + 0.8815331010452961, + 0.8786764705882353, + 0.8706293706293706, + 0.8815331010452961, + 0.8582089552238806, + 0.8560885608856088, + 0.8924731182795699, + 0.8767605633802817, + 0.8929889298892989, + 0.8835616438356164, + 0.8922558922558923, + 0.8781362007168458, + 0.9114754098360656, + 0.8993055555555556, + 0.9007352941176471, + 0.8821428571428571, + 0.8873239436619719, + 0.867595818815331, + 0.872791519434629, + 0.8786764705882353, + 0.8888888888888888, + 0.8954248366013072, + 0.8766666666666667, + 1.0, + 0.9111842105263158, + 0.888135593220339, + 0.8770764119601329, + 0.8682432432432432, + 0.8971061093247589, + 0.8763250883392226, + 0.875, + 1.0, + 1.0, + 0.8778877887788779, + 0.8733766233766234, + 1.0, + 0.9, + 0.8571428571428571, + 0.864963503649635, + 0.8805460750853242, + 0.8774834437086093, + 0.8620689655172413, + 0.8624161073825504, + 0.991869918699187, + 0.8788819875776398, + 0.8625429553264605, + 0.9924812030075187, + 0.8359375, + 1.0, + 0.8545454545454545, + 0.8736842105263158, + 0.8732394366197183, + 0.8686868686868687, + 0.9927536231884058, + 0.8659420289855072, + 0.8767605633802817, + 0.8807017543859649, + 1.0, + 0.8754208754208754, + 0.9923076923076923, + 0.8928571428571429, + 1.0, + 0.89, + 0.8825503355704698, + 0.9927007299270073, + 0.8773234200743495, + 0.8754098360655738, + 1.0, + 1.0, + 0.8617021276595744, + 0.8725490196078431, + 0.8986928104575164, + 0.8922558922558923, + 0.984, + 0.872791519434629, + 0.9922480620155039, + 0.8831168831168831, + 0.8694158075601375, + 1.0, + 0.8692307692307693, + 0.8745980707395499, + 0.8727272727272727, + 0.9927536231884058, + 0.85, + 0.882943143812709, + 1.0, + 0.898360655737705, + 0.8701754385964913, + 0.8779661016949153, + 1.0, + 0.9922480620155039, + 1.0, + 1.0, + 0.8770764119601329, + 1.0, + 0.984251968503937, + 0.8707482993197279, + 0.8581560283687943, + 0.8878205128205128, + 1.0, + 0.8703071672354948, + 0.886986301369863, + 0.976, + 1.0, + 0.8850174216027874, + 0.8695652173913043, + 0.8695652173913043, + 0.8952702702702703, + 0.8771331058020477, + 0.8673469387755102, + 0.8873720136518771, + 0.8754325259515571, + 0.8677966101694915, + 0.8576512455516014, + 0.8980891719745223, + 0.8671586715867159, + 1.0, + 0.8842105263157894, + 0.8918918918918919, + 0.8642857142857143, + 0.8821548821548821, + 0.9047619047619048, + 0.8993055555555556, + 0.8807017543859649, + 0.8620689655172413, + 0.8754098360655738, + 0.8817567567567568, + 0.8804347826086957, + 0.8703071672354948, + 0.991869918699187, + 0.8906752411575563, + 0.8885135135135135, + 0.8846153846153846, + 0.9906542056074766, + 0.8913738019169329, + 0.8673835125448028, + 1.0, + 0.8797250859106529, + 0.9776119402985075, + 0.875, + 0.875, + 0.8745874587458746, + 0.8586956521739131, + 0.8775510204081632, + 0.8698630136986302, + 0.8811188811188811, + 1.0, + 1.0, + 0.8856209150326797, + 0.8707482993197279, + 0.9925373134328358, + 0.8700361010830325, + 1.0, + 0.8870431893687708, + 0.8859060402684564, + 0.8727272727272727, + 0.8803986710963455, + 1.0, + 0.9921875, + 0.992, + 1.0, + 0.8938356164383562, + 0.9927007299270073, + 0.8685121107266436, + 0.8825396825396825, + 0.991304347826087, + 1.0, + 0.8908450704225352, + 0.8870431893687708, + 0.8777777777777778, + 0.984, + 0.8904109589041096, + 0.8720538720538721, + 0.851063829787234, + 0.9838709677419355, + 0.910828025477707, + 0.9925373134328358, + 1.0, + 0.8839590443686007, + 0.9918032786885246, + 0.889967637540453, + 0.984251968503937, + 1.0, + 0.8494208494208494, + 0.8798701298701299, + 0.8797250859106529, + 0.8762886597938144, + 0.8661971830985915, + 0.9767441860465116, + 0.8774834437086093, + 0.8120300751879699, + 0.8105263157894737, + 0.8115942028985508, + 0.8107142857142857, + 0.8734177215189873, + 0.8149466192170819, + 0.8257839721254355, + 0.7811320754716982, + 0.8498293515358362, + 0.8309352517985612, + 0.8235294117647058, + 0.8231046931407943, + 0.8321917808219178, + 0.8356164383561644, + 0.8029739776951673, + 0.7985347985347986, + 0.8063380281690141, + 0.8185185185185185, + 0.8214285714285714, + 0.8361774744027304, + 0.8216783216783217, + 0.8374558303886925, + 0.8327759197324415, + 0.8270676691729323, + 0.8142857142857143, + 0.8181818181818182, + 0.8245033112582781, + 0.8466898954703833, + 0.8351254480286738, + 0.7944664031620553, + 0.8245614035087719, + 0.8272058823529411, + 0.8024193548387096, + 0.8333333333333334, + 0.8168498168498168, + 0.8439716312056738, + 0.7689243027888446, + 0.8156028368794326, + 0.8076923076923077, + 0.7985611510791367, + 0.8133802816901409, + 0.8284671532846716, + 0.8268551236749117, + 0.807843137254902, + 0.8144329896907216, + 0.8237410071942446, + 0.8045112781954887, + 0.7977099236641222, + 0.8113207547169812, + 0.8192307692307692, + 0.8152610441767069, + 0.8104089219330854, + 0.7992424242424242, + 0.8291814946619217, + 0.8120300751879699, + 0.8215613382899628, + 0.8381294964028777, + 0.8207885304659498, + 0.7900355871886121, + 0.7971014492753623, + 0.8153310104529616, + 0.8277153558052435, + 0.863481228668942, + 0.8278388278388278, + 0.8267148014440433, + 0.8051470588235294, + 0.825, + 0.7683397683397684, + 0.7908745247148289, + 0.8100358422939068, + 0.8394160583941606, + 0.8188405797101449, + 0.823728813559322, + 0.8172043010752689, + 0.8207885304659498, + 0.8043478260869565, + 0.8068965517241379, + 0.8091872791519434, + 0.7985611510791367, + 0.8226415094339623, + 0.8477508650519031, + 0.8013698630136986, + 0.8098859315589354, + 0.7962962962962963, + 0.8226950354609929, + 0.8161764705882353, + 0.8129770992366412, + 0.8229166666666666, + 0.781021897810219, + 0.8058608058608059, + 0.8327526132404182, + 0.797153024911032, + 0.8275862068965517, + 0.8215613382899628, + 0.7929824561403509, + 0.8064516129032258, + 0.8200692041522492, + 0.8066914498141264, + 0.8047945205479452, + 0.8064516129032258, + 0.8275862068965517, + 0.8138686131386861, + 0.8188405797101449, + 0.8214285714285714, + 0.8260869565217391, + 0.8229166666666666, + 0.8113207547169812, + 0.8333333333333334, + 0.8210526315789474, + 0.8256227758007118, + 0.8098591549295775, + 0.8127340823970037, + 0.8211920529801324, + 0.8100358422939068, + 0.8243727598566308, + 0.7923076923076923, + 0.8197879858657244, + 0.8, + 0.7865612648221344, + 0.7785977859778598, + 0.8210526315789474, + 0.7777777777777778, + 0.7802197802197802, + 0.833922261484099, + 0.8215613382899628, + 0.975, + 0.8475177304964538, + 0.8285714285714286, + 0.8333333333333334, + 0.7984189723320159, + 0.8158844765342961, + 0.8181818181818182, + 0.8127208480565371, + 0.7857142857142857, + 0.8222996515679443, + 0.7786259541984732, + 0.9719626168224299, + 0.8397212543554007, + 0.7830882352941176, + 0.8671328671328671, + 0.8710801393728222, + 0.79182156133829, + 0.7978723404255319, + 0.8076923076923077, + 0.8049645390070922, + 0.7732342007434945, + 0.7798507462686567, + 0.8052434456928839, + 0.8086642599277978, + 0.7985347985347986, + 0.7769230769230769, + 0.7933579335793358, + 0.8315412186379928, + 0.8413284132841329, + 0.8037735849056604, + 0.8413793103448276, + 0.8315018315018315, + 0.8345864661654135, + 0.8209459459459459, + 0.8222222222222222, + 0.8043478260869565, + 0.7948717948717948, + 0.8543689320388349, + 0.8146853146853147, + 0.8164794007490637, + 0.7992125984251969, + 0.8047945205479452, + 0.8637992831541219, + 0.7551020408163265, + 0.8104089219330854, + 0.8260869565217391, + 0.7888888888888889, + 0.8178294573643411, + 0.823321554770318, + 0.8118081180811808, + 0.8008130081300813, + 0.8014184397163121, + 0.8285714285714286, + 0.8356643356643356, + 0.8175182481751825, + 0.8222996515679443, + 0.8141263940520446, + 0.7977941176470589, + 0.7948717948717948, + 0.8450704225352113, + 0.8189300411522634, + 0.8104089219330854, + 0.8165467625899281, + 0.7746478873239436, + 0.8392857142857143, + 0.8247422680412371, + 0.8265682656826568, + 0.8621908127208481, + 0.7896678966789668, + 0.7813620071684588, + 0.8237410071942446, + 0.795774647887324, + 0.7807692307692308, + 0.864516129032258, + 0.8498293515358362, + 1.0, + 0.8333333333333334, + 0.8552188552188552, + 0.7798507462686567, + 0.8205128205128205, + 0.8493150684931506, + 0.8197278911564626, + 0.8056537102473498, + 0.9920634920634921, + 1.0, + 0.896551724137931, + 0.8664495114006515, + 0.976, + 0.8210526315789474, + 0.7889273356401384, + 0.8115384615384615, + 0.8430034129692833, + 0.8362369337979094, + 0.8256227758007118, + 0.8268551236749117, + 0.9921875, + 0.8530465949820788, + 0.8074074074074075, + 1.0, + 0.7686567164179104, + 1.0, + 0.7898305084745763, + 0.8469750889679716, + 0.8547297297297297, + 0.8350877192982457, + 0.9925925925925926, + 0.8058608058608059, + 0.8035087719298246, + 0.8161764705882353, + 0.9922480620155039, + 0.800711743772242, + 0.9908256880733946, + 0.8021201413427562, + 0.9841269841269841, + 0.8277153558052435, + 0.8254545454545454, + 0.9924812030075187, + 0.8145454545454546, + 0.8279569892473119, + 0.9851851851851852, + 0.992, + 0.8586206896551725, + 0.8445229681978799, + 0.8322368421052632, + 0.8316498316498316, + 0.9915254237288136, + 0.8287937743190662, + 0.9841269841269841, + 0.8226950354609929, + 0.832089552238806, + 0.9761904761904762, + 0.7675276752767528, + 0.8181818181818182, + 0.8104089219330854, + 0.9922480620155039, + 0.8297101449275363, + 0.8345070422535211, + 0.9831932773109243, + 0.8395904436860068, + 0.8528428093645485, + 0.8458781362007168, + 0.9779411764705882, + 0.9838709677419355, + 1.0, + 0.972972972972973, + 0.8373702422145328, + 0.981651376146789, + 0.9724770642201835, + 0.8191489361702128, + 0.8401360544217688, + 0.8648648648648649, + 0.9922480620155039, + 0.8566433566433567, + 0.8377483443708609, + 0.9747899159663865, + 0.96, + 0.8339100346020761, + 0.8281786941580757, + 0.8254545454545454, + 0.8481848184818482, + 0.8385964912280702, + 0.8211678832116789, + 0.7830882352941176, + 0.8299319727891157, + 0.8037735849056604, + 0.8, + 0.8576271186440678, + 0.8156028368794326, + 1.0, + 0.8576388888888888, + 0.8172043010752689, + 0.8231292517006803, + 0.8175438596491228, + 0.8251748251748252, + 0.8181818181818182, + 0.8082191780821918, + 0.8388278388278388, + 0.8624161073825504, + 0.8268551236749117, + 0.8074074074074075, + 0.8, + 0.984251968503937, + 0.8285714285714286, + 0.8464285714285714, + 0.8766666666666667, + 0.975609756097561, + 0.8408304498269896, + 0.8060200668896321, + 0.9836065573770492, + 0.8360655737704918, + 0.9918032786885246, + 0.8471760797342193, + 0.8469750889679716, + 0.8581081081081081, + 0.807843137254902, + 0.8424657534246576, + 0.7947761194029851, + 0.779467680608365, + 0.9754098360655737, + 0.983739837398374, + 0.8274647887323944, + 0.7885304659498208, + 0.9922480620155039, + 0.8327402135231317, + 1.0, + 0.8327645051194539, + 0.8235294117647058, + 0.8496503496503497, + 0.8345588235294118, + 0.9708029197080292, + 0.984251968503937, + 1.0, + 0.9906542056074766, + 0.8204225352112676, + 0.9826086956521739, + 0.8076923076923077, + 0.8417508417508418, + 0.975609756097561, + 0.9846153846153847, + 0.8422939068100358, + 0.8666666666666667, + 0.825, + 0.9747899159663865, + 0.8207885304659498, + 0.8327526132404182, + 0.825938566552901, + 0.9906542056074766, + 0.8831168831168831, + 0.983739837398374, + 1.0, + 0.8363636363636363, + 0.9916666666666667, + 0.8188679245283019, + 0.963302752293578, + 0.976, + 0.8228782287822878, + 0.8477508650519031, + 0.8453947368421053, + 0.8321678321678322, + 0.8271186440677966, + 0.9831932773109243, + 0.8305084745762712 + ] + }, + "contact_maps": { + "mean_pearson": 0.291844987810106, + "valid_tracks": 28, + "per_track_pearson": [ + 0.28362431477404415, + 0.28751730939281517, + 0.3435124115826578, + 0.245333058181592, + 0.1994957538760926, + 0.23288047782507057, + 0.3091124315785611, + 0.23001686665371543, + 0.3870214670808095, + 0.33832011778070825, + 0.2359001634095449, + 0.3002411717358585, + 0.33265453149154794, + 0.22221791636093818, + 0.2622484995027094, + 0.39723511649174187, + 0.08598137174223888, + 0.32653751504138295, + 0.30807693711366696, + 0.31496385407215033, + 0.2619730509545015, + 0.26770513727620354, + 0.2660459764115163, + 0.3352417281037799, + 0.35001862519824883, + 0.3567773829287114, + 0.3095593343151959, + 0.38144713780696315 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.4231233468773853, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..df2f55b46dccfb8c5d43d783e68a51939249502c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S021/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4376270662780564, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.562427223141509, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.07467288339793669, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7788222631626409, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.28406028933162125, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.23186002435742215, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.24263765003107202, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.22116493273833532, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.17527042759924516, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22344803749116265, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5277918568224231, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.19480043631800173, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.25854761884820804, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/README.md new file mode 100644 index 0000000000000000000000000000000000000000..8f0940dca922b902319d11027f9fbc48a6ca0415 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/README.md @@ -0,0 +1,106 @@ +# SeqFunc scale13-small after 2 complete epochs with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **82,040 training examples in 20,510 optimizer updates**, 2 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +296.0 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +This fixed-link follow-up uses the preselected complete two-epoch small model. +It reuses the original 256-training-window offsets without refitting or changing +network parameters, site selection, usage or junction decoding. Earlier experiment +selection pools are unchanged. Both entrypoints share the same probability link. +Research-only perturbation reuse is verified by exact code/weight checks, the +earlier complete native proof, and fresh native fixtures for this candidate. +All 200 validation windows and all 13 development metrics are scored in full. + +## Measured validation results + +Family-mean Pearson: **0.444367** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.555700 | +| dnase | 0.644481 | +| procap | 0.516338 | +| cage | 0.296582 | +| rna_seq | 0.574258 | +| chip_tf | 0.441460 | +| chip_histone | 0.469698 | +| splice_sites | 0.287262 | +| splice_usage | 0.454583 | +| junctions | 0.344881 | +| contact_maps | 0.302800 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.6192 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5299 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.1702 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7613 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3891 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.3171 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.3378 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3584 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1899 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2134 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4208 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2089 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2281 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..6fbb36253ad0b167bb556f6cb97159c8042ea6f4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch02/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..b83050ad1dfacbffb1f8d7465c4f8f3c0cc265dc --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/package_manifest.json @@ -0,0 +1,50 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "ea695c62d648264868443acbfba188e1449f3d0044a806566b47a521426e45a4", + "files": { + "README.md": "f1bbc2f604f08c9d9d8c3d5f3ab229f3a5d99c24c564d5b8a770e2069505dd62", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "775b5cebc2335ac6732b582234e6665b189ae7ca56f246c4fe3be6d0e9d1011e", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "2cce52775b363e8e2b0084a9a8aeb88bef9e4e0127b6e55da7516ba7f47dfee1", + "decoding_policy.json": "01c62ef5b4cea4e5afae34390df82f5ea67439746fa491c1f5fdba0d17c2e543", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/fixed-link46-plan.json": "9e99d351af2dbaf0d51c4a253fbc21781dfcf9dd81076c7e337f25b234c795a7", + "validation_tracks.json": "0c30649078135d1f539645c24435297a25a5b83f2ddf19ef02536f0172aafa22", + "validation_variants.json": "7809852fc6be832b9ba9a32654ae2884ecb5779198546345e79bb9a2edc7a256" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..155ef4138af508a739d6af8d7cd283308fb7b4a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/dataset.py @@ -0,0 +1,293 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/fixed-link46-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/fixed-link46-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..e031906861bdfa87bf890c32a352f8488727ffee --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/fixed-link46-plan.json @@ -0,0 +1,68 @@ +{ + "created_utc": "2026-10-25T06:08:23.071762+00:00", + "parent_evaluation": "/workspace/research/experiments/scale13-small-epoch02", + "parent_inference_signature": { + "last.pt": "775b5cebc2335ac6732b582234e6665b189ae7ca56f246c4fe3be6d0e9d1011e", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "output": "/workspace/research/experiments/calib46", + "parent_reward": 0.42285346475207175, + "only_change": "Apply the same frozen sigmoid(raw splice logits + four training-prior offsets) in both entrypoints; keep all weights, raw site selection, sparse usage/junction decoding, other families and orientation fixed.", + "parent_choice": "The current strongest fully evaluated/replay-checked raw publication, small at82,040examples. This is an independently declared transfer of the existing output link to a later checkpoint; earlier experiment selection pools remain unchanged.", + "no_new_parameter_fitting": true, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "helper_sha256": { + "run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic symlink without overwrite; retain study prepared-calib46 outputs while referenced. No cache is an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "validation": "Fresh native/recovery52-array check; exact link-only code/weight guard, prior12486-array proof plus36fresh native perturbation arrays for exact research-cache derivation; all200 validation tracks and13 official variant scores. Unaffected10family and8variant reports must equalparent before promotion.", + "caveat": "Fixed marginal prior correction is approximate for per-example class weighting, and the verifier applies its unchanged additional sigmoid to the emitted scores. Previous checkpoints show mixed variant effects; no improvement is assumed." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..3b3cd2b5c6a71de5012f7212ad8dc3ca99792a8b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/training_provenance.json @@ -0,0 +1,119 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 2, + "offset": 0, + "global_step": 82040, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/progress.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 17761.135267923004, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 20510, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..47404fb5bedf33ac45e2ad30bcdf5f8d414c3781 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.555699689779102, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6795538134480462, + 0.6969385880934814, + 0.4440310093842537, + 0.6092467419409854, + 0.4583068413078484, + 0.45684934704813646, + 0.46921178553420084, + 0.5495368524784628, + 0.5528435696877605, + 0.6120346125916151, + 0.5573155938780116, + 0.531276502751468, + 0.38074756124202097, + 0.5087872404071951, + 0.6555722436094428, + 0.47955046587971867, + 0.5307334932263243, + 0.49759118025235466, + 0.5308782297353152, + 0.5148522167106985, + 0.4840398175316835, + 0.4523606570693779, + 0.52297369935844, + 0.5027707855494526, + 0.4967889580556305, + 0.4793307566878172, + 0.4570388250258062, + 0.47714481600221526, + 0.488290065467467, + 0.5428571245052065, + 0.45974164503486525, + 0.49524625368081404, + 0.4504490916705791, + 0.3781027504524497, + 0.34543668865105376, + 0.5564300624880137, + 0.49698914670110955, + 0.5998128659396257, + 0.48474654248326204, + 0.46998517661646777, + 0.556094370459589, + 0.5550912060661156, + 0.4729794632288362, + 0.4507121974165383, + 0.5273391304274185, + 0.5354557122424057, + 0.41681414356455704, + 0.4655640064460196, + 0.4584585370621107, + 0.499924120242928, + 0.47356542768837206, + 0.4392242009058282, + 0.5287784497541221, + 0.5708533880232379, + 0.4330355525776604, + 0.6659892979427481, + 0.7856323900724217, + 0.538530900604558, + 0.5822167724612293, + 0.6059186184549128, + 0.7561229743915936, + 0.4769197141028498, + 0.6866177881294476, + 0.6430117595735819, + 0.6206079929367571, + 0.48358695044265126, + 0.5488313193968627, + 0.6099308482636568, + 0.7636831176638607, + 0.6868874643498906, + 0.5046248295892172, + 0.45424095819938953, + 0.430772128438721, + 0.5153252200179824, + 0.4835402506854693, + 0.4638802465913259, + 0.4772851501644087, + 0.5141255473093637, + 0.5546467655774563, + 0.46840862719121185, + 0.5471974261338897, + 0.5717834122465622, + 0.48985313501663585, + 0.46305346704681827, + 0.4956054734410602, + 0.4839681376865151, + 0.4640497734304261, + 0.4560724761389201, + 0.4328168649080329, + 0.4781254878643131, + 0.4844717610487863, + 0.4824795522475992, + 0.46473433365857375, + 0.48138790379717894, + 0.5433910603987465, + 0.36713667310191267, + 0.4237152692050395, + 0.4160494109006816, + 0.46912466357979127, + 0.48698192571376586, + 0.43502994659140826, + 0.44208605921444283, + 0.48375964912907277, + 0.4456864330814112, + 0.5161476252229699, + 0.5244966648472477, + 0.5156470739280334, + 0.49675754719050974, + 0.5167993229999293, + 0.5195200639190588, + 0.5771552485741119, + 0.5038545624127307, + 0.4947267314911728, + 0.423561834088826, + 0.4484542854629033, + 0.41542687325417577, + 0.4553531870422087, + 0.462857809484415, + 0.46810574364449287, + 0.45240164545469363, + 0.6367053952651185, + 0.6545193297758936, + 0.7210520743109934, + 0.6977798632979831, + 0.6912348621997658, + 0.55811676308905, + 0.8328813462270611, + 0.697557589211012, + 0.7366284994473126, + 0.664642341355222, + 0.7104285400922129, + 0.5496092168093049, + 0.5767540528876359, + 0.550035381392622, + 0.6970461408641115, + 0.7234877435628504, + 0.6817101515374793, + 0.5274203313404869, + 0.4124698376025189, + 0.791402189950009, + 0.7653762637979575, + 0.7831615656656452, + 0.775586538192909, + 0.5614768859594594, + 0.8270057585270355, + 0.7418305066269401, + 0.7103571455266648, + 0.708989295618686, + 0.8109404039193243, + 0.7831366952820619, + 0.5974991730955057, + 0.6084089500484268, + 0.686560794281736, + 0.7200529648600367, + 0.7365543480620804, + 0.5846848376413034, + 0.7202612144680126, + 0.585272719086005, + 0.5844904039628521, + 0.6612258168092195, + 0.574817918271997, + 0.7448040942188439, + 0.7553574152610796, + 0.6321620190845969, + 0.5751431910944618, + 0.5780149632166304, + 0.7157469888768733 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.6444808623736373, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5476044694082534, + 0.9220536409749914, + 0.7822655731485053, + 0.5069966567469885, + 0.515843715966536, + 0.5267192263593916, + 0.9352008408735444, + 0.7076623302479031, + 0.4433968480337856, + 0.34595364300280934, + 0.8785986027316185, + 0.7052376613973944, + 0.4262925918801496, + 0.5090747083067502, + 0.4936443281292674, + 0.8085476528121797, + 0.7936666210486059, + 0.8500529966740585, + 0.8650096537388182, + 0.8927294891666463, + 0.6870187592258593, + 0.6102114695629742, + 0.9009230216140339, + 0.5236600830992176, + 0.538860146829215, + 0.7815239134139269, + 0.4634883367855437, + 0.6757936267979822, + 0.5260107921445775, + 0.5458086823267891, + 0.6310497129739451, + 0.4693265783492545, + 0.8431326327008857, + 0.6290144146154848, + 0.8806532107591727, + 0.8878083065580693, + 0.45083378959940684, + 0.8293087546616161, + 0.8750350151176625, + 0.8604716254637204, + 0.8295448078729024, + 0.6717104365699558, + 0.7048423688732993, + 0.6349565239518796, + 0.5297780818674196, + 0.6399971223130676, + 0.7348282189162497, + 0.49459581877390224, + 0.5904289481506941, + 0.40439999808795524, + 0.37946628492256246, + 0.5621378772330625, + 0.49521679702621996, + 0.8245430625820114, + 0.4412770677480311, + 0.403474162736982, + 0.4599297511760037, + 0.5056881310767144, + 0.4764533160480643, + 0.5194525766997988, + 0.44713854157618743, + 0.6093532908016884, + 0.49383437670646746, + 0.48278240028589053, + 0.5125928648022723, + 0.5188095462301417, + 0.5517589281287756, + 0.47324739890941653, + 0.3799921804490871, + 0.4256868228361048, + 0.5038965542454875, + 0.6427325845371018, + 0.5009696933486909, + 0.6269881251407193, + 0.3585893501640966, + 0.8079250518173641, + 0.8908154389347521, + 0.6560606150160981, + 0.4413573280572679, + 0.45792928757161677, + 0.5264368271656112, + 0.5091424029347538, + 0.4654505375110446, + 0.4832292008139911, + 0.5732451760781124, + 0.4969234790702234, + 0.616090982023824, + 0.5034937668829114, + 0.6398945200490483, + 0.49539581871245253, + 0.6328578294500083, + 0.41179786469198293, + 0.7269474049091466, + 0.4718799386334037, + 0.7187064165886228, + 0.5054471099798838, + 0.7437256606847938, + 0.7081473001253181, + 0.373369010654941, + 0.38135154239393637, + 0.4749055329807269, + 0.37649042513916875, + 0.35960789241748653, + 0.7220743028228942, + 0.6902150753104896, + 0.41136131413686794, + 0.47873515698181746, + 0.5863056757000037, + 0.8472949561038591, + 0.553407826409712, + 0.5473422269707326, + 0.5764346995983021, + 0.5047296084727095, + 0.44020217394971073, + 0.7830676330912139, + 0.5669898621887828, + 0.495819436771995, + 0.5521387828188886, + 0.6501239019579396, + 0.4614390637455715, + 0.5212747187818013, + 0.4930540447652154, + 0.5582724805356384, + 0.5412066130892803, + 0.6132021733634863, + 0.5495234720466741, + 0.6470762162210137, + 0.6637667496084891, + 0.7026412860431129, + 0.6228176703880742, + 0.46426693968807703, + 0.6102355434600566, + 0.6369858421508141, + 0.7928088831653802, + 0.4490050557892734, + 0.33396596885141383, + 0.5896887497081674, + 0.4820481589628537, + 0.6449068201714369, + 0.610916237683677, + 0.6181796695237576, + 0.5653491321608476, + 0.4848077886238296, + 0.7782221880040001, + 0.6191744902755943, + 0.7595747510440988, + 0.4018882310277486, + 0.37835604059045563, + 0.4087362501545855, + 0.3669737870176347, + 0.8925573403848708, + 0.3282043661255092, + 0.8950861705574334, + 0.7461196434655616, + 0.46567798692901613, + 0.3580379312378087, + 0.3664676470160327, + 0.3772059630731444, + 0.6543850639067056, + 0.7426642766441541, + 0.7399804022977432, + 0.6614687532270291, + 0.5673581682041692, + 0.4856159839037879, + 0.6846123628784312, + 0.6172648209438387, + 0.5154044441633583, + 0.6370262133997958, + 0.4641920945269259, + 0.4941251673326543, + 0.41202662650113864, + 0.8288827024575077, + 0.46828462025810846, + 0.4675889045726103, + 0.6542196253334235, + 0.700902977881082, + 0.5217825075232285, + 0.35068193459493663, + 0.8738979533263844, + 0.3799674824396315, + 0.7706326664857508, + 0.5052922947150706, + 0.5982950387512089, + 0.7259974075770644, + 0.4579311014894978, + 0.563570133886817, + 0.6076304253137843, + 0.8284021400354006, + 0.42218354425190896, + 0.6167187041157965, + 0.5661009333930088, + 0.7047507375160099, + 0.8926601146013748, + 0.4443285366809569, + 0.6019310970253557, + 0.49940833888515446, + 0.6355625912323589, + 0.7161059891073882, + 0.5726527337608796, + 0.46275207605922714, + 0.8683869241430124, + 0.8783493099906481, + 0.7528111236937095, + 0.8352526020694678, + 0.8458785531542296, + 0.8366075623183589, + 0.848043395730784, + 0.37029722529500303, + 0.39263953469670165, + 0.5182720527491504, + 0.4707445061018228, + 0.4614009291234557, + 0.6073030969573529, + 0.8620959146588736, + 0.8432021192510514, + 0.6838578955002318, + 0.8267712081323151, + 0.8215304459701066, + 0.8544619030742213, + 0.903882529475817, + 0.5428582881159881, + 0.6133406917895499, + 0.6806065994823142, + 0.6385752831731075, + 0.8226484739451482, + 0.591399790936403, + 0.9099441155779499, + 0.8821013636155306, + 0.869730059829768, + 0.8992060835435034, + 0.8905597765075886, + 0.8744016368434928, + 0.8324350898062624, + 0.8564086044818002, + 0.9053039082522756, + 0.45805628892744726, + 0.9239061651231775, + 0.8750764216942462, + 0.8173997249940971, + 0.6094484893094803, + 0.8696588329180704, + 0.5294475350359086, + 0.7467092324003733, + 0.8700046313949807, + 0.6016822319863656, + 0.5903281397935346, + 0.6977551126214118, + 0.5321983902248667, + 0.5598745918961587, + 0.7325447222100913, + 0.8859170180787427, + 0.7493474963270648, + 0.8171507398037711, + 0.6333071255554973, + 0.8647462597495995, + 0.8066809135077525, + 0.8379942307700007, + 0.902426866696436, + 0.8290237120635756, + 0.8970856467583491, + 0.8660974586289458, + 0.5467678543143893, + 0.8749789742355356, + 0.5876928566954519, + 0.5493070702218775, + 0.8764557698691681, + 0.9009686425247682, + 0.8785799169376745, + 0.6108963759652192, + 0.6397662497510772, + 0.6685646273110398, + 0.7926910447511447, + 0.8912583917365428, + 0.7173742802984828, + 0.9045898868186972, + 0.9143336607598184, + 0.6186874337633209, + 0.34006823209582177, + 0.5307686753495129, + 0.8952432151979917, + 0.4565679534411634, + 0.6256432127333106, + 0.8839627769380021, + 0.8247547235482643, + 0.6379451144768566, + 0.860934698961649, + 0.8922164614560263, + 0.8847715863968912, + 0.8778432590348748, + 0.8627951678855565, + 0.8893532367062419, + 0.8570670215181988, + 0.9245837251012026, + 0.7793782534616701, + 0.8339325158526741, + 0.9045439580476023, + 0.568002656478271, + 0.5747832910048752, + 0.5711287676917288, + 0.5451351151339168, + 0.9034135570982827, + 0.7992050331431412, + 0.8448778074835471, + 0.7961497109695816, + 0.7660967990984716 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5163376799902533, + "valid_tracks": 12, + "per_track_pearson": [ + 0.48332803317163364, + 0.5333825216610459, + 0.49554400530964315, + 0.524017461692309, + 0.5274327196835281, + 0.5515434504139332, + 0.4685380330103378, + 0.5250416746053118, + 0.4899537843349591, + 0.524074129330367, + 0.506814544182612, + 0.5663818024873576 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.29658178305941785, + "valid_tracks": 546, + "per_track_pearson": [ + 0.29431081562401273, + 0.33903235472322707, + 0.28311741875575086, + 0.18708035007902818, + 0.10239717969489344, + 0.2902091807879035, + 0.18585184778893274, + 0.19615010924543236, + 0.3051244096142863, + 0.276347731972689, + 0.3232086458525887, + 0.189602186802896, + 0.20306988007141327, + 0.16421412766084997, + 0.15753779428094716, + 0.20424446730400297, + 0.24965367086740345, + 0.3392170099723712, + 0.3349815015770792, + 0.3241596907502086, + 0.26036241219625195, + 0.2658316168860186, + 0.3933636658518015, + 0.21110406211927604, + 0.25418442630315224, + 0.3232610301609068, + 0.2892558886188784, + 0.3139244219889659, + 0.32652212734909386, + 0.32759327463656185, + 0.21489199774021336, + 0.3604990772473488, + 0.21897649870675862, + 0.19555445888297626, + 0.360969807356984, + 0.30394457071081155, + 0.35897880591721126, + 0.38432186852412126, + 0.35828197003605966, + 0.3592013520126759, + 0.4004248473763422, + 0.3352657601331316, + 0.31413256393566136, + 0.28647707435036696, + 0.21043237931476372, + 0.26637294993740096, + 0.2823421192206994, + 0.32356122763481443, + 0.18099512063223205, + 0.22806815993469343, + 0.3434846481431463, + 0.33683789419988636, + 0.3455338416138487, + 0.24673845496447452, + 0.39138067154968, + 0.3887892102783355, + 0.22780948757819944, + 0.3956884329640291, + 0.2733762280103112, + 0.3825918461708645, + 0.36265176673113125, + 0.2444640466754687, + 0.31421198506486125, + 0.2576659434323365, + 0.3302563688772017, + 0.3816241079236915, + 0.37266372291576644, + 0.23564708508661864, + 0.17507161864412057, + 0.15011200166000171, + 0.2939847026924509, + 0.33863398175286347, + 0.3025836965540536, + 0.2940378313602963, + 0.33233909019127, + 0.27668572579799006, + 0.3282435123215587, + 0.31057885226391446, + 0.35119964436419565, + 0.3468209721932565, + 0.3006221756096001, + 0.17646749022099575, + 0.312051799360063, + 0.40403679823151384, + 0.175732204858089, + 0.3176168112602859, + 0.28329308191196984, + 0.3882477237898525, + 0.24772716861506158, + 0.32856543523936343, + 0.26116099595414644, + 0.3022518114483142, + 0.3109890020394673, + 0.16041368955422947, + 0.3381451114329861, + 0.2799736022145766, + 0.32633745941469133, + 0.3202377635057871, + 0.2982698136558957, + 0.18657466427610148, + 0.29744478927602985, + 0.31169265531872575, + 0.33815343465199443, + 0.39690162784763494, + 0.36198654337853103, + 0.344838665290026, + 0.25619029048600844, + 0.3378039319902029, + 0.268321435994332, + 0.2676219679686667, + 0.28323036590720185, + 0.2835650782865215, + 0.3044694970927618, + 0.304456584038132, + 0.2520475174476696, + 0.33200475318392086, + 0.3050959939939741, + 0.3148469228099761, + 0.30281923872650424, + 0.3398185723209062, + 0.3450692214443664, + 0.33254610696802817, + 0.22656914790809127, + 0.26159223257020475, + 0.2841059642720161, + 0.3304971455481159, + 0.30814192513107047, + 0.3291449216863091, + 0.4073012566063173, + 0.3552262922998154, + 0.3519474815045406, + 0.25776359105019475, + 0.3489542325463518, + 0.3879150448430816, + 0.4052384497931941, + 0.30228048688846076, + 0.3728346282000917, + 0.2995437617599479, + 0.17135947608398852, + 0.348182940813866, + 0.3662171291609427, + 0.30504490063593603, + 0.34524384162050314, + 0.26943854043536236, + 0.29505643712412666, + 0.306503137436147, + 0.31028015626996835, + 0.4069608260880597, + 0.2838283172037327, + 0.2843496032725823, + 0.3677268185151687, + 0.2763091917006352, + 0.32900447632742136, + 0.32876641765624287, + 0.30864631342392307, + 0.2935220755008678, + 0.38268590602123487, + 0.40281874660752676, + 0.2669145076899873, + 0.26168680130732846, + 0.22151874918465353, + 0.2809765725736944, + 0.2885055836001304, + 0.3272856380352184, + 0.29056110140224345, + 0.26229281585259073, + 0.30067025902132444, + 0.23785795440844057, + 0.18526063023115832, + 0.36304751070761476, + 0.30905472753063107, + 0.2555697678690665, + 0.3073283520355868, + 0.41075208793363654, + 0.3165481707369744, + 0.18679873344147718, + 0.20856983402924492, + 0.2814678562187295, + 0.3525623267590987, + 0.36088890997187817, + 0.3546658727733867, + 0.3686795910997609, + 0.21441868978305423, + 0.34829531312927475, + 0.3028457242013157, + 0.2474414096129804, + 0.30490955304839623, + 0.21303171305164087, + 0.3973759744111589, + 0.32611981928578715, + 0.27108520883745013, + 0.20457764805695244, + 0.2006901655762624, + 0.2852240780226703, + 0.3111810393803042, + 0.26893455097453167, + 0.3054476787774035, + 0.31646530745385315, + 0.2332128239909085, + 0.3055668573803329, + 0.15743426401625907, + 0.14980307579897434, + 0.14256245043908164, + 0.13357710465517514, + 0.18312073791963954, + 0.29646964057851677, + 0.2933384930790231, + 0.2570183648267204, + 0.1381977768485835, + 0.2737579585675885, + 0.35061704003714256, + 0.34220006024080546, + 0.3566358464301559, + 0.3285566814022385, + 0.3121235732361274, + 0.2686688134149477, + 0.3552221742282396, + 0.3629012370951557, + 0.32544434229189684, + 0.3632027127910459, + 0.3278331061963123, + 0.2578986135546672, + 0.3280265500175577, + 0.3293333521224428, + 0.37899810358065533, + 0.37436322867311556, + 0.3198495866682189, + 0.31177314919754573, + 0.33749359888641445, + 0.3755612560498848, + 0.236753104775121, + 0.24665608408911807, + 0.3874070839162002, + 0.3024150171666939, + 0.2677140645406553, + 0.2886833005892511, + 0.39016344616908977, + 0.24471924285447044, + 0.3395811981739133, + 0.32785524042937636, + 0.14489692320683023, + 0.2465567970446051, + 0.1930484198306017, + 0.3336760625486842, + 0.28480921512839913, + 0.3235008558210823, + 0.1367746162145931, + 0.3679709372707216, + 0.3088165864088479, + 0.3556145798962987, + 0.34079731306851696, + 0.23512704071351015, + 0.4017564312908249, + 0.17111730180430784, + 0.3480449204403707, + 0.25012614459653976, + 0.39868365391319255, + 0.35277473420836186, + 0.36309463606972886, + 0.41938737583644253, + 0.2847395671976817, + 0.32507137446897927, + 0.14999884920424247, + 0.3318823119721998, + 0.3571504012815924, + 0.24735550296308134, + 0.15301260350863463, + 0.159102772414247, + 0.183256218283245, + 0.301960433073365, + 0.37223173674224647, + 0.17967132652667497, + 0.1415933206224809, + 0.2819886087289225, + 0.3485847472251417, + 0.3042952019950731, + 0.20313063390565936, + 0.09438448956724606, + 0.30742996959826613, + 0.2017217617445712, + 0.23232394486259833, + 0.30589783487558353, + 0.26362802063481494, + 0.32190208239784546, + 0.18747900140946816, + 0.21132963299776406, + 0.19344918788285073, + 0.1784870226817742, + 0.19203955042318893, + 0.2526946208994984, + 0.33388552059557425, + 0.3298637820199865, + 0.31956438542440474, + 0.2570647313335555, + 0.2645032793875549, + 0.38261514994174617, + 0.20146474083694224, + 0.24173194513456472, + 0.3506475139896228, + 0.2808199814182479, + 0.30638714925684973, + 0.3233505567811543, + 0.32277418665443997, + 0.20354251526036513, + 0.35292615055635473, + 0.20398172212697674, + 0.1965361379048454, + 0.3453201504598746, + 0.31508428977457587, + 0.3876701157901894, + 0.38835707157368693, + 0.4163482620089852, + 0.35866609524210424, + 0.3951370779593734, + 0.3209126039749478, + 0.31337372140038544, + 0.30488196304530857, + 0.22482601248145925, + 0.2808967946682373, + 0.3162360129516107, + 0.36811833338172056, + 0.17794508044386204, + 0.23587450281206415, + 0.32525011806151966, + 0.3391465572482076, + 0.3290139797346022, + 0.24368883255638546, + 0.3908945906048865, + 0.3707143933942347, + 0.239592431774187, + 0.3913725144509947, + 0.2614566361236732, + 0.3947576788099082, + 0.3503175947470811, + 0.23798970164320707, + 0.31159305508075663, + 0.26534978957757527, + 0.3382958626362782, + 0.3707265828631741, + 0.35618247188072555, + 0.23063060444089623, + 0.18247130642072765, + 0.15174556220028063, + 0.28782344064773, + 0.3430711978891432, + 0.30388167629079227, + 0.2882151478614129, + 0.33566495461908713, + 0.2715624117376276, + 0.30562750855673676, + 0.29390181538755855, + 0.34400442586525726, + 0.3481240896139652, + 0.3041440161715839, + 0.18124764050262943, + 0.31865017044351296, + 0.4098169674347201, + 0.1848478812060979, + 0.3232159785789685, + 0.31812717154953146, + 0.3968017684967531, + 0.2342946586467041, + 0.34311597350411677, + 0.2595578495302339, + 0.3103269018839957, + 0.3184228510393005, + 0.16578310815847558, + 0.3288179229089924, + 0.27123311144429135, + 0.31045834825682644, + 0.33524149557625726, + 0.28811939662487845, + 0.181895420757082, + 0.29735805071070776, + 0.316692456327791, + 0.3383632917125929, + 0.38341181435747984, + 0.34867809528102633, + 0.3477142277208819, + 0.2545197041500274, + 0.3184158641677471, + 0.2523268279653475, + 0.26274281983589287, + 0.26964106400816495, + 0.27612872817947703, + 0.3083178040321609, + 0.3134593486501396, + 0.24821670500566123, + 0.32800197199112285, + 0.28880583057916354, + 0.3106961074782191, + 0.29198693965703143, + 0.3332495264884108, + 0.3291145942156279, + 0.3222948654880126, + 0.23318341610594165, + 0.272581857300199, + 0.2986994936105696, + 0.3379949642370159, + 0.2989076998879926, + 0.34233176565665496, + 0.40391292177996685, + 0.3841616902500764, + 0.3398203036724274, + 0.2496120364411495, + 0.32126163768348887, + 0.36874583681817524, + 0.39829086203946235, + 0.3042260726973865, + 0.3809109123044089, + 0.3027400704999169, + 0.19489035542646435, + 0.3582029128473562, + 0.3392219582195536, + 0.29280285978992265, + 0.36132498731637724, + 0.2593631244786737, + 0.2834107168597972, + 0.30170991444991907, + 0.2968227319138868, + 0.3959869896726131, + 0.32505240891144, + 0.2800320450537587, + 0.3690950905514342, + 0.2802478318262644, + 0.32641793133987673, + 0.33396307272341025, + 0.3016433824967918, + 0.3122679238954442, + 0.36927012762271433, + 0.4011199309807139, + 0.26206688229366737, + 0.28128861829694374, + 0.22997848845518934, + 0.26231668400138897, + 0.2952228477558542, + 0.3559937632418386, + 0.3288386629773531, + 0.2560007108645837, + 0.3005573538898805, + 0.2706767275832094, + 0.165169011073312, + 0.394006605534326, + 0.31330348539912645, + 0.2733292779650854, + 0.30611756092262027, + 0.4091800867325742, + 0.31467261045404715, + 0.1655200598037742, + 0.21739358809736475, + 0.3033113780394405, + 0.32180961433694055, + 0.38021428840792587, + 0.36388552415071346, + 0.3787904145197871, + 0.22639572520977655, + 0.35461517269221854, + 0.32789554125797393, + 0.27776494321640155, + 0.3306898746119477, + 0.24014961692666117, + 0.3907585265186868, + 0.32690018033282187, + 0.28134106973058776, + 0.24436286235900703, + 0.1937877324848296, + 0.2366739222241371, + 0.3452603354435021, + 0.29307283094169145, + 0.32988443749296004, + 0.3298869140843532, + 0.2476030155611087, + 0.31107860052392294, + 0.1470174848551227, + 0.12834883131969174, + 0.12312090076948326, + 0.11651793243423818, + 0.15528298823907338, + 0.3191560821416409, + 0.2625479936326613, + 0.25821292031993703, + 0.1286780693781908, + 0.2884211716803833, + 0.38046352693072516, + 0.4116009160065246, + 0.41586483598415175, + 0.37315520273716185, + 0.3682655469934211, + 0.2901730459699191, + 0.3908461544623336, + 0.38665364974613986, + 0.3343065243415984, + 0.3800415297026211, + 0.35752727291915193, + 0.30406768455899724, + 0.37698916545761, + 0.3337827106436465, + 0.44841982978846795, + 0.39224635422885007, + 0.35348416492209694, + 0.33069221752709843, + 0.3336641274915304, + 0.3688882800514624, + 0.1947216655255559, + 0.1871079629792816, + 0.40850583872194973, + 0.30575553554057766, + 0.2883154283729743, + 0.2955457701116457, + 0.4045974631559117, + 0.2849900647843228, + 0.33147263136195815, + 0.32208477298981164, + 0.13928615750443468, + 0.23370588884675708, + 0.19066997639665317, + 0.3620445092327174, + 0.2835707239257104, + 0.3405072108099992, + 0.14257356531975776, + 0.37162065641809955, + 0.33601695451298186, + 0.3686526945943879, + 0.3592919105048681, + 0.24592571296181978, + 0.4498466281960259, + 0.16675252305748234, + 0.3467141548929811, + 0.2169203108906197, + 0.40024570106647095, + 0.38370471082995083, + 0.4140817213463014, + 0.4211868736094094, + 0.32222921285017386, + 0.3267053185481127, + 0.13515156286787222, + 0.3195523915424757, + 0.36728683544910196, + 0.2834539679999698, + 0.14483291078472355, + 0.14960997738304704, + 0.17225129876082484, + 0.32442842893258383, + 0.38380950577717227, + 0.15651098169773905, + 0.12081444737776434 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.5742579225948128, + "valid_tracks": 613, + "per_track_pearson": [ + 0.5152430562706712, + 0.6551364264354735, + 0.5835835254993295, + 0.6232813178682971, + 0.6360177769257379, + 0.38687312745947333, + 0.4857751212589548, + 0.6238502190780358, + 0.6052219574915354, + 0.628231948684115, + 0.5612073270709564, + 0.5579516198034464, + 0.47083952258778683, + 0.5910403908442731, + 0.5517467613890696, + 0.5723004827445488, + 0.6529005729744549, + 0.5745377312033699, + 0.6086993134550102, + 0.5466771869613176, + 0.5207786353351527, + 0.6465340729884603, + 0.5143196549438541, + 0.4880222799132733, + 0.44164690116109157, + 0.6477265148738995, + 0.5572689243179547, + 0.6083907517012284, + 0.5890577785682833, + 0.6034315710852435, + 0.5518062665258098, + 0.5884590394576215, + 0.596917447879464, + 0.5496157825211451, + 0.6256822243302641, + 0.5308315226250624, + 0.6160075279927123, + 0.6144514875861538, + 0.5753177051491463, + 0.6012702775071004, + 0.6337495484883842, + 0.5601960146882878, + 0.6097482127226441, + 0.6360201623858934, + 0.528162361893196, + 0.44460338122203413, + 0.5234026399081363, + 0.620080495005681, + 0.5614556223988043, + 0.6270241608340209, + 0.6143940989306089, + 0.6071583601224575, + 0.44627994513313096, + 0.5416905760802209, + 0.4269119324766867, + 0.5749776967539593, + 0.5405798544905047, + 0.631517582701957, + 0.6723001655597489, + 0.6852242242516605, + 0.6909716244401861, + 0.6299547627392449, + 0.6642120820627636, + 0.47457673006502343, + 0.6218551628278711, + 0.6272114411052064, + 0.6496160702810078, + 0.6686222683905143, + 0.6267963299772943, + 0.6827550075621244, + 0.6182342656781095, + 0.6443513098135858, + 0.5906998105378181, + 0.6248101575449312, + 0.6077254121370921, + 0.6066115825961372, + 0.636759292606614, + 0.6574089048728743, + 0.5648669999451059, + 0.6646003283706486, + 0.6118402227341585, + 0.4997986480579266, + 0.6625153099852135, + 0.6509785633842787, + 0.5521231802107192, + 0.6220710504104896, + 0.6501242375532154, + 0.5208683938082238, + 0.6118114081906054, + 0.5376521096609056, + 0.5022621331544264, + 0.6720036042685983, + 0.6502220918070248, + 0.6341031893972064, + 0.6299423092619306, + 0.6539218895507038, + 0.6664789238137504, + 0.6362310309311018, + 0.5779904750198787, + 0.4970001831897784, + 0.45612778459533304, + 0.4649535240224927, + 0.5242525693398249, + 0.640068109311007, + 0.5021503094261427, + 0.5214320185814086, + 0.6468141317085909, + 0.5714594980988783, + 0.6313109572014994, + 0.5442134853658535, + 0.5948321525180489, + 0.5818685831736621, + 0.5468828938803937, + 0.5679468350787261, + 0.6321161126638397, + 0.6086970071644275, + 0.620011078829965, + 0.5651065664278896, + 0.5331916578745859, + 0.5893659931324696, + 0.6305131463180719, + 0.5508428671574339, + 0.5251845869502714, + 0.5744672940473774, + 0.6596036600542644, + 0.4212391083542158, + 0.4577456516426492, + 0.44084042754016395, + 0.45815168701494824, + 0.526003479313143, + 0.6354873652777216, + 0.5200398098990576, + 0.5756416528679584, + 0.5787184582285202, + 0.5764385466633469, + 0.6283898473861332, + 0.5985369416798533, + 0.6342750144552962, + 0.5347620396206079, + 0.6103944340600095, + 0.5071311357687038, + 0.5801668673693655, + 0.6321990839524844, + 0.5390257139644563, + 0.4618285431206056, + 0.5121124735464584, + 0.5256849889911237, + 0.543894595535239, + 0.503490982792157, + 0.5374192972013497, + 0.519349815413427, + 0.5196952904047369, + 0.5917464131175366, + 0.47574522031967426, + 0.5279853131675417, + 0.5012734205888196, + 0.5652561295927995, + 0.5177543356575038, + 0.5688707614304147, + 0.5859787029267445, + 0.6223249551885153, + 0.5896480029336493, + 0.5829985005132482, + 0.4937140399839313, + 0.4918362179022163, + 0.4943990433550689, + 0.4984752120311696, + 0.49666760845591096, + 0.5474970434252233, + 0.5347978857956529, + 0.6275653965835792, + 0.5569532510787356, + 0.648843657476826, + 0.5451714877555532, + 0.48562614704237794, + 0.5284461177225962, + 0.5047125248070956, + 0.6225882022184321, + 0.43674763094425123, + 0.4631596908100806, + 0.4347772903689408, + 0.6088247780800292, + 0.46584694241754515, + 0.6102337449150385, + 0.403613422174753, + 0.534846576253723, + 0.5948101419792559, + 0.39287277622347805, + 0.5266293412352746, + 0.5200995142981253, + 0.6160055200256107, + 0.577339717530589, + 0.5484386199808279, + 0.5058243125764182, + 0.6019424014235899, + 0.5391386774945635, + 0.5602004104049358, + 0.32303077568785055, + 0.5716794829499534, + 0.5715942931503496, + 0.5011503325588382, + 0.6200503944399846, + 0.6064294743685082, + 0.5906798704448091, + 0.6113314733936352, + 0.6113806835010264, + 0.5968734438873916, + 0.3116881371672569, + 0.6011736105535447, + 0.581825199513473, + 0.5095859075873604, + 0.49955285757400764, + 0.5622571014181504, + 0.5684497452671996, + 0.5208043600305873, + 0.6240154744265834, + 0.5402756210881459, + 0.4749157519595925, + 0.590243445634618, + 0.6362871853302077, + 0.616774175010133, + 0.6640475793277058, + 0.5788475594392453, + 0.5828425231155206, + 0.5733517083644261, + 0.5742764326155709, + 0.5734698444722914, + 0.5795462557943052, + 0.459811033490438, + 0.647360349658629, + 0.5497620377857865, + 0.5853310588703114, + 0.6390459550346893, + 0.4252236081660993, + 0.6667013992079317, + 0.4111604488645865, + 0.3746183747932619, + 0.6162977402390399, + 0.6351925200123806, + 0.6052767773087865, + 0.6163372331430069, + 0.6463222042919513, + 0.5928830502827974, + 0.6276766854941318, + 0.2921076046002876, + 0.5470568573672168, + 0.5535139365136026, + 0.6103033019837113, + 0.5242014379724179, + 0.5469227181718664, + 0.4930138244000646, + 0.5492477604274537, + 0.5325160083020518, + 0.5357311503309398, + 0.5958299076004727, + 0.5717908265381463, + 0.48302051163258714, + 0.6320011957002697, + 0.5648462197802183, + 0.6229427047233899, + 0.49640041961135684, + 0.5337018837776776, + 0.6481638590840951, + 0.6553617985919871, + 0.5846219584471283, + 0.6079630286924453, + 0.6602855479733486, + 0.5061205015734603, + 0.6063471690673794, + 0.5965866397277128, + 0.41374875450422915, + 0.5421738176951506, + 0.7100455094978029, + 0.6894843391056732, + 0.6910394772854052, + 0.674141511987828, + 0.38152777372952845, + 0.48707992367181585, + 0.6742521701623262, + 0.6375745599469002, + 0.6632204182384234, + 0.5827928229122838, + 0.5371759940867885, + 0.5337689117696444, + 0.5925268285614848, + 0.5540289337907287, + 0.5714541906624361, + 0.6583317720650476, + 0.6390812474698894, + 0.6894749059738832, + 0.562518994923846, + 0.5417007601238139, + 0.6959336415630556, + 0.5353381489153455, + 0.4596893346309641, + 0.46084456478087577, + 0.6660412735460147, + 0.6941352170685764, + 0.6912432569240473, + 0.623593181903968, + 0.6287554568659247, + 0.5501302083785379, + 0.6200710929235599, + 0.5383756448228832, + 0.6146499212021705, + 0.6537743859603994, + 0.6267472259476695, + 0.6872067406860931, + 0.6960392712193496, + 0.6498898741634758, + 0.6549019972950255, + 0.6520552729896212, + 0.6157443571082781, + 0.6512448573786851, + 0.6114879908493507, + 0.5158965599408777, + 0.5344854134517484, + 0.569281918172546, + 0.6569614431592127, + 0.5765799728526158, + 0.6479686088855066, + 0.6694959782770642, + 0.6640939638839384, + 0.44046134898790695, + 0.5396877495610676, + 0.4002617616062526, + 0.5847267812261344, + 0.5732973140855103, + 0.6496829861551029, + 0.699425338064035, + 0.6679020939295319, + 0.6964425901908272, + 0.6998403619226037, + 0.7137602924155096, + 0.49736220533337977, + 0.6858739915052411, + 0.7038437576971555, + 0.7269930719210459, + 0.7356768601236245, + 0.6953076694419266, + 0.7315026649955877, + 0.657254981268467, + 0.6944740009070385, + 0.6138409559710359, + 0.7151709774800037, + 0.6788278406204574, + 0.6489925878135344, + 0.6907937177838871, + 0.7264248315794366, + 0.5263540499066357, + 0.6652467054847913, + 0.6300846976987414, + 0.5509997519440284, + 0.7142801706950942, + 0.7131130559369625, + 0.6213357799133906, + 0.6386179904512357, + 0.6364921425226645, + 0.530182187261216, + 0.6499532766673481, + 0.5532289459236692, + 0.5532524462469177, + 0.6910397237090068, + 0.6354486951553261, + 0.6257139704582273, + 0.5990817855879533, + 0.6430029038705524, + 0.6585371368880796, + 0.6221442970052531, + 0.5752963628803127, + 0.5380906555370657, + 0.4477722501028527, + 0.4354728028018125, + 0.48804939413981707, + 0.719306036060735, + 0.49503669118237165, + 0.5190849637223393, + 0.6966899127834997, + 0.5779045945678388, + 0.6425181356459795, + 0.5533983246805523, + 0.6171466564198893, + 0.5792367667649821, + 0.5415974475660247, + 0.6459645012650356, + 0.7187182860592579, + 0.6238613750557797, + 0.6459223447951655, + 0.5631674891665746, + 0.5315432707218332, + 0.5814181623385847, + 0.6900424719118021, + 0.5607711839304406, + 0.5464185802492206, + 0.585036169668481, + 0.6853521185069297, + 0.4085841997988176, + 0.4552630565045978, + 0.4370583441715528, + 0.4328756495389318, + 0.5770835872545526, + 0.6803135671783831, + 0.5172783939085662, + 0.579567578935872, + 0.6754124440467456, + 0.6231854185578865, + 0.6737991521278338, + 0.6069543539810404, + 0.6816314520499426, + 0.5474376957855269, + 0.6322856650413418, + 0.5076258735296834, + 0.6022258608831333, + 0.6538311246342775, + 0.5512869108556774, + 0.46246362479887676, + 0.5176435511964581, + 0.5778570021687579, + 0.6186932652656949, + 0.6369179920769683, + 0.5282359727880166, + 0.5395613783924362, + 0.5592133824830294, + 0.668515252216083, + 0.482477878590935, + 0.5185446241178965, + 0.5473914131872474, + 0.6504362389384046, + 0.5102073868927437, + 0.6587320552736238, + 0.6023822859887463, + 0.6424011359084058, + 0.6090947004657872, + 0.637611090099015, + 0.4066070797419246, + 0.41885749724631405, + 0.41444056759392556, + 0.41363839671088387, + 0.483717194857825, + 0.5822283576404849, + 0.5767590114639436, + 0.7160696481413357, + 0.6185822490500219, + 0.7214972318487333, + 0.5592901569407055, + 0.5507473822840826, + 0.5585532188326414, + 0.5128727055008966, + 0.6765617805413959, + 0.3989605764073157, + 0.5152282205271087, + 0.42348174946253025, + 0.6758724567149829, + 0.4499310361482321, + 0.5839137833023544, + 0.42544792562385747, + 0.5656232641554819, + 0.64795414588656, + 0.45191215605524615, + 0.5665891501397438, + 0.4876284578243821, + 0.6023522928181394, + 0.6568138963617189, + 0.5442974633219902, + 0.5131630769940099, + 0.6530939652458212, + 0.6324477531113838, + 0.6269367410511832, + 0.3077795747434383, + 0.6579371316126005, + 0.6471596498545584, + 0.4926765202267902, + 0.7302505277393737, + 0.6175868548100284, + 0.6487658126191783, + 0.6656390445302589, + 0.6423168261104403, + 0.6159808113352336, + 0.28704188118072893, + 0.6739304991690881, + 0.574506594052105, + 0.6728361797198598, + 0.6517504913633616, + 0.7288131289088702, + 0.6967737055271862, + 0.5767851584720007, + 0.6751734828703144, + 0.6973726626449321, + 0.579495852343667, + 0.7399888878732341, + 0.6534404803281139, + 0.5789630738978706, + 0.7071852492519252, + 0.5467150655520024, + 0.5292146773519795, + 0.5254190603122568, + 0.5103585161415752, + 0.4783844877367802, + 0.5336334703313775, + 0.39304878325649084, + 0.7098054643346839, + 0.48128939543645344, + 0.5812356651582979, + 0.7164724297200391, + 0.42773437494916916, + 0.7117361607412418, + 0.38631294288172163, + 0.3952021048907921, + 0.6559594763717264, + 0.6465283420071299, + 0.5951009788655528, + 0.6065967779249649, + 0.6995796189728057, + 0.6853375840458971, + 0.7354301707331602, + 0.33190262405029636, + 0.5979179395283558, + 0.5722083305227651, + 0.6509554177944182, + 0.6012736595020831, + 0.569748741648174, + 0.5498406015995438, + 0.5749486337418392, + 0.5875193538653144, + 0.5847400229822436, + 0.648970455606492, + 0.5489673558700581, + 0.43781851145776496, + 0.666887943344343, + 0.6038078398588114, + 0.716290143192201, + 0.43996003593118654, + 0.4304474531719339, + 0.6477300121817703, + 0.6481534836371154, + 0.6245537192670342, + 0.6771069872547145, + 0.7290146178874313, + 0.4475909147548885, + 0.6484798227646194, + 0.6319668620744227, + 0.425170847959495, + 0.5993652669514281, + 0.30305990847626785, + 0.4849142990828387, + 0.5585012040043317, + 0.48191998246492707, + 0.5573356061783485, + 0.4897692537088938, + 0.5992250444829125, + 0.6145455533988324, + 0.4953842857489602, + 0.47972085476502097, + 0.4799720297674368, + 0.46280191047467456, + 0.4609012640680092, + 0.4569964609419042, + 0.4640299807010604, + 0.45455538186965294, + 0.43728352302921936, + 0.39596953268199536, + 0.5925640313665302, + 0.5107064586927408, + 0.5261546357058682, + 0.5134894464405388, + 0.5299180205267376, + 0.5208666748848528, + null, + 0.5562831567904238, + 0.5779074290007709, + null, + 0.44615685755017076, + 0.5971333890585154, + 0.45896776453408655, + 0.6037679939571885, + 0.6270360050761967, + 0.5983174135984026, + 0.5820111144450818, + 0.5836631007592521, + 0.5893612084728772, + 0.5623278300933595, + 0.5186217821809304, + 0.5191964945908345, + 0.4762703980972747, + 0.5661696072792674, + 0.4598282327403421, + 0.548033539265486, + 0.5272511475668603, + null, + 0.5604099507448683, + null, + null, + 0.5383970597927232, + null, + 0.5675962385024158, + 0.5191143960026047, + null, + 0.5789429451889737, + null, + null, + null, + null, + null, + null, + null, + 0.560965857571813, + null, + null, + null, + null, + 0.4718865449854979, + null, + 0.5515205886714039, + null, + 0.5747945065559223, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.5794473584897486, + 0.43482566033454634, + 0.5406445714109536, + 0.5672930736585576, + 0.4962454652688043, + 0.5984623677072239, + null, + 0.5860776914014906, + null, + 0.553050956574765, + null, + null, + 0.5800926989202718, + 0.6032629580223553, + 0.5180353314045628, + 0.5036568250390296, + null, + null, + 0.5354893056971695, + null, + null, + 0.5560024353112236, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.559583728396825, + 0.543073987507189, + 0.5657966912772519, + 0.522216971866149, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4414599085952171, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.23506747394680022, + 0.3422883560663922, + 0.30711830124007755, + 0.462690017002351, + 0.3911806145750595, + 0.2962926132374464, + 0.38239637249615593, + 0.4469773545002478, + 0.36910348794800096, + 0.3792639250870554, + 0.40341909803240616, + 0.40007759815754235, + 0.2931320029754242, + 0.4375483352098206, + 0.3743960683521726, + 0.38205104413676755, + 0.4252645623272192, + 0.3447127814054356, + 0.48963112650129154, + 0.37867139503970476, + 0.3769479558797168, + 0.34277054290520353, + 0.31943170955738287, + 0.5087003669263256, + 0.35445863693835555, + 0.39171917805781714, + 0.3186929195029159, + 0.373448158932685, + 0.40388009288499577, + 0.42995852306189636, + 0.4200379339238532, + 0.4119888756372222, + 0.48877823596243536, + 0.2880159184525157, + 0.6534362160789895, + 0.28854819449610836, + 0.24334199870296222, + 0.18269850118465403, + 0.327575407894319, + 0.2268008908218948, + 0.27916012704959303, + 0.4614033612017425, + 0.30736503175492597, + 0.402743750827695, + 0.29578991883527944, + 0.3897994291784826, + 0.29759817569229297, + 0.22745732094612134, + 0.4649463326369977, + 0.4341601565773219, + 0.38792892736622875, + 0.3685985839897551, + 0.3871698211899983, + 0.4199388244522672, + 0.4205555030052096, + 0.4017341131096716, + 0.32225711575974314, + 0.45751583063486057, + 0.47447811890114017, + 0.4752223849881905, + 0.30893682689669966, + 0.2093342331204648, + 0.3976806800761379, + 0.4157614984910799, + 0.2731906627744006, + 0.34596449737179097, + 0.34987098122267674, + 0.4550962547969975, + 0.4721936793885047, + 0.42666399697353874, + 0.36249439153738106, + 0.6854734955267263, + 0.342126965138116, + 0.36345567200525675, + 0.35640732088974325, + 0.4313490503423291, + 0.29232056301640036, + 0.630210195900413, + 0.39887330257405473, + 0.4075719184007323, + 0.3998932700582857, + 0.40428526474946525, + 0.4738575152731023, + 0.416915248955218, + 0.33744562041480786, + 0.3000682916141274, + 0.44756812656856293, + 0.41268434190698067, + 0.7990348391463606, + 0.48103125377341033, + 0.3663624856468537, + 0.34337890966584683, + 0.412833868025345, + 0.3506946450380986, + 0.40109966940376374, + 0.5248069852820508, + 0.3412483680782178, + 0.3391860175611616, + 0.3527996524614311, + 0.3049973732459785, + 0.25201505741769825, + 0.37499382530402653, + 0.3368942285813237, + 0.32600545957176413, + 0.3722865118842735, + 0.23302903755496732, + 0.2999096805147081, + 0.30207810786322864, + 0.38115769491566953, + 0.368122729320597, + 0.24632947558771506, + 0.3331638100543952, + 0.3238967055938733, + 0.3569949713081867, + 0.35803704890147986, + 0.44202741495309045, + 0.3957048590728669, + 0.383259782882038, + 0.36472125502450076, + 0.42664269304934876, + 0.3523947353858597, + 0.4132622577423815, + 0.37522260333883906, + 0.3073228943061625, + 0.37481823388173224, + 0.331444582465782, + 0.3630916254745852, + 0.33765413873120564, + 0.3617746660550193, + 0.5318567828457729, + 0.3862385047495791, + 0.35428842708332214, + 0.40405087204939594, + 0.6096940146151258, + 0.3266461630924725, + 0.3648575273941066, + 0.3779535223992647, + 0.31465154226016695, + 0.33946856684702054, + 0.3322751393602294, + 0.373948905718786, + 0.43440046083619493, + 0.5360486674019141, + 0.34743389775645445, + 0.3956800537845554, + 0.47394312915732, + 0.42458747986357537, + 0.2725787954366018, + 0.41566119195307855, + 0.446848855945185, + 0.4129821086761039, + 0.3355618254736471, + 0.41714898849420384, + 0.4752970460036431, + 0.40645517780718626, + 0.4090206151066331, + 0.32343449435491045, + 0.360490137632718, + 0.3826967651613018, + 0.423671245756157, + 0.3968588800197591, + 0.3590691831677926, + 0.38931101354161257, + 0.40073057318241595, + 0.3822168641892643, + 0.3650043002521692, + 0.353984127483711, + 0.4500193698038618, + 0.2991803016851558, + 0.41500815426261595, + 0.3858506149795252, + 0.3491454459720504, + 0.48514838176371994, + 0.3517201984999671, + 0.3314300429994686, + 0.41634257370251193, + 0.28327920276917673, + 0.41717537587186143, + 0.4293325665854598, + 0.33591216329396334, + 0.4181383854248771, + 0.3858159402745586, + 0.3205372394880508, + 0.3817551020762973, + 0.34155937975721445, + 0.3946798571514784, + 0.43211272695859027, + 0.31369454442717537, + 0.3362984782611209, + 0.4007983185055376, + 0.33066769414289143, + 0.33909143852550305, + 0.4238734625679601, + 0.4727005478063464, + 0.33159174018505194, + 0.3411998829669037, + 0.35651349457357556, + 0.35196765400797025, + 0.27582631804087715, + 0.45159066042881957, + 0.3640690213349155, + 0.3196109835077762, + 0.38074880349584045, + 0.3255194450294579, + 0.4417710486653269, + 0.3521247106610137, + 0.42934947412714675, + 0.3749015495613723, + 0.3150244324023792, + 0.49353364433792657, + 0.3468784750378915, + 0.41235049399638807, + 0.3623747266283657, + 0.39217852104581524, + 0.8303110957804036, + 0.4162935295504309, + 0.4014862611166507, + 0.43466612754234224, + 0.45871658405754334, + 0.5482680091071874, + 0.6111820552110842, + 0.6224717378458641, + 0.42234596761877186, + 0.3900273414478359, + 0.48430376143812987, + 0.7281851352536153, + 0.35915570156126486, + 0.38217627698746226, + 0.5440674434890498, + 0.5172883435600035, + 0.562764512431924, + 0.44464870005451534, + 0.31243703163999137, + 0.3948281169468098, + 0.5953512969593706, + 0.43948513965257624, + 0.3224105079955682, + 0.611025083092553, + 0.5504212278189863, + 0.44971989702318554, + 0.4914076603893225, + 0.705425634056073, + 0.4761133622600914, + 0.4977556199781933, + 0.43818674905700494, + 0.7596730211928308, + 0.32936498136959724, + 0.6372557194183384, + 0.39905884019002597, + 0.546308504553945, + 0.43901583057680044, + 0.38994080330801467, + 0.6990794535879983, + 0.5108684985201786, + 0.3361722838246322, + 0.3767326445433625, + 0.6430517973579497, + 0.38273074680082486, + 0.4397663315295973, + 0.47825193514332853, + 0.36587209507155455, + 0.4780307926361913, + 0.4741594042772008, + 0.42296069871781344, + 0.38719307859128305, + 0.4745579545165738, + 0.3565387697015105, + 0.34332364065291104, + 0.6282423921789531, + 0.3799605436414291, + 0.4527473655762135, + 0.5588284061835006, + 0.3853863960427624, + 0.6568848353683164, + 0.34714710694313017, + 0.3576919061884163, + 0.5564944920790627, + 0.48754710786335376, + 0.5804197163895546, + 0.7244555722166451, + 0.6641943598048261, + 0.40997233785768616, + 0.46520514031990734, + 0.5537898146312403, + 0.3568007773577778, + 0.3424797738754557, + 0.5622235492278733, + 0.5127076492196508, + 0.720402512640832, + 0.4831083997295064, + 0.3708474005386957, + 0.4530196542271237, + 0.3693398262642784, + 0.4714864674823406, + 0.7010629273539136, + 0.6228369140651263, + 0.5146072583788297, + 0.527658823514349, + 0.48715177183625974, + 0.5106662787262071, + 0.7368765302084428, + 0.3095926912467301, + 0.4481805069456665, + 0.29456343928728335, + 0.601471955669479, + 0.3808757245443998, + 0.7325725906627403, + 0.41542856346209833, + 0.43021085023438227, + 0.6909895286327611, + 0.5120806510826454, + 0.3328399949633583, + 0.5640487037372752, + 0.40370507809101647, + 0.5073803311772959, + 0.49669636094712427, + 0.34733959165268313, + 0.5042874146231459, + 0.3940220141227585, + 0.45274495440722834, + 0.39895509094017517, + 0.35266378213588234, + 0.41197412721759125, + 0.5167807907729934, + 0.5770502661488056, + 0.6630598272772998, + 0.4213417816804808, + 0.3103944405291242, + 0.45128457038485953, + 0.4613826539681498, + 0.4308114598777039, + 0.47550055339993114, + 0.47697924534008684, + 0.38299399900205733, + 0.33633767136471326, + 0.5085536045997971, + 0.7611748312694384, + 0.5227812173459581, + 0.5979916579018901, + 0.5908229369733629, + 0.7613871642359735, + 0.5510606344623769, + 0.3554324616408357, + 0.5812598272557017, + 0.3830476631873168, + 0.5283195517289174, + 0.46785002724517444, + 0.38805754499851214, + 0.496023429543028, + 0.777001627812741, + 0.24718538087408395, + 0.46582104716366673, + 0.4631847677505683, + 0.6195804056023163, + 0.3788964644090718, + 0.3244975304467666, + 0.4313844793918614, + 0.719929523366015, + 0.38929190244386197, + 0.4401850282514156, + 0.5052077280175524, + 0.4581445777850231, + 0.4303043229061019, + 0.49335647706545155, + 0.61110211258273, + 0.46390425831663273, + 0.4082370444982099, + 0.5405648520380951, + 0.44991607097062764, + 0.5679311334438564, + 0.5288873158149116, + 0.5190060208049175, + 0.5437004745243892, + 0.33532997571770057, + 0.3562680133202865, + 0.5460442254770331, + 0.32676236563847216, + 0.6954865525227492, + 0.3222816881166443, + 0.561348030590293, + 0.5686590192134423, + 0.6831628250598905, + 0.30996798560125965, + 0.36770835688876286, + 0.3393922829814901, + 0.3836146481594789, + 0.3285562411116285, + 0.31152235784583837, + 0.6172901130159663, + 0.3834796934296323, + 0.6788423743255143, + 0.5816274073980005, + 0.4519662878156661, + 0.568805443024637, + 0.3804283651574856, + 0.5289093961700633, + 0.5202237275329235, + 0.47730747732453666, + 0.38211486349608836, + 0.4664213410395928, + 0.4918278176051895, + 0.5254004282873698, + 0.39371883284081705, + 0.3487650118801632, + 0.6163705723153813, + 0.4726352407264261, + 0.6543312314140602, + 0.621396052929275, + 0.5160716456420825, + 0.33464469389576346, + 0.36149124029157154, + 0.3623379480901974, + 0.47037822919890226, + 0.7435686639591548, + 0.47232368762432886, + 0.382676541076133, + 0.7368520949073498, + 0.4385196794612953, + 0.620921774531115, + 0.5309514174554766, + 0.42678466520707253, + 0.4849831854526617, + 0.4491483039998574, + 0.4589400855827708, + 0.5793047407004148, + 0.5364588000336518, + 0.5253714586155485, + 0.4356828774277688, + 0.46170081612152836, + 0.4875672559455071, + 0.47772437867074324, + 0.7633620063576941, + 0.47432786260994253, + 0.48120965980885466, + 0.4294097234599596, + 0.31041829647352875, + 0.6617995443200804, + 0.41556177136972167, + 0.3448827376536166, + 0.5136969989154201, + 0.4396045840176475, + 0.33790202833282906, + 0.4824362020215267, + 0.43388313950580304, + 0.7730507811900802, + 0.31500644506342135, + 0.39389001907990884, + 0.34917172942746083, + 0.4279617314344341, + 0.38910889684135647, + 0.3190543353718688, + 0.4016898480423829, + 0.281214713575449, + 0.4169525207727743, + 0.7313938785113685, + 0.319229928321934, + 0.40532624348200313, + 0.49859705639539675, + 0.4152720045990162, + 0.4021147995663312, + 0.47521534830794426, + 0.3475137042082608, + 0.5147332191959668, + 0.514570313758415, + 0.2660246617895734, + 0.6295254877306903, + 0.6800079546088417, + 0.5033698943576784, + 0.4812924484857193, + 0.5344610476333804, + 0.4258831199484963, + 0.447823822644632, + 0.5564735785744387, + 0.7762475504127687, + 0.38995527142305536, + 0.6025554631759169, + 0.33863241569472396, + 0.459167719325342, + 0.4749734055043428, + 0.47394155590038706, + 0.3965080861833165, + 0.3037072975980833, + 0.34085174819734965, + 0.38686223311569257, + 0.5662998256310687, + 0.45816425626602175, + 0.3325458828384536, + 0.40571081073974863, + 0.4604867030904958, + 0.30670597084412166, + 0.5374789138094137, + 0.4767734409421903, + 0.4020866443829543, + 0.44342104247538316, + 0.3261806840448596, + 0.4229248321898393, + 0.4880562104377161, + 0.5804885926026773, + 0.42761151333613273, + 0.7762188513940697, + 0.3603936830291841, + 0.554145103173105, + 0.7371245520212185, + 0.6298782153725457, + 0.3166007570835226, + 0.4958483638143899, + 0.41331762329189625, + 0.3488021695545355, + 0.3636707950654013, + 0.378328290749159, + 0.4562379316939263, + 0.4964828437800659, + 0.3685230972877922, + 0.6092977977770089, + 0.7581624146919201, + 0.42361088548337733, + 0.428482582194258, + 0.4297096965984624, + 0.488736120353091, + 0.5023521744281129, + 0.48647171255851035, + 0.495509141127562, + 0.35743581803695307, + 0.6487152077671784, + 0.4794767132309024, + 0.5470598718336126, + 0.7065853572164498, + 0.3251540080832852, + 0.5897974017290802, + 0.37883433667657296, + 0.4509907354596212, + 0.34225953396803716, + 0.5144460898601524, + 0.4887675340935965, + 0.33441526499037383, + 0.41934892111843575, + 0.3994478112199798, + 0.5335863575256329, + 0.5861374920990683, + 0.44922362766889323, + 0.4522544099819673, + 0.40706020861552306, + 0.7336964578226545, + 0.4529619618700222, + 0.5110923586456109, + 0.458083163144107, + 0.45803618624134196, + 0.6012436653076318, + 0.3439455943178023, + 0.5678002831043996, + 0.3718042773537551, + 0.5404061400645778, + 0.4203113292902737, + 0.343462069867278, + 0.35788275099460887, + 0.46270252065618556, + 0.36701838242738577, + 0.5052075916653762, + 0.7240298069747193, + 0.7180238831149154, + 0.396579482871477, + 0.34944143446527526, + 0.697299799235059, + 0.44792711531094975, + 0.3895496070326613, + 0.5107346693870441, + 0.3574435819782197, + 0.59379812020871, + 0.4265693638611862, + 0.563432458059295, + 0.5722334306340539, + 0.4299442950950354, + 0.5043125219077408, + 0.6503721109704493, + 0.5120172280547173, + 0.358116986163955, + 0.41332104948115456, + 0.30545022721461834, + 0.531065857732878, + 0.3488874293717495, + 0.5129666211907787, + 0.4929454028960776, + 0.38184614322332433, + 0.7173139903924617, + 0.3318552943855518, + 0.7233046815999525, + 0.4350730148942997, + 0.5279303826976262, + 0.742155892185052, + 0.47157957666444145, + 0.5495846185608259, + 0.36716516016213874, + 0.6979971135514454, + 0.40708190092169205, + 0.3275480448700797, + 0.5518937513782085, + 0.5153868864388004, + 0.634148942321344, + 0.5475986613886548, + 0.39462610584812224, + 0.47925777964502164, + 0.3039827449146017, + 0.44680781073138653, + 0.563328862662936, + 0.5072385437663948, + 0.35461317444659685, + 0.5351516418122418, + 0.5797626445608147, + 0.5705535633351485, + 0.4372782127770814, + 0.38296372413654334, + 0.5061342214641152, + 0.4577617769581206, + 0.554004245824079, + 0.4758432407523163, + 0.3596891903254092, + 0.4050309941885572, + 0.5031412717556921, + 0.39513448251406047, + 0.7301982571459976, + 0.5168606328828124, + 0.4494822525203608, + 0.4930114799488823, + 0.3960453108368266, + 0.44084434983542536, + 0.5385514573300595, + 0.5464864782383935, + 0.7068847882573123, + 0.5342208448974535, + 0.5721332603405641, + 0.5226693704221548, + 0.4853421880872876, + 0.5733452179818607, + 0.3979944291516326, + 0.6238664778382652, + 0.5907117294022171, + 0.721056594596524, + 0.5152367302320372, + 0.5420564694857486, + 0.5372386147010608, + 0.42163679618828687, + 0.5350669308161268, + 0.3400952834255679, + 0.6775141326353156, + 0.4790536984313556, + 0.7325337136898892, + 0.7511609258075738, + 0.46977899195792744, + 0.44654866109364466, + 0.4977051694814722, + 0.35574562125053055, + 0.491626475558898, + 0.7534950046873595, + 0.4882976751195547, + 0.5085465548427662, + 0.5824559809734992, + 0.5092474556957299, + 0.35226004766368296, + 0.6068357785102902, + 0.42993190853484065, + 0.38726096119359904, + 0.7286169016787594, + 0.411138950202062, + 0.4270972181856306, + 0.3975620607975715, + 0.4830051221991299, + 0.5572267161170601, + 0.6188870606108778, + 0.6744249476875868, + 0.6180720375796375, + 0.5870473696037806, + 0.45358766848419063, + 0.6059892598898143, + 0.55212855548331, + 0.7759238125732574, + 0.7304251241417117, + 0.3524567779043602, + 0.3772408766178647, + 0.35290872528300204, + 0.5116283135802537, + 0.41416112933536936, + 0.7818839370207065, + 0.37666216314794204, + 0.41833012922987606, + 0.5680523705026319, + 0.48175725184548085, + 0.5539044298037846, + 0.6204153442497847, + 0.4612362546085149, + 0.6160945371430508, + 0.38025455307115036, + 0.6763456910904989, + 0.7757463217050449, + 0.7617480051537221, + 0.5847351908129903, + 0.43124545597882935, + 0.5524958161991413, + 0.734744554773829, + 0.6853672167482082, + 0.6208136279283267, + 0.48376262241245993, + 0.4608068904925999, + 0.4923942450612678, + 0.732740830275524, + 0.4215425670353029, + 0.4318360405073968, + 0.3218581727077247, + 0.5939610960024844, + 0.31752602781307576, + 0.40842392350702594, + 0.6639706688417492, + 0.5403654337126585, + 0.6798737334614698, + 0.5334391541021133, + 0.516761391105535, + 0.6627462585995435, + 0.6294464782146787, + 0.563673534595111, + 0.3828310311491214, + 0.641918790188346, + 0.5184703236142126, + 0.4056140093571065, + 0.3574958827613943, + 0.4824229739751922, + 0.6376508914651439, + 0.5089425479105912, + 0.6097774786498377, + 0.5704973788481804, + 0.42467142396632634, + 0.40558938503565795, + 0.6838406505074054, + 0.4993013466284739, + 0.5982709494014439, + 0.7206243601268412, + 0.4170784989984204, + 0.3788116182490404, + 0.37422482081646863, + 0.3717891300110492, + 0.7097465644081515, + 0.732268263905096, + 0.5638695557983415, + 0.42047199476270186, + 0.46422439792998244, + 0.38777247550029065, + 0.7745189426798121, + 0.3460555663522603, + 0.48438826670037577, + 0.33947860667395363, + 0.5074112201169118, + 0.37639179482058366, + 0.5622799704500583, + 0.48011564777112625, + 0.3499908132824828, + 0.6914809508031986, + 0.5066108973078425, + 0.41458003676977684, + 0.34359341675818456, + 0.3527765326172663, + 0.40243827846302893, + 0.4130984215277159, + 0.37042590791148533, + 0.36462742340968973, + 0.3975544188716498, + 0.3896324762330257, + 0.40719607616516657, + 0.3353572796553862, + 0.43487752671368995, + 0.21017035862494143, + 0.251894614544875, + 0.29699514281396916, + 0.3873836367008545, + 0.40939915301506474, + 0.4160102656443206, + 0.31915615081986437, + 0.31510114852141313, + 0.3074232845003868, + 0.6601463642192955, + 0.31479807016348466, + 0.2948601186360099, + 0.7665862823338264, + 0.361496662561418, + 0.41151259383775857, + 0.3325556081984921, + 0.4435791567362348, + 0.28388269253171017, + 0.39255695280698805, + 0.35162254851682456, + 0.4541815470265408, + 0.2411010548258946, + 0.40977657454299654, + 0.3972219449370815, + 0.4991942534866156, + 0.3699730700187985, + 0.46745015420284236, + 0.3047331990825136, + 0.42953110287338697, + 0.38811306478003, + 0.3444239948076305, + 0.4692605313266361, + 0.35226323234553836, + 0.467488890139369, + 0.37333684089382907, + 0.3572375467125403, + 0.8035420889497972, + 0.2719018254046749, + 0.40368311508554083, + 0.36289850384000355, + 0.39244907873761514, + 0.3180549450223987, + 0.3508877379082119, + 0.44939380214150443, + 0.5600369523062303, + 0.34112612791769914, + 0.3828926604012901, + 0.27410135651574635, + 0.40577878399902073, + 0.40366334018052147, + 0.41190836869041314, + 0.4498196981812466, + 0.35913794034781804, + 0.3546379840460451, + 0.4553423405864537, + 0.37498541860791124, + 0.43869462252416713, + 0.275107041696292, + 0.31339729876785544, + 0.4678513302372344, + 0.44769016237002934, + 0.3964134216567662, + 0.3528607191443824, + 0.3513914988991654, + 0.8147527396703372, + 0.31697636158889814, + 0.3119050287911109, + 0.3480177053724628, + 0.2885618494453209, + 0.3041045545588008, + 0.3220493645421199, + 0.2697927596045328, + 0.3527185501808704, + 0.345839287990941, + 0.4702407873488593, + 0.4123330740117471, + 0.44344530651492664, + 0.46411886713580014, + 0.4397150955190382, + 0.38591846903177296, + 0.5605186383004633, + 0.42692100107041797, + 0.4351995260971065, + 0.4908100933259927, + 0.45410429099390476, + 0.38872561556063406, + 0.3188544086057547, + 0.40638130294576635, + 0.5139060195446538, + 0.4921816349199388, + 0.34360482854190577, + 0.38116853651528587, + 0.41117567669831157, + 0.4600772607965956, + 0.3594946160153152, + 0.38643138594542736, + 0.4839974517554839, + 0.529398165014111, + 0.34169069411207076, + 0.42508629799117337, + 0.3501243722472547, + 0.3482245765865478, + 0.34959020365949467, + 0.7832345320360585, + 0.41668081137360286, + 0.43557630382411894, + 0.3430404806963485, + 0.3780927825707894, + 0.3907268225559379, + 0.69479216292814, + 0.42746812798709155, + 0.41546688443008734, + 0.4452946742055064, + 0.41608956678885095, + 0.4270601730911228, + 0.45164947249793297, + 0.49457085410154544, + 0.2980896170531275, + 0.3632406995402413, + 0.4527343807953421, + 0.44481962859353624, + 0.4082886873242465, + 0.4070598588448346, + 0.40123081330979726, + 0.3343877166099594, + 0.3917936008106865, + 0.3979476532013508, + 0.3904102128710722, + 0.6004999381532338, + 0.36662327621159185, + 0.4171652704962383, + 0.3369296864996459, + 0.5682429486980609, + 0.4012456843904405, + 0.3690759737230328, + 0.43385558821996334, + 0.4185695153008128, + 0.42669239031212763, + 0.352273133415859, + 0.5461570076978194, + 0.46797634551017964, + 0.3509626211439112, + 0.4072309842664301, + 0.44401870366947677, + 0.4624799699872625, + 0.3942299908615597, + 0.4677109519741166, + 0.35443173083387425, + 0.6531595157419027, + 0.3910468355942947, + 0.4569247120625835, + 0.46915575157783546, + 0.40563498230935807, + 0.41072863267569965, + 0.4119508016125004, + 0.33582032273152096, + 0.40479956305936227, + 0.596970922202158, + 0.39404543115538976, + 0.3663737695650752, + 0.5175293701277932, + 0.4202006889398754, + 0.409989551446308, + 0.6108039691540723, + 0.39066568201360613, + 0.470034117529108, + 0.40995463265035753, + 0.39767958234926043, + 0.48718377567380117, + 0.3737727069093413, + 0.4798280997516236, + 0.41202712827977045, + 0.4572321534519499, + 0.3699761014885504, + 0.45958562929435215, + 0.4109564813592012, + 0.35086950847728765, + 0.328794674685437, + 0.3587898455479623, + 0.43255071472433415, + 0.5470424766850146, + 0.46168470098915687, + 0.446828744983603, + 0.4762346418790166, + 0.37799564516185413, + 0.37338235708709167, + 0.46739385317663773, + 0.36948267220007636, + 0.4057378474258762, + 0.485365764174556, + 0.39682797935749664, + 0.46813616932462404, + 0.5049242179714845, + 0.39859985683986837, + 0.3983614991330576, + 0.516119373852299, + 0.40905912564068075, + 0.4311700880362108, + 0.333677137866849, + 0.4858097511996795, + 0.601477946317179, + 0.7890152534171613, + 0.42505384113246863, + 0.4086469907651201, + 0.42644772332348596, + 0.36997283216606425, + 0.36021201648928985, + 0.37234408092305915, + 0.44985027297675007, + 0.43710264205615895, + 0.4062751022045809, + 0.4445428165003817, + 0.4356591032570776, + 0.47947042082311414, + 0.3582892882076501, + 0.3465776840334498, + 0.4537245065348542, + 0.40589591275221515, + 0.3871810378361713, + 0.404420943744581, + 0.39995459157106633, + 0.39628141547103646, + 0.29976391071107833, + 0.45509016280474274, + 0.4389076692229332, + 0.3895708310345054, + 0.40182682552011667, + 0.31614112109537573, + 0.42428218808900675, + 0.2974803403454002, + 0.5224881106159194, + 0.3827783451876972, + 0.3546788902203488, + 0.4417382994869998, + 0.413141126502251, + 0.34841517860695, + 0.6000653516199509, + 0.5094110094349558, + 0.347625250579366, + 0.4850481909340746, + 0.41481694834196237, + 0.4295276105337574, + 0.4358782442301181, + 0.31054632206182325, + 0.37505245958894207, + 0.48897235863689886, + 0.37229536772858784, + 0.44164863715240427, + 0.41543066813567087, + 0.4580433146557825, + 0.48213859383808755, + 0.387654733439039, + 0.49915980065431614, + 0.46904726856533596, + 0.39100989337533915, + 0.3434454696181428, + 0.4370582159175184, + 0.3595220418258275, + 0.40561383284559926, + 0.40815970071575136, + 0.47933490677001195, + 0.5220750033372771, + 0.4447479616842667, + 0.4160776740478869, + 0.44670328160997996, + 0.44277554908719347, + 0.43259387142030636, + 0.5199136875623462, + 0.3969033107139684, + 0.3365952741617994, + 0.4910810314222572, + 0.4147454268157038, + 0.4439829156476068, + 0.43354584682926245, + 0.42450540718047086, + 0.4957951700924011, + 0.3613766770457205, + 0.41346370853229675, + 0.3540991463090008, + 0.35442454989602196, + 0.35347111267254616, + 0.4143627360492723, + 0.35498282184605523, + 0.4326590439780495, + 0.5250088276776085, + 0.43280482297063017, + 0.48276022912379146, + 0.7720240602129973, + 0.5394745616282545, + 0.43132826932057944, + 0.40924435466232995, + 0.4456291452779506, + 0.43279589893524933, + 0.3374729174162184, + 0.5683372591871224, + 0.4630345060772116, + 0.4093763928669486, + 0.46688579449190054, + 0.3951482760578102, + 0.49544081315303595, + 0.514033744476223, + 0.4015868040443054, + 0.5396949429963656, + 0.4911671205475939, + 0.36894011854519515, + 0.4487869174763796, + 0.4980039906286237, + 0.4020690528678452, + 0.421226389401242, + 0.3462448499283906, + 0.5590435024842271, + 0.5050657145855977, + 0.8154912904567359, + 0.5209990835202201, + 0.44365840740761964, + 0.2845912569952488, + 0.4699632597914101, + 0.4307168876843307, + 0.6055900625126112, + 0.3456226258660438, + 0.5198940434330774, + 0.33410696438912446, + 0.3779648515964963, + 0.35889633514407243, + 0.34438401491174425, + 0.4376824606869121, + 0.3626521294370115, + 0.42520535051144465, + 0.44875720388264106, + 0.40766514563870143, + 0.35138513271414246, + 0.35747019472351954, + 0.41326631675601055, + 0.3444471971546922, + 0.522135176073662, + 0.42243181301363486, + 0.4927776617807492, + 0.41598553634443414, + 0.38067626046903097, + 0.5584131733755815, + 0.466622690568927, + 0.36823581627524293, + 0.588352295216724, + 0.31825036257189826, + 0.4443869184446313, + 0.36642423270112734, + 0.3546201207747407, + 0.3367889282986031, + 0.3838662888567503, + 0.47988335304160723, + 0.3430858199861577, + 0.6869899565701822, + 0.43570724780594633, + 0.3738120637738894, + 0.39477125329587154, + 0.3569043988031845, + 0.581019142110864, + 0.3219340513584127, + 0.3780108728868585, + 0.38336363671776397, + 0.43314303276545096, + 0.34848369290419473, + 0.4176675992318217, + 0.3140810172849974, + 0.43266425104226425, + 0.7624525282517548, + 0.45498200520092025, + 0.38476885035533065, + 0.40135852870639077, + 0.4236341681228129, + 0.3240826156260747, + 0.37645918583114407, + 0.4404845073592483, + 0.7471614800046629, + 0.5512520689860609, + 0.44523709934082006, + 0.39058413670059455, + 0.3510527214983374, + 0.40481984094871576, + 0.32529836541118934, + 0.4376321302611577, + 0.5095704403065104, + 0.3254895027071674, + 0.37858193154140474, + 0.3760018823337645, + 0.46537035788504816, + 0.3574137176999619, + 0.5182727332749382, + 0.477834988951795, + 0.30595683037619836, + 0.4111554826350052, + 0.4214004175005238, + 0.5161574021642698, + 0.5144399275840327, + 0.2418743359622363, + 0.711635296666277, + 0.7043190114784332, + 0.20405525278721098, + 0.3099135359674172, + 0.38850816450666686, + 0.4471802876331205, + 0.28689866837174455, + 0.3886845927743964, + 0.45517903506512186, + 0.4661553988582704, + 0.46720348968618186, + 0.36621971066717013, + 0.3749309007905307, + 0.34998929655475625, + 0.3180818341297564, + 0.4274793538757972, + 0.3850551131953162, + 0.3764523193328948, + 0.4908482837155606, + 0.41179349410087196, + 0.5194966715632169, + 0.434546083077568, + 0.6008612543263209, + 0.42480087465281996, + 0.4365328699227394, + 0.48125835311822956, + 0.44208663584082597, + 0.32928631666847025, + 0.6147899722000231, + 0.6015847941839029, + 0.4367749797517611, + 0.36422208790539257, + 0.38406362051339066, + 0.5145015096457973, + 0.3967754150926034, + 0.45691268713131655, + 0.44381509750983683, + 0.4708358608848941, + 0.42675605771129704, + 0.3676642768817904, + 0.3821615499101836, + 0.5377702627708292, + 0.4763421948194272, + 0.4550495738630009, + 0.476086255960824, + 0.38098878402733055, + 0.2614597670611096, + 0.30052892243205814, + 0.3344008523183058, + 0.5154841994448766, + 0.45243981547372714, + 0.3026592207180323, + 0.45577776097611294, + 0.35990746350891445, + 0.32554545860900597, + 0.3656313379720618, + 0.38571935621511494, + 0.35646387875337654, + 0.4292903343355035, + 0.6165300679923973, + 0.35244777969043883, + 0.435000167260033, + 0.566238615187761, + 0.39390364682304235, + 0.508646962163032, + 0.5817095539592309, + 0.4446141532596193, + 0.46714857753743405, + 0.4486080014002608, + 0.41403857179918546, + 0.3099840673548793, + 0.3593405214172165, + 0.475886813597744, + 0.5017652042217762, + 0.4658111567896932, + 0.33869505940969324, + 0.4569323296387595, + 0.46537441089947057, + 0.4963798549838091, + 0.40371319148648943, + 0.46285490993846634, + 0.37821595437139877, + 0.4647309760357942, + 0.4305633355382973, + 0.4087314679714068, + 0.3338790901637554, + 0.44785498723310396, + 0.4653529656906876, + 0.2952034136450152, + 0.36688407637394554, + 0.43816435038608087, + 0.39218044206797287, + 0.41704774571378617, + 0.4825737976332283, + 0.49802879577037945, + 0.43780714980484126, + 0.46060893523310287, + 0.35885637210133187, + 0.44864249882795953, + 0.42167440597465583, + 0.39776158225474123, + 0.407781116103941, + 0.4401080276650293, + 0.4090997954655266, + 0.4643367522132077, + 0.4756798381270012, + 0.7151552256618808, + 0.5387369290561485, + 0.4962776910615659, + 0.5346118579838656, + 0.5077737347571822, + 0.49197735009880844, + 0.5306499932354956, + 0.3565103476850586, + 0.3969504137356053, + 0.4612390148617836, + 0.3448086608897205, + 0.4467974706799465, + 0.491237687896508, + 0.4366578037419856, + 0.44325359431145456, + 0.4582718875068136, + 0.323053445092993, + 0.5247984874914619, + 0.48681304961771604, + 0.38187669878524166, + 0.42920133123918575, + 0.3833664520313884, + 0.3559704019755758, + 0.4467899010456256, + 0.4397898530778588, + 0.4217201469833285, + 0.4555381175262689, + 0.3904964487098678, + 0.47501637637465743, + 0.3571166873155241, + 0.38295403492209823, + 0.43161067858808094, + 0.49739664598757005, + 0.4914935057056789, + 0.4628329036142741, + 0.4025942056243457, + 0.35059341957751367, + 0.3539913799245288, + 0.437464690451922, + 0.4128981022852694, + 0.4469400434989196, + 0.47230044236096386, + 0.42776836331208823, + 0.4025160980104294, + 0.3554938568750314, + 0.4093466356854035, + 0.5189063736195189, + 0.40285806627905946, + 0.20973991412304432, + 0.35087096558160813, + 0.44395875749092156, + 0.30650357187477345, + 0.295641790083179, + 0.4565719301324631, + 0.45321120652500857, + 0.3280632376117724, + 0.3116014214041635, + 0.3979506135918833, + 0.38829400886074783, + 0.5137656584294331, + 0.3051609122855236, + 0.31141959352311244, + 0.3284242300872484, + 0.3442833641445462, + 0.2684686639270837, + 0.3255804183674078, + 0.3090678183755031, + 0.33257401505566997, + 0.3238094101036576, + 0.43225383138977486, + 0.3478525448841358, + 0.3908831775489488, + 0.39375984856529, + 0.48284568541588113, + 0.3401476691250242, + 0.39897924632537735, + 0.39325011491440603, + 0.4024907787268453, + 0.3963342373123936, + 0.40603353614288684, + 0.3941961817359008, + 0.5994913984330307, + 0.3154202201752053, + 0.4272930799646164, + 0.5507260538384264, + 0.47203824420005136, + 0.36224965002946186, + 0.4504583151195134, + 0.4041717993023597, + 0.4437761342384833, + 0.46296237391043243, + 0.3558552523123049, + 0.5667067463688196, + 0.4188562911648837, + 0.4119839503247844, + 0.35270747917848705, + 0.4262889306132423, + 0.38391059343551986, + 0.44548550046526564, + 0.47203105228694076, + 0.43694880118861873, + 0.4616070732455641, + 0.5653377146473816, + 0.42821236375589306, + 0.3989387836170987, + 0.5012011480948783, + 0.5713357748153283, + 0.5135588881040135, + 0.4380748677700974, + 0.54610033178827, + 0.538076692661185, + 0.3760057728140517, + 0.37986688180818595, + 0.42860281381090143, + 0.34077343840758617, + 0.6285817896828888, + 0.4821369123844199, + 0.4151230666347793, + 0.3475579231119172, + 0.45709330773378415, + 0.39431532317297036, + 0.5004892515090617, + 0.4325808310387795, + 0.41114511976573304, + 0.5167459707870526, + 0.3173923844911934, + 0.40190789185230097, + 0.5213953254460083, + 0.33882021583292615, + 0.4143059987943092, + 0.31673270123559133, + 0.36031037236326885, + 0.3828335161311349, + 0.38947070742187323, + 0.4511757038506873, + 0.3398739603870646, + 0.31523452057993906, + 0.421155052601683, + 0.3869018559421007, + 0.361415027914783, + 0.43790993992322946, + 0.45937359260095917, + 0.412648571762584, + 0.5141746192600885, + 0.5550566553210584, + 0.44451478977182235, + 0.3458372747135924, + 0.4694914038266204, + 0.4295001618559262, + 0.3834901516157567, + 0.41682359465853613, + 0.31801473408506514, + 0.31855445581040165, + 0.3007696697071766, + 0.35847116288741904, + 0.4241065341863381, + 0.6418801289241761, + 0.4244229654601202, + 0.4127844018413648, + 0.29531968300021194, + 0.35446209413089314, + 0.31545812683827545, + 0.4045703265543587, + 0.36463020706615124, + 0.38503207928182687, + 0.34889265506522926, + 0.4157605927669199, + 0.3778373639687359, + 0.41911081277404116, + 0.39996214600564406, + 0.36160312830285035, + 0.490255506864759, + 0.4881019333341418, + 0.37243992573979895, + 0.4223272583506213, + 0.42512916725814764, + 0.38043359472483074, + 0.35315398653886604, + 0.5053887831273512, + 0.4407776639905223, + 0.572442771090006, + 0.4425939323695977, + 0.4344631315718188, + 0.4414518925186781, + 0.44455328437042996, + 0.34229254570189654, + 0.5780210711422236, + 0.39485962219129717, + 0.23201986791830762, + 0.45364853635853314, + 0.5744082215683648, + 0.359412995137383, + 0.4542909294941787, + 0.7278571293446744, + 0.43430087717573834, + 0.5062006547385698, + 0.30512937475602764, + 0.47633826341069296, + 0.31684859617679023, + 0.3146556583465347, + 0.32281646837710753, + 0.28084640538732075, + 0.4332483363175529, + 0.22885853780405485, + 0.3715711382421613, + 0.43227172906271505, + 0.42718662616201575, + 0.38929990796636305, + 0.3858887431450717, + 0.33036721872681013, + 0.4331622145398575, + 0.5076383162172651, + 0.5284550161883238, + 0.33860779991150863, + 0.44153069733026556, + 0.39681011930054105, + 0.7197796918975259, + 0.38302596612129924, + 0.5509851040185606, + 0.7304311016063988, + 0.7084068564596221, + 0.4580245519712455, + 0.35408685731339407, + 0.3022275807545951, + 0.44036164776816616, + 0.483651614598495, + 0.48357804404019755, + 0.5532292059447668, + 0.41273997108377286, + 0.5415527230915046, + 0.4086368120147909, + 0.3793437258769786, + 0.6908135230210734, + 0.6744110979385395, + 0.45994791145381125, + 0.40503492176001593, + 0.45284281139273574, + 0.4084896847186553, + 0.43871070979652016, + 0.6367408468040754, + 0.6516982124128302, + 0.4141685293100692, + 0.42581861885661865, + 0.4258099676635724, + 0.35941223068107914, + 0.4713531154780257, + 0.33855376908334345, + 0.4710581849802589, + 0.4857079614865136, + 0.34074021913653285, + 0.4307719130652262, + 0.3952594639553078, + 0.3261635997213606, + 0.5152906538493687, + 0.44283387494941884, + 0.39289165983961755, + 0.361113017710807, + 0.466703406144271, + 0.41317466663081515, + 0.3300482546893578, + 0.3353143798750907, + 0.2849876358933387, + 0.24087833773167125, + 0.28136673339699936, + 0.4427014955787184, + 0.4844619714750543, + 0.34851400703675667, + 0.3197706315925404, + 0.34985250622150876, + 0.46517755351481976, + 0.41287412143441665, + 0.4803942764214093, + 0.35118102854109823, + 0.45893558327202555, + 0.5087811102545998, + 0.37577194860973784, + 0.5048557173902095, + 0.4656482424834399, + 0.42415031472188114, + 0.41268399308525155, + 0.3525536325755183, + 0.2303488162583641, + 0.2857030914423679, + 0.27002031382543906, + 0.5069955091645334, + 0.43275888219604497, + 0.42368044075943356, + 0.4840445283407476, + 0.5006381625930821, + 0.32805133017058596, + 0.4071511024918365, + 0.354900659469099, + 0.7140209664412628, + 0.4277448040392385, + 0.44028134069458646, + 0.3522834565155501, + 0.43790564802424875, + 0.4734043812076242, + 0.42221147017879784, + 0.3944850258314945, + 0.46773233550043347, + 0.4367121374375035, + 0.5400714149540825, + 0.40869765474540176, + 0.4739013537508259, + 0.42933296420232975, + 0.4196965119822368, + 0.40197271174954774, + 0.45647886406886157, + 0.4235088782799683, + 0.4739107028303046, + 0.3393247189173349, + 0.3309420439688674, + 0.3256151732190015, + 0.40273320336289414, + 0.4440585064430785, + 0.3934066381387973, + 0.42295118697077294, + 0.4618698712460071, + 0.4136885057633291, + 0.4225911377039474, + 0.42514499937418027, + 0.2658270807735076, + 0.4125645063177218, + 0.31429244982607835, + 0.42057534913087263, + 0.22419956609099792, + 0.6615428837232024, + 0.31247377350539535, + 0.44249327523454096, + 0.5498832386281618, + 0.46940564493518405, + 0.27182663531232915, + 0.4180200733670056, + 0.33415404262347287, + 0.7677344006597833, + 0.28026340998082955, + 0.5100909882396957, + 0.49308499369862646, + 0.5077240870885004, + 0.6300900630129532, + 0.4245020767673464, + 0.5737197610727971, + 0.43596285561813897, + 0.3796730326487695, + 0.32929493931602183, + 0.44025947839419727, + 0.3066720228582651, + 0.7077617125467475 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.46969800453339533, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.5474424069533981, + 0.7467119826392321, + 0.7243318542498888, + 0.8096748491851733, + 0.7185287084696462, + 0.6534755304361215, + 0.708451570720467, + 0.6546322992059388, + 0.4260217764120566, + 0.7006380579753476, + 0.35005418809566946, + 0.49059771976457683, + 0.410151827304283, + 0.3698236458808318, + 0.40794140740987306, + 0.34078360238121147, + 0.3868035210639765, + 0.562092399647556, + 0.5656498202425744, + 0.45618599455869946, + 0.3466639668313664, + 0.6742181273003842, + 0.46005402123246747, + 0.6049251693720739, + 0.3758693102755983, + 0.3218029900640583, + 0.40342413210190564, + 0.4939710257728302, + 0.6096509757684188, + 0.4944580915542442, + 0.33008328389805913, + 0.41250450318658544, + 0.42063751123772647, + 0.37447050637041784, + 0.47712966611152874, + 0.3668556049257026, + 0.48193935446918557, + 0.5339587445054333, + 0.5895101170659891, + 0.6234595399838107, + 0.7819696744340288, + 0.7738917728687106, + 0.7674947934829647, + 0.6459572286864016, + 0.4668046963297513, + 0.5283161502328495, + 0.7720550486898112, + 0.41977648704809295, + 0.3844051813556676, + 0.5319891084171561, + 0.42685008226803633, + 0.7712070990600024, + 0.7877280194151065, + 0.4727959208844214, + 0.464160278024664, + 0.40027307691879466, + 0.3786069010892389, + 0.43322051222190244, + 0.4203127798978411, + 0.536118379251527, + 0.495927652819004, + 0.46085657568580723, + 0.4712491474352024, + 0.5104363438006662, + 0.5894689401591549, + 0.4216178360214837, + 0.4031901985905383, + 0.584120625246884, + 0.42406747139273043, + 0.4832970624115218, + 0.4879614086620671, + 0.4900843904427079, + 0.4622734112711138, + 0.5169564443850012, + 0.4319230999534448, + 0.5286604410357681, + 0.3328222449559247, + 0.37665311734619183, + 0.40982455688503533, + 0.34926451034986106, + 0.44785734384646697, + 0.5724712213312038, + 0.4030189976043789, + 0.403064537607534, + 0.3989164474006617, + 0.4817388335536478, + 0.46012039446514125, + 0.43892241941676724, + 0.3662191027574034, + 0.39483145785391804, + 0.3533586835087929, + 0.36102560592702504, + 0.5136319809741818, + 0.5461054640968582, + 0.4442273581130771, + 0.5816255250419828, + 0.3908522826013536, + 0.5532216023563027, + 0.42943044650213774, + 0.47097801506814874, + 0.4578528147775352, + 0.5304008060854425, + 0.32773981167758165, + 0.48776006834614044, + 0.3577988306531283, + 0.37243729732466546, + 0.48471284929195957, + 0.5388195469128967, + 0.36256123957734526, + 0.8017434842229937, + 0.7966018124562895, + 0.8019647943910354, + 0.8213909605716182, + 0.5789170211479562, + 0.697805127466733, + 0.7115622306040463, + 0.7507478910378561, + 0.695666609732016, + 0.5826956865137377, + 0.6531753838889041, + 0.568993896887227, + 0.7670467908355129, + 0.6587833534806437, + 0.43551550750004564, + 0.35121168606577835, + 0.3269481189467253, + 0.3569865563657992, + 0.5575906279699023, + 0.44850596838894197, + 0.3100890555562949, + 0.47204808955125965, + 0.4370037421728388, + 0.4919836255662392, + 0.3861889777527388, + 0.4205823508165346, + 0.4829311827593811, + 0.4734881190561036, + 0.37892528649558455, + 0.4074334492640315, + 0.405616151272061, + 0.37729031191984447, + 0.32897897532508436, + 0.44058700425329933, + 0.607274551383629, + 0.3209713704800163, + 0.5638902295930076, + 0.36647733218820105, + 0.48355549552953836, + 0.44696311182340615, + 0.5783493647287499, + 0.33458973935298947, + 0.4934225601794873, + 0.34158097436907014, + 0.3526752650745808, + 0.4145449207037831, + 0.36221208664386306, + 0.2888914788507063, + 0.2138506074185173, + 0.374558553520131, + 0.3781178416630587, + 0.5080667540883422, + 0.47869169385786636, + 0.5456463375051119, + 0.48891359093953085, + 0.3906205756861582, + 0.4851905755267554, + 0.508480440478473, + 0.49920763354529946, + 0.4333095307647449, + 0.4677707658836938, + 0.4210708196618706, + 0.32651973214052926, + 0.4835320802412503, + 0.46101441040203417, + 0.29756860332161367, + 0.32107548845550143, + 0.5001106038825176, + 0.5569945535955934, + 0.5725513785673042, + 0.47973965845221217, + 0.34436054243277636, + 0.5439774549222647, + 0.5060424746908594, + 0.5137357165671961, + 0.3217983684486791, + 0.5477744218415975, + 0.4753783386480914, + 0.5156866925490249, + 0.27306078258574396, + 0.4398746692920347, + 0.35138802242244555, + 0.2851200801443185, + 0.5263357779358079, + 0.45290415212004226, + 0.5508329607906369, + 0.39675702745736197, + 0.5865637261334528, + 0.35730772262086824, + 0.46867276902660304, + 0.4219855872507988, + 0.35594351331226387, + 0.28793996395076865, + 0.39319425993077267, + 0.3869034637831793, + 0.6250896078990585, + 0.37594657458865977, + 0.3055468024835225, + 0.4560870890092174, + 0.5575473396744094, + 0.4217313517186797, + 0.46571909036613657, + 0.4224405052577976, + 0.35378288487367504, + 0.3934276691733247, + 0.3865454882068809, + 0.518799556561851, + 0.5944809332263404, + 0.5961495831326304, + 0.5599250826761318, + 0.35124965315665907, + 0.5196969323217931, + 0.4673332734984717, + 0.2967573161390823, + 0.34929827586546786, + 0.3683167411047439, + 0.23210172646097085, + 0.5490246226284328, + 0.35050781315936536, + 0.3332193325722421, + 0.5368008064280221, + 0.3048452915384514, + 0.2631959124203035, + 0.533754560914293, + 0.42037900957873175, + 0.3320499127120128, + 0.6032385253279644, + 0.5522356945771348, + 0.3177696797255922, + 0.31470674905173324, + 0.48809764560674557, + 0.33553826997723735, + 0.46011877273114954, + 0.33732220662352447, + 0.5026207859050013, + 0.4656665519295208, + 0.4088424287346347, + 0.37157917974602644, + 0.34782608699726175, + 0.4779708023491439, + 0.36349897497275513, + 0.4418726232530924, + 0.34200352161255704, + 0.4108778616645648, + 0.3709421529686354, + 0.3782547632393932, + 0.44659379758399, + 0.42220830265909426, + 0.3887074797352048, + 0.435431162114738, + 0.4502169381140237, + 0.3393625595962907, + 0.5379358661054768, + 0.3450260542183952, + 0.3653813836184627, + 0.4056766869084942, + 0.3324811595719864, + 0.3700067993893174, + 0.5232412151065801, + 0.2886969629962881, + 0.3588152959167939, + 0.3504944734584414, + 0.3627970148348814, + 0.37368437392172493, + 0.41810124418710537, + 0.38854456653236985, + 0.49160491981084176, + 0.3118893787035597, + 0.4558948309643367, + 0.510502120351095, + 0.32764252536374056, + 0.3221034818132668, + 0.6032607178983512, + 0.49342809960728334, + 0.2955711052364365, + 0.5377744264244385, + 0.475845911442774, + 0.44479346041874124, + 0.4365105493738324, + 0.506486415837849, + 0.49246377152852955, + 0.5143715331915, + 0.5052380917134824, + 0.5104243400828525, + 0.4216007723374515, + 0.4504921465584557, + 0.5636814460215165, + 0.48220143809755983, + 0.4766383871110487, + 0.647286555341537, + 0.343965852390954, + 0.6027890403195822, + 0.6536590177058159, + 0.7184404810121877, + 0.6920410699599131, + 0.5362110314749556, + 0.5733507276259406, + 0.41040113589910254, + 0.6850578878543884, + 0.4957278687789255, + 0.5068982023230892, + 0.3307644339366304, + 0.5275426522811388, + 0.4358759793580535, + 0.4196061661948803, + 0.5264167312192686, + 0.36002301934657904, + 0.4308219070073882, + 0.33463205470287294, + 0.4616918849130237, + 0.3904029110439583, + 0.35961096586375324, + 0.3949089263383433, + 0.2906348125106056, + 0.35594284501339546, + 0.3917761545068411, + 0.40129110761812115, + 0.385172186280241, + 0.3981174127896842, + 0.5991418838245861, + 0.4923119712602356, + 0.453013757458661, + 0.4985419259698821, + 0.44909068530817914, + 0.34156182704138, + 0.32049701774693623, + 0.5558428248236229, + 0.3986385244289473, + 0.3497321075480728, + 0.4537693317127838, + 0.5633315336642348, + 0.5893158370792713, + 0.447446150981705, + 0.36597764361241664, + 0.4881738467107494, + 0.3793638983039564, + 0.3536725154632614, + 0.5104782990495609, + 0.3509768544856082, + 0.8159217763983909, + 0.541230020922975, + 0.4460721772705979, + 0.4472251150043499, + 0.4757768121252575, + 0.4698951346353495, + 0.42745478544491444, + 0.5482568197526109, + 0.4154738885899649, + 0.5674343998891265, + 0.4742979094638701, + 0.529054797474156, + 0.5783549828545358, + 0.49238187589744725, + 0.503311791114436, + 0.4135314858401914, + 0.43419437560804613, + 0.6196806446065365, + 0.5968216058765927, + 0.4458371921568653, + 0.6108051708188758, + 0.4691317591564793, + 0.5742287979423104, + 0.4694075976177652, + 0.5093387704611609, + 0.45547658044279743, + 0.5461645890189181, + 0.5625463197603947, + 0.5943102312783116, + 0.5365225335463438, + 0.5280710771581535, + 0.520156795412281, + 0.4882452825405846, + 0.40613486685749867, + 0.4101851527140994, + 0.3319326510558966, + 0.3886346324508858, + 0.3341145239202383, + 0.3503830066575222, + 0.3708002368280661, + 0.5101698535230027, + 0.36483684597386856, + 0.35070591248045824, + 0.4109925732579608, + 0.5564198070621211, + 0.6325801241143624, + 0.5826015589297928, + 0.5723877218647498, + 0.43003062365839156, + 0.5845615051087846, + 0.4618593091487409, + 0.6447382973003853, + 0.4825148537303435, + 0.5752708179611599, + 0.45096972258697565, + 0.5578590995235483, + 0.577937129852851, + 0.49453460772422914, + 0.4858973424297757, + 0.5242926513167148, + 0.5849853304906225, + 0.5366838732000504, + 0.3803692948932409, + 0.4065669668394246, + 0.3568720329681659, + 0.3806135443153779, + 0.5275268173840536, + 0.5629335011003876, + 0.3306427370633338, + 0.6128901268994522, + 0.5806003215304311, + 0.4878288322806147, + 0.41149911581989085, + 0.47554737604651637, + 0.5596591371031535, + 0.5179272135962507, + 0.46374626569327515, + 0.8365009852738573, + 0.45825543197690627, + 0.5433521574688631, + 0.5133729016666768, + 0.5506153024685481, + 0.46923894842738323, + 0.408405986804917, + 0.5813013453427114, + 0.46319971447889685, + 0.4243186806605387, + 0.45580039668052746, + 0.45377775449645835, + 0.4609346874576062, + 0.427187905616931, + 0.4588604908281525, + 0.44996491998510896, + 0.48092812051723755, + 0.3295461472315277, + 0.49254039526226473, + 0.3372784453793168, + 0.39892668124038383, + 0.30206881053809254, + 0.4943563005332423, + 0.3878151239193806, + 0.33710897203729134, + 0.5337923851605368, + 0.40910939928175827, + 0.2942765838590372, + 0.5417687546499739, + 0.4843804174303122, + 0.46549734153650263, + 0.5241265413873507, + 0.41682987812971667, + 0.4231285995108825, + 0.6181871909447137, + 0.4374496076323794, + 0.5493621041169229, + 0.40012517877109866, + 0.3126123373831167, + 0.28680174216725507, + 0.411963666627898, + 0.49459514650924064, + 0.580060312116641, + 0.5710359099374249, + 0.4835796542871498, + 0.5351608751129173, + 0.47674500343659737, + 0.505598515872193, + 0.43144370447549013, + 0.41641701574507867, + 0.4297653922135434, + 0.39747180116262365, + 0.4450788283812231, + 0.3990421605466431, + 0.4258732891337633, + 0.5103352463498849, + 0.44010314934554995, + 0.4394350290395068, + 0.372699198939751, + 0.3844440633172706, + 0.4971769168277936, + 0.5410867182345349, + 0.5344870148480378, + 0.5432761094792194, + 0.5662499403184235, + 0.6020591580030853, + 0.6579298806574037, + 0.6410904708317537, + 0.4729098443657255, + 0.5179005503749661, + 0.4866028199311697, + 0.5319523118637879, + 0.4146440033247464, + 0.4992790208397255, + 0.46920227533931264, + 0.570454735246942, + 0.3953078409009475, + 0.5661782025871543, + 0.6408940639297405, + 0.6097653273620359, + 0.5683695072788625, + 0.6349859216580314, + 0.5368530792087224, + 0.47193536529979907, + 0.38850873217709564, + 0.4298100385818947, + 0.34394311071251976, + 0.38516242742155016, + 0.5449725084009394, + 0.564514028662071, + 0.4510976570460101, + 0.3646467799839839, + 0.39583177980935036, + 0.4764018921579322, + 0.46744150092649966, + 0.4025517427808195, + 0.5187026432522859, + 0.49976716024120155, + 0.382923469128279, + 0.4350787648907984, + 0.37901870581737834, + 0.4535545767765994, + 0.46668391852781704, + 0.3583674617973066, + 0.399518417162086, + 0.44257902180715186, + 0.3887963281826643, + 0.379893537380984, + 0.41192226159723716, + 0.3696204761347809, + 0.3896895442700401, + 0.4095871483327368, + 0.5116506000754976, + 0.40341265461576414, + 0.3080116077176654, + 0.5663742748436041, + 0.4039263900127973, + 0.5362248517867573, + 0.5362465511366786, + 0.3832692648531014, + 0.4989000970476725, + 0.5759302669476137, + 0.4950358954534502, + 0.6227341858290019, + 0.44050333803159875, + 0.5073134670908395, + 0.48586622284403913, + 0.4727238052996597, + 0.5296944529370955, + 0.5065937504858418, + 0.3774374272413462, + 0.47319782446280445, + 0.5056402204065349, + 0.4344495299421501, + 0.33141008114202075, + 0.46011548938480934, + 0.49155464951460326, + 0.4611832314995447, + 0.6204065759381578, + 0.5332332944948289, + 0.5414374247399636, + 0.5527959709338188, + 0.6160303407365744, + 0.5443573456611976, + 0.5920612268507104, + 0.5978714017001966, + 0.592991527720053, + 0.4031458002907635, + 0.4342875500356432, + 0.38960489385896047, + 0.3409140012115782, + 0.3112146734583273, + 0.36279330200359305, + 0.3985312362591885, + 0.3540321922905103, + 0.3131894076652937, + 0.4845276327778721, + 0.3089759346012755, + 0.45935018095484664, + 0.3536333500411953, + 0.5311675646111297, + 0.5740778555526358, + 0.506877869702038, + 0.34264388070339846, + 0.4022435748299378, + 0.5765449650283575, + 0.35887679071643236, + 0.4949319845127842, + 0.3293667157551308, + 0.8123017811936264, + 0.5292746064170097, + 0.4280570477489769, + 0.5994390276279769, + 0.5468699271881701, + 0.49389013225917333, + 0.4346788743828227, + 0.49107128248344745, + 0.5591789642918599, + 0.4496030756334192, + 0.5676356106590018, + 0.5361250000899249, + 0.3105063662015885, + 0.49407929230856673, + 0.3212334582683439, + 0.3839606683727805, + 0.3291264632268386, + 0.5159563341178454, + 0.4336627963708862, + 0.3899473592915388, + 0.5284440426290195, + 0.5545811705158558, + 0.3948649900078022, + 0.4918474900828648, + 0.4463694872648879, + 0.4697503600976986, + 0.49915097521039276, + 0.4729959782934994, + 0.3601511673652553, + 0.35123612298663315, + 0.5658417599412965, + 0.43428443129800065, + 0.38885802460762137, + 0.5182337894571961, + 0.4143775196285714, + 0.43665124665688176, + 0.33389058756865775, + 0.3699087915669617, + 0.41808429990107476, + 0.47391540347453603, + 0.4663200262325008, + 0.3518750146645853, + 0.4748536491555094, + 0.29638949628081024, + 0.4891023502345876, + 0.4490679597736011, + 0.4828237849272816, + 0.48282379196915665, + 0.4060062859040798, + 0.47061058712747367, + 0.36818997874123705, + 0.4951194399519952, + 0.40986591735969347, + 0.4543522902210939, + 0.4541792415129536, + 0.36129895155284975, + 0.4975394405527521, + 0.2701616799711853, + 0.6624314485804172, + 0.6658401635659138, + 0.5641173177922243, + 0.5560115399982761, + 0.5992802214971289, + 0.6015800267824113, + 0.2899943487849548, + 0.3903843242906931, + 0.36390318644219755, + 0.39523009559205247, + 0.3471818174303452, + 0.3472245963291774, + 0.39868798419517637, + 0.4655932306295937, + 0.4403924258200095, + 0.3585554819181254, + 0.3813094731615705, + 0.3103568996206726, + 0.4651784723123104, + 0.2560539503421999, + 0.4198235032133469, + 0.46789181919293454, + 0.47948513307324553, + 0.49374082410782905, + 0.42466481563414127, + 0.5223139795050112, + 0.2874865824284717, + 0.5238056935875651, + 0.4512878612425109, + 0.4088319132443495, + 0.3907842353048988, + 0.4235687801553192, + 0.43215040191422377, + 0.4423754813400907, + 0.4866001654786643, + 0.5071083816034022, + 0.538611878569455, + 0.355704277073728, + 0.40013871275704616, + 0.4250179706309993, + 0.47413272083392577, + 0.48519795863017434, + 0.4823882076979435, + 0.3530446850166845, + 0.557385353507064, + 0.4533338241253373, + 0.43999568602295547, + 0.5778631360355105, + 0.39066976625374894, + 0.32958163789108386, + 0.30834228229011135, + 0.37920156143848155, + 0.4278695467241543, + 0.3942988945237637, + 0.40353098250958847, + 0.405908862746723, + 0.5120227593288099, + 0.5793007738641237, + 0.6415933977801321, + 0.5317634796055452, + 0.5553958476040236, + 0.497674017798492, + 0.5509703832620031, + 0.5040403014566543, + 0.5614031144545876, + 0.3778810410277276, + 0.48094295866139447, + 0.5943657262657732, + 0.6468183147393192, + 0.21278577782619246, + 0.3709057966783847, + 0.4276882675070191, + 0.4928663897002946, + 0.4631016903246452, + 0.5268642344504499, + 0.40170411028970476, + 0.5007759622136071, + 0.7346976969959406, + 0.6109626704257068, + 0.7633957161701064, + 0.7775303133843814, + 0.648041076628953, + 0.7387401755862228, + 0.7456338795690668, + 0.3920565990273247, + 0.4088947982527166, + 0.4170588584512105, + 0.518178341303287, + 0.3570903931616992, + 0.47228996811436547, + 0.3814161585653345, + 0.4818560435575031, + 0.41608595503782075, + 0.41124264699773416, + 0.3229566538346853, + 0.4459496556976577, + 0.41130182461331327, + 0.45958669937757485, + 0.44110283740004136, + 0.32013829840348196, + 0.36166342107785865, + 0.34540426250903217, + 0.35994006529290845, + 0.27704468332907684, + 0.3643090170247833, + 0.3152168589588935, + 0.5242081547443125, + 0.5665417967900488, + 0.37561407610389597, + 0.3260648676013039, + 0.2543383764449207, + 0.5302657839748578, + 0.6292562204999355, + 0.6415212273189516, + 0.3858707230602575, + 0.3713244808579177, + 0.26315067386227153, + 0.4941646113744346, + 0.33738932187537257, + 0.39508327225717693, + 0.4860363683400464, + 0.5355673427701633, + 0.47292853445215083, + 0.3267859242373751, + 0.4672673074904411, + 0.5190002969407356, + 0.5135212854322637, + 0.46476620699107013, + 0.5099906292944233, + 0.41708011987105154, + 0.4839964342816569, + 0.4608751261727601, + 0.5201697982213916, + 0.3654470950896973, + 0.4496416063562978, + 0.614396939794398, + 0.43922417328776914, + 0.635156007544204, + 0.3555720451936483, + 0.5465592104404936, + 0.7080389768092388, + 0.5474296353070187, + 0.40137030259829237, + 0.37981423624070093, + 0.3500566696515521, + 0.32791961896260957, + 0.4500997997748047, + 0.584653755127301, + 0.4638088310170849, + 0.410736891585116, + 0.32943552575940926, + 0.49883420429124986, + 0.5107151133704164, + 0.5631602711438631, + 0.5959953388327468, + 0.5083708547823895, + 0.50393254906647, + 0.4078090605624857, + 0.5194863110457849, + 0.3135497789259056, + 0.38954897115876846, + 0.4603543969594217, + 0.398198270219252, + 0.6029784527364093, + 0.5937072493552824, + 0.6111853872188029, + 0.6698146328734992, + 0.7475323489371483, + 0.5956756837486546, + 0.7579469693231372, + 0.4920644278208222, + 0.4279656502935888, + 0.4156198340478532, + 0.5403348853077853, + 0.3829279907049298, + 0.5149029393909316, + 0.5198569666532661, + 0.5301439049961589, + 0.48017869290424553, + 0.4517467629182927, + 0.45521044565859925, + 0.4855452500616024, + 0.402438392358586, + 0.5247604982251434, + 0.4816045357028875, + 0.4880466426828392, + 0.5334763048636867, + 0.587921464664884, + 0.5620040035966002, + 0.41340669434523436, + 0.5326595522156546, + 0.5397516398817689, + 0.3521160758442071, + 0.44717719231468805, + 0.4272884056672072, + 0.42275212053018074, + 0.4094765385375354, + 0.43371346267135635, + 0.4630749342260545, + 0.45626764187232566, + 0.46958590215806917, + 0.44973568478155224, + 0.31899562079475946, + 0.4699664002084758, + 0.5039204660139147, + 0.4457605883280962, + 0.5582798409828419, + 0.5820785755168675, + 0.7392211964753577, + 0.3508823115398727, + 0.34151348760637296, + 0.5044598593274819, + 0.39398881439430655, + 0.28633236737353707, + 0.3495672108011638, + 0.34122852677748217, + 0.6763207350641, + 0.6239839466542489, + 0.6522851823628757, + 0.36139486653979724, + 0.387884520658144, + 0.3341407969101211, + 0.5272549027682433, + 0.45470364201535735, + 0.267307711987798, + 0.29191109143153354, + 0.4590609424948123, + 0.42267664715356656, + 0.4712332680152232, + 0.48802770199666484, + 0.4427641798807603, + 0.40958784412585975, + 0.3145323258957467, + 0.683245167417712, + 0.559993143595928, + 0.6879828256161309, + 0.5757496702610021, + 0.29074742256720587, + 0.5239945254367888, + 0.5288733290275892, + 0.5340181501824494, + 0.48484472249328286, + 0.404128703053707, + 0.506197580490092, + 0.5384046474364325, + 0.5421014792774818, + 0.48478440728380484, + 0.3771826462172494, + 0.46725981394098004, + 0.39065568965130004, + 0.47019310018609395, + 0.6381179932134607, + 0.5319270111225806, + 0.5809502852818971, + 0.45877428025839045, + 0.4850872231345771, + 0.46898086209294476, + 0.4349258363329811, + 0.498184394067674, + 0.4252801084098977, + 0.2956587173531057, + 0.4230193839581131, + 0.5118616279462591, + 0.27560206511214463, + 0.7581404096648607, + 0.4387942055248675, + 0.6021197427250624, + 0.7287343600006561, + 0.6759033353754715, + 0.3913576879978855, + 0.8011843053166177, + 0.49928683575831495, + 0.6691962986526728, + 0.39641759272302846, + 0.5630601684598077, + 0.548915261136906, + 0.6154249077292074, + 0.5919884555814908, + 0.6441802491505114, + 0.576923694521656, + 0.4220686683753942, + 0.5466620150238757, + 0.48492926711337825, + 0.43461907925180643, + 0.6464728399879044, + 0.3351185716216797, + 0.5766684334289033, + 0.5458592380228748, + 0.6299310396127207, + 0.5838855369658155, + 0.6437617108328616, + 0.5352662219239065, + 0.41542535558795557, + 0.5233810611815404, + 0.5373772955061554, + 0.5083923741176397, + 0.41508293738874097, + 0.37817621527991685, + 0.47266040252891806, + 0.5397743686234225, + 0.5551129124499046, + 0.4289302290493509, + 0.5752296272159482, + 0.5834830594063394, + 0.4387575550722453, + 0.5509390509289771, + 0.5463929294937283, + 0.5205377635167133, + 0.466651586698057, + 0.40432534239668555, + 0.41093003184031235, + 0.47958846763153495, + 0.4287781455082256, + 0.35470685858071255, + 0.46017885000766606, + 0.3676792366792236, + 0.47219638085285465, + 0.41372725992520587, + 0.35416784879536123, + 0.36291545818723087, + 0.3717995889245436, + 0.46745544329217664, + 0.4540349382205142, + 0.3722920974116622, + 0.3364217206316763, + 0.4249922422526805, + 0.5754270565024462, + 0.6430872132806764, + 0.3338668681659551, + 0.5454183099822715, + 0.5759295932558564, + 0.5708772640745962, + 0.5640011946346692, + 0.4347438288712142, + 0.5114664805674366, + 0.38947958097939006, + 0.5444337044336782, + 0.44165868488635934, + 0.6375852975265919, + 0.5152368168020611, + 0.5321923077434771, + 0.31557648276647493, + 0.33538297917811966, + 0.40934323189312, + 0.49214751144035157, + 0.48381441306519446, + 0.5735149403572399, + 0.35148871814033145, + 0.5992049540659611, + 0.5161692365854644, + 0.5413400968879624, + 0.5417193974738813, + 0.5062303080326828, + 0.433755144584501, + 0.5460261702283873, + 0.5571012673739, + 0.5499261223171448, + 0.5274704841406103, + 0.4500912512042266, + 0.3637220270445126, + 0.3846038048643928, + 0.39537173843068096, + 0.2604826620240907, + 0.4092064710734457, + 0.5136438866304324, + 0.44180907391962765, + 0.4272674114083601, + 0.4533309431483029, + 0.42144883068266653, + 0.5172079779409265, + 0.5419875354442494, + 0.37702014928596955, + 0.4894653950593814, + 0.5652808133785447, + 0.3930740047272049, + 0.41133989777704133, + 0.49608475755921755, + 0.48085664695370406, + 0.5740768053670016, + 0.3472445394622876, + 0.5361130073716387, + 0.3754991499248277, + 0.5939705859660949, + 0.5613623577731988, + 0.6125473947119593, + 0.30859327174239676, + 0.33918430230265634, + 0.32441134012340794, + 0.29904860407696576, + 0.3318363811182247, + 0.4361897462219194, + 0.49046432618311897, + 0.29219822389111866, + 0.38570393352291626, + 0.44828757567289157, + 0.45382873178321204, + 0.5283201822568804, + 0.472490156556828, + 0.614132818520603, + 0.2993433888012328, + 0.46424188838109226, + 0.45648603538442684, + 0.41564200195311884, + 0.4175034901032516, + 0.5088291552305796, + 0.49136276408666074, + 0.2788343346747981, + 0.53606612187169, + 0.45392551751482996, + 0.4326125789355081, + 0.5695013972826555, + 0.4213059813256233, + 0.38445699090188573, + 0.40322385774785663, + 0.44381515720273923, + 0.4431636166351342, + 0.41367310614102043, + 0.4442921941091741, + 0.5576263579842093, + 0.4505228409155514, + 0.5882728987678363, + 0.4998831299412951, + 0.4772722898747944, + 0.5187538687198169, + 0.5114847098207327, + 0.4966715042599703, + 0.4779785908986417, + 0.608688301571655, + 0.5844703818636517, + 0.5498692037304586, + 0.5730725337681424, + 0.6802662457983703, + 0.4668956924713372, + 0.47605655813721554, + 0.35981087853046895, + 0.47839618172455795, + 0.540040493614047, + 0.49837257478434127, + 0.4636206419498812, + 0.5387749583717312, + 0.3784083664868785, + 0.48544233997850234, + 0.4138661011898353, + 0.3098164917186392, + 0.4660544333099061, + 0.45410654818459584, + 0.4573148977786005, + 0.462295526735734, + 0.3564852570052599, + 0.607237079586482, + 0.6266445705989704, + 0.5279640089543265, + 0.5477718919712187, + 0.619869396075167, + 0.4920373247695959, + 0.3349116237139828, + 0.43187226586353383, + 0.4057236041377525, + 0.38731468403787317, + 0.40120071639218396, + 0.42697125787707374, + 0.473486485059938, + 0.44977620498054216, + 0.33574632925188197 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.2872617490990843, + "valid_tracks": 4, + "per_track_pearson": [ + 0.33112456446983113, + 0.24455226224363413, + 0.32155422270157247, + 0.2518159469812996 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.45458277872816183, + "valid_tracks": 626, + "per_track_pearson": [ + 0.4334304095522734, + 0.4461754164824834, + 0.4266981114343037, + 0.4417627986156857, + 0.43654092479389806, + 0.4554308365650389, + 0.40548943325364867, + 0.4166363961180644, + 0.4536611700999099, + 0.4537323844899621, + 0.4508667626653591, + 0.4272746319360356, + 0.4194423545402379, + 0.41660236627790276, + 0.4493581773439414, + 0.44032938557446205, + 0.4254470971535002, + 0.43735142638749086, + 0.4345383087635383, + 0.4419660282003718, + 0.44703247665867996, + 0.455418179610785, + 0.4472267142776794, + 0.4364789747490667, + 0.4356066700746259, + 0.45199649292222854, + 0.42569966592813246, + 0.4392752047639863, + 0.440224206450296, + 0.441333018203828, + 0.453962990727174, + 0.43698314161673935, + 0.4234406823702778, + 0.44465268973317806, + 0.4270199588971853, + 0.44247029697861534, + 0.4187922772701781, + 0.445430426960323, + 0.4241935957256719, + 0.4442221983340841, + 0.44064531837151477, + 0.43266286862497433, + 0.4040063988502233, + 0.4361575100690518, + 0.4322337515844536, + 0.42358951952443563, + 0.4424731353838235, + 0.44821687712085323, + 0.4348213454653833, + 0.4554417336437903, + 0.4259254906550131, + 0.429578753474223, + 0.44069437923294036, + 0.4518670506013683, + 0.4441127623447048, + 0.43323803501737934, + 0.42904962276868885, + 0.4498580559093767, + 0.45524077921665923, + 0.44296234191126, + 0.40544052549034887, + 0.4334934006252081, + 0.45558427126547824, + 0.4416049848197319, + 0.4447617385348493, + 0.44279909642336013, + 0.4490986814786128, + 0.461427228120802, + 0.43958539860281004, + 0.43853692959170926, + 0.4352262486629322, + 0.4548590856387308, + 0.44097549044441375, + 0.43851202990601507, + 0.46520012396184884, + 0.4756661126599262, + 0.44703216037554494, + 0.43959404677562397, + 0.4444061463336938, + 0.4434368476201071, + 0.4772779725917325, + 0.4501162840854849, + 0.4414268900559791, + 0.42281855108732525, + 0.4392800501598295, + 0.44871300488720944, + 0.4358817500055162, + 0.43769153487648865, + 0.44514182252388174, + 0.4415062744382467, + 0.4364423412159092, + 0.44378463745643615, + 0.4503250883850182, + 0.45860947461766716, + 0.4548310577769645, + 0.4628869354214965, + 0.4433813376776337, + 0.4387397446274938, + 0.43141792248318345, + 0.43649273571097835, + 0.41597685898946096, + 0.44830010069981113, + 0.4614553898562966, + 0.4435524482057834, + 0.45151308150479386, + 0.4409186204155038, + 0.4336839513801235, + 0.421049042180918, + 0.47103139173374825, + 0.4582728681239217, + 0.42747042314689054, + 0.44696518238905564, + 0.4487753822455327, + 0.4437818983095374, + 0.45265168075398116, + 0.44651776830136075, + 0.43383394091110256, + 0.42217962803016673, + 0.42487127281473797, + 0.45458052329890847, + 0.46708875342027395, + 0.4391695666660429, + 0.44826719930449094, + 0.43881882427968216, + 0.43714269501935016, + null, + 0.4494565677566386, + 0.4415009015329089, + 0.4141188828952958, + 0.4522910968646264, + 0.433451616616834, + 0.42439272325956484, + 0.43329457891172996, + 0.4460446379746536, + 0.44818367462081193, + 0.44782398828627207, + null, + 0.4382076681122741, + 0.44989352009479516, + 0.4179110900987176, + 0.42553574154161083, + 0.44187246708637645, + 0.442856091373799, + 0.4351679653189554, + 0.4362874160622467, + 0.4336319563950033, + 0.44276554479646424, + 0.44484072305645045, + 0.4348865627989177, + 0.4501480638190745, + 0.450330975097296, + 0.4581079660112174, + 0.44012395925214226, + 0.44101223593509437, + 0.4483550700868065, + 0.4593048145027335, + 0.4488744797950097, + 0.4517285463629711, + 0.4449374835828445, + 0.4504719934203846, + 0.45213742471560203, + 0.4434012329942444, + 0.4608077202576336, + 0.45014524766653957, + 0.4448746861306998, + 0.45101764807700495, + 0.44834432849188305, + 0.4372133817374265, + 0.42370481030873386, + 0.43499534130059225, + 0.458583206606469, + 0.45598596877298264, + 0.4438990832730851, + 0.4362253409720115, + 0.4506012844551995, + 0.4075917454725001, + 0.4537548406623157, + 0.4486966712156521, + 0.43911818007810033, + 0.4388191196272268, + 0.44509129372107475, + 0.4464173334354854, + 0.4333992041340998, + 0.43859594522621975, + 0.4304245840527499, + 0.39856489118873567, + 0.448677145330026, + 0.4232550356469508, + 0.44558812679661297, + 0.43956601323965444, + 0.4371284165009452, + 0.43769886690128096, + 0.4363768457999856, + 0.4268926265513363, + 0.4308807153051529, + 0.43882552011892, + 0.43673324760564225, + 0.41562273569691616, + 0.4577006711175478, + 0.45102156992180154, + null, + 0.431512655254907, + 0.43179786453057184, + 0.41734522995709883, + 0.4376788289992171, + 0.44552297346925634, + 0.4446237212187627, + 0.432073919951743, + null, + null, + 0.4466261281416659, + 0.4480244845002216, + null, + 0.4337311829915441, + 0.4085437097814406, + 0.4254210171000937, + 0.4329372037409632, + 0.4395551414630577, + 0.43764494778180274, + 0.43070810877754134, + null, + 0.4392602563600891, + 0.42831578287250294, + null, + 0.39794736653720003, + null, + 0.41914718869795736, + 0.4377638600813337, + 0.42849748066011556, + 0.4384671488693096, + null, + 0.4350482297369387, + 0.4323337259488531, + 0.4262179479349604, + null, + 0.43044924110670235, + null, + 0.4157317467222938, + null, + 0.43252775670852023, + 0.4251792116349935, + null, + 0.4114772732202884, + 0.4359695473925936, + null, + null, + 0.4351115452944483, + 0.4193468096208432, + 0.42923554652047413, + 0.4211548369699863, + null, + 0.4346723856358565, + null, + 0.4346941864297139, + 0.4318263824675469, + null, + 0.3712880801055137, + 0.4251246152284408, + 0.4328694067490732, + null, + 0.4184892145119555, + 0.441419188978533, + null, + 0.4297852139480239, + 0.4440183283307439, + 0.4147534367607786, + null, + null, + null, + null, + 0.43407809073143044, + null, + null, + 0.44012441099663746, + 0.4433904239010138, + 0.4377484324013779, + null, + 0.4270986287629546, + 0.42567886668347366, + null, + null, + 0.4408978126540265, + 0.43848516235478063, + 0.4237910538811532, + 0.44224997142862693, + 0.44239868723768727, + 0.4320196547731826, + 0.429394998918006, + 0.4375213534931189, + 0.42725604860076444, + 0.4375306737722581, + 0.4281922089217414, + 0.444902239775178, + null, + 0.424796091122725, + 0.4318085277821412, + 0.4302086450228904, + 0.4243664383811008, + 0.43018594152871137, + 0.42369322384035074, + 0.4420663980402845, + 0.4157300617790739, + 0.4083761524619527, + 0.42557958570397875, + 0.44259859474135677, + 0.4347369238389175, + null, + 0.4342106800964679, + 0.44037094466729826, + 0.42087167583681523, + null, + 0.4343653291320016, + 0.3680871863502172, + null, + 0.4486222956682509, + null, + 0.4245149596535121, + 0.4329320533323839, + 0.42717301107002714, + 0.4381437439383499, + 0.45065162280356447, + 0.41738439924256204, + 0.4197565766148758, + null, + null, + 0.4250690784331244, + 0.4236295097575994, + null, + 0.4165004060284247, + null, + 0.4190016263743178, + 0.42926584855617755, + 0.4313058094781913, + 0.4235515827018497, + null, + null, + null, + null, + 0.44139443149017477, + null, + 0.4283873595881641, + 0.45166051567775173, + null, + null, + 0.4228427166910623, + 0.45640712347319595, + 0.44140061867212527, + null, + 0.42846492394642766, + 0.43015985386532096, + 0.42220380258175044, + null, + 0.4364242904217976, + null, + null, + 0.4406883361802935, + null, + 0.4402098139469845, + null, + null, + 0.43781413860352736, + 0.42135229238841754, + 0.42923464724493426, + 0.42618152403597404, + 0.42204210616462634, + null, + 0.41766719034567806, + 0.48817459523921974, + 0.49420587115492653, + 0.4466836062219474, + 0.44558111687951196, + 0.4642764939645928, + 0.48375981963787085, + 0.47655630655339115, + 0.5036649405277416, + 0.4867984302746722, + 0.5046092592172899, + 0.49897767161060663, + 0.4719141307969956, + 0.45481513192376943, + 0.4569384664176812, + 0.4469274696808006, + 0.44435572763686587, + 0.4954057145411066, + 0.5079555859290508, + 0.48609141904087905, + 0.4897290569427657, + 0.48352415333649934, + 0.47614643888971175, + 0.447337005542406, + 0.43462792490800484, + 0.49816116570373575, + 0.48179521146711046, + 0.4961778293542326, + 0.49481743923531185, + 0.450716645586809, + 0.4446611844691474, + 0.46289995053072547, + 0.4506238553420141, + 0.4407799194590931, + 0.45985603002841485, + 0.4249449806208075, + 0.4834573534795286, + 0.4214722200411463, + 0.49405270010616387, + 0.49484162307896323, + 0.4620485600615444, + 0.46425847615765864, + 0.4548795809906527, + 0.4584483460447698, + 0.47443516085245124, + 0.4728729065037485, + 0.4499194470138658, + 0.46341407288305014, + 0.4535087445805246, + 0.46788699147660473, + 0.44784799362312455, + 0.46447160980002555, + 0.42776282763730156, + 0.4801877959374246, + 0.4752923331633132, + 0.44751958506022443, + 0.4718161736253796, + 0.4756871462757795, + 0.47337257829824275, + 0.48651760922102355, + 0.5008901575314049, + 0.4916451986249156, + 0.4922888146331415, + 0.47306848065561896, + 0.5032941891072542, + 0.4874209635706774, + 0.48781261996450287, + 0.46268220231859036, + 0.5226451749446215, + 0.5138641969327644, + 0.45868950651578494, + 0.4833504824996095, + 0.49674850220518124, + 0.47451047979466543, + 0.4806928884175709, + 0.47501031512058833, + 0.47306654630613676, + 0.4738541954679012, + 0.48484117627802925, + 0.4878905970214183, + 0.4931365521726501, + 0.49605927773571773, + 0.4943261241316671, + 0.47193336518325, + 0.4969806971635569, + 0.48836955778303764, + 0.4744494869279759, + 0.4665295908300695, + 0.4878148189875204, + 0.4896142601788936, + 0.4964006786098972, + 0.4825622300549001, + 0.47877337620237836, + 0.49722829805569485, + 0.4832037114245017, + 0.49022289567190974, + 0.4954221274509107, + 0.4927066428766349, + 0.5068554035627038, + 0.46846001940983134, + 0.5127174971131827, + 0.4949591565939307, + 0.5012734464367895, + 0.4778942747814321, + 0.48486191997047046, + 0.49196281136646014, + 0.5046073492304238, + 0.4726891841153705, + 0.4876826738109486, + 0.46930428627348064, + 0.46856645863038726, + 0.5000933747321166, + 0.44481897651190627, + 0.4745605327842782, + 0.478882022706775, + 0.47020932732706183, + 0.4920064678342158, + 0.4925353319157164, + 0.5045580378755424, + 0.4572788116716366, + 0.48644087690148624, + 0.5097842046525041, + 0.50576326600359, + 0.499944348032607, + 0.4702516975401583, + 0.5101618748734064, + null, + 0.4798943080324595, + 0.45562517013312753, + 0.4777215159480783, + 0.5021585258740632, + 0.4954653316332216, + 0.49305974311512224, + 0.4964337264582407, + 0.5110872732290898, + 0.49191060711599177, + 0.5029387853478715, + null, + 0.4793919964146492, + 0.5072226065392422, + 0.5116256365187513, + 0.506390500007479, + 0.5026374628925889, + 0.5020766148792739, + 0.493396581915767, + 0.5277470146479298, + 0.5198858567971189, + 0.5251368553841441, + 0.4970779870583325, + 0.5313464435054672, + 0.5160262944006656, + 0.5235615931325756, + 0.5204710117566287, + 0.4974442810149846, + 0.46110121245904256, + 0.44906667428435176, + 0.4497876932303235, + 0.44613063291998784, + 0.4993070292333736, + 0.47501618032325615, + 0.48581327387970047, + 0.490860495313759, + 0.5004949730728204, + 0.48484430348011437, + 0.47876329101795195, + 0.5066218030535805, + 0.48640778621593445, + 0.4898488113142862, + 0.48390930848072344, + 0.4989809206676004, + 0.5081397455914572, + 0.4945685239428039, + 0.5212801593309406, + 0.5144920397138828, + 0.49258107561441083, + 0.4763731070522199, + 0.4959494822047333, + 0.48659313137165294, + 0.493202873900857, + 0.4992655216590373, + 0.46371328628293634, + 0.4984523915250081, + 0.4983856456426633, + 0.5036540278058277, + 0.4817502334449438, + 0.5017811150904177, + 0.5053302965418526, + 0.4986262769246674, + 0.4657754021971386, + 0.48914097077377106, + 0.4846886851877734, + 0.4666760164569698, + 0.5096641528098405, + 0.47335071193750206, + 0.4723491415967112, + 0.46905509450265376, + 0.46629512509672405, + 0.4706132885998453, + 0.48493383403820495, + 0.49186892146188294, + 0.4543135665615692, + null, + 0.45315028742392305, + 0.45925613882947713, + 0.45071476804941485, + 0.45518202071477426, + 0.4644579493434694, + 0.47327248901039526, + 0.4637317566787347, + null, + null, + 0.452923344636407, + 0.4479311743925223, + null, + 0.46363892905222553, + 0.43323443684679275, + 0.4645845589733376, + 0.4427780827860655, + 0.4530641428550365, + 0.4531307784863402, + 0.48943667952941944, + null, + 0.45614590298858815, + 0.46050598956206845, + null, + 0.44870157303585284, + null, + 0.46176111635566736, + 0.4455066044507288, + 0.44928967431195554, + 0.46131689334027526, + null, + 0.4229916371770505, + 0.44002393607061874, + 0.44735671242402614, + null, + 0.4152732110070271, + null, + 0.46125924350598385, + null, + 0.46886544306528066, + 0.45800309311597787, + null, + 0.44303792750386134, + 0.47028218626413043, + null, + null, + 0.47283770518377316, + 0.4530820461468641, + 0.4312206094226448, + 0.4234682967787275, + null, + 0.4366805933195229, + null, + 0.448674391918406, + 0.4586679831066927, + null, + 0.39259257278412385, + 0.43805054019904377, + 0.4344779863410144, + null, + 0.4515054752621788, + 0.4557012789214504, + null, + 0.4839303063956054, + 0.4650770059236862, + 0.4676170126991757, + null, + null, + null, + null, + 0.4710983955645936, + null, + null, + 0.4637266331380257, + 0.44206465750387525, + 0.4612166307301157, + null, + 0.4737714703569917, + 0.4604818271217742, + null, + null, + 0.44329611583375633, + 0.43371772528318225, + 0.45257499890440867, + 0.45216659958423683, + 0.4653518487642998, + 0.45922620260578195, + 0.4443215415670724, + 0.4487858913597531, + 0.43703970049613494, + 0.4738865080905239, + 0.4324860973036446, + 0.464681729325162, + null, + 0.44853663043519554, + 0.4411127396676389, + 0.4484235639359847, + 0.44474542360913055, + 0.4733739003652537, + 0.45692537420604323, + 0.4477489153130214, + 0.44565704262504596, + 0.4565773312822645, + 0.4452996049091616, + 0.43799976183589956, + 0.430415144219206, + null, + 0.4419404794337879, + 0.47501168371533414, + 0.47279166932364475, + null, + 0.45043649042505723, + 0.44003008048157377, + null, + 0.46300761769508164, + null, + 0.44323612603781615, + 0.4452402318830937, + 0.46337233446628817, + 0.45853431575137277, + 0.45704719708757446, + 0.45480387367900066, + 0.45232674607217316, + null, + null, + 0.44444222144383516, + 0.4841439763693546, + null, + 0.4572394690108802, + null, + 0.47277773928098477, + 0.45731060077529406, + 0.4422756770103288, + 0.46553682119556217, + null, + null, + null, + null, + 0.4661608819845101, + null, + 0.4638727788925592, + 0.46709939161056524, + null, + null, + 0.4607626293247352, + 0.4508634596125109, + 0.45421016203041975, + null, + 0.4347006133121743, + 0.43801344129395003, + 0.4566301634039907, + null, + 0.4566255204134544, + null, + null, + 0.45308578488685397, + null, + 0.43804216506129157, + null, + null, + 0.4529419283167563, + 0.4609544149989717, + 0.46924324141586865, + 0.46077104314810946, + 0.468702236036624, + null, + 0.44264052629384004 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561, + 0.5849754624386561 + ] + }, + "junctions": { + "mean_pearson": 0.34488132615344674, + "valid_tracks": 626, + "per_track_pearson": [ + 0.2559316043178584, + 0.34837026703549545, + 0.48905864401066146, + 0.45704160940842914, + 0.2867897159959314, + 0.29866447619770325, + 0.10014490092632873, + 0.29058908315890664, + 0.21611048337629898, + 0.28024356689581664, + 0.4342786442125091, + 0.21475341944778953, + 0.32361472836715705, + 0.322614125196344, + 0.2651322943003084, + 0.3174686772490166, + 0.44738683499482534, + 0.24802233723045228, + 0.24856480926246796, + 0.4524438530755271, + 0.17945364271136427, + 0.2790409140396278, + 0.36863023740476336, + 0.41213349429526697, + 0.27838193955677937, + 0.18686020479390353, + 0.3532692282357198, + 0.43196299171477914, + 0.20803500080606263, + 0.3347491126613901, + 0.42735722708664287, + 0.5482328391817398, + 0.4921454983423005, + 0.3539723092382131, + 0.4322858436531703, + 0.2956512609948724, + 0.4278117233890752, + 0.3975412804831968, + 0.38480408209624806, + 0.188057024958777, + 0.40462055175503736, + 0.2853812200726645, + 0.30908755090525536, + 0.4605432069422193, + 0.2821439482217234, + 0.3504622139914354, + 0.24807389714339198, + 0.46884834649518514, + 0.30216482036027364, + 0.47636705324673445, + 0.32212749738302193, + 0.2577571517314573, + 0.28455245568952015, + 0.2769806861722925, + 0.39790088776192506, + 0.34522863124799597, + 0.2212011444422484, + 0.3830162952578762, + 0.29452863494566567, + 0.25942835486282234, + 0.23591603832604727, + 0.1799272068428366, + 0.2098479439849446, + 0.10662482122039846, + 0.25139580710107284, + 0.2002172962686035, + 0.4621507138309745, + 0.4592158977564199, + 0.35904705880921867, + 0.27256908674928004, + 0.23975231314636822, + 0.23614077845308032, + 0.2446143090080887, + 0.259733379607988, + 0.28939496139912724, + 0.27146533446652854, + 0.35687503267307374, + 0.3256845974681895, + 0.14777656236268463, + 0.27507228247511534, + 0.3633875253801859, + 0.3349895039593282, + 0.21884037772552695, + 0.31137520322203016, + 0.46972167709927304, + 0.22800087767030564, + 0.19017747811851676, + 0.30304544977696507, + 0.16714317409155735, + 0.3546311937506294, + 0.3031857903461013, + 0.4714659108612144, + 0.2087906836133724, + 0.25823312300859064, + 0.25105820900993736, + 0.40377135770387446, + 0.3110321665768263, + 0.27345732044385124, + 0.2713794575955941, + 0.19542374548222158, + 0.27128031497876814, + 0.42523360660947573, + 0.30797485439461697, + 0.1293668984360217, + 0.14205272748397646, + 0.16184745468050188, + 0.3209222514135886, + 0.3675366280303168, + 0.23390460204763966, + 0.18604720062364138, + 0.44059742546559993, + 0.1987469094079483, + 0.392600679338549, + 0.360736642371874, + 0.3012806851343896, + 0.2571617358062406, + 0.3855456878860856, + 0.350267467265966, + 0.4031715468314776, + 0.45907369560989475, + 0.36906736817074365, + 0.3310525751666513, + 0.4104327649394957, + 0.16154696995916973, + 0.36485343487754623, + null, + 0.28003664349120017, + 0.23531990019378493, + 0.14336097711063803, + 0.33747333753253106, + 0.34057454565515394, + 0.26178552317618003, + 0.25309703564248986, + 0.22732667255889596, + 0.24955444106736052, + 0.33962393868734136, + null, + 0.21211249463105528, + 0.35686688022221774, + 0.1931328668734319, + 0.29129048358154114, + 0.2263099858715917, + 0.35574645849476944, + 0.26009934116980005, + 0.23802266682905995, + 0.2314631411893492, + 0.2950146882243862, + 0.44270587583678045, + 0.3193033797958467, + 0.28245584814426344, + 0.3205610291540399, + 0.1704790168376378, + 0.28018355104624604, + 0.28789133653184057, + 0.30139207358558284, + 0.40797094467635575, + 0.31243991031527474, + 0.24120818617897166, + 0.20122603484762, + 0.1404647057918786, + 0.3180492753390893, + 0.22281391951147092, + 0.2976839451548313, + 0.24225988407031296, + 0.2521454470598458, + 0.2524380001074994, + 0.2587279843567767, + 0.20414275934568324, + 0.3668017120235106, + 0.3215307267930644, + 0.3518233999417982, + 0.28660273894204913, + 0.4207608469242821, + 0.2543723092057755, + 0.2704360983728332, + 0.5076787958931849, + 0.23950864430540947, + 0.23819117591692987, + 0.17943865805952985, + 0.3218774147300119, + 0.41770217694474226, + 0.3295800119463145, + 0.2548666123789358, + 0.2971474885793124, + 0.23735667465560453, + 0.293050937277261, + 0.23161978443916545, + 0.2759704294449195, + 0.36177237185130695, + 0.2565871094574584, + 0.2938127344361555, + 0.2956073830314631, + 0.2563762565135492, + 0.42156749562217144, + 0.3797352209736178, + 0.4306634010253232, + 0.31166916386249777, + 0.4421497785392493, + 0.2730701156466709, + 0.2359847710099068, + null, + 0.24667730128259074, + 0.445858038135192, + 0.4218112331294084, + 0.3567710442522913, + 0.23261158754304745, + 0.20309434836462817, + 0.31851864288097137, + null, + null, + 0.23184642438157824, + 0.4088835949918907, + null, + 0.3595850881383475, + 0.34982478710452497, + 0.2782907410066455, + 0.32078353859218167, + 0.4265495030053555, + 0.3698508026515432, + 0.15486150796348522, + null, + 0.42192270090987194, + 0.34043367208168285, + null, + 0.28543027484587863, + null, + 0.30440295109142046, + 0.4628839955610525, + 0.22542158764153936, + 0.3427478578123062, + null, + 0.18756199342644833, + 0.28334803955825705, + 0.2856282382830804, + null, + 0.3143097604402524, + null, + 0.3142968453933996, + null, + 0.17502184210986568, + 0.4068131337638057, + null, + 0.305673020106684, + 0.2719433582907134, + null, + null, + 0.22030978263272952, + 0.4111701309780898, + 0.30532616830421416, + 0.2967991700457057, + null, + 0.39962521703555004, + null, + 0.355726390961645, + 0.4070924905212742, + null, + 0.2526371691142393, + 0.4273343057159191, + 0.31963793971569576, + null, + 0.41239307953911863, + 0.29056475022156414, + null, + 0.38487469689193765, + 0.4585327087715384, + 0.27372463608213554, + null, + null, + null, + null, + 0.29377413842254807, + null, + null, + 0.3331229149725731, + 0.2722270502681236, + 0.4022704478288482, + null, + 0.38014320167728205, + 0.3527656255122085, + null, + null, + 0.3497071247905734, + 0.3943803923532649, + 0.385772851628348, + 0.4767738122885877, + 0.4378403420997869, + 0.38566663118074684, + 0.47255686041243694, + 0.2850864997739347, + 0.46212965321214367, + 0.2832076661720531, + 0.4005531858241446, + 0.4197059763527207, + null, + 0.5357002814218581, + 0.3312182050130941, + 0.23784717356954066, + 0.3212149191396982, + 0.35236140449070086, + 0.36453203778452553, + 0.18441394941997016, + 0.44636718573077633, + 0.5232352833537224, + 0.19582542497738722, + 0.4114476471467419, + 0.3878291276185403, + null, + 0.3132751517448953, + 0.2536576048708387, + 0.36611951095610185, + null, + 0.39459072154580904, + 0.24072019988000837, + null, + 0.3592735102068166, + null, + 0.26708850261370365, + 0.4314708356529268, + 0.2852366845881874, + 0.24225086699072135, + 0.26088864381996113, + 0.4328928760634073, + 0.39870504553255887, + null, + null, + 0.24479361929763443, + 0.403789778898936, + null, + 0.3384167523297807, + null, + 0.2548728495697405, + 0.3468444458796366, + 0.2884490785170135, + 0.48968884704698995, + null, + null, + null, + null, + 0.23303758379876946, + null, + 0.3002757588281841, + 0.2992460075828306, + null, + null, + 0.326901726500807, + 0.2684087163344797, + 0.42791169613910424, + null, + 0.35522080432386016, + 0.3870777638862374, + 0.21442509460783402, + null, + 0.2571074246976896, + null, + null, + 0.2561410684627153, + null, + 0.1477031649794011, + null, + null, + 0.3139598579991989, + 0.3172707685082246, + 0.4450807020035019, + 0.46044762493258234, + 0.43720902586850424, + null, + 0.28263008889255453, + 0.41840280651136347, + 0.2602177760194457, + 0.5588345224752939, + 0.35114615380793446, + 0.43102218305570195, + 0.294803833437589, + 0.29694195134080453, + 0.2887566714925215, + 0.4297703317017825, + 0.36221166627022344, + 0.33703687452722536, + 0.4382758105949771, + 0.37553792979300266, + 0.24754050056798785, + 0.29705493084297274, + 0.280995152148612, + 0.3514163720767322, + 0.48162361301858986, + 0.42592844505704264, + 0.38962702849112274, + 0.4301496924397625, + 0.44470428823025443, + 0.2760141929621049, + 0.6028814345354596, + 0.3530781736416768, + 0.41481466878290946, + 0.3995728928980224, + 0.3714164842771083, + 0.45695748329731406, + 0.19070545869920277, + 0.3750127163027355, + 0.5135546781024186, + 0.4204111796949472, + 0.42873965951334686, + 0.37437174949581054, + 0.27637466170814806, + 0.29364603394883904, + 0.3459846704560781, + 0.39755039022473543, + 0.42615283526868075, + 0.38196356107754564, + 0.1836375537985672, + 0.31418581227178616, + 0.3052135144601059, + 0.3189351220373243, + 0.32113267490702385, + 0.3401723292180387, + 0.3220618154857669, + 0.26183593172103614, + 0.43672354419752646, + 0.27256759196529706, + 0.3592460795987378, + 0.31761799218428854, + 0.3186577523515941, + 0.3100488098128666, + 0.3197435844008859, + 0.36865094226817247, + 0.37183752720143737, + 0.25932991278835976, + 0.34274085041172997, + 0.4437804510975448, + 0.35030758665300393, + 0.4551296275375366, + 0.35160607422633516, + 0.294591387594547, + 0.4334162977712167, + 0.327223205291296, + 0.34180255412423616, + 0.30406001036170094, + 0.3403717417732884, + 0.36148863482296395, + 0.3760274657420263, + 0.4265460147627553, + 0.4232887577950448, + 0.38223508754277424, + 0.3942112756052244, + 0.4135727335725438, + 0.3437093944348898, + 0.42585512903482364, + 0.3963829309878271, + 0.3630816752696277, + 0.3698615905311351, + 0.3730233085145492, + 0.3175269245301341, + 0.3834595537665978, + 0.35693610559542405, + 0.41663716681037105, + 0.4392972852296904, + 0.35443745665930254, + 0.4146058746622169, + 0.457019749961642, + 0.3523590233084211, + 0.3858097204039292, + 0.415653090293301, + 0.34217437806606693, + 0.3728113021311434, + 0.3490994472869909, + 0.4825475361643612, + 0.49003108175232174, + 0.5365188778149854, + 0.26280430631751545, + 0.4054708267088439, + 0.3917996111619315, + 0.37225366391175324, + 0.3336822953278382, + 0.3698622955778324, + 0.24233096168516524, + 0.37320785648391813, + 0.24709455952521864, + 0.2937483847442969, + 0.21286871866431098, + 0.31726133872046175, + 0.34228037292807156, + 0.338424713310973, + 0.3185998875711287, + 0.3304606093108716, + 0.35476148024279824, + 0.22004781791669914, + 0.2634692773434864, + 0.33436373714239603, + 0.40645467795428003, + 0.37104818045774757, + 0.360227503274914, + 0.3391507110437583, + 0.5126927160134336, + null, + 0.37084138358882235, + 0.4744070205887172, + 0.3366804194299037, + 0.40489555291604823, + 0.3623714908812497, + 0.36947534006777116, + 0.42306354582362304, + 0.41037571826470576, + 0.3924265991028822, + 0.4244871845940992, + null, + 0.399888264303568, + 0.3779611306375985, + 0.3249933481279548, + 0.29278678649203654, + 0.4059928835650672, + 0.2947317355726799, + 0.427447637137114, + 0.4508016048091512, + 0.2228265623324216, + 0.3319532231221404, + 0.45522115534323276, + 0.25912458587080156, + 0.3837787480303337, + 0.42055336669034266, + 0.47917267320276474, + 0.34087051811050356, + 0.4838141087616484, + 0.5752093429128466, + 0.5350247277917256, + 0.43465670066377, + 0.47692920172522213, + 0.4270248533735972, + 0.3447031032384474, + 0.4201231659866591, + 0.3965725818883605, + 0.3257137897515884, + 0.48742033131251145, + 0.4831566107716255, + 0.3960177225932103, + 0.4066311620477489, + 0.35721106746986653, + 0.33016239984672696, + 0.3618971875359128, + 0.34962997845411165, + 0.48247338960994635, + 0.3208381314179731, + 0.17940636721592834, + 0.3573062153959585, + 0.38825623169126583, + 0.4987419071674503, + 0.38016108772511986, + 0.3827353606400869, + 0.3126539805786967, + 0.2835865457767785, + 0.2980681774723379, + 0.5099351569556468, + 0.5100226482277104, + 0.4390290128339392, + 0.5481186818162797, + 0.5183867370221468, + 0.3045555919687915, + 0.4501514800800057, + 0.39409258185368595, + 0.629681706702426, + 0.32019370152602855, + 0.29430507877805395, + 0.28478685109415053, + 0.3625963477823325, + 0.32147439309443215, + 0.33099384942081306, + 0.3803002552795183, + 0.40613290357227316, + 0.4748521076878596, + null, + 0.4090797221116169, + 0.2991011469582217, + 0.6063605680846015, + 0.45999095013594904, + 0.42855383539545694, + 0.31622244474502925, + 0.38829235320644434, + null, + null, + 0.47481033035642756, + 0.4991508903333168, + null, + 0.19335578832592337, + 0.43335633954827835, + 0.14817655980231498, + 0.2697693912442023, + 0.3582640342549194, + 0.2711394871684065, + 0.40620923003368037, + null, + 0.3847009366991562, + 0.3244162016378869, + null, + 0.3202437509336878, + null, + 0.6055120964418765, + 0.3707972656100312, + 0.32633349666129197, + 0.30678865171143604, + null, + 0.45434995709722786, + 0.36041495937390405, + 0.3748968803789514, + null, + 0.4174504612262602, + null, + 0.33157871918328774, + null, + 0.3821691663870415, + 0.38421891602142927, + null, + 0.3034769239755353, + 0.3755044379685228, + null, + null, + 0.4201954041859911, + 0.3412236018869255, + 0.14964150778499957, + 0.28619855475441075, + null, + 0.4943599855685663, + null, + 0.2280475915853449, + 0.4173201176169765, + null, + 0.22194674748170018, + 0.47937949939698427, + 0.3170427991075299, + null, + 0.31845034088191715, + 0.49120113671505555, + null, + 0.303095436241422, + 0.40491736553416086, + 0.3728163668469216, + null, + null, + null, + null, + 0.3124891853748298, + null, + null, + 0.4616101459363706, + 0.35085174792176405, + 0.3566748223307407, + null, + 0.29498377337662435, + 0.382298229589358, + null, + null, + 0.2848485493471035, + 0.09226385902822894, + 0.3794160545872936, + 0.35883987919525134, + 0.26935282057002274, + 0.39205992472138357, + 0.1611085724210903, + 0.2530847862793843, + 0.26553866953286637, + 0.3072677248852913, + 0.48935912803078, + 0.3055195924980222, + null, + 0.1905903376101438, + 0.40325806327387664, + 0.4196649211158228, + 0.3909306343281722, + 0.16914514258364377, + 0.5197680476401084, + 0.4610861267690294, + 0.2800091704388859, + 0.3566719869347209, + 0.4318043217494497, + 0.4275052532072484, + 0.29970743152262325, + null, + 0.38032323325694967, + 0.31946851612572985, + 0.23354121069174213, + null, + 0.450906420245076, + 0.36358364124274706, + null, + 0.4130429469483774, + null, + 0.3925172815718172, + 0.4158346882916725, + 0.22749462491980127, + 0.514407251951523, + 0.4886305346244353, + 0.559539050947901, + 0.6304130989445093, + null, + null, + 0.4533340684321599, + 0.5479433258870943, + null, + 0.35341713881876724, + null, + 0.3474637809071528, + 0.4222190050549112, + 0.4606925281693323, + 0.3838046828288385, + null, + null, + null, + null, + 0.3344662583635466, + null, + 0.43076176203719196, + 0.47274875194784494, + null, + null, + 0.47887611775785693, + 0.17138444744592155, + 0.3939312558171545, + null, + 0.35903832555610815, + 0.3145589332853221, + 0.4032645380468409, + null, + 0.25113501323718534, + null, + null, + 0.35506825044205764, + null, + 0.3303842922787535, + null, + null, + 0.45651714066189586, + 0.5425008960417897, + 0.5811391602989, + 0.6099107096980916, + 0.3886393084159742, + null, + 0.4182295897090252 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.8543046357615894, + 0.8585526315789473, + 0.8398576512455516, + 0.8164556962025317, + 0.8509933774834437, + 0.8372881355932204, + 0.8274647887323944, + 0.8472222222222222, + 0.8621794871794872, + 0.8475177304964538, + 0.8282828282828283, + 0.8135593220338984, + 0.8376623376623377, + 0.853125, + 0.8566666666666667, + 0.8453947368421053, + 0.8146853146853147, + 0.8269230769230769, + 0.8439490445859873, + 0.8464052287581699, + 0.8309352517985612, + 0.8483870967741935, + 0.8409090909090909, + 0.8215613382899628, + 0.8220640569395018, + 0.8651315789473685, + 0.8621794871794872, + 0.8508474576271187, + 0.8581081081081081, + 0.8310344827586207, + 0.815068493150685, + 0.8478964401294499, + 0.8246268656716418, + 0.8601823708206687, + 0.8058608058608059, + 0.8454258675078864, + 0.8091603053435115, + 0.8495297805642633, + 0.8076923076923077, + 0.8322147651006712, + 0.825938566552901, + 0.8556701030927835, + 0.8129770992366412, + 0.8296296296296296, + 0.8491803278688524, + 0.8412162162162162, + 0.8191881918819188, + 0.8172043010752689, + 0.8049645390070922, + 0.8639455782312925, + 0.8138686131386861, + 0.8492647058823529, + 0.8181818181818182, + 0.826530612244898, + 0.8208955223880597, + 0.8299319727891157, + 0.8304498269896193, + 0.8576158940397351, + 0.8287671232876712, + 0.8204225352112676, + 0.8206896551724138, + 0.8622950819672132, + 0.8576158940397351, + 0.863481228668942, + 0.8456375838926175, + 0.8765822784810127, + 0.8412162162162162, + 0.8422818791946308, + 0.8532423208191127, + 0.8529411764705882, + 0.8469750889679716, + 0.8290322580645161, + 0.8451178451178452, + 0.8543046357615894, + 0.8422939068100358, + 0.8406779661016949, + 0.8488745980707395, + 0.842809364548495, + 0.8620689655172413, + 0.8571428571428571, + 0.8410596026490066, + 0.8512658227848101, + 0.8595890410958904, + 0.8543689320388349, + 0.8666666666666667, + 0.8544891640866873, + 0.8595890410958904, + 0.8464052287581699, + 0.8535714285714285, + 0.8512658227848101, + 0.8362369337979094, + 0.8316151202749141, + 0.856140350877193, + 0.8503401360544217, + 0.8175675675675675, + 0.8298611111111112, + 0.8440677966101695, + 0.8235294117647058, + 0.8609271523178808, + 0.8544303797468354, + 0.8433333333333334, + 0.8426229508196721, + 0.8256227758007118, + 0.865814696485623, + 0.8600682593856656, + 0.8633333333333333, + 0.8202247191011236, + 0.8695652173913043, + 0.8604651162790697, + 0.847972972972973, + 0.8285714285714286, + 0.8422818791946308, + 0.8409090909090909, + 0.8372881355932204, + 0.8571428571428571, + 0.831081081081081, + 0.8482758620689655, + 0.8412162162162162, + 0.8158844765342961, + 0.8498293515358362, + 0.8107142857142857, + 0.8542372881355932, + 0.8382352941176471, + 0.8618421052631579, + 0.8097165991902834, + 0.9838709677419355, + 0.8512110726643599, + 0.8606271777003485, + 0.8636363636363636, + 0.8544776119402985, + 0.8297101449275363, + 0.8620689655172413, + 0.8501742160278746, + 0.8345864661654135, + 0.8491228070175438, + 0.8193979933110368, + 0.9919354838709677, + 0.8610169491525423, + 0.8283582089552238, + 0.8519736842105263, + 0.843065693430657, + 0.8059701492537313, + 0.8602941176470589, + 0.8586206896551725, + 0.8421052631578947, + 0.8206106870229007, + 0.8267716535433071, + 0.8194444444444444, + 0.8198529411764706, + 0.8445229681978799, + 0.8074074074074075, + 0.8226950354609929, + 0.8315018315018315, + 0.8603896103896104, + 0.8413793103448276, + 0.8350515463917526, + 0.8181818181818182, + 0.855595667870036, + 0.8515901060070671, + 0.8540145985401459, + 0.8068181818181818, + 0.8402777777777778, + 0.8557993730407524, + 0.8327526132404182, + 0.8188679245283019, + 0.8345323741007195, + 0.8373702422145328, + 0.8741721854304636, + 0.8226415094339623, + 0.8421052631578947, + 0.8131868131868132, + 0.8191126279863481, + 0.8458646616541353, + 0.8493150684931506, + 0.8346153846153846, + 0.8249027237354085, + 0.8442028985507246, + 0.8466898954703833, + 0.8529411764705882, + 0.8356643356643356, + 0.8327526132404182, + 0.8246268656716418, + 0.8265682656826568, + 0.8136200716845878, + 0.8556338028169014, + 0.8597785977859779, + 0.8184931506849316, + 0.8249158249158249, + 0.8136200716845878, + 0.8459016393442623, + 0.8263888888888888, + 0.8529411764705882, + 0.8321428571428572, + 0.8098591549295775, + 0.8222996515679443, + 0.8127208480565371, + 0.8125, + 0.8298611111111112, + 0.8496732026143791, + 0.85, + 1.0, + 0.875, + 0.8440677966101695, + 0.8239202657807309, + 0.8378378378378378, + 0.8713826366559485, + 0.8303886925795053, + 0.8289473684210527, + 1.0, + 1.0, + 0.8547854785478548, + 0.8506493506493507, + 0.9923076923076923, + 0.8517241379310345, + 0.7692307692307693, + 0.8248175182481752, + 0.8430034129692833, + 0.8311258278145696, + 0.8137931034482758, + 0.8389261744966443, + 0.991869918699187, + 0.8416149068322981, + 0.8350515463917526, + 1.0, + 0.76953125, + 1.0, + 0.8, + 0.8140350877192982, + 0.8204225352112676, + 0.835016835016835, + 0.9927536231884058, + 0.8115942028985508, + 0.823943661971831, + 0.8280701754385965, + 1.0, + 0.8047138047138047, + 0.9923076923076923, + 0.8474025974025974, + 0.9915254237288136, + 0.8433333333333334, + 0.8422818791946308, + 0.9854014598540146, + 0.8141263940520446, + 0.839344262295082, + 0.9844961240310077, + 0.984375, + 0.8297872340425532, + 0.8464052287581699, + 0.8366013071895425, + 0.8114478114478114, + 1.0, + 0.833922261484099, + 1.0, + 0.8214285714285714, + 0.8109965635738832, + 0.9861111111111112, + 0.8192307692307692, + 0.8070739549839229, + 0.8218181818181818, + 0.9927536231884058, + 0.8142857142857143, + 0.8160535117056856, + 0.9917355371900827, + 0.8459016393442623, + 0.8105263157894737, + 0.8271186440677966, + 0.9776119402985075, + 1.0, + 1.0, + 1.0, + 0.8239202657807309, + 1.0, + 0.9921259842519685, + 0.8503401360544217, + 0.8156028368794326, + 0.8269230769230769, + 1.0, + 0.8156996587030717, + 0.8287671232876712, + 0.992, + 0.9923076923076923, + 0.8466898954703833, + 0.822463768115942, + 0.8394648829431438, + 0.8378378378378378, + 0.8395904436860068, + 0.8197278911564626, + 0.8430034129692833, + 0.8096885813148789, + 0.8203389830508474, + 0.8113879003558719, + 0.8535031847133758, + 0.8191881918819188, + 1.0, + 0.8385964912280702, + 0.8378378378378378, + 0.8285714285714286, + 0.8383838383838383, + 0.8333333333333334, + 0.8680555555555556, + 0.8245614035087719, + 0.8068965517241379, + 0.8295081967213115, + 0.7972972972972973, + 0.8188405797101449, + 0.8156996587030717, + 0.983739837398374, + 0.8327974276527331, + 0.8412162162162162, + 0.842948717948718, + 1.0, + 0.8274760383386581, + 0.8064516129032258, + 0.9834710743801653, + 0.8144329896907216, + 1.0, + 0.8493589743589743, + 0.8263888888888888, + 0.834983498349835, + 0.7971014492753623, + 0.8503401360544217, + 0.8013698630136986, + 0.8076923076923077, + 0.9858156028368794, + 0.9829059829059829, + 0.826797385620915, + 0.8095238095238095, + 0.9850746268656716, + 0.8411552346570397, + 0.9923076923076923, + 0.8172757475083057, + 0.8389261744966443, + 0.8109090909090909, + 0.8338870431893688, + 0.9785714285714285, + 1.0, + 1.0, + 0.9920634920634921, + 0.8321917808219178, + 1.0, + 0.8235294117647058, + 0.8571428571428571, + 0.9739130434782609, + 0.9929078014184397, + 0.852112676056338, + 0.8239202657807309, + 0.8333333333333334, + 0.984, + 0.8356164383561644, + 0.8383838383838383, + 0.8014184397163121, + 0.9838709677419355, + 0.8598726114649682, + 0.9925373134328358, + 1.0, + 0.8395904436860068, + 0.9918032786885246, + 0.8446601941747572, + 0.9921259842519685, + 1.0, + 0.7915057915057915, + 0.8376623376623377, + 0.8247422680412371, + 0.8213058419243986, + 0.8133802816901409, + 0.9922480620155039, + 0.8344370860927153, + 0.8120300751879699, + 0.8140350877192982, + 0.8115942028985508, + 0.8392857142857143, + 0.9082278481012658, + 0.8327402135231317, + 0.8013937282229965, + 0.7396226415094339, + 0.8361774744027304, + 0.802158273381295, + 0.8235294117647058, + 0.8267148014440433, + 0.8184931506849316, + 0.8493150684931506, + 0.828996282527881, + 0.8021978021978022, + 0.8063380281690141, + 0.7851851851851852, + 0.8428571428571429, + 0.8464163822525598, + 0.8321678321678322, + 0.8374558303886925, + 0.8528428093645485, + 0.8233082706766918, + 0.7892857142857143, + 0.8363636363636363, + 0.8443708609271523, + 0.8606271777003485, + 0.8351254480286738, + 0.782608695652174, + 0.8280701754385965, + 0.8345588235294118, + 0.7903225806451613, + 0.8368794326241135, + 0.8021978021978022, + 0.8297872340425532, + 0.7808764940239044, + 0.7943262411347518, + 0.8006993006993007, + 0.7985611510791367, + 0.7922535211267606, + 0.8467153284671532, + 0.8127208480565371, + 0.8, + 0.8281786941580757, + 0.8309352517985612, + 0.7857142857142857, + 0.8320610687022901, + 0.8, + 0.8076923076923077, + 0.7991967871485943, + 0.828996282527881, + 0.7803030303030303, + 0.8220640569395018, + 0.7894736842105263, + 0.8029739776951673, + 0.8345323741007195, + 0.8136200716845878, + 0.7864768683274022, + 0.8007246376811594, + 0.8048780487804879, + 0.8352059925093633, + 0.8532423208191127, + 0.8241758241758241, + 0.8014440433212996, + 0.8529411764705882, + 0.8214285714285714, + 0.7374517374517374, + 0.8174904942965779, + 0.8028673835125448, + 0.8211678832116789, + 0.7789855072463768, + 0.8338983050847457, + 0.8207885304659498, + 0.8100358422939068, + 0.822463768115942, + 0.8, + 0.8162544169611308, + 0.8093525179856115, + 0.8264150943396227, + 0.8166089965397924, + 0.8184931506849316, + 0.8212927756653993, + 0.7962962962962963, + 0.8191489361702128, + 0.8161764705882353, + 0.8015267175572519, + 0.8333333333333334, + 0.7883211678832117, + 0.8131868131868132, + 0.8257839721254355, + 0.7935943060498221, + 0.8310344827586207, + 0.8066914498141264, + 0.775438596491228, + 0.8100358422939068, + 0.8166089965397924, + 0.8327137546468402, + 0.8184931506849316, + 0.8243727598566308, + 0.7816091954022989, + 0.8248175182481752, + 0.8079710144927537, + 0.8321428571428572, + 0.8260869565217391, + 0.8263888888888888, + 0.8075471698113208, + 0.837037037037037, + 0.8035087719298246, + 0.8291814946619217, + 0.7852112676056338, + 0.8089887640449438, + 0.8344370860927153, + 0.8064516129032258, + 0.8172043010752689, + 0.7923076923076923, + 0.7985865724381626, + 0.7653846153846153, + 0.7786561264822134, + 0.7749077490774908, + 0.8210526315789474, + 0.7849462365591398, + 0.7802197802197802, + 0.8480565371024735, + 0.7881040892193308, + 0.9916666666666667, + 0.8546099290780141, + 0.8178571428571428, + 0.8368794326241135, + 0.7944664031620553, + 0.8050541516245487, + 0.8461538461538461, + 0.8162544169611308, + 0.8082706766917294, + 0.8397212543554007, + 0.7862595419847328, + 0.9906542056074766, + 0.8292682926829268, + 0.7757352941176471, + 0.8566433566433567, + 0.8885017421602788, + 0.7955390334572491, + 0.7907801418439716, + 0.8216783216783217, + 0.7801418439716312, + 0.7657992565055762, + 0.7947761194029851, + 0.797752808988764, + 0.7906137184115524, + 0.7948717948717948, + 0.7423076923076923, + 0.7785977859778598, + 0.7992831541218638, + 0.8154981549815498, + 0.8377358490566038, + 0.8517241379310345, + 0.8278388278388278, + 0.849624060150376, + 0.8277027027027027, + 0.8296296296296296, + 0.7753623188405797, + 0.7728937728937729, + 0.8446601941747572, + 0.8076923076923077, + 0.8052434456928839, + 0.7992125984251969, + 0.8013698630136986, + 0.8458781362007168, + 0.746938775510204, + 0.7881040892193308, + 0.8260869565217391, + 0.7777777777777778, + 0.7906976744186046, + 0.8056537102473498, + 0.8154981549815498, + 0.8048780487804879, + 0.8191489361702128, + 0.8321428571428572, + 0.8216783216783217, + 0.8248175182481752, + 0.8118466898954704, + 0.8029739776951673, + 0.7794117647058824, + 0.8131868131868132, + 0.8345070422535211, + 0.8106995884773662, + 0.8252788104089219, + 0.8273381294964028, + 0.7852112676056338, + 0.85, + 0.8556701030927835, + 0.7933579335793358, + 0.8445229681978799, + 0.8007380073800738, + 0.7992831541218638, + 0.802158273381295, + 0.7922535211267606, + 0.7615384615384615, + 0.8774193548387097, + 0.8430034129692833, + 0.9914529914529915, + 0.851063829787234, + 0.8653198653198653, + 0.7723880597014925, + 0.8424908424908425, + 0.863013698630137, + 0.8367346938775511, + 0.8056537102473498, + 1.0, + 0.9916666666666667, + 0.9090909090909091, + 0.8762214983713354, + 1.0, + 0.8385964912280702, + 0.7923875432525952, + 0.7846153846153846, + 0.8327645051194539, + 0.8397212543554007, + 0.8256227758007118, + 0.8409893992932862, + 0.9921875, + 0.8422939068100358, + 0.8296296296296296, + 0.9923076923076923, + 0.7686567164179104, + 1.0, + 0.7898305084745763, + 0.8540925266903915, + 0.847972972972973, + 0.8140350877192982, + 0.9777777777777777, + 0.8205128205128205, + 0.8070175438596491, + 0.8051470588235294, + 0.9922480620155039, + 0.8042704626334519, + 1.0, + 0.8197879858657244, + 0.9682539682539683, + 0.8164794007490637, + 0.8145454545454546, + 0.9849624060150376, + 0.8218181818181818, + 0.8243727598566308, + 1.0, + 0.976, + 0.8517241379310345, + 0.8480565371024735, + 0.8322368421052632, + 0.8282828282828283, + 1.0, + 0.7937743190661478, + 0.9920634920634921, + 0.8226950354609929, + 0.832089552238806, + 1.0, + 0.7785977859778598, + 0.8076923076923077, + 0.7992565055762082, + 0.9922480620155039, + 0.8188405797101449, + 0.8450704225352113, + 0.9915966386554622, + 0.8395904436860068, + 0.8494983277591973, + 0.8494623655913979, + 1.0, + 1.0, + 0.9920634920634921, + 0.990990990990991, + 0.8581314878892734, + 1.0, + 0.9908256880733946, + 0.8368794326241135, + 0.8503401360544217, + 0.8513513513513513, + 1.0, + 0.8461538461538461, + 0.8576158940397351, + 0.9915966386554622, + 1.0, + 0.8304498269896193, + 0.845360824742268, + 0.8145454545454546, + 0.8382838283828383, + 0.8315789473684211, + 0.8211678832116789, + 0.7830882352941176, + 0.8401360544217688, + 0.8150943396226416, + 0.8113207547169812, + 0.8508474576271187, + 0.7943262411347518, + 0.9834710743801653, + 0.8506944444444444, + 0.8172043010752689, + 0.8333333333333334, + 0.8280701754385965, + 0.8181818181818182, + 0.8076923076923077, + 0.821917808219178, + 0.8351648351648352, + 0.8523489932885906, + 0.8197879858657244, + 0.8, + 0.8072727272727273, + 0.9921259842519685, + 0.8321428571428572, + 0.85, + 0.88, + 0.991869918699187, + 0.8442906574394463, + 0.822742474916388, + 0.9836065573770492, + 0.8459016393442623, + 1.0, + 0.8538205980066446, + 0.8362989323843416, + 0.875, + 0.788235294117647, + 0.8321917808219178, + 0.7910447761194029, + 0.7870722433460076, + 1.0, + 0.959349593495935, + 0.8380281690140845, + 0.7777777777777778, + 0.9922480620155039, + 0.8362989323843416, + 0.991304347826087, + 0.8498293515358362, + 0.8166089965397924, + 0.8146853146853147, + 0.8235294117647058, + 0.9927007299270073, + 1.0, + 0.9915254237288136, + 1.0, + 0.8204225352112676, + 1.0, + 0.8192307692307692, + 0.8552188552188552, + 0.975609756097561, + 0.9692307692307692, + 0.8387096774193549, + 0.8491228070175438, + 0.8035714285714286, + 0.9915966386554622, + 0.8279569892473119, + 0.8362369337979094, + 0.8395904436860068, + 0.9813084112149533, + 0.8831168831168831, + 0.991869918699187, + 0.9917355371900827, + 0.8581818181818182, + 0.9833333333333333, + 0.8075471698113208, + 0.9908256880733946, + 1.0, + 0.7822878228782287, + 0.8373702422145328, + 0.8486842105263158, + 0.8496503496503497, + 0.8305084745762712, + 1.0, + 0.8372881355932204 + ] + }, + "contact_maps": { + "mean_pearson": 0.3028004276186084, + "valid_tracks": 28, + "per_track_pearson": [ + 0.29594374193192946, + 0.29666371379837375, + 0.34793474041586564, + 0.24617767512524513, + 0.21183498639841025, + 0.2506119916747931, + 0.32477126329468264, + 0.23211651793594693, + 0.39770249702340266, + 0.34909875529463025, + 0.2533450314117726, + 0.30650893118524286, + 0.33838731689674684, + 0.2363695706405294, + 0.280342790625254, + 0.40907188899950847, + 0.08712570240143322, + 0.3419529705052659, + 0.31782700467947717, + 0.3294870300441284, + 0.2765462803204789, + 0.286010612592544, + 0.27474851107505477, + 0.34712133819648905, + 0.3594301253850397, + 0.36943163486772723, + 0.3226942275016211, + 0.3891551230994415 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.44436746659319426, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..251cf0a10d89c3067018fd5c4d584d4cf1654789 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S022/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.6192065420989417, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.52992277992278, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.17024781779397408, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7612514089079802, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.38913285526917785, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.317090943116532, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3378381424400757, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.3583924685949039, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.18985298166686143, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.21344859335547622, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.4208306206196988, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.2088797811679252, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.22805518766693694, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/README.md new file mode 100644 index 0000000000000000000000000000000000000000..43e233cb7c5c5eb226fe0510f8f1d9673108700c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/README.md @@ -0,0 +1,85 @@ +# SeqFunc scale13-small after 4 complete epochs with transcript-direction junction decoding + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. At the saved small96k/large78k boundaries, an optional training-only byte cache was enabled after exact256example and integrated16example tensor/ID checks. Source compressed bytes are verified on every cache hit, with read-only source fallback. Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. The cache is disposable and is disabled for validation/public inference. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **164,080 training examples in 41,020 optimizer updates**, 4 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +672.8 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +## Measured validation results + +Family-mean Pearson: **0.446108** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.575423 | +| dnase | 0.653354 | +| procap | 0.528898 | +| cage | 0.315120 | +| rna_seq | 0.594899 | +| chip_tf | 0.454794 | +| chip_histone | 0.483808 | +| splice_sites | 0.055513 | +| splice_usage | 0.531936 | +| junctions | 0.381239 | +| contact_maps | 0.332206 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.6107 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.6237 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.1355 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7683 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3944 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.2817 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2330 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.2074 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1542 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.1831 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4550 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1585 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2128 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..66effa982bcac090635a521be6b0c6f5f37debe7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch04/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..e0416692d2ee7deb1a25927397f725eec54d16a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.py @@ -0,0 +1,101 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = logits[:, row.splice_start:row.splice_end] + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..d945900cf1ab96abb3981b6ff2771f02f9446b6a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S023/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.6106889288637979, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.6237237237237238, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.13551315995135027, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7683294984197494, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.39437605544666954, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.2816961849795354, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.23296970341127543, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.2073928125804208, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.15421218585352367, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.1831344038460025, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.45501275758906606, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.1585073990625231, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.2127724951274715, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/README.md new file mode 100644 index 0000000000000000000000000000000000000000..80c6563d74689f9a771b1c3b77f57544862a3e87 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/README.md @@ -0,0 +1,122 @@ +# SeqFunc rc58: two-epoch small with fixed splice probabilities and equal forward/RC inference + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **82,040 training examples in 20,510 optimizer updates**, 2 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +296.0 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +This fixed-link follow-up uses the preselected complete two-epoch small model. +It reuses the original 256-training-window offsets without refitting or changing +network parameters, site selection, usage or junction decoding. Earlier experiment +selection pools are unchanged. Both entrypoints share the same probability link. +Research-only perturbation reuse is verified by exact code/weight checks, the +earlier complete native proof, and fresh native fixtures for this candidate. +All 200 validation windows and all 13 development metrics are scored in full. + +Every output family and both entry points average this same checkpoint on the provided DNA +and its reverse complement, mapped back using vendored public-metadata strand permutations. +There are 13,290,504 unique learned parameters and two deterministic views, +weighted equally without fitting development labels. No additional flanking DNA is read. +Dense and contact values average in the contract's asinh space; splice scores average the +training-prior-corrected probabilities. This approximately doubles inference work. + +Each view selects sparse coordinates independently. Their union averages missing-view values +as zero. Usage retains at most4,096 positions; junctions retain16,384 pairs by averaged own-strand +strength, with deterministic coordinate tie handling. Track arrays are averaged in float32 and +then stored in float16; perturbation arrays remain float32. Sparse recall and score dilution +can both change. The frozen decoder passed the RC17 synthetic, native offline and partial-resume +checks; its exact parent-extraction and source fingerprints are enforced by the builder. +This candidate requires its own complete validation track and development variant reports +before publication. Architecture, learned weights and training history are inherited unchanged. + +## Measured validation results + +Family-mean Pearson: **0.456509** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.561921 | +| dnase | 0.647846 | +| procap | 0.524396 | +| cage | 0.300687 | +| rna_seq | 0.584476 | +| chip_tf | 0.443282 | +| chip_histone | 0.473751 | +| splice_sites | 0.319829 | +| splice_usage | 0.493984 | +| junctions | 0.334946 | +| contact_maps | 0.336485 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5093 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.4459 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0851 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7564 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.4003 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.3192 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.3280 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3772 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1789 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2183 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.5161 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1703 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2530 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..6fbb36253ad0b167bb556f6cb97159c8042ea6f4 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch02/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..a8d93f996cb49cb57e85ec98a8226ebb48f64f5e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/package_manifest.json @@ -0,0 +1,54 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "ea695c62d648264868443acbfba188e1449f3d0044a806566b47a521426e45a4", + "files": { + "README.md": "3a668b7fc3c5e7e2008228bf3a99fa0c486c5c5c121cdc917ceafe56ae272afe", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "775b5cebc2335ac6732b582234e6665b189ae7ca56f246c4fe3be6d0e9d1011e", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "3fac4df3e9e6f91583c4d7df7e5c2574ba89939fa10ff2a07dc51c237961ceb5", + "decoding_policy.json": "01c62ef5b4cea4e5afae34390df82f5ea67439746fa491c1f5fdba0d17c2e543", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/fixed-link46-plan.json": "9e99d351af2dbaf0d51c4a253fbc21781dfcf9dd81076c7e337f25b234c795a7", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60", + "rc_ensemble_policy.json": "ba917bb81513b4be00472ad481de1acd25f76750c6d1044592dd6e03b3bd4404", + "validation_tracks.json": "2780fb9379ac4356f8eae388ec801bb8eafdcbb706974faa0a00f46a975d3634", + "validation_variants.json": "dbb6e4c50ca9517c9025ab4dc3111456aa77ae43777f74cc384f838f299badf4" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..b8ee571ad9dab90aded5077affbb42419869da16 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/perturb.py @@ -0,0 +1,92 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import predict_requested + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + group = root.create_group(f"perturbations/p{index:06d}") + for family, values in predict_requested(model, dna, row).items(): + values = np.asarray(values, dtype=np.float32) + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/rc_ensemble.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/rc_ensemble.py new file mode 100644 index 0000000000000000000000000000000000000000..f076804c66e481eb823e6ca5e721f39fce50bad5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/rc_ensemble.py @@ -0,0 +1,111 @@ +"""One shared checkpoint evaluated on DNA and its reverse complement for every output.""" +import json +from pathlib import Path +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from single_prediction import single_prediction +from splice_output import splice_output_scores + + +def contact_reversal(bins): + row, column = np.triu_indices(bins) + lookup = np.empty((bins, bins), dtype=np.int64) + lookup[row, column] = np.arange(len(row)) + lookup[column, row] = np.arange(len(row)) + return lookup[bins-1-row, bins-1-column] + + +def reverse_readout(arrays, permutations, input_length, target_length, contact_bins): + """Map a reverse-complement prediction back into the original genomic coordinates.""" + out = {name: arrays[name][::-1, permutations[name]].copy() for name in DENSE_TRACKS} + out['splice_sites'] = arrays['splice_sites'][::-1, permutations['splice_sites']].copy() + positions = target_length-1-arrays['splice_usage_positions'] + order = np.argsort(positions) + out['splice_usage_positions'] = positions[order].astype(np.int32) + out['splice_usage'] = arrays['splice_usage'][order][:, permutations['splice_usage']].copy() + out['junction_half'] = (1-arrays['junction_half']).astype(np.int8) + for key in ['junction_donor', 'junction_acceptor']: + out[key] = (input_length-1-arrays[key]).astype(np.int32) + out['junctions'] = arrays['junctions'][:, permutations['junctions']].copy() + out['contact_maps'] = arrays['contact_maps'][contact_reversal(contact_bins)][:, permutations['contact_maps']].copy() + return out + + +def combine_readouts(a, b, track_halves): + """Average sparse functions with zero at unlisted coordinates, then enforce output budgets.""" + out = {name: (a[name]+b[name])*np.float32(.5) + for name in [*DENSE_TRACKS, 'splice_sites', 'contact_maps']} + usage_positions, inverse = np.unique(np.concatenate([a['splice_usage_positions'], b['splice_usage_positions']]), return_inverse=True) + assert len(usage_positions) <= MAX_USAGE_SITES + usage = np.zeros((len(usage_positions), a['splice_usage'].shape[-1]), dtype=np.float32) + np.add.at(usage, inverse, np.concatenate([a['splice_usage'], b['splice_usage']])*np.float32(.5)) + out.update(splice_usage_positions=usage_positions.astype(np.int32), splice_usage=usage) + keys = np.concatenate([np.stack([x['junction_half'], x['junction_donor'], x['junction_acceptor']], axis=1) for x in [a, b]]) + keys, inverse = np.unique(keys, axis=0, return_inverse=True) + values = np.zeros((len(keys), len(track_halves)), dtype=np.float32) + np.add.at(values, inverse, np.concatenate([a['junctions'], b['junctions']])*np.float32(.5)) + strength = np.empty(len(keys), dtype=np.float32) + for half in [0, 1]: + keep = keys[:, 0] == half + strength[keep] = values[keep][:, track_halves == half].max(axis=1) + # Stable sorting makes ties deterministic in the canonical coordinate order. + order = np.argsort(-strength, kind='stable')[:MAX_JUNCTION_PAIRS] + keys, values = keys[order], values[order] + out.update(junction_half=keys[:, 0].astype(np.int8), + junction_donor=keys[:, 1].astype(np.int32), + junction_acceptor=keys[:, 2].astype(np.int32), junctions=values) + return out + + +class ReverseComplementMean: + def __init__(self, model): + self.model = model + data = json.loads(Path(__file__).with_name('strand_permutations.json').read_text()) + self.permutations = {name: np.asarray(values, dtype=np.int64) for name, values in data['permutations'].items()} + for name, indices in self.permutations.items(): + assert np.array_equal(indices[indices], np.arange(len(indices))), name + + def tracks(self, dna): + assert len(dna) == 1 + config = self.model.config + original = single_prediction(self.model, dna) + reverse = single_prediction(self.model, dna.flip((1, 2))) + reverse = reverse_readout(reverse, self.permutations, config.input_length, config.target_length, config.contact_bins) + return combine_readouts(original, reverse, self.model.junction_track_half.cpu().numpy()) + + def dense_and_sites(self, dna, families): + """Return central readouts for both alleles; this same probability/asinh averaging defines tracks.""" + config = self.model.config + start = (config.input_length-config.target_length)//2 + predictions = [] + for reverse in [False, True]: + trunk = self.model._trunk(dna.flip((1, 2)) if reverse else dna) + features = None + values = {} + for family in families: + if family == 'splice_sites': + logits = self.model._splice_site_logits(trunk)[:, start:start+config.target_length] + array = splice_output_scores(logits).float().cpu().numpy() + else: + if features is None: + features = self.model._dense_features(trunk) + array = self.model.dense_heads[family](features).float().cpu().numpy() + if reverse: + array = array[:, ::-1][:, :, self.permutations[family]].copy() + values[family] = array + predictions.append(values) + return {family: (predictions[0][family]+predictions[1][family])*np.float32(.5) for family in families} + + +def engine_for(model): + if not hasattr(model, '_rc_prediction_engine'): + # Plain helper, not a child nn.Module; the backbone state dictionary stays unchanged. + model._rc_prediction_engine = ReverseComplementMean(model) + return model._rc_prediction_engine + + +def predict_requested(model, dna, row): + values = engine_for(model).dense_and_sites(dna, row.families) + return {family: array[:, row.splice_start:row.splice_end] if family == 'splice_sites' + else array[:, row.dense_bin_start:row.dense_bin_end] for family, array in values.items()} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..155ef4138af508a739d6af8d7cd283308fb7b4a9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/dataset.py @@ -0,0 +1,293 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/fixed-link46-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/fixed-link46-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..e031906861bdfa87bf890c32a352f8488727ffee --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/fixed-link46-plan.json @@ -0,0 +1,68 @@ +{ + "created_utc": "2026-10-25T06:08:23.071762+00:00", + "parent_evaluation": "/workspace/research/experiments/scale13-small-epoch02", + "parent_inference_signature": { + "last.pt": "775b5cebc2335ac6732b582234e6665b189ae7ca56f246c4fe3be6d0e9d1011e", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "output": "/workspace/research/experiments/calib46", + "parent_reward": 0.42285346475207175, + "only_change": "Apply the same frozen sigmoid(raw splice logits + four training-prior offsets) in both entrypoints; keep all weights, raw site selection, sparse usage/junction decoding, other families and orientation fixed.", + "parent_choice": "The current strongest fully evaluated/replay-checked raw publication, small at82,040examples. This is an independently declared transfer of the existing output link to a later checkpoint; earlier experiment selection pools remain unchanged.", + "no_new_parameter_fitting": true, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "helper_sha256": { + "run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic symlink without overwrite; retain study prepared-calib46 outputs while referenced. No cache is an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "validation": "Fresh native/recovery52-array check; exact link-only code/weight guard, prior12486-array proof plus36fresh native perturbation arrays for exact research-cache derivation; all200 validation tracks and13 official variant scores. Unaffected10family and8variant reports must equalparent before promotion.", + "caveat": "Fixed marginal prior correction is approximate for per-example class weighting, and the verifier applies its unchanged additional sigmoid to the emitted scores. Previous checkpoints show mixed variant effects; no improvement is assumed." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..9a1a939759aadcbbf055141ba324252daeb98ef3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S024/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.509338513036824, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.4458923208923209, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.08513165285451701, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7564011316068845, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.4003161595277227, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.31922744776012935, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3280305270933055, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.37718924150492894, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.17891907672450724, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2182563107115556, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.5160508760368996, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.17032665551673193, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.2530095097041453, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/README.md new file mode 100644 index 0000000000000000000000000000000000000000..93f9e3f0d04e7d4b34f9892c1132c3c964051e4d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/README.md @@ -0,0 +1,122 @@ +# SeqFunc rc85: three-epoch small with fixed splice probabilities and equal forward/RC inference + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Unchanged small control from the original matched scale13 study, trained from local random initialization with seed 17, batch four, fixed variance-normalized family losses and the original 14-epoch learning-rate schedule. This additional immutable three-epoch evaluation is the fixed equal-exposure control for context43; training is not restarted or altered. It retains six local blocks (10,941-bp receptive field), the local-to-trunk bridge, residual RMS cap 0.1, raw splice logits and original sparse decoding. Complete native/recovery and full validation checks precede any publication. + +This checkpoint records **123,060 training examples in 30,765 optimizer updates**, 3 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +494.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. In preliminary +diagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed. + +This fixed-link follow-up uses the preselected complete three-epoch small model. +It reuses the original 256-training-window offsets without refitting or changing +network parameters, site selection, usage or junction decoding. Earlier experiment +selection pools are unchanged. Both entrypoints share the same probability link. +Research-only perturbation reuse is verified by exact code/weight checks, the +earlier complete native proof, and fresh native fixtures for this candidate. +All 200 validation windows and all 13 development metrics are scored in full. + +Every output family and both entry points average this same checkpoint on the provided DNA +and its reverse complement, mapped back using vendored public-metadata strand permutations. +There are 13,290,504 unique learned parameters and two deterministic views, +weighted equally without fitting development labels. No additional flanking DNA is read. +Dense and contact values average in the contract's asinh space; splice scores average the +training-prior-corrected probabilities. This approximately doubles inference work. + +Each view selects sparse coordinates independently. Their union averages missing-view values +as zero. Usage retains at most4,096 positions; junctions retain16,384 pairs by averaged own-strand +strength, with deterministic coordinate tie handling. Track arrays are averaged in float32 and +then stored in float16; perturbation arrays remain float32. Sparse recall and score dilution +can both change. The frozen decoder passed the RC17 synthetic, native offline and partial-resume +checks; its exact parent-extraction and source fingerprints are enforced by the builder. +This candidate requires its own complete validation track and development variant reports +before publication. Architecture, learned weights and training history are inherited unchanged. + +## Measured validation results + +Family-mean Pearson: **0.478074** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.574384 | +| dnase | 0.653846 | +| procap | 0.537602 | +| cage | 0.314840 | +| rna_seq | 0.602918 | +| chip_tf | 0.457175 | +| chip_histone | 0.487492 | +| splice_sites | 0.351592 | +| splice_usage | 0.538465 | +| junctions | 0.397637 | +| contact_maps | 0.342865 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4533 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5506 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.0458 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7640 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.3883 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.3533 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.3125 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.3965 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1621 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2041 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4494 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1817 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2830 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..ebb1bbde8898494f04485b6fdb6d2e5756809e86 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch03/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..f4be9b8793a61ea16a9d3b3d2f0cd4999c5e96f3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/package_manifest.json @@ -0,0 +1,64 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "6e9091a83e34d161017bb7662329d849ce2bd54fe3dd9870336645a63da1f71f", + "files": { + "README.md": "d998134cd87823b637578e0cbdf4fdfa10905cd6e4badc07eebfbe75f096a826", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "bb5df0fc3e1827e183047345bf246fd33348c79558a51467fb2811528ab3e418", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "72131abbba8ee5744773b844dc863bf9c4cf1a8aa46abd14d1366f6b1c637a43", + "training_assets_manifest.json": "7e4a33e58edeb9c25fb59ec5b1f3b65ff2b1ca83abb9c3fc0770ec4f57af06c9", + "decoding_policy.json": "cd2960a0fd5e5c6a94f4fa026ae1a17f19dccf73ae2fda146a906e246f5dfe62", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/fixed-link57-plan.json": "ae2908ab7195f3c3f0dd93b2aa9112fc103ac1e215f9652809da741414c7d25d", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60", + "rc_ensemble_policy.json": "707cdebc095eb292124c56512149cea2ba0e8261edbd76705fda46a843ada5e5", + "validation_tracks.json": "03dd4a3e4839a09df89a12c3a1b82765cbe7d7e40e484fa2edfdde24be3ba8ab", + "validation_variants.json": "c0e650d8c948b232b74643b5ceba1b373098b13784ff2f9968c77bc8e70efa31" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..b8ee571ad9dab90aded5077affbb42419869da16 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.py @@ -0,0 +1,92 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import predict_requested + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + group = root.create_group(f"perturbations/p{index:06d}") + for family, values in predict_requested(model, dna, row).items(): + values = np.asarray(values, dtype=np.float32) + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..665fb186f3a62b1ff130abb034ac25a3a924c7c8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/predict.py @@ -0,0 +1,52 @@ +"""Stream one shared-checkpoint forward/reverse-complement mean for every output family.""" +from __future__ import annotations +import argparse +import json +from pathlib import Path +from numcodecs import Blosc +import numpy as np +import torch +import zarr +from prediction_contract import PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import engine_for + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + model = CompactModel(ModelConfig(**payload['config'])) + model.load_state_dict(payload['model'], strict=True) + device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + engine = engine_for(model) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample['example_id'] + values = engine.tracks(sample['dna'][None].to(device)) + group = root.create_group(f"examples/{example_id}") + for name, array in values.items(): + dtype = np.int8 if name == 'junction_half' else np.int32 if name in ( + 'splice_usage_positions', 'junction_donor', 'junction_acceptor') else np.float16 + group.create_dataset(name, data=array.astype(dtype), compressor=COMPRESSOR) + ids.append(example_id) + print(json.dumps({'event': 'predict', 'example': len(ids), 'total': len(dataset)}), flush=True) + root.attrs.update({'schema': PREDICTION_SCHEMA, 'example_ids': ids}) + (output_root/'example_ids.json').write_text(json.dumps(ids)+'\n') + + +if __name__ == '__main__': + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('input_root', type=Path) + p.add_argument('output_root', type=Path) + p.add_argument('--checkpoint', type=Path, default=Path(__file__).resolve().parent/'last.pt') + p.add_argument('--split', default='test', choices=['valid', 'test']) + a = p.parse_args() + main(a.input_root, a.output_root, a.checkpoint, a.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble.py new file mode 100644 index 0000000000000000000000000000000000000000..f076804c66e481eb823e6ca5e721f39fce50bad5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble.py @@ -0,0 +1,111 @@ +"""One shared checkpoint evaluated on DNA and its reverse complement for every output.""" +import json +from pathlib import Path +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from single_prediction import single_prediction +from splice_output import splice_output_scores + + +def contact_reversal(bins): + row, column = np.triu_indices(bins) + lookup = np.empty((bins, bins), dtype=np.int64) + lookup[row, column] = np.arange(len(row)) + lookup[column, row] = np.arange(len(row)) + return lookup[bins-1-row, bins-1-column] + + +def reverse_readout(arrays, permutations, input_length, target_length, contact_bins): + """Map a reverse-complement prediction back into the original genomic coordinates.""" + out = {name: arrays[name][::-1, permutations[name]].copy() for name in DENSE_TRACKS} + out['splice_sites'] = arrays['splice_sites'][::-1, permutations['splice_sites']].copy() + positions = target_length-1-arrays['splice_usage_positions'] + order = np.argsort(positions) + out['splice_usage_positions'] = positions[order].astype(np.int32) + out['splice_usage'] = arrays['splice_usage'][order][:, permutations['splice_usage']].copy() + out['junction_half'] = (1-arrays['junction_half']).astype(np.int8) + for key in ['junction_donor', 'junction_acceptor']: + out[key] = (input_length-1-arrays[key]).astype(np.int32) + out['junctions'] = arrays['junctions'][:, permutations['junctions']].copy() + out['contact_maps'] = arrays['contact_maps'][contact_reversal(contact_bins)][:, permutations['contact_maps']].copy() + return out + + +def combine_readouts(a, b, track_halves): + """Average sparse functions with zero at unlisted coordinates, then enforce output budgets.""" + out = {name: (a[name]+b[name])*np.float32(.5) + for name in [*DENSE_TRACKS, 'splice_sites', 'contact_maps']} + usage_positions, inverse = np.unique(np.concatenate([a['splice_usage_positions'], b['splice_usage_positions']]), return_inverse=True) + assert len(usage_positions) <= MAX_USAGE_SITES + usage = np.zeros((len(usage_positions), a['splice_usage'].shape[-1]), dtype=np.float32) + np.add.at(usage, inverse, np.concatenate([a['splice_usage'], b['splice_usage']])*np.float32(.5)) + out.update(splice_usage_positions=usage_positions.astype(np.int32), splice_usage=usage) + keys = np.concatenate([np.stack([x['junction_half'], x['junction_donor'], x['junction_acceptor']], axis=1) for x in [a, b]]) + keys, inverse = np.unique(keys, axis=0, return_inverse=True) + values = np.zeros((len(keys), len(track_halves)), dtype=np.float32) + np.add.at(values, inverse, np.concatenate([a['junctions'], b['junctions']])*np.float32(.5)) + strength = np.empty(len(keys), dtype=np.float32) + for half in [0, 1]: + keep = keys[:, 0] == half + strength[keep] = values[keep][:, track_halves == half].max(axis=1) + # Stable sorting makes ties deterministic in the canonical coordinate order. + order = np.argsort(-strength, kind='stable')[:MAX_JUNCTION_PAIRS] + keys, values = keys[order], values[order] + out.update(junction_half=keys[:, 0].astype(np.int8), + junction_donor=keys[:, 1].astype(np.int32), + junction_acceptor=keys[:, 2].astype(np.int32), junctions=values) + return out + + +class ReverseComplementMean: + def __init__(self, model): + self.model = model + data = json.loads(Path(__file__).with_name('strand_permutations.json').read_text()) + self.permutations = {name: np.asarray(values, dtype=np.int64) for name, values in data['permutations'].items()} + for name, indices in self.permutations.items(): + assert np.array_equal(indices[indices], np.arange(len(indices))), name + + def tracks(self, dna): + assert len(dna) == 1 + config = self.model.config + original = single_prediction(self.model, dna) + reverse = single_prediction(self.model, dna.flip((1, 2))) + reverse = reverse_readout(reverse, self.permutations, config.input_length, config.target_length, config.contact_bins) + return combine_readouts(original, reverse, self.model.junction_track_half.cpu().numpy()) + + def dense_and_sites(self, dna, families): + """Return central readouts for both alleles; this same probability/asinh averaging defines tracks.""" + config = self.model.config + start = (config.input_length-config.target_length)//2 + predictions = [] + for reverse in [False, True]: + trunk = self.model._trunk(dna.flip((1, 2)) if reverse else dna) + features = None + values = {} + for family in families: + if family == 'splice_sites': + logits = self.model._splice_site_logits(trunk)[:, start:start+config.target_length] + array = splice_output_scores(logits).float().cpu().numpy() + else: + if features is None: + features = self.model._dense_features(trunk) + array = self.model.dense_heads[family](features).float().cpu().numpy() + if reverse: + array = array[:, ::-1][:, :, self.permutations[family]].copy() + values[family] = array + predictions.append(values) + return {family: (predictions[0][family]+predictions[1][family])*np.float32(.5) for family in families} + + +def engine_for(model): + if not hasattr(model, '_rc_prediction_engine'): + # Plain helper, not a child nn.Module; the backbone state dictionary stays unchanged. + model._rc_prediction_engine = ReverseComplementMean(model) + return model._rc_prediction_engine + + +def predict_requested(model, dna, row): + values = engine_for(model).dense_and_sites(dna, row.families) + return {family: array[:, row.splice_start:row.splice_end] if family == 'splice_sites' + else array[:, row.dense_bin_start:row.dense_bin_end] for family, array in values.items()} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..227241078b2fa73249ebaeb53d53b25d3f579b13 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/rc_ensemble_policy.json @@ -0,0 +1,35 @@ +{ + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-25T17:46:19.540817+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib57/package", + "parent_inference_signature": { + "last.pt": "bb5df0fc3e1827e183047345bf246fd33348c79558a51467fb2811528ab3e418", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 13290504, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/cached_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/cached_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3677c1e8518ea97db0dd6759f3d636f5fff27c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/cached_read_store.py @@ -0,0 +1,71 @@ +"""Optional local byte cache for immutable training Zarr stores. + +Source storage stays read-only. Cached entries contain a key fingerprint, length, +and SHA256 of the original compressed bytes; every hit verifies all three. Writes +replace a private temporary inode atomically. A missing/corrupt/unwritable cache +falls back to the original store, so this layer changes neither values nor keys. +Only training loaders may opt into it; it is not an inference asset. +""" +import hashlib +import os +from pathlib import Path +import struct +import tempfile + +from .direct_read_store import ParallelDirectReadDirectoryStore + +MAGIC = b'SFBC0001' +HEADER = struct.Struct('<8sQ32s32s') + + +class CachedReadDirectoryStore(ParallelDirectReadDirectoryStore): + def __init__(self, path, *, cache_root, namespace, io_workers=4): + super().__init__(path, io_workers=io_workers) + if not namespace: + raise ValueError('An immutable input namespace is required') + self.cache_root = Path(cache_root) + self.namespace = namespace + self.store_key = hashlib.sha256((namespace+'\0'+os.path.abspath(path)).encode()).digest() + self.statistics = {'hit': 0, 'miss': 0, 'corrupt': 0, 'write_failure': 0} + + def cache_location(self, key): + key = self._normalize_key(key) + identity = hashlib.sha256(self.store_key+key.encode()).digest() + name = identity.hex() + return self.cache_root/name[:2]/name[2:], identity + + def __getitem__(self, key): + key = self._normalize_key(key) + destination, identity = self.cache_location(key) + try: + blob = destination.read_bytes() + if len(blob) >= HEADER.size: + magic, size, cached_key, checksum = HEADER.unpack_from(blob) + payload = blob[HEADER.size:] + if (magic == MAGIC and size == len(payload) and cached_key == identity + and hashlib.sha256(payload).digest() == checksum): + self.statistics['hit'] += 1 + return payload + self.statistics['corrupt'] += 1 + except OSError: + pass + self.statistics['miss'] += 1 + payload = super().__getitem__(key) + temporary = None + try: + destination.parent.mkdir(parents=True, exist_ok=True) + with tempfile.NamedTemporaryFile(prefix='.pending-', dir=destination.parent, delete=False) as stream: + temporary = stream.name + stream.write(HEADER.pack(MAGIC, len(payload), identity, hashlib.sha256(payload).digest())) + stream.write(payload) + os.replace(temporary, destination) + temporary = None + except OSError: + self.statistics['write_failure'] += 1 + finally: + if temporary is not None: + try: + os.unlink(temporary) + except OSError: + pass + return payload diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/single_prediction.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/single_prediction.py new file mode 100644 index 0000000000000000000000000000000000000000..a47022029461e5c8097b3a653af8b3407196d67d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/single_prediction.py @@ -0,0 +1,97 @@ +"""The parent single-orientation decoder, preserving float32 until ensemble averaging.""" +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from splice_output import splice_output_scores + +COMPRESSOR = None + +class ArrayCollector: + def __init__(self): + self.arrays = {} + def create_dataset(self, name, *, data, **kwargs): + self.arrays[name] = np.asarray(data) + +def single_prediction(model, dna): + config = model.config + group = ArrayCollector() + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy(), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy(), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy(), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy(), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy(), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + return group.arrays diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/strand_permutations.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/strand_permutations.json new file mode 100644 index 0000000000000000000000000000000000000000..044c090773f20347601dcd4f0983c4dfd4c0ecc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/strand_permutations.json @@ -0,0 +1,5959 @@ +{ + "schema": "seqfunc-strand-mapping/v1", + "source": "public track metadata; no target values", + "source_sha256": "23450e6ff9463c64d271d7014f8a798875bc95186c36aeb34609c7dc1d62cd5f", + "permutations": { + "atac": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166 + ], + "dnase": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304 + ], + "procap": [ + 6, + 7, + 8, + 9, + 10, + 11, + 0, + 1, + 2, + 3, + 4, + 5 + ], + "cage": [ + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272 + ], + "rna_seq": [ + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666 + ], + "chip_tf": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115, + 1116, + 1117, + 1118, + 1119, + 1120, + 1121, + 1122, + 1123, + 1124, + 1125, + 1126, + 1127, + 1128, + 1129, + 1130, + 1131, + 1132, + 1133, + 1134, + 1135, + 1136, + 1137, + 1138, + 1139, + 1140, + 1141, + 1142, + 1143, + 1144, + 1145, + 1146, + 1147, + 1148, + 1149, + 1150, + 1151, + 1152, + 1153, + 1154, + 1155, + 1156, + 1157, + 1158, + 1159, + 1160, + 1161, + 1162, + 1163, + 1164, + 1165, + 1166, + 1167, + 1168, + 1169, + 1170, + 1171, + 1172, + 1173, + 1174, + 1175, + 1176, + 1177, + 1178, + 1179, + 1180, + 1181, + 1182, + 1183, + 1184, + 1185, + 1186, + 1187, + 1188, + 1189, + 1190, + 1191, + 1192, + 1193, + 1194, + 1195, + 1196, + 1197, + 1198, + 1199, + 1200, + 1201, + 1202, + 1203, + 1204, + 1205, + 1206, + 1207, + 1208, + 1209, + 1210, + 1211, + 1212, + 1213, + 1214, + 1215, + 1216, + 1217, + 1218, + 1219, + 1220, + 1221, + 1222, + 1223, + 1224, + 1225, + 1226, + 1227, + 1228, + 1229, + 1230, + 1231, + 1232, + 1233, + 1234, + 1235, + 1236, + 1237, + 1238, + 1239, + 1240, + 1241, + 1242, + 1243, + 1244, + 1245, + 1246, + 1247, + 1248, + 1249, + 1250, + 1251, + 1252, + 1253, + 1254, + 1255, + 1256, + 1257, + 1258, + 1259, + 1260, + 1261, + 1262, + 1263, + 1264, + 1265, + 1266, + 1267, + 1268, + 1269, + 1270, + 1271, + 1272, + 1273, + 1274, + 1275, + 1276, + 1277, + 1278, + 1279, + 1280, + 1281, + 1282, + 1283, + 1284, + 1285, + 1286, + 1287, + 1288, + 1289, + 1290, + 1291, + 1292, + 1293, + 1294, + 1295, + 1296, + 1297, + 1298, + 1299, + 1300, + 1301, + 1302, + 1303, + 1304, + 1305, + 1306, + 1307, + 1308, + 1309, + 1310, + 1311, + 1312, + 1313, + 1314, + 1315, + 1316, + 1317, + 1318, + 1319, + 1320, + 1321, + 1322, + 1323, + 1324, + 1325, + 1326, + 1327, + 1328, + 1329, + 1330, + 1331, + 1332, + 1333, + 1334, + 1335, + 1336, + 1337, + 1338, + 1339, + 1340, + 1341, + 1342, + 1343, + 1344, + 1345, + 1346, + 1347, + 1348, + 1349, + 1350, + 1351, + 1352, + 1353, + 1354, + 1355, + 1356, + 1357, + 1358, + 1359, + 1360, + 1361, + 1362, + 1363, + 1364, + 1365, + 1366, + 1367, + 1368, + 1369, + 1370, + 1371, + 1372, + 1373, + 1374, + 1375, + 1376, + 1377, + 1378, + 1379, + 1380, + 1381, + 1382, + 1383, + 1384, + 1385, + 1386, + 1387, + 1388, + 1389, + 1390, + 1391, + 1392, + 1393, + 1394, + 1395, + 1396, + 1397, + 1398, + 1399, + 1400, + 1401, + 1402, + 1403, + 1404, + 1405, + 1406, + 1407, + 1408, + 1409, + 1410, + 1411, + 1412, + 1413, + 1414, + 1415, + 1416, + 1417, + 1418, + 1419, + 1420, + 1421, + 1422, + 1423, + 1424, + 1425, + 1426, + 1427, + 1428, + 1429, + 1430, + 1431, + 1432, + 1433, + 1434, + 1435, + 1436, + 1437, + 1438, + 1439, + 1440, + 1441, + 1442, + 1443, + 1444, + 1445, + 1446, + 1447, + 1448, + 1449, + 1450, + 1451, + 1452, + 1453, + 1454, + 1455, + 1456, + 1457, + 1458, + 1459, + 1460, + 1461, + 1462, + 1463, + 1464, + 1465, + 1466, + 1467, + 1468, + 1469, + 1470, + 1471, + 1472, + 1473, + 1474, + 1475, + 1476, + 1477, + 1478, + 1479, + 1480, + 1481, + 1482, + 1483, + 1484, + 1485, + 1486, + 1487, + 1488, + 1489, + 1490, + 1491, + 1492, + 1493, + 1494, + 1495, + 1496, + 1497, + 1498, + 1499, + 1500, + 1501, + 1502, + 1503, + 1504, + 1505, + 1506, + 1507, + 1508, + 1509, + 1510, + 1511, + 1512, + 1513, + 1514, + 1515, + 1516, + 1517, + 1518, + 1519, + 1520, + 1521, + 1522, + 1523, + 1524, + 1525, + 1526, + 1527, + 1528, + 1529, + 1530, + 1531, + 1532, + 1533, + 1534, + 1535, + 1536, + 1537, + 1538, + 1539, + 1540, + 1541, + 1542, + 1543, + 1544, + 1545, + 1546, + 1547, + 1548, + 1549, + 1550, + 1551, + 1552, + 1553, + 1554, + 1555, + 1556, + 1557, + 1558, + 1559, + 1560, + 1561, + 1562, + 1563, + 1564, + 1565, + 1566, + 1567, + 1568, + 1569, + 1570, + 1571, + 1572, + 1573, + 1574, + 1575, + 1576, + 1577, + 1578, + 1579, + 1580, + 1581, + 1582, + 1583, + 1584, + 1585, + 1586, + 1587, + 1588, + 1589, + 1590, + 1591, + 1592, + 1593, + 1594, + 1595, + 1596, + 1597, + 1598, + 1599, + 1600, + 1601, + 1602, + 1603, + 1604, + 1605, + 1606, + 1607, + 1608, + 1609, + 1610, + 1611, + 1612, + 1613, + 1614, + 1615, + 1616 + ], + "chip_histone": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115 + ], + "contact_maps": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27 + ], + "splice_sites": [ + 2, + 3, + 0, + 1 + ], + "splice_usage": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ], + "junctions": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ] + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8e314c842cca039da1d2be8da7fff6a146178651 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_assets_manifest.json @@ -0,0 +1,21 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-integration_check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-integration_check.json new file mode 100644 index 0000000000000000000000000000000000000000..1377348921feb42310e66b133df8ebee9f0d591c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-integration_check.json @@ -0,0 +1,166 @@ +{ + "created_utc": "2026-10-25T03:48:07.725319+00:00", + "preparation_sha256": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "source_sha256": "b0921bfd3d3dc47f3cb82debc0b492d8291693c40e0cef028764c713c1257e85", + "original_full256_report_sha256": "e6620ff13b00a36857dad22fa0c9d1ad5c2b266eca70773650d280998ba20005", + "examples": 16, + "worker_processes": 4, + "ordered_indices": [ + 38509, + 39126, + 38926, + 17432, + 6551, + 22599, + 24749, + 10790, + 7022, + 30789, + 17340, + 40258, + 26222, + 22674, + 23658, + 17167 + ], + "all_returned_training_ids_and_tensor_bytes_exact": true, + "reference_and_cache_both_match_original_report": true, + "public_inference_and_validation_cache_disabled_despite_environment": true, + "default_cache_disabled": true, + "models_and_trainers_unchanged": true, + "rows": { + "reference": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ], + "cached": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ] + }, + "no_active_source_or_training_changes": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/fixed-link57-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/fixed-link57-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..a6b7bf5ecc16d4f887d74509c5b8632f6069fb71 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/fixed-link57-plan.json @@ -0,0 +1,84 @@ +{ + "created_utc": "2026-10-25T09:02:04.055682+00:00", + "parent_evaluation": "/workspace/research/experiments/scale13-small-epoch03", + "parent_inference_signature": { + "last.pt": "bb5df0fc3e1827e183047345bf246fd33348c79558a51467fb2811528ab3e418", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "output": "/workspace/research/experiments/calib57", + "parent_reward": null, + "only_change": "Apply the same frozen sigmoid(raw splice logits + four training-prior offsets) in both entrypoints; keep all weights, raw site selection, sparse usage/junction decoding, other families and orientation fixed.", + "parent_choice": "The already immutable rawsmall3 matched control, declared before its full result. Admit only after its fullnative/full200/13 evaluation completes and reward exceeds rawsmall2 by1e-6. Original candidate pools remain unchanged.", + "no_new_parameter_fitting": true, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "helper_sha256": { + "run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "run_fixed_link_followup57.py": "18e93ffcce73900b4c79e6f5b724b571d197a0003cf816bfcca2fb4dfa2494eb" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic symlink without overwrite; retain study prepared-calib57 outputs while referenced. No cache is an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "validation": "Fresh native/recovery52-array check; exact link-only code/weight guard, prior12486-array proof plus36fresh native perturbation arrays for exact research-cache derivation; all200 validation tracks and13 official variant scores. Unaffected10family and8variant reports must equalparent before promotion.", + "caveat": "Fixed marginal prior correction is approximate for per-example class weighting, and the verifier applies its unchanged additional sigmoid to the emitted scores. Previous checkpoints show mixed variant effects; no improvement is assumed.", + "parent_native_report": "/workspace/research/studies/context43/native-control/report.json", + "parent_gain_gate": { + "previous_raw_reward": 0.42285346475207175, + "minimum_gain": 1e-06, + "failed_gate_action": "Finish with rejected_at_parent_gain_gate; do not build or evaluate the candidate." + }, + "evidence_sha256": { + "/workspace/research/studies/fixed_link57/construction-proof.json": "ec9561dda0b29ab844340f1a9486173419f33848a5b43f4bbd6f2d59f3441dd7", + "/workspace/research/experiments/calib46/evaluation.json": "adc6b1f35d643d044dcfc4acc53d861d131f7a79b2f34896e3b51cde00e9e876", + "/workspace/research/studies/fixed_link46/matched-comparison.json": "00565bd9b9f26479b9cdf2926814287cc0ed6779d6bc1ad2ee1aa3e31728b490", + "/workspace/research/analysis/calib46-vs-small-epoch02-splice-variants.json": "60e1ef096b62482b87ad416bde3d5c87fb92db199143e8eb8f05d0baef341579", + "/workspace/research/experiments/scale13-small-epoch02/valid-score/report.json": "db460f3cc22b24250402b8c8362983a0c456d4b1c466675c40673f19391c2302", + "/workspace/research/experiments/scale13-small/epoch-03.pt": "6e9091a83e34d161017bb7662329d849ce2bd54fe3dd9870336645a63da1f71f" + }, + "motivation": "Calib46 completed with +.021514reward solelyfromsites and five positive splicevariant pointchanges; onlyMFASS pairedintervals excludezero bothschemes. Test transferof the exact frozen trainingprior to the next completed stronger rawcheckpoint. Do not refit or assume variant gains." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..4c83676ebb49d11e0865cb2de03caa5a6193b1f1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/training_provenance.json @@ -0,0 +1,154 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 3, + "offset": 0, + "global_step": 123060, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 29675.902750284004, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 30765, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + }, + "inference_ensemble": { + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-25T17:46:19.540817+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib57/package", + "parent_inference_signature": { + "last.pt": "bb5df0fc3e1827e183047345bf246fd33348c79558a51467fb2811528ab3e418", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 13290504, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..229e3b94a8eecaa1dc98350e0e1e347b919e68ef --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5743844951710724, + "valid_tracks": 167, + "per_track_pearson": [ + 0.6957353860428591, + 0.7092630306696002, + 0.45451563535273354, + 0.6390579697153944, + 0.464933058052433, + 0.4635697179709902, + 0.48032737860123753, + 0.5614561944409215, + 0.5731216834523751, + 0.6362656282645907, + 0.5751266277222301, + 0.550551538731975, + 0.39351212535231744, + 0.5206534132398324, + 0.6752517531787227, + 0.49949069667039, + 0.5513377019104274, + 0.5185086997863014, + 0.5539247031685435, + 0.5371271000068742, + 0.50476602619879, + 0.47064740083929907, + 0.5458677926780732, + 0.521095059910754, + 0.5194164498309527, + 0.5010636284785981, + 0.4780478897708932, + 0.49747765014579987, + 0.5074840803215358, + 0.5671312365129272, + 0.48301078490501925, + 0.5197098559648811, + 0.4738568801658744, + 0.39711921359542013, + 0.3651015683602756, + 0.5796781459469268, + 0.5182997347984812, + 0.6249062980585369, + 0.5084799493653263, + 0.49465314027427004, + 0.5792431155305647, + 0.5800667476242767, + 0.49468938093056425, + 0.4718902274741405, + 0.5463833319560577, + 0.5586967630526704, + 0.4333499579775184, + 0.4854446303431067, + 0.48014749628573994, + 0.5206906118079904, + 0.4926632994730835, + 0.46138679528800636, + 0.552367065210425, + 0.593089523704501, + 0.45256167716383666, + 0.6813262843700324, + 0.8002339626326592, + 0.5490454485924825, + 0.5923788691821364, + 0.628568574037675, + 0.7692716693411966, + 0.4938328477255295, + 0.6936239769550426, + 0.665167772253618, + 0.635843913154512, + 0.5031745054273932, + 0.5671610658454069, + 0.6306603154486412, + 0.7761388622972071, + 0.6995359180384751, + 0.521158081151603, + 0.4727370705370921, + 0.45010578308672394, + 0.5374998138553739, + 0.5010843534501569, + 0.4846918437277839, + 0.49455897870081056, + 0.5337183547948554, + 0.57701337099364, + 0.48831063036661293, + 0.5698319355388946, + 0.5929464719508005, + 0.5093071641583816, + 0.48268626835455813, + 0.5157292141420885, + 0.506450546943294, + 0.4861794993006857, + 0.477888561339638, + 0.44911390338338525, + 0.5008192383606757, + 0.5052687548125948, + 0.5008871736538217, + 0.4831147447150411, + 0.5005913890473996, + 0.5629405543670974, + 0.387692631405122, + 0.4411128738524907, + 0.4324776528594302, + 0.48884086379643094, + 0.5054966509633332, + 0.45643879880949084, + 0.45912488979118116, + 0.5026805262726186, + 0.46476018048717066, + 0.5403451249913633, + 0.5472417896483334, + 0.5363072126974526, + 0.5178146325627108, + 0.539406014157964, + 0.5455632256035732, + 0.6012003732793644, + 0.5252279431747958, + 0.5161800864003923, + 0.4411229770766478, + 0.46629363840038396, + 0.4362866623087981, + 0.4754890737343135, + 0.4822514722300489, + 0.4901558905972402, + 0.47268033216470423, + 0.6522784656921756, + 0.6663933847139105, + 0.7323975879924929, + 0.7196194830384944, + 0.7100843526591094, + 0.5730327289282174, + 0.848984892149444, + 0.7181401939041898, + 0.7557384814112628, + 0.6785687002153479, + 0.722835763060516, + 0.5624462821431508, + 0.5890737778476122, + 0.5609003650455276, + 0.707629244645276, + 0.7384091290260029, + 0.6926569562501371, + 0.5379587961420694, + 0.4249926516071647, + 0.8075069924144341, + 0.7792389751805511, + 0.7985302595927181, + 0.79123123483265, + 0.5746740651104316, + 0.8566319660001295, + 0.767662806635998, + 0.7335418010821618, + 0.7257175956669101, + 0.830721383537086, + 0.8058380009742011, + 0.6150931113516945, + 0.6200895894125121, + 0.7122591750168659, + 0.7425983186960363, + 0.7587543527885056, + 0.6026449597137741, + 0.73614322588476, + 0.5969525608948072, + 0.5980690182579579, + 0.6820118596277251, + 0.5947755223025856, + 0.7688034758920615, + 0.7726322857441009, + 0.6443416792291263, + 0.5917423009012928, + 0.5890775355982012, + 0.7357827491814665 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.653846058305967, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5607484303801804, + 0.9241072656624721, + 0.7879553773272169, + 0.5215498226222285, + 0.5319076831244028, + 0.540479419466409, + 0.938882243292685, + 0.7182351096854374, + 0.4567522072298517, + 0.35341159994261523, + 0.8844852062929898, + 0.7139820388772549, + 0.43850183198746173, + 0.5237995445752986, + 0.5056918744603146, + 0.8129459513045095, + 0.7966687307568652, + 0.8515646022842447, + 0.867849546786195, + 0.8951299838758248, + 0.699421925168662, + 0.623226401264008, + 0.9040364684596607, + 0.5347534746068663, + 0.5486915844344101, + 0.7840928933662471, + 0.46709202643541126, + 0.6817586812409832, + 0.5337579298045226, + 0.5560543086094754, + 0.6385117928913636, + 0.4769724657126996, + 0.8445579315526343, + 0.6313820384729585, + 0.8826828945114502, + 0.8903406106878144, + 0.4551803231591698, + 0.8313218475576143, + 0.8772041058913974, + 0.8631382068873246, + 0.8316218397484254, + 0.6737383989746761, + 0.7080431605675059, + 0.6443529300746181, + 0.5385637375085917, + 0.6428511505817479, + 0.7369902752763097, + 0.5059052295799898, + 0.6016394741123332, + 0.4136744675028047, + 0.39233388730243823, + 0.5736451782228434, + 0.5093921366873344, + 0.8283612530560763, + 0.4488992895037183, + 0.4133302073592845, + 0.47167912591348754, + 0.5208432425571183, + 0.4909894176201022, + 0.5281638423921683, + 0.4598469383300673, + 0.6216017988275201, + 0.5060831397132224, + 0.4960760251142462, + 0.5245046700006921, + 0.5322813798321653, + 0.5629319871704302, + 0.4905667148906886, + 0.39219438097500375, + 0.4368821171528839, + 0.5137247386790365, + 0.6546989556150727, + 0.5155359238821962, + 0.6439272165906464, + 0.37128822779608794, + 0.8124558664377841, + 0.8947077035614261, + 0.666797945900111, + 0.451643545200134, + 0.468191591729566, + 0.5415183044890679, + 0.5196831327434093, + 0.47193391683646313, + 0.49702272655440266, + 0.5834692384408897, + 0.5069315896894907, + 0.6269624834724286, + 0.5141063160924345, + 0.6489370275282793, + 0.5093877864443546, + 0.6426890393322979, + 0.4202049768822087, + 0.7314596187712692, + 0.4839167276113077, + 0.7218384043786986, + 0.5145013909681726, + 0.7567556772770624, + 0.7241532595337775, + 0.3846248882384665, + 0.3944379856398529, + 0.4856158111685696, + 0.38631165868893425, + 0.368210481889109, + 0.7341665411895283, + 0.7053211116510096, + 0.4201710002109346, + 0.49276785351159536, + 0.6047161694204675, + 0.8503318420798577, + 0.5691988549544357, + 0.5722722723742557, + 0.5908686474779355, + 0.5198144330250134, + 0.45352771816880233, + 0.7863739933787826, + 0.5833748266706312, + 0.5061708988786409, + 0.5646474221004225, + 0.6584140356878277, + 0.4747890571688444, + 0.5443389785369277, + 0.5101541250322497, + 0.5661428416232669, + 0.5605932274683835, + 0.6275242986600316, + 0.5669558364391071, + 0.6566242711074699, + 0.6809723799429317, + 0.704723937014408, + 0.6433616892662339, + 0.47518638087030163, + 0.6233908359281181, + 0.6502910180840829, + 0.7978026141640544, + 0.4588431707657522, + 0.3438735659955104, + 0.6034773696041564, + 0.4932583488538993, + 0.6538695900029988, + 0.6182600690286856, + 0.6280982088214941, + 0.5730820301715067, + 0.4970308278655824, + 0.7854273845630243, + 0.6340511714497609, + 0.7644432491412988, + 0.4133474690415828, + 0.3910538923684975, + 0.415170002593296, + 0.37297941024403947, + 0.8965807073621167, + 0.34005203357237135, + 0.900090037478295, + 0.7603911279088874, + 0.4771754355706893, + 0.37250343867572705, + 0.37291361618127206, + 0.3848968283161913, + 0.66520394738729, + 0.7599925874029907, + 0.7549801224654099, + 0.6760763264532709, + 0.5767682664385445, + 0.4946026700888096, + 0.6919022009980826, + 0.6299958446389313, + 0.5309976083075589, + 0.651505278896538, + 0.4723735663052468, + 0.507147215484201, + 0.42460311831469383, + 0.8325547296973493, + 0.4806365921366095, + 0.4862174044790505, + 0.6672483369884561, + 0.7063666211300251, + 0.5329521570953641, + 0.3619459019314777, + 0.8763074024978021, + 0.3906545798843341, + 0.7802911004647496, + 0.5184354712464241, + 0.6105970025023487, + 0.7403990592422731, + 0.4666430094329833, + 0.5847009849381898, + 0.6203737193540062, + 0.8309271593903874, + 0.429396841274162, + 0.6249506374780656, + 0.5792041177695602, + 0.7195467934109102, + 0.8966748013511014, + 0.4562695688426807, + 0.6111251989499968, + 0.5084670968264298, + 0.6501090884502654, + 0.7302791324204373, + 0.5804547297533184, + 0.4744235818302358, + 0.8720882074541875, + 0.8833667382677585, + 0.7571215478916073, + 0.8396341320081994, + 0.8500822637961908, + 0.8404935419576599, + 0.8517257338753739, + 0.3813076435402446, + 0.4041433155676719, + 0.5290411966705768, + 0.48043030390315744, + 0.4731692984357428, + 0.620625567209176, + 0.8666459953906944, + 0.8482724168286166, + 0.6962797321980885, + 0.8315434352017471, + 0.8276822728353814, + 0.8595483791318517, + 0.9075298384702573, + 0.5571498700034838, + 0.6356091601246577, + 0.6991444854771176, + 0.6608894196452368, + 0.8277092076556103, + 0.6077076155109475, + 0.9132314551381516, + 0.8851482695698669, + 0.8741094893838239, + 0.9023040870825696, + 0.8931394474211563, + 0.8788131992974452, + 0.8355580493290296, + 0.8603573927870577, + 0.9092875450104858, + 0.47171797900823265, + 0.9268694074299577, + 0.8798120155447682, + 0.8226343112121844, + 0.6216228399352424, + 0.8725989967817597, + 0.5410672282480105, + 0.7514355132375067, + 0.8745943562239732, + 0.6175548275577429, + 0.6052496759984802, + 0.7112950130217683, + 0.5417855655635301, + 0.5706042087226287, + 0.7409033349966777, + 0.8892272600493341, + 0.7618210890908433, + 0.8210320777833582, + 0.6458726272703269, + 0.8688005221029532, + 0.8101440543929335, + 0.8408852865365983, + 0.9057149049387273, + 0.8321895882667251, + 0.9013524716518533, + 0.8707484931842149, + 0.5572659690910832, + 0.878568290994332, + 0.5977105396702076, + 0.5631084497615748, + 0.8825294466615927, + 0.9056452942429126, + 0.8832213657758655, + 0.6315564340959263, + 0.653014111754044, + 0.6769907413386368, + 0.7992753694739285, + 0.8958715052042194, + 0.7322212513923265, + 0.9102416138306269, + 0.9185656954383798, + 0.6308341607471885, + 0.35056156942996425, + 0.5402577950741193, + 0.898892182810213, + 0.4737613602354505, + 0.6394518423910713, + 0.8869551245363655, + 0.8306481247604695, + 0.6560707171878866, + 0.8654731067347967, + 0.8956524079665402, + 0.8888206166320182, + 0.8814769012123838, + 0.8661474969856039, + 0.8941746730097162, + 0.8605860061835907, + 0.929183597952342, + 0.7836743226594087, + 0.836527142480414, + 0.9090695831581825, + 0.582705505925942, + 0.5905023633792131, + 0.5884099877272921, + 0.5620256958381028, + 0.9073692768044219, + 0.8042400646230532, + 0.8494037215239183, + 0.800838234065083, + 0.7705066128639907 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.537602205514598, + "valid_tracks": 12, + "per_track_pearson": [ + 0.49956856398269756, + 0.5549903574876275, + 0.5208031854335569, + 0.5499281238780179, + 0.5477011494244037, + 0.5791221497541675, + 0.4839620283150018, + 0.5450268509646334, + 0.5112810576332448, + 0.5423821403827758, + 0.5238575505876982, + 0.5926033083313508 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.31483950388099885, + "valid_tracks": 546, + "per_track_pearson": [ + 0.314883942853255, + 0.3574597785416706, + 0.29655140648106076, + 0.19361490144280966, + 0.11221490964863626, + 0.3073412866491889, + 0.20891497777489954, + 0.2049712041365392, + 0.32182568999892586, + 0.3008022884756961, + 0.34324842665979927, + 0.20587229726225653, + 0.21511211289393878, + 0.1711988007927719, + 0.17562050107762486, + 0.22101079987858943, + 0.26514788500970776, + 0.3620266505411897, + 0.3561665373657401, + 0.3496743195347625, + 0.27778238946047207, + 0.2862202974580102, + 0.4145300603010499, + 0.23088576671676603, + 0.2744018387706224, + 0.3453568993814037, + 0.30715004057755907, + 0.337509106168625, + 0.35235745212914216, + 0.3546597072542836, + 0.2288137722492354, + 0.38337051215939233, + 0.23648979812949977, + 0.20788834328587116, + 0.3851878183474438, + 0.32411618186101915, + 0.3805997570974334, + 0.4098197288391829, + 0.36790765309410894, + 0.38141012054662077, + 0.4254830753445905, + 0.3579294924917208, + 0.33416714845201456, + 0.3025211897778543, + 0.21223137819837629, + 0.28293871618149835, + 0.2958245780687705, + 0.34840141584451284, + 0.19059426789173897, + 0.2419625453851997, + 0.36857136297976106, + 0.357806228216754, + 0.3632261183594962, + 0.26357703836681834, + 0.4172908154830116, + 0.41027605262658695, + 0.245790707451828, + 0.4213377123004623, + 0.2940539335356425, + 0.40801479894058146, + 0.3876741188987486, + 0.2603318991055353, + 0.33393618237609773, + 0.2749914379733729, + 0.34951901081230846, + 0.40823105027342677, + 0.3932424061850031, + 0.24970490528236106, + 0.19205089100758316, + 0.16259701647009342, + 0.3100650021875334, + 0.36009389335710845, + 0.3254087549330561, + 0.3109753929644097, + 0.3544931314909874, + 0.2923465689908376, + 0.3437998439468956, + 0.3297598229044323, + 0.3763687554207835, + 0.3671117755038327, + 0.3234177949290535, + 0.18840418999117095, + 0.3328868158498185, + 0.429630026235273, + 0.18932268071397182, + 0.33747987771855886, + 0.30375834036988236, + 0.41446938334777494, + 0.26890993842279176, + 0.3504902409542265, + 0.28032400544116776, + 0.32754990534760864, + 0.3362450338789871, + 0.1725622963783102, + 0.36047649539642185, + 0.29862918782184233, + 0.34855642042200724, + 0.34225599825194797, + 0.3191818868701706, + 0.20156669664005553, + 0.32609320971673156, + 0.3340393104448756, + 0.3617279644443226, + 0.4175930167053664, + 0.38483942250799197, + 0.36765404395835943, + 0.27079072560462253, + 0.3606167848405789, + 0.28460150183302363, + 0.2857783452419488, + 0.30104165874046707, + 0.3059283640446529, + 0.32580190952527993, + 0.32474865340776754, + 0.27195589813475457, + 0.35276661739869564, + 0.3239888492600901, + 0.33448696381819154, + 0.32266326696529507, + 0.35700786319801675, + 0.3651327648593374, + 0.3506770379615916, + 0.24823543780383237, + 0.28706945242839793, + 0.3060884159538794, + 0.3539106890876101, + 0.3298400577634003, + 0.35700205388005124, + 0.4260122963299559, + 0.37570632575302587, + 0.3727064393314507, + 0.2791438474378899, + 0.370956629167132, + 0.4130708865192664, + 0.42694713197221507, + 0.32304935088597614, + 0.3998767278069988, + 0.3260047093635751, + 0.18115921794050086, + 0.3704743314077373, + 0.3842568728422082, + 0.324156523065549, + 0.36932048159263614, + 0.29075101737201486, + 0.3131310424984259, + 0.32780707233849005, + 0.32604600677493445, + 0.4386171881651105, + 0.308713028312876, + 0.31029990927971324, + 0.391755578974162, + 0.3040513137387174, + 0.35119471659715473, + 0.34973221556609124, + 0.33350065560271813, + 0.3177305257855592, + 0.40751539055344654, + 0.43322603464311643, + 0.2951958175659252, + 0.2851424337197586, + 0.23888854896631587, + 0.3052002057521907, + 0.31236751746111646, + 0.35281469752111855, + 0.31310955442308497, + 0.2836332457074222, + 0.32432186682651293, + 0.25586989352698, + 0.20211170019030744, + 0.3758361417600323, + 0.33098115902609515, + 0.2759269933760191, + 0.3317738559990059, + 0.4287067186286367, + 0.3382823648910813, + 0.20151721709608952, + 0.2251020122546327, + 0.3041446767707802, + 0.3754110336724799, + 0.38256161437549185, + 0.37510604491027055, + 0.3851309327860221, + 0.24248297718403908, + 0.37219478757051383, + 0.3260226412679222, + 0.2659606859769473, + 0.3208544543528883, + 0.2309323097883269, + 0.42144195568042325, + 0.3482992121824413, + 0.29128322184112954, + 0.2247803257063739, + 0.22011658070087448, + 0.31320345084657236, + 0.33534414690894687, + 0.2878124622915654, + 0.3292588543572118, + 0.34448344313430623, + 0.2526049212030386, + 0.32933675089146647, + 0.17066812171113221, + 0.16793986464890237, + 0.15711584252194263, + 0.1464897433849187, + 0.20506553536739436, + 0.31930780497124694, + 0.3162278095702408, + 0.27432604532450844, + 0.1503183341222737, + 0.2959576773435498, + 0.37323097475802747, + 0.3698184018986717, + 0.38519696647212814, + 0.34914605981123, + 0.33932657172447783, + 0.28103949822396435, + 0.3778294922524143, + 0.3852723871187372, + 0.3471068801010547, + 0.3863230244376616, + 0.34782473242649325, + 0.27821719697985636, + 0.35246419011479346, + 0.3560577272862702, + 0.4160214240283038, + 0.3949271086024254, + 0.33844989951808546, + 0.3356338322167101, + 0.3683856468767912, + 0.41287084563204535, + 0.26216000801546, + 0.2674830537737139, + 0.4064406373748661, + 0.32091571692613574, + 0.2910427101560342, + 0.310925784943549, + 0.41604367724693764, + 0.2720803408371827, + 0.3633261474724241, + 0.3531728550353978, + 0.16104350885468102, + 0.268157491759451, + 0.21417961376441141, + 0.3577012766215229, + 0.309292666634858, + 0.34804068317947795, + 0.1499419377514399, + 0.38688531606620585, + 0.337303080276515, + 0.3789894828221014, + 0.3648595605327656, + 0.255303476417134, + 0.42045826486428456, + 0.186556403067688, + 0.3669730155387181, + 0.2753829325214132, + 0.42024782279623185, + 0.37285192132408734, + 0.3911841950068546, + 0.44186657346817426, + 0.3092368685791271, + 0.3505299477088734, + 0.16294496380897916, + 0.35944294553524087, + 0.3839103220420733, + 0.26799439192845065, + 0.16907131877075862, + 0.17424225674842225, + 0.2013924871095917, + 0.32317381189813094, + 0.3929416601556063, + 0.1969518736830683, + 0.15569338664666987, + 0.308150654112841, + 0.3636799201117955, + 0.3129849352822311, + 0.2176041935711126, + 0.10276128833876923, + 0.32895939332378366, + 0.22103093871567772, + 0.24021146817073466, + 0.3278300887302006, + 0.28996561108093716, + 0.34742358579150623, + 0.2058659903761391, + 0.22372526682400873, + 0.20053857221544483, + 0.1964119254427816, + 0.21289963321899674, + 0.26810949353473285, + 0.3504846343563122, + 0.34807677693470085, + 0.33427332471008236, + 0.2728470484943318, + 0.2785948900004669, + 0.40277570729642015, + 0.22128474706309384, + 0.26193952334973597, + 0.36366343535055, + 0.29905303614997636, + 0.3258132604173428, + 0.34754466151999547, + 0.34389328519868206, + 0.21641795563352423, + 0.3690728252551157, + 0.22846754231971594, + 0.207161181943367, + 0.3662940856506068, + 0.33704941387414517, + 0.3955257631520372, + 0.4057381369738064, + 0.4276472631385523, + 0.37739630761866644, + 0.4177813678616393, + 0.3405978114946491, + 0.3313881506521038, + 0.320422614074221, + 0.2312813263206817, + 0.2989560285759007, + 0.33170319825745437, + 0.38416767522923323, + 0.18846509604916778, + 0.24636294375491535, + 0.3371431856639354, + 0.3603350645743943, + 0.34543491642318197, + 0.25815783877915444, + 0.4099749209708685, + 0.39168468282853897, + 0.256723972252687, + 0.41006531649139966, + 0.28200660992653104, + 0.4090788615627199, + 0.3713156350503067, + 0.2529051005027423, + 0.3325587257287504, + 0.28400494932316916, + 0.3597840908018543, + 0.3885678672831931, + 0.374993450568595, + 0.24062250604659222, + 0.20025585311060512, + 0.16490270221991887, + 0.308811057665301, + 0.3616743910134391, + 0.32389878570411645, + 0.3034007368182481, + 0.3562678647986401, + 0.2862755683250544, + 0.32240282028574846, + 0.310607271578734, + 0.3642465788556673, + 0.3644061107992175, + 0.3219109298266366, + 0.19628516319388312, + 0.3376902401209043, + 0.4275078107604676, + 0.19818582905605484, + 0.34825263264751377, + 0.3476363799047916, + 0.41810874749823157, + 0.25258017020044465, + 0.3675306714331405, + 0.27769933532784324, + 0.3283403430013076, + 0.3411220527622403, + 0.17916136484926926, + 0.3465037955876911, + 0.2858117360464224, + 0.3276096344964009, + 0.35417073029017315, + 0.3072108806789103, + 0.1987263549090798, + 0.3262954422583578, + 0.33540097851819944, + 0.3586066700769768, + 0.40497366088043313, + 0.36776950227724303, + 0.3713489536734118, + 0.262741528263197, + 0.33704969287385556, + 0.2756632279153415, + 0.28098509222216433, + 0.28906548485394296, + 0.2957784099550987, + 0.32666479274736465, + 0.33234695499325745, + 0.2641567515678078, + 0.3426457304288331, + 0.3059959957862587, + 0.32905767547899073, + 0.31114894036044133, + 0.3531766423932931, + 0.35007130447222934, + 0.3432910399864894, + 0.2500259619558779, + 0.29496121872167697, + 0.314365901522803, + 0.3616692398213671, + 0.3199427680713452, + 0.36648034290874004, + 0.42166492452070536, + 0.4005265231266731, + 0.35702411092511066, + 0.26560694836707016, + 0.3379304755793301, + 0.385200114650369, + 0.41502550400803745, + 0.3206504605112225, + 0.4000326334676453, + 0.31975524347877954, + 0.20423439253734796, + 0.3714628147028475, + 0.35944581843896073, + 0.3126371132329464, + 0.3796697367092844, + 0.277135867093328, + 0.308912043539055, + 0.3220639821684084, + 0.3189275490850696, + 0.4237307429179317, + 0.3501238732756484, + 0.3002597835486602, + 0.39456570595126395, + 0.3008979449785905, + 0.3443015479674404, + 0.3491912264634218, + 0.3242093837895421, + 0.3309823846704981, + 0.3903614474114009, + 0.426903587271014, + 0.29279075680891253, + 0.3060634912227379, + 0.24724724527000527, + 0.2804893413529555, + 0.31666941789334585, + 0.36780485293623333, + 0.3418791189243056, + 0.2687264753514999, + 0.3190024517780385, + 0.2820456979794577, + 0.1955626291171106, + 0.4018488325853932, + 0.3302853399136496, + 0.2893799916924838, + 0.31299387333543, + 0.424854375302142, + 0.32756118693365593, + 0.19064729833993746, + 0.23411455895713823, + 0.31219120651168936, + 0.3279825323952001, + 0.38881154526819817, + 0.37680575849049225, + 0.38807836961389874, + 0.22749183020661834, + 0.37206432196554706, + 0.33926889693772516, + 0.29730660706706413, + 0.31934780872989094, + 0.2443881718941929, + 0.4086752898707333, + 0.3431264698169373, + 0.29855612978655083, + 0.2499996536852493, + 0.1966513494163297, + 0.25106644670382594, + 0.35818873328277323, + 0.3031325524150826, + 0.3415585264352286, + 0.3464827213582034, + 0.2661755938188634, + 0.3101359691109536, + 0.15375652478001045, + 0.13271172644965326, + 0.13084384449892897, + 0.1264153816626775, + 0.1645736727673134, + 0.3295131758100027, + 0.27764993056443993, + 0.26775129289214933, + 0.14810957488526527, + 0.30864736124354997, + 0.38827869224649675, + 0.41334857111134804, + 0.42022187484734536, + 0.37888422386765014, + 0.3711109772643694, + 0.29552814698730506, + 0.39793708135934025, + 0.39377981640665327, + 0.3451041439108233, + 0.38930224896700116, + 0.3618318655807928, + 0.3155558926440856, + 0.38037253716319025, + 0.3469899474340753, + 0.4563967806214008, + 0.39975646225908407, + 0.3680149613489474, + 0.3385269464637381, + 0.33927803441443055, + 0.3758140545099815, + 0.20303563024017043, + 0.2024231915643145, + 0.4192508031765572, + 0.32763242100834505, + 0.30173931179704466, + 0.31176824408852305, + 0.42095575523564926, + 0.2944291268505786, + 0.34872180380852635, + 0.3379279334287486, + 0.15404844366372203, + 0.23897028222012856, + 0.20446587295301044, + 0.3715678648824158, + 0.2889929979912051, + 0.3500946087923454, + 0.15212655570747596, + 0.3833525617373257, + 0.3490785810246621, + 0.37857372417551216, + 0.3773818925272645, + 0.2692095049728772, + 0.4595095080863562, + 0.18460783303182096, + 0.3582261955682806, + 0.2657621474272467, + 0.41016687720865913, + 0.39062012809208224, + 0.4237814476280261, + 0.4361519298811735, + 0.32952222565569766, + 0.3380209590959786, + 0.1532565012963366, + 0.34166497758417563, + 0.381221450817397, + 0.2970721206427658, + 0.15983194638645815, + 0.15863405582436543, + 0.19468160443423088, + 0.33423340874826857, + 0.3912233542357626, + 0.1842461953308905, + 0.1383951521744973 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.6029183031115146, + "valid_tracks": 613, + "per_track_pearson": [ + 0.568754259247188, + 0.677078811621102, + 0.6113292562171921, + 0.6489202468453291, + 0.6607054086684558, + 0.41792664377625804, + 0.545445137618712, + 0.6554778209408756, + 0.636589618079961, + 0.6677688672054256, + 0.6214899704638243, + 0.5793668355063529, + 0.509076934406272, + 0.6386738178592327, + 0.5967091867905121, + 0.6166200883762216, + 0.6826505986984907, + 0.6183811565848444, + 0.6376907479958195, + 0.5853973710581953, + 0.5606863297719175, + 0.6674403000366943, + 0.5690492930964501, + 0.5350842338876821, + 0.47597145933957646, + 0.6769347411477098, + 0.585161569355847, + 0.626711112725884, + 0.6073331641404462, + 0.6219638113371162, + 0.5923057544289126, + 0.6231965495485814, + 0.6252866301298594, + 0.5577036831549665, + 0.6401637752459022, + 0.5458942706763082, + 0.6415405618429331, + 0.6435358584718224, + 0.589363241455633, + 0.6313175613393682, + 0.6725235733026323, + 0.5915618235682971, + 0.6519065505727488, + 0.6703071062006523, + 0.5807279636018763, + 0.4905999001074339, + 0.5652389040787549, + 0.6509630092960678, + 0.60431339751872, + 0.6651724665353435, + 0.6534719031058749, + 0.6584108701409168, + 0.49955955435768307, + 0.5878430730454033, + 0.462238250414108, + 0.6048755438815305, + 0.5781317649987, + 0.6626851358832321, + 0.6955085044431942, + 0.7262687848334511, + 0.7311309833143198, + 0.6509260147165824, + 0.6826361175840268, + 0.525780100286843, + 0.6461361303769243, + 0.6478518003979801, + 0.676066904558718, + 0.6849847763236808, + 0.6599255955283327, + 0.7099371697130933, + 0.6625457384429283, + 0.6722558111077107, + 0.6181181501185937, + 0.6471629890400367, + 0.6343221203741998, + 0.6316729918659071, + 0.6614531359705044, + 0.6788178253634771, + 0.6051370335055066, + 0.7013082866684593, + 0.6511925823632829, + 0.5484178357952283, + 0.7054180544401359, + 0.6874368575405, + 0.5913706997095809, + 0.6424704776029809, + 0.6927862914648717, + 0.5770917522530332, + 0.6580497004345579, + 0.5873896597179861, + 0.5348540178260079, + 0.6925766040319665, + 0.6945390076721196, + 0.6709099009191267, + 0.6675976089744402, + 0.6922988278240364, + 0.7070685202794741, + 0.676762336024574, + 0.6128294832171801, + 0.5324448575916961, + 0.5006416721598055, + 0.5079026557685468, + 0.5668521676040315, + 0.664849892825257, + 0.550213652691601, + 0.5715077105819906, + 0.6834091634789548, + 0.6203560667327467, + 0.6633740909611491, + 0.5934203264419561, + 0.6417674935971901, + 0.6206467467364735, + 0.6002719882584368, + 0.610642188111776, + 0.6601241715997636, + 0.6466115700423912, + 0.6540359508958938, + 0.6192293816354987, + 0.5785128921133157, + 0.6333880999152228, + 0.6617274101071755, + 0.5877070579851326, + 0.562115385572779, + 0.6110783815541433, + 0.6871486925634832, + 0.47291992409996875, + 0.5064115088276205, + 0.4914671134675041, + 0.5026170661032262, + 0.5631375977509158, + 0.6671474503841947, + 0.5604505014909289, + 0.6163793862887443, + 0.5995224506040986, + 0.613719690821142, + 0.6739833072252525, + 0.6351920424423088, + 0.6651007030064703, + 0.5834833877023772, + 0.6445944541435644, + 0.5608979586362108, + 0.6007277540544115, + 0.6662076278492898, + 0.589461302499141, + 0.5093914033338341, + 0.5497009813927829, + 0.5611539413226981, + 0.5782257931687199, + 0.536988503784285, + 0.5910146837119981, + 0.5616866284933152, + 0.5735139521600187, + 0.6264197347244042, + 0.5133245044746606, + 0.5680075528803851, + 0.5448464438683011, + 0.6052482958904933, + 0.5713102981456443, + 0.5972546605437206, + 0.6304720233963814, + 0.6518977041352894, + 0.6318124856754543, + 0.6333127478111064, + 0.5266649740501146, + 0.5239619189658322, + 0.5280223630650384, + 0.530539253891323, + 0.5465025348060732, + 0.5932022531555221, + 0.590064543376732, + 0.688030293691873, + 0.5984506881089563, + 0.6989335524895223, + 0.5945805379970837, + 0.5382713510616212, + 0.5792996138445197, + 0.5407548994230029, + 0.6529506668240334, + 0.4843330428248157, + 0.4978684759419599, + 0.4740522893973261, + 0.652525518197582, + 0.5111880518127734, + 0.6563588026125062, + 0.46208030468687983, + 0.5744486649129391, + 0.6310521732072917, + 0.43063080626284755, + 0.5649841083784226, + 0.5541102061410214, + 0.650098524560973, + 0.6179799317537032, + 0.5984789074881868, + 0.5447798158732153, + 0.6338979262921974, + 0.5878738894478219, + 0.6085148580910521, + 0.35967028546188845, + 0.6146891098505345, + 0.6049399804452704, + 0.5494973619983348, + 0.6773285721803477, + 0.6459260497816961, + 0.6390349070755166, + 0.6471277727714371, + 0.6493028221091557, + 0.6331791296091048, + 0.3482731656121814, + 0.6354846992869101, + 0.6168501313260266, + 0.5823289456489614, + 0.5594482409608739, + 0.635135096674115, + 0.6416288616234259, + 0.5539681094308105, + 0.6518782937582965, + 0.6127908194916546, + 0.5419131235204983, + 0.6712285924275581, + 0.6850323499797242, + 0.6492445045856187, + 0.7300516873019397, + 0.6077509160213291, + 0.6159711954083054, + 0.6081974562714294, + 0.5937996943274841, + 0.6039220392198839, + 0.6006956835345106, + 0.4974174051253134, + 0.7042679381439739, + 0.5771561020683921, + 0.6231289207635553, + 0.6732633285666098, + 0.476120374493943, + 0.7040084758676506, + 0.4524005116006979, + 0.4334937264576633, + 0.6560383000507909, + 0.6740160734641797, + 0.6440548728717548, + 0.6558004005705002, + 0.6948115864319078, + 0.649966514834251, + 0.7023709205840177, + 0.3281084951806218, + 0.5882060403944818, + 0.5955707126856938, + 0.6476128690128277, + 0.577370675019664, + 0.5822782377164987, + 0.5425499017171637, + 0.590568417504092, + 0.5751107286650261, + 0.5782211777934022, + 0.6406213860993217, + 0.6127208739160562, + 0.52281529909293, + 0.6731864967205106, + 0.6027237804509291, + 0.6613693180673926, + 0.5303177560293046, + 0.5555150229888133, + 0.6980299980806938, + 0.6975331985473467, + 0.6320924436678599, + 0.6667438353134902, + 0.7089772878322068, + 0.5252215029479138, + 0.6418461373787161, + 0.6357370257578919, + 0.4573444827876454, + 0.5774483384514989, + 0.7223282491664935, + 0.6971014902041864, + 0.6969027788750247, + 0.68427682822047, + 0.4050654491252037, + 0.5221296191212405, + 0.6823506909653331, + 0.6519925582424438, + 0.6778883112673147, + 0.6089676116863008, + 0.5505649187166091, + 0.5417781573077118, + 0.6215219465512961, + 0.5852903599517011, + 0.5984099579816397, + 0.6738650598909489, + 0.656481045763302, + 0.6977339727503058, + 0.5803354469921574, + 0.5737609048509041, + 0.7050735002229979, + 0.5747047850963457, + 0.499359230026282, + 0.4719581426738879, + 0.6755535382772451, + 0.6995569266045176, + 0.6970866359142496, + 0.6279839913574717, + 0.6373323704715947, + 0.5767136399104063, + 0.6468259455235293, + 0.5498199260562355, + 0.6181351594167657, + 0.6774169278233443, + 0.6269150107179525, + 0.6918003989861952, + 0.7027923228295332, + 0.6497196842966446, + 0.6724442846772425, + 0.6717020531751573, + 0.6294831861245572, + 0.6689464571880519, + 0.6294261233079336, + 0.5451323174524318, + 0.5608421201730378, + 0.5979156681214277, + 0.6742149010789348, + 0.6012150454346847, + 0.6626556731707797, + 0.6862260492876482, + 0.6782252933434938, + 0.473001879810385, + 0.5628391526246096, + 0.4198512321709191, + 0.6086777107793684, + 0.595824774373694, + 0.659897579522525, + 0.7101266028552479, + 0.6846927502822416, + 0.7145892169592825, + 0.701891112670114, + 0.7256513826107438, + 0.5301133585888888, + 0.6967703418566392, + 0.7127244161619785, + 0.7399224672828648, + 0.7455451888038788, + 0.7058324044941864, + 0.7438792142898879, + 0.6746095771089823, + 0.706504639388157, + 0.6308760679112565, + 0.728741110800434, + 0.695364025172734, + 0.6582183846562979, + 0.70008797936074, + 0.739510002675538, + 0.5618769473915912, + 0.6877549952365014, + 0.6427046498174023, + 0.5700523864076773, + 0.7242880383299171, + 0.7131763561853762, + 0.6439022827921549, + 0.647383305150878, + 0.6548946464038854, + 0.5714570091041348, + 0.6695630080813139, + 0.5820706001684443, + 0.5743880507872867, + 0.7012594500632491, + 0.6509301668570578, + 0.640207963332901, + 0.6132524929182236, + 0.6614029426269122, + 0.6745084231361728, + 0.6376577523698889, + 0.5834269668082084, + 0.5524144686228134, + 0.47080719296632806, + 0.4643447225077201, + 0.5097475452785771, + 0.7332774887388291, + 0.5238420244082869, + 0.5483982775455981, + 0.7120608090629696, + 0.6012478021247666, + 0.667827941845961, + 0.579422418674119, + 0.6410675417560218, + 0.6019617156111805, + 0.5774621353534554, + 0.6687621834910056, + 0.7321577889610736, + 0.6530437212012464, + 0.6613979930523474, + 0.599344369930636, + 0.5595691469839655, + 0.6129488603403735, + 0.7148480903406601, + 0.5754582866704983, + 0.5697860513417313, + 0.5998181963512034, + 0.6987808577019137, + 0.4481217597628471, + 0.47994255294271326, + 0.46538325828278815, + 0.4655732792715058, + 0.5958827842428706, + 0.6915045075546126, + 0.5460802384687224, + 0.605392813511821, + 0.6924501545128028, + 0.6419273166224571, + 0.7006896113589789, + 0.6247206741389886, + 0.6948801885453417, + 0.5837311709837582, + 0.6494639677919418, + 0.5335370793474183, + 0.6098524386609333, + 0.671422126883414, + 0.5887039747000687, + 0.4983772371990239, + 0.5402011517791357, + 0.6011834146765133, + 0.6357562376942856, + 0.6571209397015343, + 0.5495193372754328, + 0.5458371953233614, + 0.5967639093650005, + 0.6735648790997872, + 0.4944686944839029, + 0.545793062300462, + 0.5739780627203961, + 0.6737010312695116, + 0.5484046657657319, + 0.6692822325875143, + 0.6256727003971508, + 0.6575848376501395, + 0.6415479020498791, + 0.6513674641884413, + 0.42027195974823023, + 0.4264718815774773, + 0.4258357115161457, + 0.42472216441752875, + 0.5349384030798501, + 0.6056999031023218, + 0.5970996355109801, + 0.7354461120323528, + 0.6224088357032781, + 0.7168525719854857, + 0.5878633926936978, + 0.5740444965228103, + 0.5730041885170621, + 0.5493923392622715, + 0.6863711751916464, + 0.43972915343184865, + 0.522838182405318, + 0.45091981125289, + 0.6801218660635793, + 0.48232793804604907, + 0.5806660154708527, + 0.4733148160293422, + 0.5854749657603819, + 0.6521516283091308, + 0.4633708014817761, + 0.5719864985928004, + 0.509707842343715, + 0.6254457418339557, + 0.6613647341384477, + 0.5607198109379682, + 0.5306582116749338, + 0.6459417404600449, + 0.6404252067490527, + 0.635017572653451, + 0.33883264546572006, + 0.6654997773924387, + 0.6576467166583587, + 0.5283193018943055, + 0.730380090913769, + 0.6368242012860922, + 0.6620904472170821, + 0.667811245465721, + 0.650316055328585, + 0.6152281147978128, + 0.3008560070160964, + 0.6761876378224924, + 0.5759544338633967, + 0.6777822366114296, + 0.6557400276283251, + 0.7296151451643174, + 0.7123269142581186, + 0.5931174477272457, + 0.6890764618228842, + 0.7032488523839527, + 0.5971588246719138, + 0.7459181519391032, + 0.654514396884353, + 0.5948199958916217, + 0.703183911487142, + 0.557004543519507, + 0.5342042800039567, + 0.5331946278332059, + 0.5177238062658823, + 0.48344939046195323, + 0.5344854917839958, + 0.4059759999248811, + 0.7154461478642131, + 0.4809515167861678, + 0.6063953627910688, + 0.7153405766215691, + 0.4634741508433299, + 0.7220791339257915, + 0.420568963390869, + 0.438674465585029, + 0.6660562670955275, + 0.6463299339367345, + 0.5954898315501514, + 0.6080855828902811, + 0.7113134743132539, + 0.6933784042015061, + 0.7450718676954823, + 0.347138641026362, + 0.602060859682901, + 0.5979040279941878, + 0.6580078598023082, + 0.6175383656573821, + 0.5792724879703246, + 0.5742301993370589, + 0.5794291158572473, + 0.5931948530622903, + 0.5907391914026395, + 0.6589592621724036, + 0.5689171523220391, + 0.4475961579344578, + 0.6752839512270338, + 0.6262732650947258, + 0.7210168506036218, + 0.474453075758897, + 0.44753464971830786, + 0.6501232365930052, + 0.6503927636720847, + 0.6367118486855883, + 0.6829238786184142, + 0.7364430780854287, + 0.46583107884916436, + 0.6606025409636525, + 0.634332279393319, + 0.4518041053985736, + 0.6320306104387631, + 0.3591733142075935, + 0.5221337183162675, + 0.5817049906930797, + 0.517785204339316, + 0.5935079992160118, + 0.5298476162602965, + 0.6326025110813495, + 0.6471353420185034, + 0.5213499805445906, + 0.5101360970544303, + 0.5109921178016807, + 0.4931335807609445, + 0.49343830084077356, + 0.4917929974175738, + 0.4989240654186452, + 0.4915122561975449, + 0.47334287039911377, + 0.4347423816552656, + 0.6175322494760362, + 0.5491081963834109, + 0.560159323949169, + 0.5525813854150012, + 0.5651315808458828, + 0.5622026640344093, + null, + 0.5909522649294551, + 0.6200673821911817, + null, + 0.4804437552672554, + 0.6221812069492173, + 0.49271612668424863, + 0.6294734634733641, + 0.6540365733155102, + 0.6306914390861628, + 0.6180385362431181, + 0.6203539730397544, + 0.6207500757243709, + 0.5915938444539692, + 0.5562797675975739, + 0.5536048361073271, + 0.5069758503522386, + 0.5946760564073862, + 0.4927174560552426, + 0.5894434437593409, + 0.5632991799122862, + null, + 0.5936798337645488, + null, + null, + 0.5721797592046596, + null, + 0.6068469538254513, + 0.5533896361862737, + null, + 0.6148253050985643, + null, + null, + null, + null, + null, + null, + null, + 0.5974342001078193, + null, + null, + null, + null, + 0.5013555527716975, + null, + 0.5823592225017166, + null, + 0.5887024982203317, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6122982070027821, + 0.4687244028986452, + 0.5749262948309055, + 0.6019267823961446, + 0.5273093247692625, + 0.630718658795814, + null, + 0.6164208489304739, + null, + 0.586463566489332, + null, + null, + 0.614582984726239, + 0.6381867488925752, + 0.548273635293977, + 0.5331688082138606, + null, + null, + 0.5701882890889568, + null, + null, + 0.5931227101889399, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.5972161266468886, + 0.5825899974322387, + 0.6046895001111419, + 0.5624241506599805, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.45717487152706243, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.24251860133638242, + 0.36544055247311297, + 0.3191418364215735, + 0.4788718791440506, + 0.4152375858621847, + 0.30583227913362304, + 0.40428811384257934, + 0.4742558948932971, + 0.38772664801857043, + 0.41068331965266897, + 0.4082305041179278, + 0.4182536005603557, + 0.2999757233513651, + 0.4756474864154499, + 0.4025023999195654, + 0.40766932085078966, + 0.45922906108608896, + 0.37251424062003663, + 0.5105240305954675, + 0.40102492846486637, + 0.39814684954810753, + 0.3535321315736433, + 0.3304654215410757, + 0.533208884222347, + 0.37705314649295407, + 0.42011464190006625, + 0.3362128224883118, + 0.3860140300550579, + 0.41743934640820946, + 0.4414707268237168, + 0.4390859634437572, + 0.4267540804625911, + 0.5004708575978378, + 0.2959146352781493, + 0.669318685968225, + 0.2922462740564804, + 0.25224678398048633, + 0.18844822364413166, + 0.3405564004240537, + 0.2358021112953651, + 0.288990835167429, + 0.4721102575265178, + 0.31805492705189237, + 0.4169220353007937, + 0.305701229705803, + 0.4036765977274136, + 0.3090323283555701, + 0.23439998787499006, + 0.4785294000019893, + 0.4492694151057396, + 0.4000793940487039, + 0.3749271924480973, + 0.3934782140482199, + 0.4331336858771111, + 0.43202985361591323, + 0.42524133398945374, + 0.3449326509470608, + 0.4917582388793029, + 0.5116748036375793, + 0.5135442397030837, + 0.32228347641757177, + 0.21627016552914374, + 0.4050516932646238, + 0.43204616810557805, + 0.2817837573083324, + 0.3595470816251773, + 0.3599174648360551, + 0.4813605074569348, + 0.5005236509379218, + 0.4488425560277219, + 0.3763700994035243, + 0.7123442130830836, + 0.3670242218448222, + 0.3887092443786985, + 0.38994709001259237, + 0.442723529822262, + 0.31866035467648823, + 0.6558377170913718, + 0.42273679763262545, + 0.4411181137730299, + 0.411176789626087, + 0.4258749048736661, + 0.49089511781650325, + 0.45114302388873745, + 0.3692455029476125, + 0.32800019800084923, + 0.4611537622816081, + 0.4632123635476625, + 0.8058503921850934, + 0.49385042932493617, + 0.3968666889374938, + 0.35724374494553923, + 0.45721545378458417, + 0.38875980367582796, + 0.43140528713074616, + 0.55007370500474, + 0.35350547553862544, + 0.3869912011704778, + 0.3953759762951401, + 0.33707572202882297, + 0.26030213272936215, + 0.38526617087353776, + 0.3437438046088764, + 0.3304667523486293, + 0.37696409374895495, + 0.2542884590243428, + 0.3194038703581809, + 0.3368436333119266, + 0.3861385643281596, + 0.40900700390046246, + 0.2607771930845627, + 0.3623604365145796, + 0.340437130720715, + 0.3811654152981786, + 0.41042189335770596, + 0.45082880438102196, + 0.41047904218646003, + 0.4128822387539385, + 0.3864415848387769, + 0.44638354306162714, + 0.3878573227893656, + 0.4326297800354992, + 0.40935753944762476, + 0.33287450573471605, + 0.3959686199074171, + 0.35263608097102367, + 0.3827603455359107, + 0.35684247450102624, + 0.38288735663090745, + 0.5412643439894624, + 0.4276043998625644, + 0.36952477101115927, + 0.42758337571056254, + 0.633040657785606, + 0.35671734952929346, + 0.4094702605483887, + 0.40994653710128726, + 0.3276136747677144, + 0.36969741907428216, + 0.35667345667129674, + 0.38722018364810123, + 0.44649497064633925, + 0.5576734909437483, + 0.3704210290380832, + 0.427901082111089, + 0.5004953330543445, + 0.4497712870090018, + 0.3108253469183358, + 0.4398966820539978, + 0.46295085438756334, + 0.43876331092142656, + 0.3674437629889843, + 0.4562730988869961, + 0.49446934305956763, + 0.4275978239915705, + 0.43152915312168005, + 0.34056582175245453, + 0.3854007085229983, + 0.39763140810803144, + 0.4474788807595586, + 0.4383606612178442, + 0.38420966606273355, + 0.4105826040879398, + 0.4234720457671208, + 0.43081084206950415, + 0.38996238044408504, + 0.379924921545652, + 0.476575725808072, + 0.32385342851542476, + 0.43325735666717924, + 0.4101026381027716, + 0.3682351525887411, + 0.5001543086726912, + 0.3760471239445375, + 0.3458028704156237, + 0.4358555706179126, + 0.3058838882598434, + 0.45209326511567405, + 0.4592493538184406, + 0.35167587500943454, + 0.44139262141044666, + 0.42074340688764295, + 0.3503288965825701, + 0.42068861773436117, + 0.3650919197916563, + 0.43517453331842765, + 0.45369156004352357, + 0.3273735261162026, + 0.35616867965391, + 0.42646413195061966, + 0.35305774815465335, + 0.37619406050817616, + 0.4475460892011909, + 0.4890919833176112, + 0.36470070815035416, + 0.36510879781069816, + 0.37373246373361535, + 0.372289941643477, + 0.29484020227858476, + 0.4804366932456555, + 0.386367450167, + 0.35545917585038694, + 0.4035906570802472, + 0.3514617796302879, + 0.46530350689724864, + 0.3879023019042821, + 0.46078991573642164, + 0.3956356410683261, + 0.3441728280440024, + 0.5168781335825459, + 0.38095493427668176, + 0.4430164580646821, + 0.39556279170542324, + 0.42005759773842694, + 0.8479903438343673, + 0.4585978469408909, + 0.4129954239142079, + 0.45255706811776664, + 0.4719799530619911, + 0.554334607322375, + 0.6182613085691702, + 0.6305848940598018, + 0.441014431043402, + 0.4032850849066544, + 0.4969320738629059, + 0.7348321845651551, + 0.372370088499036, + 0.3961924780326298, + 0.5533595554440985, + 0.5271841450975902, + 0.5804802280075112, + 0.4652567248842482, + 0.3385728023482545, + 0.41210790149059995, + 0.6040502343018668, + 0.4583744728018787, + 0.355483746959399, + 0.6176324885545941, + 0.5590275742921745, + 0.46155800201128405, + 0.5027591575300382, + 0.712947600746771, + 0.4896101681842501, + 0.5163159284413223, + 0.4509459713803348, + 0.764012775029202, + 0.3454396933672012, + 0.6436279763371439, + 0.41457254668261334, + 0.5546177356710017, + 0.4491725279574062, + 0.3956561390971248, + 0.7032914498670483, + 0.5214411030573305, + 0.35165025444215187, + 0.3843200132237559, + 0.6494909818984617, + 0.39803582527356507, + 0.450152823263501, + 0.494160868148276, + 0.3871093260829568, + 0.49361212381443664, + 0.49580800375457806, + 0.4362335781721035, + 0.39838322086674677, + 0.4810368283474499, + 0.37379400580257444, + 0.35666828997034317, + 0.6345224715758825, + 0.39310344548494086, + 0.461413603763326, + 0.5730172775139106, + 0.4029957168169718, + 0.6631706023180536, + 0.367068272640304, + 0.3730878444088476, + 0.563576208103694, + 0.5015467055068284, + 0.5865943560097885, + 0.7305300494710667, + 0.6727363544463001, + 0.42525027365541435, + 0.4772072116395716, + 0.5672345373214442, + 0.37010149878763293, + 0.35883383229772686, + 0.5708001770533624, + 0.531549820938563, + 0.7329456954602905, + 0.4951610077439349, + 0.3838517786168919, + 0.4694073954218, + 0.38449042109257375, + 0.48107744940774655, + 0.7070320754461014, + 0.6312641200944016, + 0.538501418038067, + 0.5397064247242399, + 0.5016232306191278, + 0.5205767058444047, + 0.7409677330973033, + 0.3226708176177118, + 0.4564976264205292, + 0.3230739248022568, + 0.6122804325751043, + 0.39504102460170365, + 0.7377270574773441, + 0.4333059260016945, + 0.44152624742233754, + 0.6982569431318184, + 0.5238812856609222, + 0.3469903004783464, + 0.5724525442531565, + 0.4172069358333311, + 0.5207418481681214, + 0.5040114649050439, + 0.35934291407337665, + 0.5102859358967246, + 0.4098248080331433, + 0.46458618674905805, + 0.41729621039109094, + 0.3684392855031779, + 0.4282096028622037, + 0.5332703555686161, + 0.5925575614487134, + 0.669240469525433, + 0.4374339018025271, + 0.3234143564156886, + 0.4685835839285504, + 0.4761650696764921, + 0.43941677738468776, + 0.489280991727108, + 0.4931915429398678, + 0.39630592690006283, + 0.34795201156964245, + 0.5202759364211776, + 0.7650377431810224, + 0.5338981992438648, + 0.6087104970371031, + 0.596654411257467, + 0.7679858133425511, + 0.5688668860587917, + 0.3693558980464137, + 0.5915172909837946, + 0.39866979895133103, + 0.5426015012099087, + 0.48484325574096115, + 0.4040190192984992, + 0.5070710351696758, + 0.7822982041161398, + 0.2548345208323866, + 0.4761673701326468, + 0.4791108360212029, + 0.6282131533083263, + 0.38926569120472304, + 0.3373038746886946, + 0.4440806702830713, + 0.7239870948182495, + 0.40348109543158783, + 0.45394579364148285, + 0.5144302140385281, + 0.4666449480460933, + 0.4402868440611094, + 0.5080138253953422, + 0.6186760616208172, + 0.4727907008973608, + 0.42294396748545643, + 0.5554347842202604, + 0.46417774655227323, + 0.5777428249262578, + 0.5512151234400491, + 0.534043346935634, + 0.555028722238445, + 0.35160238939512867, + 0.37347714457452685, + 0.5627510351115169, + 0.3361870926075204, + 0.6991139895268441, + 0.33562158650395996, + 0.5695685289524871, + 0.5884768635481276, + 0.6886684793779511, + 0.321381447255967, + 0.38437635450010454, + 0.35651330401186176, + 0.4025349879673704, + 0.3408205553665815, + 0.32455700868481135, + 0.6257667745045769, + 0.39919978311950605, + 0.6833839456033348, + 0.5893123689735397, + 0.4667469897224573, + 0.5901629787579029, + 0.3916536181539172, + 0.5471577027541439, + 0.5372833387913696, + 0.4848069147178832, + 0.3946929467099509, + 0.47524160415180344, + 0.5105479600039626, + 0.5368054644796146, + 0.4049881928048883, + 0.36532339583036855, + 0.6312770780889116, + 0.48563616749161254, + 0.6597669763489061, + 0.64458412409154, + 0.5280343720942465, + 0.3481809537647006, + 0.3770083283644065, + 0.37725668548255287, + 0.4913841026208206, + 0.7472732212578769, + 0.4798701768329656, + 0.3949176884629491, + 0.7429006804938802, + 0.4545822569839783, + 0.6277763741666593, + 0.5382750561446151, + 0.4404452644843375, + 0.5011094770990414, + 0.46596857898672583, + 0.4749057646289808, + 0.5833114932168902, + 0.549405504020434, + 0.5387768520444042, + 0.4550977420280887, + 0.47816524457781284, + 0.5112603696003338, + 0.49372562862310704, + 0.7672024440718375, + 0.49067993793271586, + 0.49439276738880406, + 0.44601723585302766, + 0.32294723912783296, + 0.6655602516971892, + 0.4251092956910889, + 0.36022859444533273, + 0.5241274163868456, + 0.45910397111338075, + 0.3570671509289035, + 0.4914601845688834, + 0.4486724209309481, + 0.7773493958783432, + 0.32370253728519055, + 0.4085149320951137, + 0.3636660135362076, + 0.4404170538900299, + 0.40658896049196896, + 0.33054179210678336, + 0.4228542674112746, + 0.2939585111925219, + 0.43043596284066177, + 0.7407947949623841, + 0.3361974610685375, + 0.41574280793204155, + 0.5075424526992025, + 0.4270692143131865, + 0.4143084160241904, + 0.4859927102026593, + 0.35633591311339724, + 0.5270525798043743, + 0.5334748083330212, + 0.2761499919809231, + 0.6386393355485982, + 0.6907904645793475, + 0.5141868709780534, + 0.4933165984454211, + 0.5457322240238947, + 0.4392102104479545, + 0.46533081360997386, + 0.5737814754386736, + 0.7803630290052321, + 0.40819160684770794, + 0.6098888896632145, + 0.3554372320868898, + 0.4673825617952799, + 0.490137669856389, + 0.489511167158127, + 0.4052006088161949, + 0.3398845941493744, + 0.35310340315008504, + 0.4042137491537399, + 0.5818874360183329, + 0.4698411366450295, + 0.3472181334584933, + 0.41585968848806315, + 0.470240445933624, + 0.31843037831928456, + 0.5515027407957128, + 0.49775408484745876, + 0.4221712113889387, + 0.45597626612387077, + 0.3402226276720315, + 0.4342852248820519, + 0.5080253790860945, + 0.5886487745668498, + 0.43552594228816116, + 0.7800977426504421, + 0.3754402308919976, + 0.5614427993080676, + 0.7405257957999685, + 0.6366875169043102, + 0.327448921820161, + 0.5116962274875311, + 0.4239551425662597, + 0.3636788577540679, + 0.38134253697756315, + 0.39064822040251784, + 0.47159840809578624, + 0.5045084440495963, + 0.3824923968780673, + 0.6167481732097662, + 0.76443729699494, + 0.4393546186035467, + 0.44953509081505383, + 0.4448103350905991, + 0.5055475653485052, + 0.5214540464247973, + 0.49508649684288836, + 0.5066339822493489, + 0.3716760334446735, + 0.6619214999780535, + 0.48966839888940583, + 0.5576223078216281, + 0.7145824461429662, + 0.3382919634193488, + 0.5958822106214317, + 0.39161165630816813, + 0.46626198930267126, + 0.3508142889211602, + 0.5239773322277538, + 0.502537561502235, + 0.3478690983038224, + 0.44093418240693455, + 0.4136978988322543, + 0.5399712215259647, + 0.5956783983602544, + 0.466078622475836, + 0.46790713489209884, + 0.4204650463807771, + 0.740572590208007, + 0.47898220038417416, + 0.5320903577667648, + 0.4711516368918514, + 0.47028691683818413, + 0.6144840757308169, + 0.3612676266827731, + 0.573875534744756, + 0.38737364189208306, + 0.5511615746674955, + 0.4308666773581781, + 0.35877174085541413, + 0.3735274541985518, + 0.4764536104656019, + 0.38104963323907937, + 0.5134543605525799, + 0.733674055435665, + 0.7243664460710075, + 0.4112086159887532, + 0.36791219825459215, + 0.7027877761779825, + 0.456411248003403, + 0.4009788011330406, + 0.5242893504483549, + 0.36561535609225404, + 0.6013258189805313, + 0.43651866781591747, + 0.5710785109670848, + 0.5868548332878294, + 0.454470688049003, + 0.5173818879208366, + 0.6566899948859836, + 0.5237174013579353, + 0.37313107119872696, + 0.4291395274258063, + 0.3198737211595844, + 0.5426022160706426, + 0.3608051691154201, + 0.5243366278575512, + 0.5041101196724461, + 0.399311357227373, + 0.7221767126131955, + 0.34458654524767457, + 0.7274375895556306, + 0.45205370347422835, + 0.5437234698754694, + 0.7452571689484243, + 0.4860092520352436, + 0.5581046283356943, + 0.40549508311038335, + 0.7053409384105619, + 0.42427790634938745, + 0.34545734370821185, + 0.5644073465590729, + 0.5260366768527401, + 0.6429703525657132, + 0.5544918528200673, + 0.40953494275187247, + 0.4906963013810615, + 0.317051465475116, + 0.4654588951017218, + 0.5703734445086214, + 0.5176027959721358, + 0.3691515804711226, + 0.5479139030347022, + 0.5888584618315253, + 0.5755144139176605, + 0.44643693139833746, + 0.4012419925764735, + 0.5135989692296197, + 0.4693132137843293, + 0.5616262867973327, + 0.49193647742448016, + 0.3749260243956354, + 0.4191387169656683, + 0.5160752165225038, + 0.4094213024109814, + 0.7362655735719289, + 0.5273415322138623, + 0.45943917541922075, + 0.5043291213364302, + 0.40987828307840113, + 0.4532659544769539, + 0.5454625162138036, + 0.556002598671887, + 0.7112893469945653, + 0.5438148968561158, + 0.580101974866277, + 0.5339406058928997, + 0.4990629759641461, + 0.5809920127377141, + 0.41328383177249906, + 0.6385622767309085, + 0.6000443505991249, + 0.7250263837253109, + 0.5269240768817897, + 0.5503002293417026, + 0.5450754907477497, + 0.44127526236510284, + 0.5417493050034099, + 0.3569630063209288, + 0.6844968854437187, + 0.49211451162920505, + 0.7369074598841748, + 0.760121111730257, + 0.4896147939511099, + 0.4574160986542373, + 0.5075415851932724, + 0.37216196502783455, + 0.5070986512889643, + 0.7583110689881082, + 0.4952650698290551, + 0.5209475825406066, + 0.5905893263169425, + 0.523498025091827, + 0.3649757864546288, + 0.6151707575947721, + 0.44233911045159635, + 0.4026008459302329, + 0.7336776566381829, + 0.4243345718935977, + 0.44359528780365803, + 0.4119850087953878, + 0.4921235437055437, + 0.5656272019650229, + 0.632359457628226, + 0.6806917902662617, + 0.6261868200748562, + 0.5932286733541162, + 0.46774583876656595, + 0.6113295760961851, + 0.5660414457373943, + 0.7798575925471052, + 0.7396301119611861, + 0.3719979161170586, + 0.3931093042959004, + 0.36943194314235167, + 0.5242922762427723, + 0.4248400184864105, + 0.7868494975823057, + 0.38975240510621767, + 0.4301491093109951, + 0.578906862390266, + 0.4930435590626272, + 0.5688376238370615, + 0.6286902488257561, + 0.4768771451042446, + 0.6209449770998515, + 0.3931241744086707, + 0.6877089163622913, + 0.7803591081496537, + 0.7651368293989725, + 0.5911576675064897, + 0.44381264545021387, + 0.5655246985883929, + 0.7419948037579497, + 0.6929675317911859, + 0.6300673983670965, + 0.49660343118258704, + 0.47407777745984, + 0.5008652358534041, + 0.7373113958697318, + 0.43259288019713044, + 0.44488285679791495, + 0.3338967877062094, + 0.604943697837606, + 0.3308601936430249, + 0.42643575015267143, + 0.6725621037735945, + 0.5465391865755558, + 0.6865296414959696, + 0.5506510326199878, + 0.5266753457748342, + 0.6707378910259, + 0.6375745381649086, + 0.5717541653472493, + 0.3940007352479803, + 0.6502029118973275, + 0.5292471891172832, + 0.41711782579406353, + 0.3699690337201155, + 0.49140044706195984, + 0.6422196462546979, + 0.524293769274247, + 0.6193255773599622, + 0.5815970394196893, + 0.43493838784742256, + 0.42028930564576444, + 0.6913181166147817, + 0.5116401580032102, + 0.614563221664267, + 0.7255007513250274, + 0.4273139319015499, + 0.3931154342157917, + 0.38954223137544375, + 0.39000056907162756, + 0.7157917044290896, + 0.7363733667639056, + 0.5708513581897399, + 0.43220082691158224, + 0.47862963889167476, + 0.4022924170490725, + 0.780759879702527, + 0.3614402025730854, + 0.49378771068954297, + 0.35388738336498954, + 0.5188776789253664, + 0.3900522343976014, + 0.573392165440272, + 0.4959888965430928, + 0.3639181242259264, + 0.7008689634596003, + 0.5182602988887874, + 0.43031521710935855, + 0.35290951246084723, + 0.3649368998003748, + 0.415336689239234, + 0.428021990114716, + 0.38186850384759574, + 0.37602700701485103, + 0.4104660368216666, + 0.40400017374865316, + 0.4209797233545616, + 0.3466757598282232, + 0.4507203480912289, + 0.21751817173958674, + 0.2928375868749654, + 0.3567546179836887, + 0.39646063359647815, + 0.4521599390654826, + 0.44903419442279885, + 0.36734577396356954, + 0.37953834031858086, + 0.3575015644137741, + 0.6754591931506768, + 0.35827605500019577, + 0.33538197348914794, + 0.7779998716263806, + 0.37464026540596695, + 0.4459400117139559, + 0.36923502724471446, + 0.45358015422643233, + 0.30920045044442834, + 0.4241765064697159, + 0.36470831593187536, + 0.48992547761870103, + 0.27918460840234544, + 0.42185112333772945, + 0.40696208981796433, + 0.5144628586570544, + 0.4125846379599105, + 0.4766858522745669, + 0.3174075048225539, + 0.4736726596224461, + 0.422304235020702, + 0.37799569214992684, + 0.4775608786554911, + 0.3953646208415129, + 0.47904834945170477, + 0.4112685080820058, + 0.3882622915963776, + 0.8069357817036735, + 0.32955198902228766, + 0.46088614082224355, + 0.37404987319323013, + 0.4262900946741547, + 0.3450168812560842, + 0.39879789124294174, + 0.4615906238426876, + 0.5677922989724592, + 0.35076929700825205, + 0.3946141325005354, + 0.3134482781606723, + 0.41745248974858784, + 0.45348914981500865, + 0.4489780299377487, + 0.4771567318835986, + 0.38047172452520117, + 0.3725779872109774, + 0.46921978763691047, + 0.4235657957261938, + 0.4528876177689952, + 0.3217351402091021, + 0.3289063112261344, + 0.49010043654601243, + 0.45973247151748603, + 0.42359128095131465, + 0.3921486028152664, + 0.408215139116615, + 0.8304611959393566, + 0.3377089875942524, + 0.36295606576085554, + 0.37298426678956936, + 0.33759714902777477, + 0.3731763849832236, + 0.37969087037864135, + 0.32625387701174374, + 0.3687759245029909, + 0.3541140333095197, + 0.48619868260215154, + 0.436689502850158, + 0.4545784676676063, + 0.4804464218085731, + 0.463516971861229, + 0.3983120291885024, + 0.5690540303480982, + 0.44154992608810517, + 0.4615859555505751, + 0.5059080490724946, + 0.4641338601515011, + 0.40510663725309143, + 0.33185299545446867, + 0.419029705447843, + 0.5326838432199521, + 0.5005676619748031, + 0.36175986633190094, + 0.39366404268799654, + 0.42178325043864445, + 0.4760550153313333, + 0.3706695374876945, + 0.40200975140319745, + 0.4981336378122374, + 0.5445501762776258, + 0.3587017019016525, + 0.43686745920935444, + 0.368688053517093, + 0.3648309267082388, + 0.3669416868110402, + 0.7899686590111643, + 0.42524884727268014, + 0.4545005957061429, + 0.3634484762758518, + 0.3904499848254764, + 0.40473879042714345, + 0.7035411118854948, + 0.4454895552406106, + 0.43202750605025725, + 0.4562458026109136, + 0.42830166143751414, + 0.4460339632234239, + 0.4655629299998797, + 0.5044705930067453, + 0.308026997113374, + 0.3754636741720363, + 0.46079615729074086, + 0.4557727813055633, + 0.4222938360461708, + 0.4229183033323785, + 0.4168673263438459, + 0.3536473579829292, + 0.40359375264438646, + 0.41470041466695606, + 0.4039666150412559, + 0.6157948942845981, + 0.387162829609068, + 0.42940954999292447, + 0.34918670127573126, + 0.5873521810969189, + 0.4146328629843338, + 0.38081957606851424, + 0.4465282227951272, + 0.44425185139452855, + 0.44833755750113996, + 0.36892232314450835, + 0.5575086146913076, + 0.48472801096159657, + 0.36141393123528937, + 0.43741956126745357, + 0.4568578876528855, + 0.4863341869421264, + 0.40573498488426113, + 0.4734427061165762, + 0.36157489793732606, + 0.6617107689747753, + 0.40294045018657054, + 0.46266916186258655, + 0.4831730594736489, + 0.4184665746449499, + 0.42714784777361686, + 0.43721944278808617, + 0.34783021838462924, + 0.41236855718914855, + 0.6047663988000854, + 0.4068214635618967, + 0.3836325165857249, + 0.5264396408246649, + 0.4335414154874777, + 0.42143208063700566, + 0.6265589672682095, + 0.4104627526398473, + 0.48427071272702854, + 0.4314773112044669, + 0.40906596220827796, + 0.5072242514479716, + 0.39835167433394936, + 0.49234092548122727, + 0.42465913534632604, + 0.4794303131568033, + 0.3803378303501209, + 0.47677439366991453, + 0.4284562036511616, + 0.36216009572168456, + 0.3445871239932667, + 0.3875352802432142, + 0.44502420501247664, + 0.5596226351149268, + 0.47308519320555587, + 0.4608350259090913, + 0.49098640499184143, + 0.39617212795528023, + 0.3897507216336862, + 0.48464830003197146, + 0.37972032623271906, + 0.41672996941280893, + 0.4994059844873657, + 0.41193716486476056, + 0.49291146007723213, + 0.5112996670282112, + 0.413496088220693, + 0.41029509240005324, + 0.5266418360711542, + 0.4237222019001533, + 0.4467082797781836, + 0.3495359933743843, + 0.5028728170016533, + 0.6070662608419574, + 0.7922577131017978, + 0.43676901104257276, + 0.4232646113058639, + 0.4375488200106937, + 0.39138350133153677, + 0.37330989207982845, + 0.38652309268395624, + 0.4666244449497376, + 0.4597288807147415, + 0.42317502093072434, + 0.45649143593686214, + 0.4734568589778856, + 0.49820543537630146, + 0.3762289634534157, + 0.36693151343535835, + 0.4660265918488228, + 0.4201470285706008, + 0.3998651959403923, + 0.4170658180940995, + 0.41062067038321304, + 0.4144478634176612, + 0.320221461540004, + 0.48359824572773125, + 0.44930272510615593, + 0.40509695961445136, + 0.4131683762232144, + 0.3266500252486254, + 0.4478253730639762, + 0.30971950844800505, + 0.537184730723362, + 0.40645243396034897, + 0.36246631142396546, + 0.4607016945606884, + 0.4316687304110571, + 0.36254069363527197, + 0.6090660546338557, + 0.5281894303763627, + 0.35724957069829927, + 0.496324970239613, + 0.4289083072756557, + 0.4440220986078451, + 0.451097761886176, + 0.3189599822886128, + 0.39497911766120064, + 0.508911697606761, + 0.3869194232271083, + 0.45457180290295324, + 0.43166985018999565, + 0.4755699288258256, + 0.4925497860338554, + 0.40609464290617214, + 0.5131452423636265, + 0.48081798483610577, + 0.4085171955790592, + 0.3535173184731163, + 0.4509250248888943, + 0.37163065623843505, + 0.4162128134352704, + 0.41875496084273683, + 0.4902507382607676, + 0.5374534245515816, + 0.45833465502055604, + 0.42786566804068277, + 0.47322036445171095, + 0.455134719252503, + 0.44619451890193745, + 0.5323994508109007, + 0.4098142549785965, + 0.34578172603038093, + 0.5011224636120866, + 0.42493119075636937, + 0.45752868910127714, + 0.446735332807758, + 0.43510880940634994, + 0.5065862112170905, + 0.3766470776819207, + 0.43315370222264304, + 0.3636472293265125, + 0.36886075707861565, + 0.36189665025195206, + 0.42858157229808314, + 0.3677805941004515, + 0.44684087968220165, + 0.5316829128550605, + 0.4450624502529229, + 0.4944627612126675, + 0.7761357429697047, + 0.5590319358341201, + 0.45045415332742195, + 0.4197760510120009, + 0.46483636199086764, + 0.46242777298060506, + 0.3526181856389085, + 0.5776600916908001, + 0.4796962810152896, + 0.437683132231655, + 0.4781680104993309, + 0.4064880517497924, + 0.5091845779237746, + 0.5213882442484572, + 0.4151563070751278, + 0.5474767179042513, + 0.5104471043025107, + 0.3865554366715724, + 0.45363839116698923, + 0.5133014620170926, + 0.4212860061817079, + 0.4427431680058866, + 0.3579701528976619, + 0.5674320568491877, + 0.5190813702165409, + 0.8194729914957504, + 0.5349383991457515, + 0.4562100509777783, + 0.2913267777464516, + 0.47756379228628354, + 0.4365357415557917, + 0.6122500531719415, + 0.35864681052702885, + 0.544901720043863, + 0.3517051898761574, + 0.3951238722855006, + 0.3787188692214114, + 0.365511976999, + 0.4547097872634255, + 0.37643231529560744, + 0.4404816323158553, + 0.4549079128602696, + 0.419214047051255, + 0.3699905252497523, + 0.37100797455602735, + 0.4252407809804809, + 0.36052493422544374, + 0.5447974736531289, + 0.4332292073171098, + 0.5142964008966303, + 0.4288154288697182, + 0.39128937863476515, + 0.5665578319019162, + 0.48110295396315894, + 0.379011929755855, + 0.5967486410037292, + 0.32866034137280686, + 0.4608712179780079, + 0.37864333932982475, + 0.38348028349014124, + 0.34694153250324794, + 0.3991283298724723, + 0.4924019958805071, + 0.36186543853743075, + 0.69327246284976, + 0.45427661947788966, + 0.38400014600112425, + 0.4119312829834086, + 0.38130848658630023, + 0.5985720741326849, + 0.3314313718381665, + 0.39177968533664137, + 0.3961523286412467, + 0.4483129964031055, + 0.37132877667889885, + 0.42761348896545426, + 0.3276824332947621, + 0.44745780051287765, + 0.7660865870439245, + 0.48204086153127884, + 0.39586803139908805, + 0.41503021730999373, + 0.4353660523573788, + 0.3454002716943021, + 0.39021360728353666, + 0.45728949921698026, + 0.7510342272946038, + 0.5595605725715859, + 0.46237378330867795, + 0.40603584423318095, + 0.3783429112066444, + 0.42181622833050936, + 0.34054511832287226, + 0.4605326401094667, + 0.523819141289066, + 0.3617889177129153, + 0.39672899736791406, + 0.39198479971219713, + 0.49028168959001045, + 0.36665931619444203, + 0.5439925744809616, + 0.4846442017194329, + 0.31772074348201984, + 0.4193280899212272, + 0.43407947224326304, + 0.5262726832433212, + 0.5343023091940404, + 0.2481105183376677, + 0.7191739583925536, + 0.7110258631980481, + 0.20971339118674248, + 0.3240761468326918, + 0.41201905323693216, + 0.46230658634833804, + 0.30576626298701925, + 0.40311256538902435, + 0.46557502948383706, + 0.4749681059929037, + 0.48224004552949024, + 0.3777035347914044, + 0.4022004834057006, + 0.3627604808717782, + 0.32733406767737533, + 0.4400741637616197, + 0.3940302318390911, + 0.3885121991808871, + 0.5071004856517075, + 0.4283003824837106, + 0.534145654934709, + 0.4454411860533996, + 0.6067005070473221, + 0.43360069630413217, + 0.45748630298814885, + 0.4970901637753066, + 0.45619965948654984, + 0.3407715071240187, + 0.6275601519590113, + 0.6180337316502191, + 0.45168821994175185, + 0.37771541983628, + 0.3951231795402926, + 0.5306311510505277, + 0.4259855670916177, + 0.47578948794570963, + 0.4676081566638806, + 0.48056242473313215, + 0.44325215050091166, + 0.38056856606595024, + 0.3992756137942306, + 0.5559964543688686, + 0.49088383848551376, + 0.4685210054242896, + 0.4905341514290114, + 0.3966313125320144, + 0.26966997167484474, + 0.328342581840295, + 0.36252527444262883, + 0.5308411196629054, + 0.4678568878124668, + 0.33586553205198244, + 0.4735756247610524, + 0.3731922659683949, + 0.33912968854094455, + 0.3873551682891007, + 0.40754818430110634, + 0.3744368428120132, + 0.4438716613536707, + 0.6267152488324256, + 0.3758351893277113, + 0.4505645562036744, + 0.5726703359828049, + 0.4036546340424113, + 0.5164141016534346, + 0.5959580429460117, + 0.45791047298411003, + 0.485092576756007, + 0.46728624643704164, + 0.43214437462946675, + 0.3243349636713914, + 0.36847616851900383, + 0.49385191669935846, + 0.5122322615761585, + 0.48262586262058693, + 0.3621195261795481, + 0.4710855285174863, + 0.48176812285719195, + 0.5123028125228158, + 0.4108548001165532, + 0.4705812889221215, + 0.38657577544822647, + 0.4785552801065051, + 0.4451038798661538, + 0.4217975641113334, + 0.3604066491461673, + 0.46249586816591176, + 0.4768987134989715, + 0.30533067824800264, + 0.3826495922062927, + 0.44321538388235443, + 0.40941888713688623, + 0.442080152986347, + 0.4959520296676009, + 0.5044489530685824, + 0.4550014966682858, + 0.4842973686816342, + 0.38352889890120617, + 0.4653216612907758, + 0.44756811096400434, + 0.41576840585784647, + 0.4275458483235146, + 0.4541240402522811, + 0.43215018490398976, + 0.47339573206093516, + 0.48356084835098, + 0.7217311584489208, + 0.5470893328929134, + 0.5079702823956999, + 0.5405042335957472, + 0.5196331616238267, + 0.5012426462989779, + 0.540064989207996, + 0.3660523173150516, + 0.4063958804817842, + 0.4742542159012608, + 0.35250377728592697, + 0.45220703887975955, + 0.5011935538594429, + 0.44513760732203234, + 0.45027610868119117, + 0.4696005759068787, + 0.33545242674666054, + 0.5418405528476905, + 0.4937896471769025, + 0.3911384226827138, + 0.4402559949671268, + 0.3928427405924223, + 0.3645046253211907, + 0.4602137276164786, + 0.4518270436077228, + 0.43065326752711214, + 0.4648028982485353, + 0.39967939313753653, + 0.4874612631538568, + 0.37321793029116995, + 0.3971719430981407, + 0.4397259365977417, + 0.5064052488299026, + 0.4969964481743149, + 0.475728193204399, + 0.4124983498083566, + 0.3578281358562146, + 0.36132627074951557, + 0.44969196709840437, + 0.41908564880330484, + 0.45558310182589157, + 0.48293591134376024, + 0.43805637617930987, + 0.41415287390497796, + 0.36513033502598446, + 0.4185399738702555, + 0.5312065254473824, + 0.41285312959456494, + 0.2155404982679314, + 0.3663000516274418, + 0.4509281234831253, + 0.3156197207219661, + 0.3060093487154306, + 0.46757042697784884, + 0.46468418617587604, + 0.33738201429009695, + 0.31931315157911905, + 0.4086869360018249, + 0.400178396015363, + 0.5308042716300557, + 0.31853978959111523, + 0.32424509653786077, + 0.3428827441826288, + 0.3614482699088313, + 0.27869354749827774, + 0.33967938921294366, + 0.31919804982313355, + 0.3482910488532544, + 0.33596845566461414, + 0.4479819892348233, + 0.36203440608235304, + 0.40351125620496847, + 0.4055856527970454, + 0.4944013780092566, + 0.3511195135467099, + 0.4115180514250562, + 0.40489175794468973, + 0.41435461627732745, + 0.4097987651594895, + 0.41549285967585453, + 0.40485323008358953, + 0.6212495405244628, + 0.3432219835479928, + 0.44290844384592315, + 0.5708496650267436, + 0.49005248950867797, + 0.37074078959134826, + 0.4671143312057617, + 0.41920627046288467, + 0.46083748193180946, + 0.474535759477449, + 0.36301328256402776, + 0.5788349191028905, + 0.4318304003928683, + 0.4279246314196395, + 0.3652803945380044, + 0.43747085017421594, + 0.39581316025277474, + 0.46100173264566946, + 0.4907425579544688, + 0.45256542318128073, + 0.4714693276736187, + 0.5729467661682555, + 0.44377773171955637, + 0.41196489936656155, + 0.5098171166900807, + 0.5793024459257237, + 0.5281061584366875, + 0.4473435644894185, + 0.5706410804931838, + 0.5581239379421278, + 0.3897477857268756, + 0.3951257842400775, + 0.444618685375577, + 0.3521577946488128, + 0.6558446133948321, + 0.4897252980155602, + 0.429944342789216, + 0.35711646534802555, + 0.46767150325485507, + 0.4057096551847874, + 0.5172254059026118, + 0.44235574312479553, + 0.44209328686238913, + 0.5288509489645371, + 0.3343508334825231, + 0.41853568331736474, + 0.5433408521540253, + 0.35159020582629796, + 0.4293513311659262, + 0.32927089986258207, + 0.37558225986839483, + 0.3992080889232697, + 0.4047202190535114, + 0.47189587371609987, + 0.35145653911114044, + 0.3305940619706765, + 0.4364665080291861, + 0.4012454165254459, + 0.3763069998184036, + 0.45322041213510983, + 0.47188498636711657, + 0.42731669776444203, + 0.524786939954793, + 0.5626865779854949, + 0.46190400664461856, + 0.36094765515785576, + 0.4814170038974369, + 0.4429160112838419, + 0.3976673501948979, + 0.4332586751127949, + 0.3306640130962209, + 0.33199773127835513, + 0.3140592464190256, + 0.37141297432151743, + 0.43708236221503205, + 0.6500359282892187, + 0.4379974651509567, + 0.42666705500225865, + 0.3028630756233425, + 0.36119223913979465, + 0.32351761449032046, + 0.41264830714489736, + 0.37241297437155607, + 0.3943061959793052, + 0.3575841621167441, + 0.4233472396342176, + 0.3933960771909747, + 0.4245770940421911, + 0.4166191477500685, + 0.37922898253134957, + 0.5034380222760821, + 0.5003052134707534, + 0.39037091923394324, + 0.4362875269104582, + 0.44424123087251205, + 0.3943968966181833, + 0.368070688378745, + 0.5160489358892355, + 0.4529706176306261, + 0.5827905290982032, + 0.4565084531837179, + 0.44959330431550987, + 0.4514856250224764, + 0.4559571386290743, + 0.3535156829785204, + 0.6067546166297257, + 0.406224814497108, + 0.23587220949762713, + 0.4733280546054864, + 0.5943498745623752, + 0.3727189668093015, + 0.480463594939999, + 0.7528157513468045, + 0.4503993982506686, + 0.5254568478173661, + 0.31551713640061557, + 0.48884179739140066, + 0.3330026116026705, + 0.3425106112097761, + 0.3443412200145751, + 0.29680775890423255, + 0.446568044743158, + 0.24783519472386037, + 0.4078414352860242, + 0.45102789165306223, + 0.436100439529927, + 0.4018667091920383, + 0.39181057288493226, + 0.3527856226127046, + 0.4458680475973061, + 0.521073774709927, + 0.5424525807356961, + 0.3584976269106247, + 0.45410257256926123, + 0.4168492618515408, + 0.7310061722569743, + 0.4066403138622959, + 0.5592626121445348, + 0.7351419876462649, + 0.7137365963325792, + 0.47272579262256637, + 0.3838649330850111, + 0.32992175225649034, + 0.4563497071810913, + 0.5002355892882614, + 0.4942075273657982, + 0.5727474230115867, + 0.4257285418163047, + 0.5513695070540524, + 0.4468187957905582, + 0.3953374863910868, + 0.7001842759681974, + 0.6809672090639258, + 0.48161059882663004, + 0.41852424485078493, + 0.4691848829780948, + 0.42071624398854546, + 0.45535798648877784, + 0.6433785016150221, + 0.666193058034987, + 0.42677112415690777, + 0.43876286422363725, + 0.4405802257956282, + 0.37915442962682017, + 0.4906363765366919, + 0.35501321262027813, + 0.508773428744266, + 0.5157713671907902, + 0.36102768571103533, + 0.44673208825132965, + 0.43032726922287245, + 0.3627099256021973, + 0.5286223747050587, + 0.48269281581984175, + 0.41490267765264666, + 0.37744189205249273, + 0.4835214750925854, + 0.43716927635291264, + 0.3511102881909473, + 0.3473830736094596, + 0.2916300009893597, + 0.25355027781477757, + 0.29870641079996396, + 0.4693761625097416, + 0.5135301250510251, + 0.373281405459832, + 0.3257151436232246, + 0.3782241123862678, + 0.4795077682510239, + 0.45524481208606454, + 0.5042719581195828, + 0.38307731061448536, + 0.47910876809772696, + 0.5283632975925056, + 0.39682038646722095, + 0.5497493482097867, + 0.5046185904360162, + 0.44218890808816025, + 0.42600119298434674, + 0.37340030615535863, + 0.24725059460273593, + 0.2992955635439496, + 0.2860932277643949, + 0.5390689906793598, + 0.45278104304530076, + 0.4319845433112435, + 0.5167996145384413, + 0.5373439202172341, + 0.33426079282151905, + 0.4446419150194638, + 0.3772685974383883, + 0.7202241179024373, + 0.4452230890389591, + 0.455637902084332, + 0.36770873140115634, + 0.45566252875512714, + 0.4943511667664162, + 0.44266679340976517, + 0.4130513443230208, + 0.4893141609525534, + 0.45612234325824497, + 0.5643384068427946, + 0.4243605115290637, + 0.4954050249456373, + 0.447818020354486, + 0.4358662194968785, + 0.4159887184799567, + 0.4769525924591619, + 0.4419697842791473, + 0.4957951717245607, + 0.36379627348412447, + 0.34608489366110906, + 0.3443011628227616, + 0.41867673345793954, + 0.4607030761876151, + 0.40511298073881014, + 0.4527545107622981, + 0.48580865589954514, + 0.43837058688202996, + 0.432196516054693, + 0.43904470920112004, + 0.2748621002646977, + 0.429096290907425, + 0.32197702609225964, + 0.4395988912982628, + 0.23207130790176558, + 0.6671782881099932, + 0.3312262061249891, + 0.47846174592301516, + 0.5682656780081995, + 0.4903349229417555, + 0.2796126390729696, + 0.4457621898806477, + 0.35699633066531217, + 0.7742151541329414, + 0.29092064369330206, + 0.5342618008892166, + 0.5123642624037624, + 0.5294948092049808, + 0.6360502489552134, + 0.4566348824041327, + 0.6095409169314486, + 0.45945192866491275, + 0.404885196441629, + 0.35462633295886054, + 0.4600065457144224, + 0.3222040537756203, + 0.7169623095059964 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.48749237829666486, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.56912258444131, + 0.7573476951466652, + 0.7322982626559701, + 0.8156949719285007, + 0.7374038550374896, + 0.6767880690811475, + 0.7286093424403789, + 0.6690592644385993, + 0.4322608438614718, + 0.7137123429969136, + 0.3689377118288546, + 0.5075596862094595, + 0.42474796254855834, + 0.39240591404317005, + 0.42957961766240654, + 0.35213216881196224, + 0.39929754744505214, + 0.5811833784943641, + 0.5884718096405991, + 0.4713051886700745, + 0.3586389365298068, + 0.6863497968769027, + 0.4740243591735051, + 0.6042638278127136, + 0.387874946140192, + 0.3409153063072016, + 0.42789572317692426, + 0.5198902465560585, + 0.6338629559002531, + 0.5079135459401782, + 0.33920971501751856, + 0.43127975425977877, + 0.4369941598620987, + 0.38456650890008015, + 0.4881815818742208, + 0.3791296097690962, + 0.498239825383871, + 0.5496765069135637, + 0.6118505101848644, + 0.6287081828714465, + 0.7883689868921315, + 0.7803417543590897, + 0.7733361610291795, + 0.6622884755071788, + 0.4962738926293794, + 0.5520820035335724, + 0.7787456397995083, + 0.42852028422502153, + 0.4104856925476855, + 0.5466827771182505, + 0.45114742291722054, + 0.7772470576167761, + 0.7936178374817014, + 0.5014085496345143, + 0.4900178317518565, + 0.4115816344272369, + 0.40022767626688366, + 0.44867386195765724, + 0.44739653933858653, + 0.5633838248837022, + 0.5155159871612355, + 0.481239808240316, + 0.49055392741013704, + 0.5383451926988093, + 0.6304439275365931, + 0.43509348627852545, + 0.4107051326009348, + 0.6097483259582993, + 0.44121162235685796, + 0.5166712681015074, + 0.5183032441988888, + 0.5111924728517737, + 0.4866092439055256, + 0.5473821899274093, + 0.45284175541406946, + 0.5540430158026728, + 0.3532410808249268, + 0.3945706950368485, + 0.4258588777495086, + 0.3593326968912295, + 0.46453096193319576, + 0.5969597106901469, + 0.41661274200246845, + 0.41302687494371065, + 0.41939293650399445, + 0.4991070607497259, + 0.4885197923060582, + 0.4643718046799888, + 0.37843831309110587, + 0.4185605005562758, + 0.375168498442975, + 0.37369628825396856, + 0.5375082721206426, + 0.5698922106991742, + 0.45589549244467464, + 0.59950055203916, + 0.40079496623005234, + 0.5756891144728092, + 0.4428888194753585, + 0.4910710475750021, + 0.47980040354578574, + 0.5505048527582931, + 0.3411455203969169, + 0.500832784239845, + 0.367578161676492, + 0.38120565862706046, + 0.5052185948327026, + 0.5605899236474654, + 0.3712138384639378, + 0.8083484870877604, + 0.803386536299521, + 0.8082880430344249, + 0.8274477279634028, + 0.5858236084062783, + 0.7041687447236781, + 0.7179817666213767, + 0.7572849101963103, + 0.7065320044668362, + 0.5888650099168337, + 0.6600840612235375, + 0.5755802602453014, + 0.773397037408953, + 0.6647335627495715, + 0.44828249851531377, + 0.3574017780753619, + 0.3330614750569474, + 0.36585762974526853, + 0.5741447019543949, + 0.45850694629790917, + 0.3166457819110712, + 0.49217876785739045, + 0.4614902275236674, + 0.513287584909265, + 0.399159336016668, + 0.4350548609515315, + 0.5070562617598187, + 0.4923981503927936, + 0.38506998620848726, + 0.41503299056766896, + 0.4191897269614472, + 0.38571370019009066, + 0.33726013592658594, + 0.45837867688293576, + 0.6086583348723635, + 0.3314823973270704, + 0.5750865151837197, + 0.37995096265238926, + 0.5048227086100657, + 0.46032239667141844, + 0.5776730934041311, + 0.343222734765333, + 0.508789330413979, + 0.354209274035403, + 0.3764709717889596, + 0.43697888324235745, + 0.373225820002126, + 0.297523347849684, + 0.22105401144234893, + 0.38678000892536774, + 0.384894533739214, + 0.5308108516350988, + 0.5170412715100499, + 0.5882826905741024, + 0.5164597128822631, + 0.3974118313904244, + 0.49956087605870364, + 0.5401971962973359, + 0.527619427899456, + 0.4441521016158845, + 0.4914320749538682, + 0.4396318454430473, + 0.33177612285605673, + 0.5081217784324099, + 0.4793839873215079, + 0.3017522642614497, + 0.32560034879901034, + 0.5224087710265908, + 0.5843173874127755, + 0.5830034381312285, + 0.5039838693984832, + 0.35093407230390294, + 0.5724839019397331, + 0.5189976571577515, + 0.5373843152395861, + 0.33214040093987013, + 0.5742106416401199, + 0.4855521554661252, + 0.538348860322185, + 0.2798409529778903, + 0.4592362313707702, + 0.36571687803911623, + 0.2969782303104186, + 0.5527247377159382, + 0.46581444960267315, + 0.5757735955497969, + 0.40579945378885485, + 0.5944988417093879, + 0.37975178899484846, + 0.4828158585662617, + 0.42217272535071165, + 0.3600353193267219, + 0.29772397985910315, + 0.4062219759301369, + 0.3963089794015173, + 0.6385840916930651, + 0.38320490670555823, + 0.31506048531950587, + 0.48715630852707825, + 0.5802377747301, + 0.42976477743334635, + 0.48029335287966063, + 0.4500617777494673, + 0.36851525543450636, + 0.4081231093170027, + 0.3963562563184824, + 0.5412331993724322, + 0.6201444677474787, + 0.6215733385061639, + 0.5821953629651935, + 0.37186648533264166, + 0.5368594518025338, + 0.4743409499094822, + 0.3015910894806002, + 0.3594801791034793, + 0.3882589728826691, + 0.25741632104995427, + 0.5626682973034837, + 0.35424635295309964, + 0.3488943835930392, + 0.5499114877957471, + 0.3137191051884399, + 0.2900193909808754, + 0.5535518306219914, + 0.4293486775210664, + 0.34207674836957735, + 0.6307137187562851, + 0.5782356842487792, + 0.3262762555773771, + 0.32739096286569436, + 0.4995556255670945, + 0.3447924836010273, + 0.4703832235701307, + 0.3429443845124752, + 0.5323922160471495, + 0.47741371763134954, + 0.41541714022241527, + 0.3788359605642366, + 0.35463532018039007, + 0.4854377897525968, + 0.37155580833543916, + 0.45813342690220465, + 0.35112420662490024, + 0.4192825096367596, + 0.3790926918728127, + 0.3900776484213594, + 0.46601381393095387, + 0.4327882623989002, + 0.39936974574504397, + 0.45210356059912726, + 0.4685271009060396, + 0.3494705145024178, + 0.5538375204966673, + 0.35446663066989376, + 0.38084934284541405, + 0.418281690369512, + 0.34321112796222025, + 0.3793126923044866, + 0.5452862433979495, + 0.2967562241112188, + 0.3693890724573101, + 0.3607806303250659, + 0.3720247817260587, + 0.38624265558954773, + 0.42528624629474104, + 0.3973442283340379, + 0.50554570475768, + 0.3221801960760654, + 0.4790181928501842, + 0.5241875296291748, + 0.33670277090952205, + 0.33079757342360994, + 0.6284211169527723, + 0.517451517582951, + 0.3038812644384597, + 0.5618918451277506, + 0.4970484005140165, + 0.47496174279431214, + 0.4780691669064784, + 0.5430963188116814, + 0.5216459587980065, + 0.5496388754835805, + 0.5369506307732493, + 0.5372654735986558, + 0.4392902125632175, + 0.4820018131872105, + 0.577726057247221, + 0.4991261660127552, + 0.48331936408106513, + 0.6566780498337491, + 0.362506870027256, + 0.6116915001900534, + 0.6687822994548793, + 0.7297672879605006, + 0.7002840817883765, + 0.5414646210807488, + 0.5798484677831596, + 0.42719513549441035, + 0.7000205849887196, + 0.5060629032373347, + 0.5152334606658808, + 0.34829483959640367, + 0.5380732039376637, + 0.45266437407426097, + 0.4436713782289095, + 0.5562502870582973, + 0.3688821937921446, + 0.4426265470588384, + 0.34997973239409963, + 0.4722705847145355, + 0.3997682633562113, + 0.36579134097873434, + 0.4045528879469499, + 0.30445748350440643, + 0.3672475666548492, + 0.4004888418288382, + 0.4091064203258669, + 0.39474448002080864, + 0.40643340757262164, + 0.610555597286502, + 0.5084846693917089, + 0.46628509694780457, + 0.529857125909072, + 0.4854743580831763, + 0.3516475048499183, + 0.3277196499650672, + 0.5913708721214593, + 0.4166044775598945, + 0.3603373207190448, + 0.47289220412723604, + 0.5886076078815807, + 0.6181757322015159, + 0.46325689357950023, + 0.3764595519879509, + 0.5036809062066383, + 0.4030496287468534, + 0.36607424255320553, + 0.5324041908205235, + 0.366487496277669, + 0.8222017313419211, + 0.5473202915988595, + 0.4516181099666484, + 0.45376966410230324, + 0.48104320485110674, + 0.47599070099763413, + 0.44106399255222484, + 0.5540172668165416, + 0.42540225867138565, + 0.5743254724418186, + 0.48147625027282875, + 0.5483470069764149, + 0.591370912578172, + 0.49920272564735024, + 0.5122469342296734, + 0.42322801316243513, + 0.44237591079464206, + 0.6337291109274865, + 0.6068976614245271, + 0.45201615893927355, + 0.6188644294470113, + 0.48838086563928773, + 0.58347152396176, + 0.479937424616414, + 0.5176705831233656, + 0.4620800362192126, + 0.5782717380683311, + 0.587872839225232, + 0.6305682180673523, + 0.5556135591262595, + 0.5516427197548953, + 0.5448239111730534, + 0.5080121120940965, + 0.4239476877600818, + 0.4213549183327307, + 0.3468282454846931, + 0.40972774305155923, + 0.3437111020867657, + 0.3707961064834279, + 0.39887194474865073, + 0.5372964559184569, + 0.38218980924349527, + 0.35919719522578275, + 0.42129567470497575, + 0.5777673862184886, + 0.6530379374929621, + 0.604638644954979, + 0.586391602742077, + 0.43722765824697046, + 0.6036714318246589, + 0.47560755972714175, + 0.6503990574498535, + 0.4948754759996833, + 0.5919124711248445, + 0.4638941081170655, + 0.5833662974277479, + 0.6039255190759714, + 0.5083245778525982, + 0.5005295495667543, + 0.5436866453723087, + 0.6074599206108077, + 0.5523320646530991, + 0.41696425554329186, + 0.4179593168666107, + 0.36166676330300107, + 0.3986005227657068, + 0.5479582987877261, + 0.5863827660268978, + 0.3452374541351479, + 0.637194799177627, + 0.6132432368786662, + 0.5005164619188626, + 0.4269162458904042, + 0.4957386695839204, + 0.584182757382379, + 0.5453947748189925, + 0.49496327123063594, + 0.8517927343893892, + 0.47558856500089186, + 0.573481475241053, + 0.5283990984448471, + 0.5736234045623282, + 0.4925217384534745, + 0.42814338320935036, + 0.6214746004553267, + 0.48489830848183924, + 0.4514899314515298, + 0.4779712949893348, + 0.4634054443453587, + 0.47979718671648824, + 0.4386944783806409, + 0.4718972785286087, + 0.461668851763032, + 0.4917099739443036, + 0.3369239283945528, + 0.5042783837803285, + 0.34566341123486155, + 0.4129852026119337, + 0.3083005396415404, + 0.5069573545637209, + 0.3959121283798843, + 0.3426338789460043, + 0.560735229439397, + 0.4209169310847148, + 0.3044080653114917, + 0.5592772819622837, + 0.493061317152165, + 0.4895552828012227, + 0.5443173928174023, + 0.43013386160178685, + 0.4375732082178629, + 0.6450645735133146, + 0.4495767924799949, + 0.5585864949728383, + 0.4316526925267203, + 0.3229961918035833, + 0.3009907807056817, + 0.4222962482025188, + 0.5173790628850164, + 0.6088947795861056, + 0.5910841138478448, + 0.4983527913942193, + 0.5581919675651098, + 0.4972705395745575, + 0.5213235138771806, + 0.4619143361228686, + 0.43941674364883393, + 0.45778685444092404, + 0.4387986014306396, + 0.4805746649067488, + 0.4239309780754757, + 0.46316535692738375, + 0.5497110180667211, + 0.4774327036639046, + 0.46848012374391435, + 0.3996006753097606, + 0.4239297085351453, + 0.5130970828611449, + 0.5671640755097523, + 0.5496132521533322, + 0.5655184140626297, + 0.5837111603503208, + 0.6236391617830171, + 0.6666148637516355, + 0.650208249198213, + 0.48635716083542985, + 0.5328117276167892, + 0.5067146124471436, + 0.5445622093760832, + 0.42299989556846784, + 0.5193253445929918, + 0.48844612304268387, + 0.5830386382863465, + 0.4174377278184781, + 0.586148332643806, + 0.6526929664144762, + 0.6234220060396473, + 0.5741557135929529, + 0.6466815963093824, + 0.5478125358084976, + 0.4778806415261836, + 0.4044804431511744, + 0.4506603817777552, + 0.35895157472284234, + 0.41406943347077724, + 0.5663979587374257, + 0.5865545778683013, + 0.4789888515129366, + 0.3928519802178589, + 0.4178896815147334, + 0.5083913983636169, + 0.4974737935430497, + 0.420152515386138, + 0.538296817462774, + 0.5121152844094484, + 0.39502661568429964, + 0.4454063803280907, + 0.4072260527688124, + 0.4833177009489347, + 0.4744598362236328, + 0.368554236673627, + 0.42613212892763913, + 0.4780364614084602, + 0.40987171181803844, + 0.40441174297627497, + 0.4413678730669745, + 0.39780722897674087, + 0.3995173405811839, + 0.43469654986146467, + 0.5475523389978227, + 0.4237803435912286, + 0.31759645518813223, + 0.5901648436848471, + 0.41858186325444996, + 0.5713583921154985, + 0.55394173586261, + 0.39991781188006376, + 0.531766038583929, + 0.6018082109989887, + 0.5147664119088319, + 0.6624607706040089, + 0.46398924905389516, + 0.531728823858798, + 0.5013836326357797, + 0.4909186552291853, + 0.5495889805689831, + 0.5218315737497219, + 0.38782830197349005, + 0.5017202437158192, + 0.5399810013068106, + 0.4495343628442757, + 0.3412700011996801, + 0.48063150385785675, + 0.5130582739509781, + 0.4772101849599515, + 0.6471449424391591, + 0.5537394123807694, + 0.5621310125449811, + 0.5766030420356923, + 0.6415655106241963, + 0.5648286986004725, + 0.6180857888867989, + 0.6236418627712353, + 0.6187973535062212, + 0.4541519978070827, + 0.47020086244385023, + 0.4157615144304055, + 0.35483722152480474, + 0.32316155083817677, + 0.38428991595987794, + 0.4144542839088835, + 0.36456095024977975, + 0.32759302367436854, + 0.5125907524988248, + 0.3167886111539428, + 0.47970513283996014, + 0.3673997585891458, + 0.5519067040186694, + 0.6019265010475613, + 0.512461927336467, + 0.3628315038650653, + 0.4111818711372623, + 0.5852265880656293, + 0.3655324225031292, + 0.5173124110078052, + 0.3358965394044285, + 0.8249120093824928, + 0.5459053289928507, + 0.44434126415073466, + 0.6235651635000125, + 0.5833057105854335, + 0.5080938345926785, + 0.44842490101730975, + 0.5122073864867629, + 0.5835100696209033, + 0.46456342778634907, + 0.5928794422653961, + 0.55840268685158, + 0.32081424805642195, + 0.5179776691644953, + 0.3271897752832055, + 0.39635050948948297, + 0.34155497843885313, + 0.533804692989774, + 0.4466840707590408, + 0.40268526000912075, + 0.5607113410132039, + 0.5842558722252952, + 0.4070231351670037, + 0.5087564269990988, + 0.4657057643070324, + 0.4849175120923566, + 0.5127680837965454, + 0.49313840109849316, + 0.3690011576679526, + 0.3635749821466068, + 0.5973578886626197, + 0.44605288170428303, + 0.39885806031462406, + 0.5317099824632812, + 0.42644610108537695, + 0.44973519506649434, + 0.3436993072636898, + 0.38337533841197513, + 0.44140961108298443, + 0.49500044826935297, + 0.4832226615651143, + 0.37220885166229883, + 0.49270976334379335, + 0.32390135356702315, + 0.5135481839336691, + 0.47841263944725426, + 0.5022587370695658, + 0.5002939749742501, + 0.42009150486599894, + 0.480264945641423, + 0.3920507522279539, + 0.5137246003448941, + 0.4333359096538318, + 0.4761480528018514, + 0.47229970543059885, + 0.3769954922527508, + 0.516351197164269, + 0.2955792855839965, + 0.6683676622447229, + 0.6705163971392978, + 0.569468064291617, + 0.5629098006419487, + 0.6108405307043533, + 0.6085516548674712, + 0.29927893619646667, + 0.41101918501602924, + 0.3753023732146222, + 0.4088410947137348, + 0.3587345178684305, + 0.35932895899895845, + 0.41823699330093633, + 0.4864501371500099, + 0.45741178844383384, + 0.3755940592760852, + 0.3937392519433453, + 0.3313135864700576, + 0.47350758135164217, + 0.26560090657746777, + 0.43545363373935375, + 0.49116502585575134, + 0.503064426822, + 0.5101823130346361, + 0.44332685016020507, + 0.542020259178586, + 0.3119425886727053, + 0.5334964561884799, + 0.46989313218608425, + 0.41604494412073156, + 0.3987697183959751, + 0.43441180037460014, + 0.44568491252332515, + 0.44921380500722796, + 0.49806248465620867, + 0.5166467139278841, + 0.5501374736800203, + 0.3671765870623856, + 0.4116324924282364, + 0.4590639611139661, + 0.5096985248222416, + 0.5133302724757789, + 0.5149697491999696, + 0.36906226785580737, + 0.5861929121188197, + 0.4770304669530895, + 0.4565100218086513, + 0.6032234742374396, + 0.4173413692937702, + 0.35874723156961824, + 0.32835933931452926, + 0.394445813898698, + 0.44617089419043565, + 0.4132518944039165, + 0.4237614602106839, + 0.42850821426620217, + 0.5403965848263529, + 0.6086851294910047, + 0.6648922136024961, + 0.5480145498169526, + 0.5691690877033446, + 0.5208010844066846, + 0.579568065812544, + 0.5366024202939427, + 0.5945198417528507, + 0.38436109238940985, + 0.5070411669696728, + 0.619338915274289, + 0.659299595964641, + 0.22228116408951484, + 0.4038541249205699, + 0.4338302366459513, + 0.5333853194140701, + 0.5070344236044197, + 0.5373603943492941, + 0.4088025870397393, + 0.5255008630732407, + 0.7443085168034448, + 0.6166935676850931, + 0.7802643504987156, + 0.7951307667425328, + 0.6563011558772542, + 0.7558254116600094, + 0.7608537234999342, + 0.4163930932714775, + 0.42923513148745085, + 0.44002847221713914, + 0.5389070607742494, + 0.36963524889391186, + 0.49700881381101614, + 0.3946214948154945, + 0.4996881666865078, + 0.4284057680424688, + 0.4191569980747466, + 0.3405946775397739, + 0.4514601753907362, + 0.42965783799963864, + 0.4733695613722028, + 0.4502148869246685, + 0.33835276471498255, + 0.3704490128015135, + 0.3559244461957917, + 0.372017495800158, + 0.28540319227795136, + 0.37003756544298716, + 0.3239642938854716, + 0.5447557200547175, + 0.5853451774312113, + 0.3841448563115709, + 0.33489881428210294, + 0.27841481356998554, + 0.5416222307217387, + 0.644765342894756, + 0.6462299204766989, + 0.3970316056795447, + 0.3809196880588628, + 0.2782999743557199, + 0.5038976007998034, + 0.3507833434892092, + 0.4084768842203873, + 0.49610855470976156, + 0.555432638165745, + 0.4826837429230492, + 0.3360566945801039, + 0.48522772325580243, + 0.538916683004925, + 0.5237597681527769, + 0.47876479443188563, + 0.5241130722854369, + 0.4365831287424494, + 0.5044303932232654, + 0.48051652397259076, + 0.5574028514064223, + 0.37412208865205065, + 0.48505250613431367, + 0.6558534620375435, + 0.4517935048669081, + 0.6574427010268725, + 0.3890409007757654, + 0.5570004009408672, + 0.715277589243833, + 0.5770712013812328, + 0.41754021633929167, + 0.39060091578331024, + 0.3662798017375343, + 0.337039940154823, + 0.4809060598420883, + 0.6095495593500112, + 0.4802909111036706, + 0.42458859304382107, + 0.3512804684586219, + 0.5153774437619997, + 0.5344680720116176, + 0.5700033274958308, + 0.60142601233806, + 0.5363859369430958, + 0.516876779491409, + 0.41887666071809787, + 0.5562074426110243, + 0.3205937699512361, + 0.4206588059553507, + 0.4751215557596119, + 0.42069598639814226, + 0.6214791098883831, + 0.6244029761547764, + 0.6145163949546448, + 0.6761184491096478, + 0.7529495490376407, + 0.6067384374266634, + 0.7641230139604323, + 0.5160392416469584, + 0.46054440958624787, + 0.4401099916661669, + 0.577856593474152, + 0.39022079714615865, + 0.5382637429824648, + 0.5428877008975191, + 0.5365270638386151, + 0.4969988346201817, + 0.4598602698537772, + 0.4684530123044788, + 0.5064139551963198, + 0.42853968349830107, + 0.5398623969284092, + 0.5243301416035888, + 0.5408449430218505, + 0.5624518071794495, + 0.6252595314374474, + 0.5713487861864591, + 0.44450539213600876, + 0.5828126711365705, + 0.5624191970023027, + 0.3704747771030826, + 0.4634786464837662, + 0.4390517683855937, + 0.4399376187761153, + 0.41934284619859347, + 0.44625790625113126, + 0.492359178520255, + 0.47474967715166294, + 0.4948508111113405, + 0.4834638056893352, + 0.32436911953529013, + 0.4790709372069643, + 0.5317219065370382, + 0.4846127501775585, + 0.5882472780296736, + 0.6163641335797375, + 0.7461298880505045, + 0.3611339865809955, + 0.38053587741930145, + 0.530307600845979, + 0.40920882330462616, + 0.30262034257510884, + 0.3745976705861211, + 0.3484116301544859, + 0.6874449024709579, + 0.6341841841457917, + 0.6578150791465203, + 0.37186565135516453, + 0.4155592191756386, + 0.34492415921150077, + 0.5688375140591547, + 0.47788016198838634, + 0.2851639752617386, + 0.327339472933435, + 0.47396434276155675, + 0.4457426588245657, + 0.4970969698263764, + 0.511025040514803, + 0.45010839515486795, + 0.4283660631753122, + 0.32104723314573846, + 0.6932715674231522, + 0.5702710671128828, + 0.698281243011446, + 0.5855711955518731, + 0.314168283970271, + 0.5383946373579678, + 0.5526145464248179, + 0.5471468557836225, + 0.5032997951481413, + 0.4140855680989085, + 0.5182417036400584, + 0.566044504020736, + 0.5555990527561125, + 0.5034265838860912, + 0.38544559695369596, + 0.4889549453468355, + 0.42568348977736664, + 0.4860604748519131, + 0.6438400218253912, + 0.5658549435242946, + 0.5915387783378852, + 0.46910455517325506, + 0.49430048384997405, + 0.49226480966926933, + 0.44678757311218875, + 0.5180157886809698, + 0.43619242512433476, + 0.30515959440249174, + 0.46098720382848274, + 0.537754273843802, + 0.2888164999816371, + 0.764085741799196, + 0.4630280349649861, + 0.6109837915230752, + 0.7342196893708091, + 0.6820168408414954, + 0.4025578704446212, + 0.8087172937710184, + 0.5029524744409832, + 0.6748980127386833, + 0.4146404861556153, + 0.5731703582720701, + 0.5930938837580249, + 0.6583772749405867, + 0.6082247429527856, + 0.6693041146580407, + 0.6106697260568331, + 0.43110844008140653, + 0.5923843269567904, + 0.5302633119482119, + 0.45694461751839593, + 0.674792537501682, + 0.36550784257208413, + 0.5916124612039434, + 0.5944184933319471, + 0.6715425681371928, + 0.606977061152751, + 0.6769022764863494, + 0.5658018840393859, + 0.428074455707959, + 0.5468283616063789, + 0.5500839781604194, + 0.534473382069687, + 0.42896859050602354, + 0.3986009050993171, + 0.4779524759623938, + 0.5540694413309752, + 0.5682630535461842, + 0.438245192995982, + 0.5857097294902545, + 0.5960280871909134, + 0.45188070577433237, + 0.5737104201909283, + 0.5623874595669694, + 0.5395671357371784, + 0.47937466569010695, + 0.41864795218124023, + 0.45053450312283344, + 0.5140124605310817, + 0.441128017496058, + 0.37240859649133107, + 0.4750271354071847, + 0.3817125590190219, + 0.5042310101510268, + 0.4296497555747094, + 0.36977921798784047, + 0.3760378221262121, + 0.3960841033597932, + 0.4969569439923279, + 0.46966737471574344, + 0.3868307446990274, + 0.345502169841327, + 0.44226060992295546, + 0.5895053677343209, + 0.6528567477128907, + 0.341907345353812, + 0.5552936930232242, + 0.6012805099230326, + 0.5834462150625513, + 0.5869381701673634, + 0.4436440169836073, + 0.5332176967866641, + 0.42373212000232974, + 0.582349074958564, + 0.46280322759940346, + 0.6540440270713072, + 0.5575438562829885, + 0.5650058640587704, + 0.334788077062891, + 0.367370409823081, + 0.421144522652133, + 0.5126182838610264, + 0.4971408092374982, + 0.5787982090330959, + 0.36294007172972426, + 0.6083590707834088, + 0.5293466397453728, + 0.5679067155847527, + 0.5575263385195032, + 0.5300657008715716, + 0.446564531216891, + 0.5603487647639462, + 0.5840812892742397, + 0.5649037333763557, + 0.5493169824996137, + 0.4602249202937165, + 0.3718235562041271, + 0.3997191497097465, + 0.4017463074924277, + 0.27183447330478216, + 0.4171376920084068, + 0.538909425284736, + 0.45310394917784463, + 0.4414430024227543, + 0.4674226228833024, + 0.4305567079053468, + 0.5395632058265453, + 0.5647932055997164, + 0.3831952482207781, + 0.5035652335959556, + 0.5808657704166645, + 0.4003371442834832, + 0.4410144627810198, + 0.5213812726683648, + 0.49086115842678935, + 0.5965787368005765, + 0.3807438850098647, + 0.5431819266593547, + 0.4012497772095828, + 0.6333875019118186, + 0.5798943751443365, + 0.6365684791016742, + 0.33994950924699796, + 0.35275723335032777, + 0.3413332864214495, + 0.31987776767186094, + 0.3401047907368778, + 0.47313041863270217, + 0.5114579473417062, + 0.3059099616047503, + 0.42134044869863263, + 0.4575773285183144, + 0.48378344615573005, + 0.555698784470877, + 0.49839150601389487, + 0.6426605423371043, + 0.33065835134910093, + 0.4719499700772957, + 0.47024987566742554, + 0.4495426407891224, + 0.44779566769022766, + 0.534622458067545, + 0.5100373994165543, + 0.30722773947670357, + 0.5462252964990216, + 0.4977545251459303, + 0.45208919475412906, + 0.591532898377704, + 0.45085140900901893, + 0.3875988798942395, + 0.4268106483706857, + 0.46450530673050744, + 0.45511092745628423, + 0.42524274731499934, + 0.45712713613078315, + 0.6001537286071266, + 0.4662666594712298, + 0.613773417079426, + 0.5181862714973775, + 0.524149302778006, + 0.5677852596329992, + 0.5281201285164472, + 0.5183772078799554, + 0.5102817138916668, + 0.6228535267844572, + 0.6431182199834545, + 0.5719789778373994, + 0.5877390904900625, + 0.7130287023485044, + 0.4780696425082264, + 0.4824079071119169, + 0.3806602611400438, + 0.5238274606608881, + 0.5586411316408609, + 0.5236315554845825, + 0.47416744865138816, + 0.561849612114842, + 0.38678844816402674, + 0.5049185952819268, + 0.4317879257295854, + 0.3158467466925612, + 0.48612858224199335, + 0.466166326820057, + 0.4763838809235383, + 0.479706981072886, + 0.3671842534510296, + 0.6429435332205343, + 0.662031103945949, + 0.5393769328837997, + 0.5744801251685668, + 0.6397808687578111, + 0.5230499249023467, + 0.3415918947726495, + 0.4455648094663599, + 0.418005999797382, + 0.39599935657613966, + 0.4198921419323007, + 0.4612848047322912, + 0.505997919559435, + 0.46466563199129995, + 0.34850136409108856 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.3515923120262079, + "valid_tracks": 4, + "per_track_pearson": [ + 0.39861709126122263, + 0.3176696086531944, + 0.3871979337949845, + 0.3028846143954301 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.5384647735855815, + "valid_tracks": 626, + "per_track_pearson": [ + 0.542205300125706, + 0.5429002764358571, + 0.5435517302326606, + 0.5474680303518085, + 0.5507979481319131, + 0.5514721129684541, + 0.48713134080252773, + 0.5093957713185456, + 0.5528056691049082, + 0.549804642804812, + 0.5447996003434304, + 0.5387890781013269, + 0.516275690447993, + 0.5283310942400813, + 0.5450195120794096, + 0.5395225367344715, + 0.5356417591242816, + 0.539765314473132, + 0.5511779115102138, + 0.5497862416770549, + 0.5580310169205955, + 0.5616422359173192, + 0.5522985873508237, + 0.5489828986695762, + 0.5343712718851474, + 0.5492039180191629, + 0.5406207292490518, + 0.541075593648219, + 0.5360571619354613, + 0.5329944442912238, + 0.5442679559591186, + 0.547599772040676, + 0.5325079618930169, + 0.5541093384167697, + 0.5439261944028614, + 0.5583904240490314, + 0.5316616187902609, + 0.5456625194446869, + 0.5426513687199435, + 0.5474875481750668, + 0.5475509096584421, + 0.542786473192497, + 0.5266430902974933, + 0.5428722209744966, + 0.5465341871712778, + 0.5206193239908531, + 0.5533790417523948, + 0.5483961542599782, + 0.549958601658005, + 0.561211716315303, + 0.5311875313595849, + 0.544927818751959, + 0.5453293655253205, + 0.5592270809459478, + 0.5454160495811842, + 0.5410131205814234, + 0.543480010450986, + 0.5504692483431255, + 0.55136361011045, + 0.5449425401051462, + 0.5277158513555299, + 0.5369397638750651, + 0.546592135202563, + 0.5449428941434022, + 0.542245888040495, + 0.5327879850950792, + 0.5466162085275188, + 0.5489129001181748, + 0.5373666447804135, + 0.5308217873390678, + 0.5413175497539724, + 0.552108302592147, + 0.5453995090181435, + 0.5389713374848212, + 0.5561146629474354, + 0.5621555911925941, + 0.5395281851624024, + 0.5457632464181648, + 0.5369972730717211, + 0.5333862599964171, + 0.5693973382607338, + 0.5513943035224802, + 0.5362536396372242, + 0.5280757281204418, + 0.5386171374702278, + 0.5391032084598427, + 0.5442278728798173, + 0.5401045605663862, + 0.5429011464081963, + 0.5410053330262105, + 0.529685496113923, + 0.5335605527869366, + 0.5464835627718201, + 0.5514356363420515, + 0.5563507167529832, + 0.5626203695190205, + 0.5419379797142413, + 0.5466018884440774, + 0.5483160906598021, + 0.5444396894853722, + 0.5246201589460312, + 0.5427035460872901, + 0.5586822200707885, + 0.5375968881372306, + 0.5485861388630076, + 0.5337922012236193, + 0.5322101664761887, + 0.5316037104960065, + 0.5561010113943456, + 0.5494901298334006, + 0.5271658094624455, + 0.5487743639571719, + 0.5620290926316767, + 0.5537591874978981, + 0.5474228918527323, + 0.5471699127474359, + 0.5360072314729326, + 0.5341697497769204, + 0.5307953875849901, + 0.5502491343826803, + 0.5593352954524726, + 0.5461612805193531, + 0.5488426194459247, + 0.5394559443182437, + 0.5409325442010489, + null, + 0.5392157066273446, + 0.5472184500998042, + 0.5301620288411231, + 0.5515949563072523, + 0.5387029729167956, + 0.5336147449958235, + 0.5440607146613703, + 0.5500798732098165, + 0.5384260710323157, + 0.5480900992686095, + null, + 0.5524425279626276, + 0.5521047211650485, + 0.5235091440553709, + 0.5361736810284768, + 0.5508047986740986, + 0.5319433806860454, + 0.5488157518439138, + 0.546849490865383, + 0.5387820792110808, + 0.5434444675499113, + 0.5524281673115564, + 0.5440753830213757, + 0.5504475165722005, + 0.551109234229882, + 0.5484429224270062, + 0.537593623672589, + 0.5434114316609838, + 0.552366806222357, + 0.5524381858154207, + 0.5555033078004762, + 0.5497911845147617, + 0.5494920667882658, + 0.5493160930738094, + 0.5461734146374485, + 0.5514593195610147, + 0.5621050789919467, + 0.5529427881160096, + 0.5346825312174551, + 0.5504789283248622, + 0.5482461221670841, + 0.5168321732680103, + 0.5262548877059002, + 0.527363772172074, + 0.5454493711081009, + 0.544704617443249, + 0.542515355354555, + 0.5428780453416072, + 0.5416302251730029, + 0.529727200550538, + 0.5476203918378874, + 0.5517124343762259, + 0.5350895007823961, + 0.5507064975267667, + 0.5452593205048607, + 0.544288353990458, + 0.5208357746067291, + 0.5529810037645398, + 0.532198474560272, + 0.5295222153410253, + 0.5588242781884947, + 0.5505612975244866, + 0.5605455602739272, + 0.5374019816579546, + 0.5445709303455118, + 0.5439599885810484, + 0.5400663583564363, + 0.5388657337736457, + 0.5380302575549337, + 0.5427195745140073, + 0.5409945516148142, + 0.5286414283109261, + 0.5622909362449989, + 0.5545855557903113, + null, + 0.5436540766981744, + 0.54970411590952, + 0.5389523829613683, + 0.5441935104056331, + 0.5563124276987416, + 0.5541294187021226, + 0.5417289883488443, + null, + null, + 0.5473450475707644, + 0.5563601809963994, + null, + 0.538295122126307, + 0.5168966117865181, + 0.5276302061842145, + 0.5452147709696336, + 0.5503967647935736, + 0.548510349628583, + 0.540612691427823, + null, + 0.5569196914869081, + 0.5373797899300651, + null, + 0.49291262203611025, + null, + 0.5233573296584314, + 0.5425307449458785, + 0.5539371594341596, + 0.5497817545872283, + null, + 0.5388508280692692, + 0.5264797736544102, + 0.5328923390723336, + null, + 0.5287968546462273, + null, + 0.5351070788767024, + null, + 0.5341758193601799, + 0.5368367621641844, + null, + 0.5248213083504769, + 0.5367339860721055, + null, + null, + 0.5470474834584258, + 0.5401564907397755, + 0.5360964779312368, + 0.5334909509029468, + null, + 0.5392851391538506, + null, + 0.5430654036609881, + 0.5542549257439384, + null, + 0.45161716159011067, + 0.53441626852681, + 0.5478513325997889, + null, + 0.5334100362423593, + 0.5507630816533052, + null, + 0.5544788353655237, + 0.5507866961910998, + 0.5457712941533129, + null, + null, + null, + null, + 0.5570180886387447, + null, + null, + 0.5486265529619476, + 0.5466123342548653, + 0.5529885877600147, + null, + 0.5561646485799486, + 0.5442042596373976, + null, + null, + 0.552576525642846, + 0.5370624325321766, + 0.5377652789621805, + 0.546643637123648, + 0.5490343772756711, + 0.544315564575344, + 0.5325912831046821, + 0.5456448370385367, + 0.5370237492304748, + 0.5375292563347224, + 0.5381426718439012, + 0.5532263083715585, + null, + 0.533213240591412, + 0.5462757432182874, + 0.5423113909211563, + 0.5353634428147633, + 0.5281914871714503, + 0.5130091937005674, + 0.5449322757328592, + 0.5267327092292642, + 0.5250533430759424, + 0.5338946830316692, + 0.5433821674632394, + 0.538138034160508, + null, + 0.5389291650589492, + 0.554982685699829, + 0.5555294188602687, + null, + 0.5385875866378591, + 0.46343496768688824, + null, + 0.550913431451748, + null, + 0.5365103168292762, + 0.5378539729487251, + 0.5482960942260433, + 0.5329554649976852, + 0.5500447007873069, + 0.5363418218971846, + 0.5237048094745, + null, + null, + 0.5373976780831331, + 0.5247760344234215, + null, + 0.543347152751075, + null, + 0.5421129229213795, + 0.5374222853706334, + 0.5363482591025011, + 0.5330772818238383, + null, + null, + null, + null, + 0.5462337766384034, + null, + 0.5466867295927673, + 0.5522192436965468, + null, + null, + 0.5408065492070918, + 0.5488345998829783, + 0.5476655311753935, + null, + 0.535773464032909, + 0.5426492113075302, + 0.5354769296967349, + null, + 0.554131102863425, + null, + null, + 0.5519296310402302, + null, + 0.5392362012261221, + null, + null, + 0.5408825368529743, + 0.5271493659992516, + 0.5308192497075012, + 0.5337165645262583, + 0.5300387882551271, + null, + 0.5286399221076025, + 0.5510147289277896, + 0.553488306708676, + 0.5111075863148045, + 0.5126141590833094, + 0.54016543704154, + 0.5528133883120308, + 0.5092989746134915, + 0.5535767430129361, + 0.5495999605441276, + 0.5648152870689558, + 0.5563743985419298, + 0.5335125181707215, + 0.5240502235368815, + 0.5265538955302054, + 0.5109349635301592, + 0.5064340714489763, + 0.5556351364397752, + 0.5633352833396481, + 0.5500821414570252, + 0.5521419124447722, + 0.5485751017570297, + 0.5421687031031472, + 0.5159798795445038, + 0.4976161269172832, + 0.5605381556257156, + 0.5501458322084385, + 0.5503655732412513, + 0.5535354951969625, + 0.5142956853951294, + 0.5058909715447448, + 0.5266464879917699, + 0.5136820755659839, + 0.4964004089818247, + 0.5294223186743048, + 0.4935217010027631, + 0.5437286793684107, + 0.4853790805650288, + 0.5516712282709644, + 0.5567306905288619, + 0.5364773169515892, + 0.5379646374046423, + 0.5196120476365185, + 0.5265427245530613, + 0.5458105595901194, + 0.5292675635896723, + 0.4968679661958138, + 0.5376335606149942, + 0.5202342180412132, + 0.5387302689646688, + 0.5208658621608359, + 0.5323594718709882, + 0.5073614855931001, + 0.5510251207661656, + 0.5377774292128827, + 0.5145740032880448, + 0.5324853106836825, + 0.5393659516395221, + 0.5426510156349136, + 0.5437776661238051, + 0.5578668224330267, + 0.5409681382503417, + 0.5590908022163492, + 0.5478439253168, + 0.5591742927214655, + 0.5486368781130612, + 0.5579485138214351, + 0.5294569229895043, + 0.5821117288132919, + 0.5702490198508838, + 0.5183082298743829, + 0.5487960592053925, + 0.5505662257492822, + 0.5359733634067705, + 0.5433520800907157, + 0.536382408222253, + 0.5410732511116171, + 0.5355081712566729, + 0.5418921918105146, + 0.538754506611411, + 0.5478634010981218, + 0.5597958106626613, + 0.5479729048235872, + 0.53589239340288, + 0.5670995621882701, + 0.5469024198712349, + 0.5397455631851973, + 0.525465552709922, + 0.5412191783848551, + 0.5395436178717591, + 0.552023246535682, + 0.5357028828898982, + 0.5386037255798617, + 0.5515693951036977, + 0.5482877233782311, + 0.5533255098440649, + 0.5561286114220113, + 0.5450162407129585, + 0.5672729459274305, + 0.5364211455402274, + 0.565848549934265, + 0.5535870131358507, + 0.5654809406360546, + 0.543772527395752, + 0.5429720213918697, + 0.5599856224146795, + 0.5541205191191746, + 0.5383559775323072, + 0.5439150787702369, + 0.5309656334472201, + 0.5316629113601842, + 0.5518283178010186, + 0.5214570788988873, + 0.5328128652671037, + 0.5378032928880572, + 0.5360026453896259, + 0.5545851087566853, + 0.5474737676400779, + 0.5589884044240936, + 0.5169044944235504, + 0.5487125100042193, + 0.5621909770421426, + 0.5682137323002479, + 0.5577930169034817, + 0.5421069031941577, + 0.5698521939824477, + null, + 0.545759783688795, + 0.5164319492489673, + 0.5407937605033204, + 0.5583901455648523, + 0.5526863225885662, + 0.5549250634436326, + 0.5606759629904148, + 0.5752765448518503, + 0.5534912506074321, + 0.5672665760895941, + null, + 0.5371584575046826, + 0.5682245608571043, + 0.5721921868565929, + 0.5633431978788194, + 0.5695883939782378, + 0.563501352029117, + 0.5524980773357699, + 0.577268459358404, + 0.5860730368345488, + 0.5834504037620059, + 0.5616268642167688, + 0.5935792722385649, + 0.5763500008745865, + 0.5765537499054951, + 0.5796988034210265, + 0.5537713933946425, + 0.5197452989128104, + 0.5104295453602241, + 0.5132280993955137, + 0.5120637737074084, + 0.5577623913013353, + 0.5435115832117758, + 0.5480940723249408, + 0.5542192390181767, + 0.5657521642802574, + 0.5450709003484779, + 0.5383479939553382, + 0.5602029209048368, + 0.552760382362587, + 0.5556240809564752, + 0.5444526830618451, + 0.5568197706038589, + 0.5569753147967569, + 0.5622480009841513, + 0.5776547969526505, + 0.5708682490794061, + 0.5540352610307518, + 0.5355465807203703, + 0.5595191149004756, + 0.5512777894611528, + 0.5592767790399836, + 0.5598951111857959, + 0.5249024112405148, + 0.5558382490936284, + 0.5562767431659671, + 0.5474466849753754, + 0.5461069331345962, + 0.5525928154298299, + 0.571390206144106, + 0.5546372105636713, + 0.5317013186137464, + 0.5470777717999973, + 0.5542235789485751, + 0.536454361279736, + 0.568641954309166, + 0.5328276392346984, + 0.5337759511436049, + 0.5286088689291022, + 0.5310720342683056, + 0.537006268160343, + 0.5435367369379301, + 0.5502176851638243, + 0.5283139398069074, + null, + 0.5140457522547804, + 0.5320528136909405, + 0.5137064474232664, + 0.5239500342761596, + 0.5330193267106194, + 0.5408842416069124, + 0.5184910585750391, + null, + null, + 0.5265944624239007, + 0.5331274993148039, + null, + 0.525204829164229, + 0.4873706188749744, + 0.5145001656222612, + 0.5107665520328545, + 0.5279174132011238, + 0.5206029878288754, + 0.5450943339567406, + null, + 0.530384353553313, + 0.5198724630689138, + null, + 0.5003373779236983, + null, + 0.5189408100485049, + 0.517536837421766, + 0.5207022394093609, + 0.5180424101491283, + null, + 0.48877902097564185, + 0.49927324343266494, + 0.514455366603331, + null, + 0.47569353703746287, + null, + 0.5218995018275449, + null, + 0.5212168131409206, + 0.5202151726626126, + null, + 0.5034205008827387, + 0.5290713909001641, + null, + null, + 0.5250372445770675, + 0.5280954189628277, + 0.49991495965711236, + 0.4912086543945544, + null, + 0.5077604403458422, + null, + 0.5171565185432745, + 0.5210018878063934, + null, + 0.4300801681421441, + 0.5069777552772019, + 0.49572844702888397, + null, + 0.5184688544685911, + 0.5256153928438396, + null, + 0.5383636997674737, + 0.5138576806007544, + 0.5297683159331159, + null, + null, + null, + null, + 0.5395680846397145, + null, + null, + 0.5245833988808266, + 0.5206518776140306, + 0.5178421881628874, + null, + 0.5374559640771818, + 0.5260009396739086, + null, + null, + 0.5211939014020932, + 0.5059757437058862, + 0.5245564595389905, + 0.5194687233219331, + 0.5348495216653756, + 0.5340805464193018, + 0.5091622309193377, + 0.5237460103589534, + 0.5031866064940468, + 0.536100424303936, + 0.5118140681212424, + 0.5304036153431814, + null, + 0.5168630011638753, + 0.505499638305992, + 0.5123153864737987, + 0.5148020241854188, + 0.5365728921693528, + 0.5159336001541422, + 0.5168597749968077, + 0.5082128962291966, + 0.5152461834440054, + 0.5145989961301111, + 0.5067124318179622, + 0.49894109611869153, + null, + 0.5103948140477439, + 0.5301831625378002, + 0.5385543798026846, + null, + 0.5188613301244281, + 0.4910882851330934, + null, + 0.5251215433664767, + null, + 0.5046166403366945, + 0.5152279375652818, + 0.5276348861937402, + 0.5231283426971478, + 0.5232275473302435, + 0.5184999660626674, + 0.5136357005795474, + null, + null, + 0.5213605002442375, + 0.533584243491818, + null, + 0.5204326230254103, + null, + 0.5395636647587432, + 0.5209166140786663, + 0.5124129724030899, + 0.5278703180803106, + null, + null, + null, + null, + 0.5335512086320325, + null, + 0.5329264572255119, + 0.5390536999576446, + null, + null, + 0.5248899672807733, + 0.5110835640821358, + 0.5230570337959025, + null, + 0.5084557406022303, + 0.5170474294698856, + 0.5168093258316653, + null, + 0.5363196765337812, + null, + null, + 0.5265459205853628, + null, + 0.5024595658034973, + null, + null, + 0.5169895248085667, + 0.5170820098959702, + 0.5249613645397839, + 0.5201964109152177, + 0.5264051598337736, + null, + 0.5147851881430668 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996, + 0.4399395998489996 + ] + }, + "junctions": { + "mean_pearson": 0.3976367025645326, + "valid_tracks": 626, + "per_track_pearson": [ + 0.3144694099004888, + 0.4461648967913132, + 0.5262036522349791, + 0.45642841319575045, + 0.5000293855535907, + 0.32425544150514085, + 0.06903258386866816, + 0.30441662447085127, + 0.3038012773077145, + 0.35954755650955744, + 0.4843254487967322, + 0.37418483982413764, + 0.28824298354063244, + 0.29314113626391475, + 0.3609278599722875, + 0.29253957121706164, + 0.3839526926082649, + 0.21571538219357672, + 0.36220502642270347, + 0.3926591513731981, + 0.3255197338960402, + 0.3643058080040621, + 0.3010654047961485, + 0.39329427144530477, + 0.40223703995864935, + 0.3140526867051788, + 0.40871562570416015, + 0.4435926282162951, + 0.26077294837969267, + 0.3871442509872395, + 0.42410249529197885, + 0.5125534331468944, + 0.4433076987868399, + 0.44061115366391146, + 0.4462170536436331, + 0.3550854546922928, + 0.4362778879775365, + 0.3715516516943737, + 0.3790700790794857, + 0.38241420440962587, + 0.3272650498281258, + 0.320376892279498, + 0.4163177040283305, + 0.4421154196396811, + 0.37080249985423724, + 0.3744403279937389, + 0.2628699350030754, + 0.44035682579079993, + 0.37685129735906303, + 0.4235602095462945, + 0.3855653366059921, + 0.33598834691864193, + 0.3125145456082201, + 0.39964151448896196, + 0.3284850629342318, + 0.40101874159766476, + 0.3470373231321684, + 0.3715432737685818, + 0.26913094553343975, + 0.3707371997043427, + 0.2715650059701851, + 0.2709530360610899, + 0.29953029370802886, + 0.23232070226207144, + 0.2948294950736506, + 0.33293339697058977, + 0.3980125778005228, + 0.39633891864817505, + 0.38358102788405196, + 0.33208768460991656, + 0.3159908831227007, + 0.2893662615813663, + 0.36389356264233236, + 0.27275554863727297, + 0.32872053924459577, + 0.43355228907061033, + 0.40885639501394505, + 0.3748977496539748, + 0.24423499536695137, + 0.45427139377138775, + 0.43131729616275793, + 0.3329921573756503, + 0.32326470582921313, + 0.29993407195952443, + 0.4376439513861658, + 0.3520707591350834, + 0.3437475703657997, + 0.4400556694330751, + 0.2916057006540688, + 0.442009457640127, + 0.377658896371429, + 0.4577828914055688, + 0.37275880336977474, + 0.28204424807632594, + 0.33100446001477146, + 0.3827180345762629, + 0.3834667546971514, + 0.36675166294456324, + 0.4100550835950479, + 0.2785393781865221, + 0.35708144703986594, + 0.48544260696334696, + 0.40634159196196473, + 0.30073295046946974, + 0.3390581622715078, + 0.25704567500237957, + 0.34842132180604646, + 0.4941190540396693, + 0.3408164044783601, + 0.38064493440660996, + 0.42810550323549734, + 0.37853863293800966, + 0.47667827360340126, + 0.4471776097463946, + 0.3717815019557101, + 0.31052892066692633, + 0.3052924158584259, + 0.40775496396483696, + 0.4146104247176592, + 0.46470206408329195, + 0.40432249498917305, + 0.31855280143339565, + 0.40713853288155755, + 0.3251798953563983, + 0.3170083181825096, + null, + 0.43041307316270416, + 0.31956662489442633, + 0.15298043355560856, + 0.283367606254081, + 0.36978673035484566, + 0.2554058431541201, + 0.2704115738763756, + 0.33125459964562126, + 0.34161904888218797, + 0.4087927251039333, + null, + 0.28388051204352294, + 0.36307636456366593, + 0.30456129950536226, + 0.37889106620411783, + 0.3776437143804962, + 0.30785022362306724, + 0.3153566582686697, + 0.3051191389168911, + 0.19176066555096902, + 0.32307547898426736, + 0.41176980983913203, + 0.2848330450128674, + 0.30048409086671696, + 0.2854824830795814, + 0.22425708736369443, + 0.29399061959498596, + 0.4787062691398811, + 0.4059474776994337, + 0.4270018670552814, + 0.405758150385265, + 0.32083294246444066, + 0.28503375074741516, + 0.16131622215964425, + 0.2810523389407348, + 0.21608524065620713, + 0.42804988405193634, + 0.30793961239454215, + 0.19870704489509114, + 0.4291561988482432, + 0.36450625732183195, + 0.3150187139978473, + 0.27930598123798345, + 0.2932238080094227, + 0.39493822236728077, + 0.32100533732966924, + 0.4692756985276812, + 0.27734219149542966, + 0.2903827021722703, + 0.49659334159754237, + 0.2618987928982211, + 0.3093653315308584, + 0.23905459062754864, + 0.3602999606752404, + 0.36903305263413966, + 0.31244057519720725, + 0.3505325029400054, + 0.3763152917711038, + 0.3197058665655048, + 0.31157447573701963, + 0.33468711495563896, + 0.36113180647252824, + 0.34558059483034387, + 0.40076767773388994, + 0.39065217922201695, + 0.3998685138762595, + 0.35490548913629427, + 0.3496906006290335, + 0.30481891753057044, + 0.3356608737685403, + 0.3249261938420738, + 0.3539012401637925, + 0.38774608769879715, + 0.3034415997348116, + null, + 0.3558551693249087, + 0.39852635187604857, + 0.4073121294142938, + 0.45378514299035727, + 0.3157164622766576, + 0.37295372852637754, + 0.3049074293528774, + null, + null, + 0.4763041767550773, + 0.4960918802807603, + null, + 0.3756981670997541, + 0.3423462331568425, + 0.36205550297462147, + 0.29704810448069185, + 0.35885113277605873, + 0.3454446816538487, + 0.21566325722680538, + null, + 0.42459341388504035, + 0.40354690788505976, + null, + 0.32839053534775403, + null, + 0.34862367526486066, + 0.44408628965270514, + 0.2781768744374192, + 0.31736751876981567, + null, + 0.19806587352836175, + 0.21147790588346138, + 0.382753836242566, + null, + 0.28001180541566983, + null, + 0.296210943968728, + null, + 0.2029895518338826, + 0.30929372021669166, + null, + 0.3511666908327803, + 0.3202638227979518, + null, + null, + 0.2720963829289451, + 0.42559222050765355, + 0.23821478884399425, + 0.3168683635076884, + null, + 0.3994648584478909, + null, + 0.39152455167328437, + 0.4324384457456739, + null, + 0.29803490242634756, + 0.43371375696530995, + 0.38347041338056803, + null, + 0.39873101115030224, + 0.43492718779354445, + null, + 0.47201878715568185, + 0.5241965747001287, + 0.4190934807185186, + null, + null, + null, + null, + 0.28275487201044286, + null, + null, + 0.3251916783414894, + 0.3040515342620018, + 0.46721809884614174, + null, + 0.3941692170043025, + 0.4564773202892017, + null, + null, + 0.37067205648242935, + 0.36210960043616053, + 0.4394489800117631, + 0.34539900587526906, + 0.39546293529154064, + 0.33919642407124484, + 0.37642119267468616, + 0.3004904393868226, + 0.3837416242405357, + 0.25589333776152606, + 0.4192563545266386, + 0.3586327538916837, + null, + 0.40766343775909464, + 0.3735953475552166, + 0.19645480490004036, + 0.34232195204440885, + 0.31210175637029547, + 0.3846638683951049, + 0.2596790285718345, + 0.3716628125702255, + 0.44652186545967576, + 0.2664975276089534, + 0.34265964857922776, + 0.3660130707927724, + null, + 0.27263171659162133, + 0.3358600288297316, + 0.3703536448610556, + null, + 0.3535563927138006, + 0.2953389957533457, + null, + 0.4071656630479159, + null, + 0.3866177462767979, + 0.40665733265840803, + 0.4532743022015819, + 0.2355817020755228, + 0.3038520485618849, + 0.45777209949835945, + 0.37506084100456877, + null, + null, + 0.26892672749055707, + 0.3692109721894391, + null, + 0.3303523982496593, + null, + 0.28511906372006085, + 0.3384879873970896, + 0.2697054631862977, + 0.4484301704188108, + null, + null, + null, + null, + 0.2012060203422726, + null, + 0.32598642234601105, + 0.38956232039736866, + null, + null, + 0.32944217229377765, + 0.2946947187879747, + 0.2869343464317844, + null, + 0.42589037477933556, + 0.4244207966131235, + 0.24291164483644773, + null, + 0.46696905022270774, + null, + null, + 0.2957991343943957, + null, + 0.14183155688674323, + null, + null, + 0.25084749351017593, + 0.3967495883911213, + 0.38382249815421327, + 0.4376419963183827, + 0.45504181819550626, + null, + 0.3144207408565899, + 0.4750439780839599, + 0.34201734862737004, + 0.538710810549719, + 0.43009442868212744, + 0.5477541198871717, + 0.40207241319424175, + 0.36522125727484817, + 0.37267226451970853, + 0.4742699240017453, + 0.4572136668876295, + 0.3038165906318429, + 0.40335263466919197, + 0.36320729669792895, + 0.3749924498305771, + 0.39279593932502127, + 0.3611087847910261, + 0.43105134632223424, + 0.5393013014822374, + 0.5016590099435629, + 0.4997265554469003, + 0.5163473574000709, + 0.5236853500442709, + 0.353358514647675, + 0.6103650469362438, + 0.4210105273163843, + 0.4579607536047262, + 0.4486158141136844, + 0.5034970170082245, + 0.5120863196144496, + 0.320574843825043, + 0.4409375872249783, + 0.5158198106567289, + 0.4629069705420879, + 0.5474076130017763, + 0.4021736836198037, + 0.38001149567987513, + 0.3170339084028415, + 0.38102809978670826, + 0.471244583702983, + 0.383084031800025, + 0.4409120489948779, + 0.299007132809879, + 0.31064859498233255, + 0.33784686959445387, + 0.3629451292877557, + 0.35825803671372636, + 0.39290567871882, + 0.433346243828452, + 0.30533373342698483, + 0.4752744405949506, + 0.28340775727966877, + 0.42202372810377375, + 0.31292556114340286, + 0.40602608842710247, + 0.3478396936465679, + 0.40402650669242324, + 0.3733538805228574, + 0.42314990583531925, + 0.4015513990340701, + 0.4807085073140839, + 0.4995167708217882, + 0.3331602639030599, + 0.47348368930209034, + 0.43872640959080533, + 0.35810249037298225, + 0.41395942335250013, + 0.4374576618546285, + 0.425272353122361, + 0.4339817881559538, + 0.38873459743780175, + 0.35474521969258516, + 0.47818633997801446, + 0.49468385296657424, + 0.4897372323904227, + 0.4835195179795705, + 0.5138863490416358, + 0.4776873014730849, + 0.4388401815283408, + 0.46657363256440676, + 0.46536202689414125, + 0.4468257911041386, + 0.4920625665179861, + 0.4622392582304336, + 0.3713337897133515, + 0.4830993936312822, + 0.4389516117962291, + 0.49884818940533926, + 0.5492389746924623, + 0.44712149517147065, + 0.4346004811730015, + 0.5455481882598214, + 0.44185809270976756, + 0.4635182540220372, + 0.4218836823295941, + 0.41809077351511076, + 0.4740471652749534, + 0.43977117875513977, + 0.4808623453573485, + 0.5334462482787028, + 0.5535291339196636, + 0.3698266850095288, + 0.505340607420977, + 0.495984242972427, + 0.4305617592735816, + 0.5073259854272275, + 0.5291299671562846, + 0.33838346889817605, + 0.4481473130771485, + 0.4140576233197297, + 0.4459631652047371, + 0.25821493868080825, + 0.462293997925254, + 0.45697693099280073, + 0.48719339311016197, + 0.470630083550714, + 0.3874979839997623, + 0.4976296515610798, + 0.26635288105508115, + 0.3612825550739203, + 0.39988137282893654, + 0.5608874335234794, + 0.3972963030684165, + 0.489580246475214, + 0.4092551019113595, + 0.5011816330438004, + null, + 0.4342974727994962, + 0.5090760176809294, + 0.3424946481154256, + 0.5284538665503381, + 0.47056788852566844, + 0.4345026147958394, + 0.5030761123926847, + 0.5215054468821061, + 0.5106168769038619, + 0.4280399812759375, + null, + 0.48764479386086484, + 0.4520776563100883, + 0.371837457134025, + 0.40336567448278204, + 0.5004525754848423, + 0.4288878725844793, + 0.5190167783474086, + 0.5569519289545053, + 0.3442492469921906, + 0.39790037885381635, + 0.5249961530376307, + 0.38539109248215897, + 0.46955258691189156, + 0.43217332634474986, + 0.5136344292325534, + 0.4727630621809196, + 0.5484657337809192, + 0.6678897815827265, + 0.4549437001218062, + 0.5140070098228396, + 0.5049712546080964, + 0.4811762263958313, + 0.4120034208887336, + 0.42150838741009805, + 0.4271395187961159, + 0.49529438941330484, + 0.5693963953813432, + 0.5211535529861712, + 0.4862197760921432, + 0.5072424662583936, + 0.4696111910406838, + 0.37755005297308586, + 0.4127726538471068, + 0.41307290083838494, + 0.45742295870546346, + 0.40061011694328613, + 0.3061417185198615, + 0.5177441993809548, + 0.4518044927198566, + 0.5399031972399334, + 0.49138416156127795, + 0.44990614359541914, + 0.4810525369156705, + 0.3538085791634094, + 0.4249089262109495, + 0.5009463717286228, + 0.6237092672356102, + 0.5049693016459866, + 0.6016489408376587, + 0.5707048646150467, + 0.35984522592060286, + 0.5358768350379572, + 0.45546020627194284, + 0.613291555168487, + 0.40312667108839, + 0.3902828884720085, + 0.2803054720148861, + 0.3406987410788332, + 0.3418499883360654, + 0.36982520043322936, + 0.4812288998988411, + 0.5844573260166995, + 0.5891760922161605, + null, + 0.42088208414002226, + 0.47455664585668417, + 0.610463030407239, + 0.5417659862436496, + 0.5044716141945315, + 0.3166783494046118, + 0.4124541985512173, + null, + null, + 0.5386133496494748, + 0.5131572048597672, + null, + 0.25617143869286885, + 0.5052780723329545, + 0.2952811874442121, + 0.3590867976231164, + 0.4194358456163155, + 0.35171064106188177, + 0.39854314311642924, + null, + 0.5005080552179756, + 0.5031033771348772, + null, + 0.37963878657546235, + null, + 0.5171666295210972, + 0.487676201735996, + 0.3652913634416742, + 0.38766924560835564, + null, + 0.5106794821511179, + 0.3960504629204382, + 0.4932456224874553, + null, + 0.4248698588719148, + null, + 0.42608890217739137, + null, + 0.39418980735574966, + 0.4924487509615067, + null, + 0.45727314807791947, + 0.4845619079245766, + null, + null, + 0.4592131305066477, + 0.4205504626603431, + 0.1828500008869811, + 0.3414632906151262, + null, + 0.5102322770367088, + null, + 0.2891906458019209, + 0.5421630516899774, + null, + 0.22012968188635762, + 0.47996023929626935, + 0.3854011263082773, + null, + 0.42933165624210434, + 0.5781525521574695, + null, + 0.4404563803733097, + 0.49002088409258965, + 0.49857229330306674, + null, + null, + null, + null, + 0.43649754209915637, + null, + null, + 0.5595513169851202, + 0.42485652747810737, + 0.5215088036900929, + null, + 0.42147919638385517, + 0.48742164772215263, + null, + null, + 0.38592828066552093, + 0.1454237767526141, + 0.4741945956893745, + 0.28710855659222323, + 0.34931936604961444, + 0.43782617482987285, + 0.20109965961996243, + 0.3494275367362168, + 0.22419887772695835, + 0.40921176922504143, + 0.5654061747222782, + 0.3230945597459767, + null, + 0.2744266599153487, + 0.4896680594287639, + 0.39176853636837605, + 0.5211423225629537, + 0.23940717954326995, + 0.45894575965541845, + 0.4745564089755678, + 0.4311255413818845, + 0.5127589018510478, + 0.4929636217695843, + 0.4868111749078781, + 0.3900789926190375, + null, + 0.4189827709793404, + 0.519416426624646, + 0.4075601852350675, + null, + 0.5571941177686126, + 0.4258765938161883, + null, + 0.4878846852969122, + null, + 0.47323124736585365, + 0.5547078286709609, + 0.355213165266559, + 0.5432854162112192, + 0.492910641661455, + 0.46606034635525845, + 0.5610578436914724, + null, + null, + 0.4873452542622871, + 0.5380811528572983, + null, + 0.41659501764013723, + null, + 0.38845849427099777, + 0.5191960967879011, + 0.4989480014959509, + 0.5024569057572398, + null, + null, + null, + null, + 0.4121356951764932, + null, + 0.5099436046934718, + 0.6091744053804027, + null, + null, + 0.5315522878226325, + 0.29416424889258547, + 0.39069074884966654, + null, + 0.3769278450707322, + 0.43377516463483307, + 0.49880881450245673, + null, + 0.32333369012943125, + null, + null, + 0.3910501170280976, + null, + 0.34447602103991526, + null, + null, + 0.4438422855739888, + 0.5073126100759654, + 0.5418942480687635, + 0.5619863227249691, + 0.5012178036766497, + null, + 0.4845272693460711 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.8543046357615894, + 0.8651315789473685, + 0.8398576512455516, + 0.8069620253164557, + 0.8807947019867549, + 0.847457627118644, + 0.7992957746478874, + 0.8506944444444444, + 0.8589743589743589, + 0.8439716312056738, + 0.835016835016835, + 0.8406779661016949, + 0.8474025974025974, + 0.859375, + 0.8533333333333334, + 0.8388157894736842, + 0.8181818181818182, + 0.8141025641025641, + 0.8662420382165605, + 0.8366013071895425, + 0.8453237410071942, + 0.8419354838709677, + 0.8344155844155844, + 0.8104089219330854, + 0.8362989323843416, + 0.8651315789473685, + 0.8846153846153846, + 0.8542372881355932, + 0.8445945945945946, + 0.8310344827586207, + 0.8116438356164384, + 0.86084142394822, + 0.8246268656716418, + 0.851063829787234, + 0.8205128205128205, + 0.8422712933753943, + 0.7977099236641222, + 0.8338557993730408, + 0.8076923076923077, + 0.8657718120805369, + 0.8395904436860068, + 0.8419243986254296, + 0.816793893129771, + 0.8296296296296296, + 0.8295081967213115, + 0.831081081081081, + 0.8339483394833949, + 0.7956989247311828, + 0.7943262411347518, + 0.8333333333333334, + 0.8248175182481752, + 0.8455882352941176, + 0.8181818181818182, + 0.8299319727891157, + 0.7873134328358209, + 0.8197278911564626, + 0.8131487889273357, + 0.8675496688741722, + 0.8424657534246576, + 0.8169014084507042, + 0.8379310344827586, + 0.8688524590163934, + 0.8609271523178808, + 0.8464163822525598, + 0.8288590604026845, + 0.8639240506329114, + 0.847972972972973, + 0.8355704697986577, + 0.8293515358361775, + 0.8496732026143791, + 0.8612099644128114, + 0.8419354838709677, + 0.8417508417508418, + 0.8675496688741722, + 0.8422939068100358, + 0.8610169491525423, + 0.8520900321543409, + 0.8528428093645485, + 0.8586206896551725, + 0.8673469387755102, + 0.8410596026490066, + 0.8639240506329114, + 0.8698630136986302, + 0.8284789644012945, + 0.8631578947368421, + 0.8606811145510835, + 0.8595890410958904, + 0.8333333333333334, + 0.8571428571428571, + 0.8512658227848101, + 0.8432055749128919, + 0.8247422680412371, + 0.8596491228070176, + 0.8299319727891157, + 0.831081081081081, + 0.8333333333333334, + 0.8610169491525423, + 0.8235294117647058, + 0.8675496688741722, + 0.8639240506329114, + 0.83, + 0.8557377049180328, + 0.8220640569395018, + 0.8722044728434505, + 0.8703071672354948, + 0.87, + 0.8127340823970037, + 0.8662207357859532, + 0.8504983388704319, + 0.8445945945945946, + 0.8464285714285714, + 0.825503355704698, + 0.8506493506493507, + 0.8271186440677966, + 0.8503401360544217, + 0.8581081081081081, + 0.8172413793103448, + 0.8344594594594594, + 0.8267148014440433, + 0.856655290102389, + 0.8357142857142857, + 0.8542372881355932, + 0.8308823529411765, + 0.8552631578947368, + 0.8097165991902834, + 0.9919354838709677, + 0.8546712802768166, + 0.8362369337979094, + 0.875, + 0.8507462686567164, + 0.8188405797101449, + 0.8689655172413793, + 0.8571428571428571, + 0.8195488721804511, + 0.8210526315789474, + 0.8327759197324415, + 0.9838709677419355, + 0.864406779661017, + 0.835820895522388, + 0.8717105263157895, + 0.864963503649635, + 0.8022388059701493, + 0.8308823529411765, + 0.8689655172413793, + 0.8350877192982457, + 0.8129770992366412, + 0.8110236220472441, + 0.8402777777777778, + 0.8051470588235294, + 0.833922261484099, + 0.8074074074074075, + 0.8262411347517731, + 0.8241758241758241, + 0.8766233766233766, + 0.8448275862068966, + 0.8350515463917526, + 0.8484848484848485, + 0.8447653429602888, + 0.8409893992932862, + 0.8503649635036497, + 0.8068181818181818, + 0.8472222222222222, + 0.877742946708464, + 0.8432055749128919, + 0.8075471698113208, + 0.8345323741007195, + 0.8546712802768166, + 0.8741721854304636, + 0.8188679245283019, + 0.8421052631578947, + 0.8351648351648352, + 0.8191126279863481, + 0.849624060150376, + 0.8321917808219178, + 0.85, + 0.8326848249027238, + 0.8478260869565217, + 0.8432055749128919, + 0.8492647058823529, + 0.8496503496503497, + 0.8257839721254355, + 0.8283582089552238, + 0.8228782287822878, + 0.8458781362007168, + 0.8450704225352113, + 0.8634686346863468, + 0.8527397260273972, + 0.8619528619528619, + 0.8387096774193549, + 0.8655737704918033, + 0.8541666666666666, + 0.8602941176470589, + 0.8535714285714285, + 0.8098591549295775, + 0.8292682926829268, + 0.8127208480565371, + 0.8051470588235294, + 0.8368055555555556, + 0.8660130718954249, + 0.8566666666666667, + 0.9922480620155039, + 0.868421052631579, + 0.823728813559322, + 0.8205980066445183, + 0.8581081081081081, + 0.8585209003215434, + 0.8374558303886925, + 0.8355263157894737, + 1.0, + 0.9924242424242424, + 0.8910891089108911, + 0.8733766233766234, + 0.9846153846153847, + 0.8517241379310345, + 0.7692307692307693, + 0.8065693430656934, + 0.8395904436860068, + 0.8377483443708609, + 0.8241379310344827, + 0.8489932885906041, + 0.983739837398374, + 0.860248447204969, + 0.8213058419243986, + 0.9849624060150376, + 0.765625, + 0.9788732394366197, + 0.7963636363636364, + 0.8421052631578947, + 0.8556338028169014, + 0.835016835016835, + 0.9782608695652174, + 0.8333333333333334, + 0.8028169014084507, + 0.8456140350877193, + 0.984251968503937, + 0.8013468013468014, + 0.9923076923076923, + 0.8344155844155844, + 0.9915254237288136, + 0.8466666666666667, + 0.8456375838926175, + 0.9635036496350365, + 0.8327137546468402, + 0.8459016393442623, + 0.9612403100775194, + 0.9609375, + 0.8368794326241135, + 0.8496732026143791, + 0.8496732026143791, + 0.835016835016835, + 1.0, + 0.8162544169611308, + 0.9689922480620154, + 0.8246753246753247, + 0.845360824742268, + 0.9930555555555556, + 0.8038461538461539, + 0.8231511254019293, + 0.8327272727272728, + 0.9710144927536232, + 0.8285714285714286, + 0.8561872909698997, + 0.9752066115702479, + 0.8524590163934426, + 0.8280701754385965, + 0.8440677966101695, + 0.9850746268656716, + 0.9922480620155039, + 0.9922480620155039, + 0.9916666666666667, + 0.8172757475083057, + 0.9858156028368794, + 1.0, + 0.8367346938775511, + 0.8262411347517731, + 0.842948717948718, + 1.0, + 0.8361774744027304, + 0.8527397260273972, + 1.0, + 0.9846153846153847, + 0.8501742160278746, + 0.8115942028985508, + 0.8494983277591973, + 0.8378378378378378, + 0.8293515358361775, + 0.8027210884353742, + 0.8361774744027304, + 0.8166089965397924, + 0.8203389830508474, + 0.8113879003558719, + 0.8535031847133758, + 0.8339483394833949, + 0.9776119402985075, + 0.8421052631578947, + 0.847972972972973, + 0.8428571428571429, + 0.8518518518518519, + 0.8401360544217688, + 0.8506944444444444, + 0.8456140350877193, + 0.8172413793103448, + 0.839344262295082, + 0.8141891891891891, + 0.7934782608695652, + 0.8088737201365188, + 0.983739837398374, + 0.8360128617363344, + 0.8513513513513513, + 0.842948717948718, + 1.0, + 0.8434504792332268, + 0.8100358422939068, + 0.9669421487603306, + 0.8384879725085911, + 0.9925373134328358, + 0.8653846153846154, + 0.8333333333333334, + 0.858085808580858, + 0.8188405797101449, + 0.8435374149659864, + 0.8287671232876712, + 0.8216783216783217, + 0.9929078014184397, + 1.0, + 0.8496732026143791, + 0.8163265306122449, + 0.9626865671641791, + 0.8339350180505415, + 0.9846153846153847, + 0.8305647840531561, + 0.8288590604026845, + 0.8181818181818182, + 0.8504983388704319, + 0.9928571428571429, + 1.0, + 0.992, + 0.9841269841269841, + 0.8047945205479452, + 0.9927007299270073, + 0.8408304498269896, + 0.8666666666666667, + 0.9826086956521739, + 1.0, + 0.8450704225352113, + 0.840531561461794, + 0.8185185185185185, + 0.992, + 0.8356164383561644, + 0.8484848484848485, + 0.8120567375886525, + 0.9838709677419355, + 0.8662420382165605, + 0.9850746268656716, + 0.9922480620155039, + 0.8430034129692833, + 0.9672131147540983, + 0.8478964401294499, + 0.984251968503937, + 0.9767441860465116, + 0.7837837837837838, + 0.8409090909090909, + 0.8419243986254296, + 0.852233676975945, + 0.8380281690140845, + 1.0, + 0.8278145695364238, + 0.793233082706767, + 0.7929824561403509, + 0.8043478260869565, + 0.8214285714285714, + 0.879746835443038, + 0.7935943060498221, + 0.7874564459930313, + 0.7245283018867924, + 0.8122866894197952, + 0.7949640287769785, + 0.8096885813148789, + 0.8050541516245487, + 0.8082191780821918, + 0.821917808219178, + 0.8104089219330854, + 0.7838827838827839, + 0.7887323943661971, + 0.762962962962963, + 0.8035714285714286, + 0.8191126279863481, + 0.8146853146853147, + 0.8056537102473498, + 0.8260869565217391, + 0.8270676691729323, + 0.775, + 0.8145454545454546, + 0.8245033112582781, + 0.8257839721254355, + 0.8207885304659498, + 0.782608695652174, + 0.8140350877192982, + 0.8125, + 0.7943548387096774, + 0.8297872340425532, + 0.8095238095238095, + 0.8120567375886525, + 0.7729083665338645, + 0.7695035460992907, + 0.7972027972027972, + 0.7769784172661871, + 0.7640845070422535, + 0.843065693430657, + 0.7950530035335689, + 0.792156862745098, + 0.8178694158075601, + 0.802158273381295, + 0.7744360902255639, + 0.8206106870229007, + 0.7849056603773585, + 0.7807692307692308, + 0.7951807228915663, + 0.8104089219330854, + 0.7613636363636364, + 0.8149466192170819, + 0.7857142857142857, + 0.8029739776951673, + 0.8309352517985612, + 0.7992831541218638, + 0.7758007117437722, + 0.7862318840579711, + 0.8048780487804879, + 0.8239700374531835, + 0.8327645051194539, + 0.8241758241758241, + 0.8086642599277978, + 0.8088235294117647, + 0.7964285714285714, + 0.7258687258687259, + 0.7908745247148289, + 0.7885304659498208, + 0.8321167883211679, + 0.7789855072463768, + 0.8101694915254237, + 0.7921146953405018, + 0.8100358422939068, + 0.7934782608695652, + 0.7827586206896552, + 0.7950530035335689, + 0.7985611510791367, + 0.8113207547169812, + 0.8062283737024222, + 0.7945205479452054, + 0.7908745247148289, + 0.7777777777777778, + 0.8014184397163121, + 0.7977941176470589, + 0.8015267175572519, + 0.8159722222222222, + 0.7773722627737226, + 0.7948717948717948, + 0.8083623693379791, + 0.797153024911032, + 0.8172413793103448, + 0.79182156133829, + 0.7684210526315789, + 0.7956989247311828, + 0.7889273356401384, + 0.7843866171003717, + 0.8013698630136986, + 0.7956989247311828, + 0.7662835249042146, + 0.7883211678832117, + 0.8079710144927537, + 0.8071428571428572, + 0.8152173913043478, + 0.8194444444444444, + 0.7962264150943397, + 0.8037037037037037, + 0.7964912280701755, + 0.800711743772242, + 0.795774647887324, + 0.8052434456928839, + 0.804635761589404, + 0.7992831541218638, + 0.8028673835125448, + 0.7692307692307693, + 0.7985865724381626, + 0.75, + 0.758893280632411, + 0.7785977859778598, + 0.8175438596491228, + 0.7849462365591398, + 0.7655677655677655, + 0.823321554770318, + 0.7620817843866171, + 0.9666666666666667, + 0.8014184397163121, + 0.8142857142857143, + 0.8156028368794326, + 0.7984189723320159, + 0.7870036101083032, + 0.8146853146853147, + 0.7950530035335689, + 0.7781954887218046, + 0.8257839721254355, + 0.7480916030534351, + 0.9813084112149533, + 0.818815331010453, + 0.7647058823529411, + 0.8496503496503497, + 0.8710801393728222, + 0.7509293680297398, + 0.7907801418439716, + 0.7867132867132867, + 0.7907801418439716, + 0.7509293680297398, + 0.7723880597014925, + 0.7715355805243446, + 0.7581227436823105, + 0.7948717948717948, + 0.7307692307692307, + 0.7601476014760148, + 0.8028673835125448, + 0.8191881918819188, + 0.8037735849056604, + 0.8482758620689655, + 0.8241758241758241, + 0.8195488721804511, + 0.8040540540540541, + 0.8037037037037037, + 0.7753623188405797, + 0.7472527472527473, + 0.8543689320388349, + 0.7832167832167832, + 0.7790262172284644, + 0.7795275590551181, + 0.7842465753424658, + 0.8279569892473119, + 0.7387755102040816, + 0.7843866171003717, + 0.8115942028985508, + 0.7666666666666667, + 0.7906976744186046, + 0.8127208480565371, + 0.8044280442804428, + 0.7642276422764228, + 0.7872340425531915, + 0.8214285714285714, + 0.7832167832167832, + 0.7992700729927007, + 0.7979094076655052, + 0.7955390334572491, + 0.7683823529411765, + 0.7912087912087912, + 0.8274647887323944, + 0.7901234567901234, + 0.7955390334572491, + 0.7985611510791367, + 0.7605633802816901, + 0.8214285714285714, + 0.8350515463917526, + 0.7822878228782287, + 0.8268551236749117, + 0.7675276752767528, + 0.7706093189964157, + 0.7841726618705036, + 0.778169014084507, + 0.7423076923076923, + 0.867741935483871, + 0.825938566552901, + 0.9914529914529915, + 0.8368794326241135, + 0.8316498316498316, + 0.7761194029850746, + 0.8168498168498168, + 0.8458904109589042, + 0.8129251700680272, + 0.7667844522968198, + 0.9920634920634921, + 0.9833333333333333, + 0.8934169278996865, + 0.8599348534201955, + 0.984, + 0.8210526315789474, + 0.7612456747404844, + 0.7730769230769231, + 0.8156996587030717, + 0.818815331010453, + 0.8042704626334519, + 0.8056537102473498, + 0.9921875, + 0.8172043010752689, + 0.7925925925925926, + 0.9923076923076923, + 0.746268656716418, + 1.0, + 0.7661016949152543, + 0.8434163701067615, + 0.8412162162162162, + 0.7859649122807018, + 0.9851851851851852, + 0.8021978021978022, + 0.7929824561403509, + 0.7794117647058824, + 0.9922480620155039, + 0.7793594306049823, + 1.0, + 0.7915194346289752, + 0.9920634920634921, + 0.7602996254681648, + 0.8109090909090909, + 0.9924812030075187, + 0.7854545454545454, + 0.8243727598566308, + 0.9851851851851852, + 1.0, + 0.8275862068965517, + 0.8197879858657244, + 0.8157894736842105, + 0.8181818181818182, + 0.9745762711864406, + 0.7859922178988327, + 1.0, + 0.7836879432624113, + 0.7985074626865671, + 0.9920634920634921, + 0.7638376383763837, + 0.7832167832167832, + 0.7806691449814126, + 0.9844961240310077, + 0.7898550724637681, + 0.8063380281690141, + 0.957983193277311, + 0.8327645051194539, + 0.8160535117056856, + 0.8351254480286738, + 0.9926470588235294, + 0.9919354838709677, + 0.9920634920634921, + 0.990990990990991, + 0.8408304498269896, + 1.0, + 0.981651376146789, + 0.8014184397163121, + 0.8299319727891157, + 0.8378378378378378, + 1.0, + 0.8216783216783217, + 0.8178807947019867, + 0.957983193277311, + 0.976, + 0.8131487889273357, + 0.8316151202749141, + 0.8109090909090909, + 0.8217821782178217, + 0.8245614035087719, + 0.8138686131386861, + 0.7647058823529411, + 0.8163265306122449, + 0.7924528301886793, + 0.8, + 0.8305084745762712, + 0.7978723404255319, + 1.0, + 0.8194444444444444, + 0.8028673835125448, + 0.8231292517006803, + 0.8070175438596491, + 0.8006993006993007, + 0.8006993006993007, + 0.8047945205479452, + 0.8205128205128205, + 0.8355704697986577, + 0.7985865724381626, + 0.8, + 0.7818181818181819, + 0.9921259842519685, + 0.8107142857142857, + 0.8285714285714286, + 0.86, + 1.0, + 0.8166089965397924, + 0.7892976588628763, + 1.0, + 0.8163934426229508, + 0.9836065573770492, + 0.8305647840531561, + 0.8220640569395018, + 0.847972972972973, + 0.7568627450980392, + 0.8184931506849316, + 0.7723880597014925, + 0.7566539923954373, + 0.9918032786885246, + 0.991869918699187, + 0.795774647887324, + 0.7455197132616488, + 0.9922480620155039, + 0.8113879003558719, + 0.9739130434782609, + 0.8020477815699659, + 0.8027681660899654, + 0.7902097902097902, + 0.8125, + 0.9854014598540146, + 0.9921259842519685, + 0.9830508474576272, + 1.0, + 0.8063380281690141, + 1.0, + 0.7846153846153846, + 0.8383838383838383, + 0.975609756097561, + 0.9923076923076923, + 0.8172043010752689, + 0.8245614035087719, + 0.8, + 0.957983193277311, + 0.7992831541218638, + 0.8153310104529616, + 0.8020477815699659, + 1.0, + 0.8603896103896104, + 0.983739837398374, + 1.0, + 0.8254545454545454, + 0.9833333333333333, + 0.8075471698113208, + 0.981651376146789, + 0.992, + 0.7822878228782287, + 0.8408304498269896, + 0.8421052631578947, + 0.8216783216783217, + 0.8135593220338984, + 0.9747899159663865, + 0.8203389830508474 + ] + }, + "contact_maps": { + "mean_pearson": 0.34286500464058006, + "valid_tracks": 28, + "per_track_pearson": [ + 0.3314899050306271, + 0.3533808678409344, + 0.39320887850597075, + 0.3127368561965061, + 0.2545524262859985, + 0.29031747692288773, + 0.3520561666360796, + 0.26252692702788816, + 0.42458795914640635, + 0.38589848840694113, + 0.3125241753928112, + 0.36937374301380677, + 0.38532350338679017, + 0.2940503908039734, + 0.3277380074184395, + 0.4434598894035502, + 0.10115480312042216, + 0.372436533813084, + 0.35400636681685954, + 0.36461355373799703, + 0.31970423034479417, + 0.32388285532171923, + 0.3159820670985887, + 0.38169500181700683, + 0.39360171668404964, + 0.4019688833602014, + 0.35978573016103127, + 0.41816272624087836 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.4780742371477073, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..a81fb207e30b9e0ee71e38b2715e3e855812d39e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S027/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4532644206566745, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5505868113010971, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.045759921671091254, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7640385643548026, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.3882757372991099, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.353250081324716, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.31253915820582545, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.3964986374957853, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.1620880116273098, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.20412277527444903, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.44938219703449467, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.18170733528093427, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.2829607293455378, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/README.md new file mode 100644 index 0000000000000000000000000000000000000000..609918513138133353b0e6d267f2caa18b8b9909 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/README.md @@ -0,0 +1,99 @@ +# SeqFunc scale13-small after 8 complete epochs with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. At the saved small96k/large78k boundaries, an optional training-only byte cache was enabled after exact256example and integrated16example tensor/ID checks. Source compressed bytes are verified on every cache hit, with read-only source fallback. Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. The cache is disposable and is disabled for validation/public inference. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **328,160 training examples in 82,040 optimizer updates**, 8 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +1231.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. The complete reports below determine this checkpoint's track and variant tradeoffs. + +This predeclared late fixed-link follow-up uses the complete 8-epoch small main model. It applies the original four training-only offsets without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. Both entry points use the same link. All200 validation windows and all13 development metrics are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. A publication additionally requires native replay and three sampled runtime estimates below four hours. + +## Measured validation results + +Family-mean Pearson: **0.493349** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.589011 | +| dnase | 0.665345 | +| procap | 0.543583 | +| cage | 0.331119 | +| rna_seq | 0.624430 | +| chip_tf | 0.467955 | +| chip_histone | 0.499526 | +| splice_sites | 0.399419 | +| splice_usage | 0.591521 | +| junctions | 0.380731 | +| contact_maps | 0.334203 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4258 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5559 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.1034 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7243 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.5741 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.5819 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2657 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4297 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1528 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2242 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4634 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1831 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2250 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..8adab1ac9a55de960ee5252eeb36c4b4b5bdf689 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch08/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..dd3ea50dc9c4192ed2edeb2941de4ec2c0f1681d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/package_manifest.json @@ -0,0 +1,64 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c", + "files": { + "README.md": "a04639b94262c5de3d5054c1a4d6b21113157f8595d4c9054b68c0c24005442b", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01", + "training_assets_manifest.json": "7e4a33e58edeb9c25fb59ec5b1f3b65ff2b1ca83abb9c3fc0770ec4f57af06c9", + "decoding_policy.json": "57d76f661dadbf958ab071b23062f5f9e6803fa596a18bf746765fc63c271aa9", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/late-fixed-links81-master-plan.json": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "training_history/late-fixed-links81-resolved-parents.json": "719225ce26773880c04343af0e6c61941eb0a0649f8145bf7067748620fe4f5b", + "training_history/late-fixed-links81-parent-check.json": "9d6c82a8f82993fad54c586b29c3bda774e3839e67352c84eb663f819d0b51ea", + "training_history/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "training_history/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "validation_tracks.json": "1b3e0f7f5861bc93b9fdaa6d8d0185b85dd02beaff15a1a88236f7e306a7cca2", + "validation_variants.json": "24213866acbf7f51bd8344c6231bba6eb9fd7067320e662287677570870bde95" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/cached_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/cached_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3677c1e8518ea97db0dd6759f3d636f5fff27c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/cached_read_store.py @@ -0,0 +1,71 @@ +"""Optional local byte cache for immutable training Zarr stores. + +Source storage stays read-only. Cached entries contain a key fingerprint, length, +and SHA256 of the original compressed bytes; every hit verifies all three. Writes +replace a private temporary inode atomically. A missing/corrupt/unwritable cache +falls back to the original store, so this layer changes neither values nor keys. +Only training loaders may opt into it; it is not an inference asset. +""" +import hashlib +import os +from pathlib import Path +import struct +import tempfile + +from .direct_read_store import ParallelDirectReadDirectoryStore + +MAGIC = b'SFBC0001' +HEADER = struct.Struct('<8sQ32s32s') + + +class CachedReadDirectoryStore(ParallelDirectReadDirectoryStore): + def __init__(self, path, *, cache_root, namespace, io_workers=4): + super().__init__(path, io_workers=io_workers) + if not namespace: + raise ValueError('An immutable input namespace is required') + self.cache_root = Path(cache_root) + self.namespace = namespace + self.store_key = hashlib.sha256((namespace+'\0'+os.path.abspath(path)).encode()).digest() + self.statistics = {'hit': 0, 'miss': 0, 'corrupt': 0, 'write_failure': 0} + + def cache_location(self, key): + key = self._normalize_key(key) + identity = hashlib.sha256(self.store_key+key.encode()).digest() + name = identity.hex() + return self.cache_root/name[:2]/name[2:], identity + + def __getitem__(self, key): + key = self._normalize_key(key) + destination, identity = self.cache_location(key) + try: + blob = destination.read_bytes() + if len(blob) >= HEADER.size: + magic, size, cached_key, checksum = HEADER.unpack_from(blob) + payload = blob[HEADER.size:] + if (magic == MAGIC and size == len(payload) and cached_key == identity + and hashlib.sha256(payload).digest() == checksum): + self.statistics['hit'] += 1 + return payload + self.statistics['corrupt'] += 1 + except OSError: + pass + self.statistics['miss'] += 1 + payload = super().__getitem__(key) + temporary = None + try: + destination.parent.mkdir(parents=True, exist_ok=True) + with tempfile.NamedTemporaryFile(prefix='.pending-', dir=destination.parent, delete=False) as stream: + temporary = stream.name + stream.write(HEADER.pack(MAGIC, len(payload), identity, hashlib.sha256(payload).digest())) + stream.write(payload) + os.replace(temporary, destination) + temporary = None + except OSError: + self.statistics['write_failure'] += 1 + finally: + if temporary is not None: + try: + os.unlink(temporary) + except OSError: + pass + return payload diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8e314c842cca039da1d2be8da7fff6a146178651 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_assets_manifest.json @@ -0,0 +1,21 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-integration_check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-integration_check.json new file mode 100644 index 0000000000000000000000000000000000000000..1377348921feb42310e66b133df8ebee9f0d591c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-integration_check.json @@ -0,0 +1,166 @@ +{ + "created_utc": "2026-10-25T03:48:07.725319+00:00", + "preparation_sha256": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "source_sha256": "b0921bfd3d3dc47f3cb82debc0b492d8291693c40e0cef028764c713c1257e85", + "original_full256_report_sha256": "e6620ff13b00a36857dad22fa0c9d1ad5c2b266eca70773650d280998ba20005", + "examples": 16, + "worker_processes": 4, + "ordered_indices": [ + 38509, + 39126, + 38926, + 17432, + 6551, + 22599, + 24749, + 10790, + 7022, + 30789, + 17340, + 40258, + 26222, + 22674, + 23658, + 17167 + ], + "all_returned_training_ids_and_tensor_bytes_exact": true, + "reference_and_cache_both_match_original_report": true, + "public_inference_and_validation_cache_disabled_despite_environment": true, + "default_cache_disabled": true, + "models_and_trainers_unchanged": true, + "rows": { + "reference": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ], + "cached": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ] + }, + "no_active_source_or_training_changes": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-master-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-master-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..60f920df90c63e6614b09745bb36dedab5f40cc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-master-plan.json @@ -0,0 +1,562 @@ +{ + "created_utc": "2026-10-25T16:20:50.709567+00:00", + "members": { + "large-epoch04": { + "capacity": "large", + "epoch": 4, + "previous_epoch": 2, + "parent": "/workspace/research/experiments/scale13-large-epoch04", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch02", + "output": "/workspace/research/experiments/calib81-large-epoch04", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch04" + }, + "small-epoch08": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-small-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch04", + "output": "/workspace/research/experiments/calib81-small-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch08" + }, + "large-epoch08": { + "capacity": "large", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-large-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch04", + "output": "/workspace/research/experiments/calib81-large-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch08" + }, + "small-epoch14": { + "capacity": "small", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-small-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/calib81-small-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch14" + }, + "large-epoch14": { + "capacity": "large", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-large-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch08", + "output": "/workspace/research/experiments/calib81-large-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch14" + } + }, + "capacities": { + "small": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 13290504 + }, + "large": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 96, + "channel_growth": 96, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 1024, + "num_sequence_blocks": 16, + "num_attention_heads": 16, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 768, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 64, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 215626632 + } + }, + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "calibration_reference": "/workspace/research/experiments/calib57/package/splice_calibration.json", + "helper_sha256": { + "/run-state/workspace/research/seqfunc/check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "/run-state/workspace/research/seqfunc/check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "/run-state/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "/run-state/workspace/research/seqfunc/diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "/run-state/workspace/research/seqfunc/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "/run-state/workspace/research/seqfunc/evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "/run-state/workspace/research/seqfunc/finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "/run-state/workspace/research/seqfunc/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "/run-state/workspace/research/seqfunc/make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "/run-state/workspace/research/seqfunc/materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "/run-state/workspace/research/seqfunc/measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "/run-state/workspace/research/seqfunc/promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "/run-state/workspace/research/seqfunc/publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "/run-state/workspace/research/seqfunc/resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "/run-state/workspace/research/seqfunc/resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "/run-state/workspace/research/seqfunc/run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "/run-state/workspace/research/seqfunc/run_fixed_link_followup69.py": "3234a647219aae3df72afee9e5d9ed671d9db8b08adafff58bbffcbebcafb40d", + "/run-state/workspace/research/seqfunc/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "/run-state/workspace/research/seqfunc/verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "evidence_sha256": { + "/workspace/research/studies/fixed_link69/plan.json": "45541635f549af8ce6c3db3fc7c857720970a5ee35e9de3c6a57f41dbe1f56bc", + "/workspace/research/studies/fixed_link61/plan.json": "100e4da9bf4faa715e901a0aa928dab65fc19f60970312e3e0eb548fb5b70860", + "/workspace/research/studies/fixed_link57/complete-summary.json": "5bba2a23162a77a523777f19de0be819d44cb44246c8c5e2f4bef913f5ee7a87", + "/workspace/research/studies/scale13/study-cached-io.json": "88aa6d95ef93776ed3551d68d98b1a31d726aa4b0270edbfc16b3f16ceeb937f", + "/workspace/research/studies/scale13/loss_scales.json": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42", + "/workspace/data/public/train.parquet": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "/workspace/research/experiments/calib57/package/splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5", + "/workspace/research/experiments/scale13-large-epoch04/package/training_provenance.json": "6589e14349cbd5419c814aa3d292c81f5328f1441bc75bc3ea7cae2b114dcdac" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "native_latency_template": { + "created_utc": "2026-10-25T09:54:57.944795+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.", + "scope": "Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "The rawlarge2 parent timing is contextual only. Require fresh candidate replay and allthree estimates <=14400seconds; no estimate is a guaranteed upper bound." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "bounded_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "native_timeout_seconds": 1800, + "materialization_timeout_seconds": 1800, + "parent_minimum_reward_gain": 1e-06, + "publication_minimum_reward_gain": 1e-06, + "hypothesis": "Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.", + "admission": "Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.", + "validation": "Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.", + "diagnostics": "After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.", + "nonimprovement_policy": "A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.", + "limitations": [ + "The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.", + "The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.", + "Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.", + "No test labels, variant-label training, external parameters or inference caches are used." + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-parent-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-parent-check.json new file mode 100644 index 0000000000000000000000000000000000000000..4e85ff1231bb0b51a26679743837f33f5732f829 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-parent-check.json @@ -0,0 +1,53 @@ +{ + "created_utc": "2026-10-25T16:21:07.054587+00:00", + "plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "actual_completed_parent_checks": [ + { + "experiment": "/workspace/research/experiments/scale13-small-epoch04", + "capacity": "small", + "epoch": 4, + "inference_signature": { + "last.pt": "ae6a15c21e8d9b5900f986f28f91b00962c0030ffc952340b1b3264eef1a6659", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.44610804541871724, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch04/evaluation.json": "3193c974189a7ad77a06d0abf64a2c3e002953470660ca20f30d97271e3ffd0a", + "/workspace/research/experiments/scale13-small-epoch04/evaluation-package.json": "e06f0c731873e9ef1a94722d29651c180eefc6c1b54650b52622bad525dd5da7", + "/workspace/research/experiments/scale13-small-epoch04/valid-score/report.json": "3a45ae11ac8ad3897b926283073b5e755b3013a7796aa9837954dec842d7a942", + "/workspace/research/experiments/scale13-small-epoch04/valid-variant-score/report.json": "19224eae63273b62b0421fc48c32704b86418a2bfb21adda2a90449cfa0988d8", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5" + } + }, + { + "experiment": "/workspace/research/experiments/scale13-large-epoch02", + "capacity": "large", + "epoch": 2, + "inference_signature": { + "last.pt": "6e7c605f2683881b9f59571f042cb7df93a3a0f224d8f2cb4e37be0a123cdb27", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4231233468773853, + "sha256": { + "/workspace/research/experiments/scale13-large-epoch02/evaluation.json": "d2336bfba9e6788cb2ab7dcf01834d825846920c21db43b763f3f5a4de056afb", + "/workspace/research/experiments/scale13-large-epoch02/evaluation-package.json": "cad26f6c8167a550d9a11057e69f2feca299cb091bfce64a95b403a627271610", + "/workspace/research/experiments/scale13-large-epoch02/valid-score/report.json": "eeec21adce86c14356ed29a5bc0e76e84f847bceba41c0faefa8e3a8d6793d4d", + "/workspace/research/experiments/scale13-large-epoch02/valid-variant-score/report.json": "f806730a18850e289cb95020db0ac6f69d4203c8ab53c2ef31845fd798a36f28", + "/workspace/research/experiments/scale13-large-epoch02/package/training_provenance.json": "1eb6eafaaec3f950708cabd7dc43a3ed22e67be602d7c036fe0fbbe0fbbff802" + } + } + ], + "gpu_used": false, + "packages_changed": false, + "no_new_predictions_or_parameter_fitting": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-resolved-parents.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-resolved-parents.json new file mode 100644 index 0000000000000000000000000000000000000000..241cd648df020f91b007aefc0cf0e8d9f188765a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late-fixed-links81-resolved-parents.json @@ -0,0 +1,58 @@ +{ + "master_plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "member": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-small-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch04", + "output": "/workspace/research/experiments/calib81-small-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch08" + }, + "receipts": { + "parent": { + "experiment": "/workspace/research/experiments/scale13-small-epoch08", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4622727224599102, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch08/evaluation.json": "048a4f8e805c8476b4240977d2e6c6e800dd8a509ec1183f099e9726cc51cd11", + "/workspace/research/experiments/scale13-small-epoch08/evaluation-package.json": "84a37d41701a015643c86a120d0c8b1794af718fce1478824f8aaf0798411cf5", + "/workspace/research/experiments/scale13-small-epoch08/valid-score/report.json": "2baaccea669daefc1fde4adec4477eeb813e1e17fc5d64ed3dc1f8b54dff676f", + "/workspace/research/experiments/scale13-small-epoch08/valid-variant-score/report.json": "e204ff744b560b4df16ad6c114205480864faffac1a5e22c519f4a1960638d76", + "/workspace/research/experiments/scale13-small-epoch08/package/training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01" + } + }, + "previous": { + "experiment": "/workspace/research/experiments/scale13-small-epoch04", + "capacity": "small", + "epoch": 4, + "inference_signature": { + "last.pt": "ae6a15c21e8d9b5900f986f28f91b00962c0030ffc952340b1b3264eef1a6659", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.44610804541871724, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch04/evaluation.json": "3193c974189a7ad77a06d0abf64a2c3e002953470660ca20f30d97271e3ffd0a", + "/workspace/research/experiments/scale13-small-epoch04/evaluation-package.json": "e06f0c731873e9ef1a94722d29651c180eefc6c1b54650b52622bad525dd5da7", + "/workspace/research/experiments/scale13-small-epoch04/valid-score/report.json": "3a45ae11ac8ad3897b926283073b5e755b3013a7796aa9837954dec842d7a942", + "/workspace/research/experiments/scale13-small-epoch04/valid-variant-score/report.json": "19224eae63273b62b0421fc48c32704b86418a2bfb21adda2a90449cfa0988d8", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5" + } + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..28c50d9e3a33f6accbd9ce3cdeca8bdbb9b481a6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/late_fixed_links81.py @@ -0,0 +1,145 @@ +"""Predeclare fixed-prior output links for the remaining main training checkpoints.""" +import argparse +from datetime import datetime, timezone +import hashlib +import json +import math +from pathlib import Path + +from verify_package import inference_signature, verify_package + +ROOT = Path('/workspace/research') +STUDY = ROOT / 'studies/late_fixed_links81' +HERE = Path(__file__).resolve().parent + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def write_json(path, value): + path = Path(path) + temporary = path.with_suffix(path.suffix + '.tmp') + temporary.write_text(json.dumps(value, indent=2, allow_nan=False) + '\n') + temporary.replace(path) + + +def parent_ready(path): + path = Path(path) + try: + return (json.loads((path / 'pipeline-state.json').read_text())['phase'] == 'complete' + and (path / 'evaluation.json').exists()) + except (FileNotFoundError, json.JSONDecodeError, KeyError): + return False + + +def parent_receipt(path, capacity, epoch, plan): + path = Path(path) + assert parent_ready(path), str(path) + package = path / 'package' + verify_package(package) + signature = inference_signature(package) + assert json.loads((path / 'evaluation.json').read_text())['package_inference_signature'] == signature + provenance = json.loads((package / 'training_provenance.json').read_text()) + assert provenance['config'] == plan['capacities'][capacity]['config'] + assert provenance['global_step'] == epoch * 41020 and provenance['optimizer_steps'] == epoch * 10255 + assert provenance['epoch'] == epoch and provenance['offset'] == 0 + assert provenance['seed'] == 17 and provenance['schedule_epochs'] == 14 + assert provenance['objective'] == 'variance_balanced' and provenance['examples'] == 41020 + assert provenance['training_arguments']['batch_size'] == 4 + assert provenance['family_loss_scales'] == plan['family_loss_scales'] + source = provenance['provenance'] + assert source['split'] == 'train' and source['pretrained'] is False and source['world_size'] == 1 + assert source['train_manifest_sha256'] == plan['train_manifest_sha256'] + manifest = json.loads((package / 'package_manifest.json').read_text()) + assert manifest['parameters'] == plan['capacities'][capacity]['parameters'] + assert not any((package / name).exists() for name in [ + 'splice_calibration.json', 'checkpoint_ensemble_policy.json', 'rc_ensemble_policy.json']) + tracks = json.loads((path / 'valid-score/report.json').read_text()) + variants = json.loads((path / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + assert all(math.isfinite(value['model']) for value in variants.values()) + return dict(experiment=str(path), capacity=capacity, epoch=epoch, inference_signature=signature, + reward=tracks['overall_mean_family_pearson'], + sha256={str(p): digest(p) for p in [path / 'evaluation.json', path / 'evaluation-package.json', + path / 'valid-score/report.json', path / 'valid-variant-score/report.json', + package / 'training_provenance.json']}) + + +def prepare(): + STUDY.mkdir(exist_ok=True) + assert not (STUDY / 'plan.json').exists() + link69_path = ROOT / 'studies/fixed_link69/plan.json' + link69 = json.loads(link69_path.read_text()) + link61_path = ROOT / 'studies/fixed_link61/plan.json' + link61 = json.loads(link61_path.read_text()) + members = {} + for capacity, epoch, previous in [('large', 4, 2), ('small', 8, 4), ('large', 8, 4), + ('small', 14, 8), ('large', 14, 8)]: + key = f'{capacity}-epoch{epoch:02d}' + members[key] = dict(capacity=capacity, epoch=epoch, previous_epoch=previous, + parent=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{epoch:02d}'), + previous_raw=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{previous:02d}'), + output=str(ROOT / 'experiments' / f'calib81-{key}'), study=str(STUDY / key)) + (STUDY / key).mkdir(exist_ok=True) + capacities, evidence = {}, [link69_path, link61_path, ROOT / 'studies/fixed_link57/complete-summary.json', + ROOT / 'studies/scale13/study-cached-io.json', ROOT / 'studies/scale13/loss_scales.json', + Path('/workspace/data/public/train.parquet'), ROOT / 'experiments/calib57/package/splice_calibration.json'] + for capacity in ['small', 'large']: + package = ROOT / 'experiments' / f'scale13-{capacity}-epoch04/package' + provenance_path = package / 'training_provenance.json' + provenance = json.loads(provenance_path.read_text()) + capacities[capacity] = dict(config=provenance['config'], parameters=13290504 if capacity == 'small' else 215626632) + evidence.append(provenance_path) + helpers = set(link69['helper_sha256']) | { + 'late_fixed_links81.py', 'run_late_fixed_links81.py', 'diagnose_track_uncertainty.py', + 'diagnose_remaining_variant_uncertainty.py'} + materialization = dict(link69['materialization']) + materialization['installation'] = 'Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.' + latency = dict(link61['native_latency_template']) + latency['scope'] = 'Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.' + latency['memory_evidence'] = 'One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.' + latency['minimum_free_GiB'] = 16 + plan = dict(created_utc=datetime.now(timezone.utc).isoformat(), members=members, capacities=capacities, + train_manifest_sha256=digest('/workspace/data/public/train.parquet'), + family_loss_scales=json.loads((ROOT / 'studies/scale13/loss_scales.json').read_text())['family_loss_scales'], + training_prior=link69['training_prior'], training_prior_sha256=link69['training_prior_sha256'], + calibration_reference=str(ROOT / 'experiments/calib57/package/splice_calibration.json'), + helper_sha256={str(HERE / name): digest(HERE / name) for name in sorted(helpers)}, + evidence_sha256={str(path): digest(path) for path in evidence}, materialization=materialization, + native_latency_template=latency, evaluation_lock=str(ROOT / 'studies/scale13/evaluation.lock'), + bounded_lock=str(ROOT / 'studies/junction_discovery39/bounded-diagnostic.lock'), + minimum_free_GiB=16, native_timeout_seconds=1800, materialization_timeout_seconds=1800, + parent_minimum_reward_gain=1e-6, publication_minimum_reward_gain=1e-6, + hypothesis='Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.', + admission='Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.', + validation='Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.', + diagnostics='After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.', + nonimprovement_policy='A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.', + limitations=['The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.', + 'The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.', + 'Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.', + 'No test labels, variant-label training, external parameters or inference caches are used.']) + write_json(STUDY / 'plan.json', plan) + print(json.dumps(dict(plan=str(STUDY / 'plan.json'), sha256=digest(STUDY / 'plan.json'), members=list(members))), flush=True) + + +def audit(): + assert not (STUDY / 'parent-check.json').exists() + plan = json.loads((STUDY / 'plan.json').read_text()) + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + receipts = [parent_receipt(ROOT / 'experiments' / name, capacity, epoch, plan) + for name, capacity, epoch in [('scale13-small-epoch04', 'small', 4), ('scale13-large-epoch02', 'large', 2)]] + write_json(STUDY / 'parent-check.json', dict(created_utc=datetime.now(timezone.utc).isoformat(), + plan_sha256=digest(STUDY / 'plan.json'), actual_completed_parent_checks=receipts, + gpu_used=False, packages_changed=False, no_new_predictions_or_parameter_fitting=True)) + print(json.dumps(dict(phase='parent_audit_complete', report_sha256=digest(STUDY / 'parent-check.json'))), flush=True) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('action', choices=['prepare', 'audit']) + globals()[parser.parse_args().action]() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..3e3973883128f4aea458210d43790edce69d380a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_late_fixed_links81.py @@ -0,0 +1,268 @@ +"""Execute one predeclared late fixed-link follow-up after its complete raw parent.""" +import argparse +from datetime import datetime, timezone +import fcntl +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from check_splice_link_reuse import check_link_only +from late_fixed_links81 import ROOT, STUDY, HERE, digest, write_json, parent_ready, parent_receipt +from verify_package import inference_signature, verify_package + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('member') + args = parser.parse_args() + plan_path = STUDY / 'plan.json' + plan = json.loads(plan_path.read_text()) + member = plan['members'][args.member] + study, parent, previous, output = [Path(member[k]) for k in ['study','parent','previous_raw','output']] + lock = (study / 'controller.lock').open('a') + fcntl.flock(lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + complete = study / 'complete.json' + if complete.exists(): + assert json.loads(complete.read_text())['master_plan_sha256'] == digest(plan_path) + print(json.dumps(dict(phase='already_complete_no_rerun', member=args.member)), flush=True) + return + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + assert digest(plan['training_prior']) == plan['training_prior_sha256'] + audit = json.loads((STUDY / 'parent-check.json').read_text()) + assert audit['plan_sha256'] == digest(plan_path) and len(audit['actual_completed_parent_checks']) == 2 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2', + PYTHONNOUSERSITE='1', PYTHONDONTWRITEBYTECODE='1') + for key in ['PYTHONPATH','PYTHONHOME','SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def status(phase, **extra): + record = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), member=args.member, **extra) + write_json(study / 'state.json', record) + print(json.dumps(record), flush=True) + + def finish(phase, **extra): + write_json(complete, dict(phase=phase, master_plan_sha256=digest(plan_path), + created_utc=datetime.now(timezone.utc).isoformat(), **extra)) + status(phase, **extra) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study / f'{phase}.log').open('a') as stream: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=stream, stderr=subprocess.STDOUT, start_new_session=True) + try: + code = process.wait(timeout=timeout) + assert code == 0, (phase, code) + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try:process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL); process.wait() + + def wait_memory(): + while int(subprocess.check_output(['nvidia-smi','--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB'] * 1024: + time.sleep(5) + + status('waiting_for_complete_declared_raw_parents') + while not (parent_ready(parent) and parent_ready(previous)): + time.sleep(10) + receipts = dict(parent=parent_receipt(parent, member['capacity'], member['epoch'], plan), + previous=parent_receipt(previous, member['capacity'], member['previous_epoch'], plan)) + resolved_path = study / 'resolved-parents.json' + resolved = dict(master_plan_sha256=digest(plan_path), member=member, receipts=receipts) + if resolved_path.exists():assert json.loads(resolved_path.read_text()) == resolved + else:write_json(resolved_path, resolved) + admitted = receipts['parent']['reward'] > receipts['previous']['reward'] + plan['parent_minimum_reward_gain'] + gate = dict(admitted=admitted, parent_reward=receipts['parent']['reward'], + previous_raw_reward=receipts['previous']['reward'], minimum_gain=plan['parent_minimum_reward_gain']) + if not (study / 'parent-gate.json').exists():write_json(study / 'parent-gate.json', gate) + else:assert json.loads((study / 'parent-gate.json').read_text()) == gate + if not admitted: + finish('rejected_at_parent_gain_gate', **gate) + return + output.mkdir(exist_ok=True) + package = output / 'package' + construction_path = study / 'construction-check.json' + if not construction_path.exists(): + if package.exists(): + try:verify_package(package);check_link_only(parent / 'package', package) + except (OSError,ValueError,KeyError,AssertionError): + package.rename(package.with_name('package-incomplete-' + str(time.time_ns()))) + if not package.exists(): + execute('build', [HERE / 'make_splice_probability_candidate.py', parent / 'package', package, + '--training-prior', plan['training_prior']]) + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] + assert json.loads((package / 'package_manifest.json').read_text())['parameters'] == plan['capacities'][member['capacity']]['parameters'] + before = (parent / 'package/predict.py').read_text() + marker = 'from reference.alphagenome_50m.model import CompactModel, ModelConfig\n' + old = 'data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16),' + new = 'data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16),' + assert before.count(marker) == before.count(old) == 1 + expected = before.replace(marker, marker + 'from splice_output import splice_output_scores\n').replace(old, new) + assert (package / 'predict.py').read_text() == expected + for name in ['infer.sh','perturb.sh']: + assert digest(package / name) == digest(parent / 'package' / name) + assert json.loads((package / 'splice_calibration.json').read_text()) == json.loads(Path(plan['calibration_reference']).read_text()) + history = package / 'training_history'; history.mkdir(exist_ok=True) + additions = {'late-fixed-links81-master-plan.json': plan_path, + 'late-fixed-links81-resolved-parents.json': resolved_path, + 'late-fixed-links81-parent-check.json': STUDY / 'parent-check.json', + 'late_fixed_links81.py': HERE / 'late_fixed_links81.py', + 'run_late_fixed_links81.py': Path(__file__)} + for name, source in additions.items(): + (history / name).write_bytes(source.read_bytes()) + readme = package / 'README.md' + body = readme.read_text().replace( + 'In preliminary\ndiagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed.', + "The complete reports below determine this checkpoint's track and variant tradeoffs.") + note = (f'\nThis predeclared late fixed-link follow-up uses the complete {member["epoch"]}-epoch ' + f'{member["capacity"]} main model. It applies the original four training-only offsets ' + 'without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. ' + 'Both entry points use the same link. All200 validation windows and all13 development metrics ' + 'are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. ' + 'A publication additionally requires native replay and three sampled runtime estimates below four hours.\n') + if note.strip() not in body:body += note + readme.write_text(body) + manifest = json.loads((package / 'package_manifest.json').read_text()) + for name in ['README.md', *['training_history/' + name for name in additions]]: + manifest['files'][name] = digest(package / name) + write_json(package / 'package_manifest.json', manifest) + verification = verify_package(package) + write_json(construction_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), resolved_parents_sha256=digest(resolved_path), + package_inference_signature=inference_signature(package), exact_link_only_checks=checks, + parameters=manifest['parameters'], network_updates=0, prior_refitted=False, package_verified=True)) + construction = json.loads(construction_path.read_text()) + assert construction['master_plan_sha256'] == digest(plan_path) + assert construction['resolved_parents_sha256'] == digest(resolved_path) + verify_package(package) + signature = inference_signature(package) + assert construction['package_inference_signature'] == signature + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] and checks['candidate'] == signature + native = study / 'native-recovery' + if not (native / 'report.json').exists() or not (output / 'valid-variants').exists(): + status('waiting_for_bounded_native_and_derivation_slot') + with Path(plan['bounded_lock']).open('a') as probe_lock: + fcntl.flock(probe_lock, fcntl.LOCK_EX);wait_memory() + if not (native / 'report.json').exists(): + execute('native', [HERE / 'check_native_recovery.py', package, native], plan['native_timeout_seconds']) + if not (output / 'valid-variants').exists(): + materialization = dict(plan['materialization'], candidates=[dict(parent=str(parent), candidate=str(output), + parent_inference_signature=checks['parent'], candidate_inference_signature=signature)]) + path = study / 'materialization-plan.json' + if path.exists():assert json.loads(path.read_text()) == materialization + else:write_json(path, materialization) + execute('materialize', [HERE / 'materialize_splice_link_outputs_v2.py', path], plan['materialization_timeout_seconds']) + proof = json.loads((native / 'report.json').read_text()) + assert proof['package_inference_signature'] == signature and proof['all_arrays_bit_identical'] + assert proof['native_vendored_entrypoints_passed'] and proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks':30,'perturbations':22} + derivation = json.loads((output / 'valid-variants/perturb-link-derivation.json').read_text()) + assert derivation['checks']['candidate'] == signature and derivation['checks']['parent'] == checks['parent'] + assert derivation['native_arrays_bit_exact'] == 36 and derivation['queries'] == 2697 + if not (output / 'evaluation.json').exists(): + status('waiting_for_full_evaluation_slot') + with Path(plan['evaluation_lock']).open('a') as evaluation_lock: + fcntl.flock(evaluation_lock, fcntl.LOCK_EX);wait_memory() + execute('evaluate', [HERE / 'evaluate_package.py', package, output, + '--resume-partial-tracks','--resume-partial-perturbations']) + assert json.loads((output / 'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output / 'valid-score/report.json').read_text()) + variants = json.loads((output / 'valid-variant-score/report.json').read_text()) + raw_tracks = json.loads((parent / 'valid-score/report.json').read_text()) + raw_variants = json.loads((parent / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None and len(variants) == 13 + affected = {'sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'} + for family, report in tracks['families'].items(): + if family != 'splice_sites':assert report == raw_tracks['families'][family],family + for name, report in variants.items(): + assert math.isfinite(report['model']) + for field in ['rows','perturbations','metric','anchors']:assert report[field] == raw_variants[name][field],(name,field) + if name not in affected:assert report == raw_variants[name],name + comparison_path = study / 'matched-comparison.json' + if not comparison_path.exists(): + write_json(comparison_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), reward=tracks['overall_mean_family_pearson'], + reward_delta=tracks['overall_mean_family_pearson']-raw_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-raw_tracks['families'][k]['mean_pearson'] for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-raw_variants[k]['model'] for k,v in variants.items()}, + unaffected_family_reports_exact=10, unaffected_variant_reports_exact=8, package_inference_signature=signature)) + decision_path = study / 'publication-decision.json' + if not decision_path.exists(): + current = json.loads(Path('/workspace/submission/validation_tracks.json').read_text())['overall_mean_family_pearson'] + eligible = tracks['overall_mean_family_pearson'] > current + plan['publication_minimum_reward_gain'] + runtime_passed = None + if eligible: + latency = study / 'native-latency';latency.mkdir(exist_ok=True) + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency / 'plan.json').exists():assert json.loads((latency / 'plan.json').read_text()) == timing_plan + else:write_json(latency / 'plan.json', timing_plan) + if not (latency / 'report.json').exists(): + execute('native-latency', [HERE / 'measure_native_latency.py', latency / 'plan.json']) + timing = json.loads((latency / 'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + names = ['development_mixture_seconds','slowest_observed_stratum_mean_seconds','pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in names) + runtime_passed = max(estimates[k] for k in names) <= 14400 + write_json(study / 'runtime-gate.json', dict(passed=runtime_passed, estimates=estimates, + report_sha256=digest(latency / 'report.json'), scope='Shared-load extrapolations, not guaranteed upper bounds.')) + if runtime_passed:execute('publish', [HERE / 'promote_if_better.py', package, output]) + write_json(decision_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward=tracks['overall_mean_family_pearson'], current_reward_at_review=current, + eligible_at_review=eligible, runtime_passed=runtime_passed, + publication='Atomic promoter rechecks the current strongest package; see publish.log if run.')) + for phase, helper in [('tracks','diagnose_track_uncertainty.py'), + ('remaining-variants','diagnose_remaining_variant_uncertainty.py')]: + report_path = study / f'paired-{phase}.json' + marker_path = study / f'paired-{phase}-complete.json' + if marker_path.exists(): + assert json.loads(marker_path.read_text())['output_sha256'] == digest(report_path) + continue + def diagnostic_complete(): + try: + report = json.loads(report_path.read_text()) + assert Path(report['candidate']).resolve() == output.resolve() + assert Path(report['control']).resolve() == parent.resolve() + assert report['replicates'] == 2000 and report['seed'] == 1729 + assert report['diagnostic_only_no_parameter_fitting'] + leaf = 'valid-score/report.json' if phase == 'tracks' else 'valid-variant-score/report.json' + assert report['report_sha256'] == {'candidate':digest(output / leaf),'control':digest(parent / leaf)} + if phase == 'tracks': + assert len(report['point_deltas']) == 12 and 'reward' in report['point_deltas'] + assert set(report['bootstrap']) == {'chromosome','chromosome_10Mb_blocks'} + for result in report['bootstrap'].values(): + assert set(result['deltas']) == set(report['point_deltas']) + else: + assert set(report['evaluations']) == affected | {'caqtl','bqtl_spi1'} + for item in report['evaluations'].values(): + assert set(item['bootstrap']) == {'variant_id','genomic_block'} + for result in item['bootstrap'].values(): + assert result['reference_replicates_checked'] >= 32 + assert result['maximum_absolute_metric_error'] < 1e-10 + return True + except (FileNotFoundError,json.JSONDecodeError,AssertionError,KeyError): + return False + if not diagnostic_complete(): + execute('paired-' + phase, [HERE / helper, output, parent, report_path, + '--replicates','2000','--seed','1729']) + assert diagnostic_complete(), ('incomplete diagnostic', phase) + write_json(marker_path, dict(output_sha256=digest(report_path), master_plan_sha256=digest(plan_path))) + finish('complete', reward=tracks['overall_mean_family_pearson'], + matched_comparison_sha256=digest(comparison_path), publication_decision=json.loads(decision_path.read_text()), + diagnostics='All11families/sevenvariants, with eight unaffected official variant reports exact; not anall13bootstrap.') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..e20a00c50c406f11d042daf7341e520ec3a1ec3b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/training_provenance.json @@ -0,0 +1,119 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 8, + "offset": 0, + "global_step": 328160, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 73897.99778079404, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 82040, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..b3067c19bfd7457bab049da699b094c11437d0e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5890112780553517, + "valid_tracks": 167, + "per_track_pearson": [ + 0.7093273994911413, + 0.7318732919264792, + 0.46123506970433775, + 0.6551390901386592, + 0.47084665163137857, + 0.4709979882578834, + 0.4895597322705998, + 0.5729392805909, + 0.5881603223191597, + 0.6535026008145658, + 0.5880360335172274, + 0.5655012303137811, + 0.40012644746048676, + 0.5303718828464126, + 0.6908757993575628, + 0.5144628566323924, + 0.565611528334607, + 0.5352080858446879, + 0.5667115288700991, + 0.5531181769381871, + 0.5213874943536934, + 0.48176019373528883, + 0.5611760560982423, + 0.5364430550251358, + 0.534794211799496, + 0.513036416546783, + 0.49120776989876763, + 0.51436396241498, + 0.5228954726446216, + 0.585254768890651, + 0.5011153460168937, + 0.5396605712916998, + 0.4948353729943055, + 0.4134583935899317, + 0.3835115380110611, + 0.5971145197402808, + 0.5342739893074467, + 0.644174379161105, + 0.5250398574424145, + 0.5173617048001232, + 0.5960349530326469, + 0.5975867352612421, + 0.5139731701266815, + 0.4908417969833281, + 0.5609381380347472, + 0.5771382728534389, + 0.4469704851865573, + 0.49840881026006273, + 0.492540227784674, + 0.5384606072955899, + 0.5089441104439321, + 0.4743795543946625, + 0.5740836079436296, + 0.6095934191960566, + 0.4696814280505234, + 0.6943316030142607, + 0.8096572955433028, + 0.5591213518185417, + 0.6037549717358162, + 0.6486354047315646, + 0.7750751054822401, + 0.5113818874757378, + 0.700445430300126, + 0.6853189043005357, + 0.6443352747877285, + 0.5207024520536523, + 0.5802935237958292, + 0.6482935970297349, + 0.7991794545797972, + 0.7193260389032188, + 0.5372381354472139, + 0.4894956502672596, + 0.46248073177179605, + 0.5501246958860466, + 0.5116537684123332, + 0.5018574817756469, + 0.5082468541862414, + 0.5485529036273684, + 0.5950740544903761, + 0.503277399587205, + 0.5847630141291886, + 0.6087223352124104, + 0.524880154070378, + 0.4982243063325506, + 0.5338192308561982, + 0.5251750646765007, + 0.5029690049917774, + 0.49240040417955216, + 0.46333638145159206, + 0.5179214503808331, + 0.5202760257512358, + 0.5153938792618152, + 0.4972452288994276, + 0.5152730234724384, + 0.5767988125467807, + 0.4035979882402515, + 0.45628691228010937, + 0.4451985497045946, + 0.5039260503248049, + 0.5193709345920866, + 0.46916810842964746, + 0.4731113551348079, + 0.5180664106974563, + 0.4762942845715202, + 0.5580820020061669, + 0.5625990139257251, + 0.5492446932312015, + 0.5321429193914696, + 0.5558970184264502, + 0.5654491017773329, + 0.617961002529419, + 0.5396210549361535, + 0.5339035471057851, + 0.452556154117759, + 0.47776474346036696, + 0.4510009971113101, + 0.49213990025299154, + 0.49409497115821116, + 0.5090175482517634, + 0.4859020822327645, + 0.6644199486133838, + 0.67902256089913, + 0.7448088766138113, + 0.7373345536501569, + 0.726618533812171, + 0.5860342986083054, + 0.8652385598261294, + 0.73260309635645, + 0.7722971815064851, + 0.6904445489781268, + 0.7351891821417844, + 0.5713172053023039, + 0.5999063029570055, + 0.5728192049461153, + 0.7195996568057094, + 0.7519941742297239, + 0.7034483350267055, + 0.5484381522589372, + 0.4347728819273804, + 0.8226614666539933, + 0.7926790275863541, + 0.8131303412400268, + 0.8074614723791677, + 0.5866062475060084, + 0.8757927013958386, + 0.7839383788440674, + 0.7502812428805127, + 0.7398117083759606, + 0.8446177041591753, + 0.8224855319820905, + 0.6288972627949393, + 0.6315120768595278, + 0.7322596978957882, + 0.7612900301235574, + 0.7766741592229991, + 0.6154899720454766, + 0.751159977946832, + 0.605002560913064, + 0.6057307892929673, + 0.6963837082925731, + 0.60622255764456, + 0.7822576784361867, + 0.7867699023646314, + 0.656003853579537, + 0.6043982131111405, + 0.5969741480993155, + 0.7485887501799842 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.66534539905906, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5774432575603706, + 0.930559916684097, + 0.796683554232992, + 0.5372504812832963, + 0.5454965304638487, + 0.5595090892947928, + 0.9446947812231203, + 0.7357872330785266, + 0.4672108241091273, + 0.3652938911488286, + 0.8954334972923758, + 0.7200323992696523, + 0.4462730195543252, + 0.5366443099500002, + 0.5160890157967898, + 0.8193662246648854, + 0.8030532111795518, + 0.8562560640994875, + 0.8736202873869685, + 0.9020501745163031, + 0.7153530347769653, + 0.6358196573024999, + 0.912302239111283, + 0.5473076529501003, + 0.5641727998656205, + 0.7891290989377293, + 0.47108069362195404, + 0.6864332939625674, + 0.5420706278427003, + 0.5686488747947083, + 0.6492122356226537, + 0.48502934962672817, + 0.8493568239732454, + 0.6353606917803386, + 0.8879341642523289, + 0.895636187590752, + 0.46037135050941835, + 0.8364688367757962, + 0.8815318598697282, + 0.8689347510574124, + 0.8370070985089936, + 0.6786405341656961, + 0.7140653129075304, + 0.6506386874047597, + 0.5503875948608001, + 0.6481901356833006, + 0.743112838833379, + 0.5147604249179064, + 0.6156140197837473, + 0.42220324188162484, + 0.40258948035465997, + 0.5864617384442599, + 0.5242519225313486, + 0.8351603507633342, + 0.4585044210702088, + 0.4280569578171463, + 0.4825333117002261, + 0.5334854337070388, + 0.4973171217054612, + 0.5395188581123425, + 0.4746776541108987, + 0.6302454652873288, + 0.5216657622265688, + 0.5097645025347295, + 0.5389961389941674, + 0.54539546511194, + 0.5758840150140413, + 0.5049181745063933, + 0.40707361276980875, + 0.4499433679491946, + 0.525187338060691, + 0.6719754170400872, + 0.530642000045719, + 0.6534111662434255, + 0.38532897294782426, + 0.8213077869602595, + 0.902145325903526, + 0.6873517621861955, + 0.4645792254498496, + 0.4805532724663886, + 0.5612563110702364, + 0.5368356214538309, + 0.4785945110596862, + 0.5138637431385507, + 0.5992043383147142, + 0.5197929439487963, + 0.6341948640715168, + 0.5262235347455146, + 0.6572559752308598, + 0.5192822136414507, + 0.6520076949034872, + 0.427211929689802, + 0.7410744923608633, + 0.49831402244594686, + 0.7312917806178041, + 0.5259361914771314, + 0.7718473811924194, + 0.7413039496903989, + 0.3951803724441513, + 0.40261169248176215, + 0.4982671507837205, + 0.3950787635776584, + 0.37790437877839217, + 0.7503401436009131, + 0.7236377603900296, + 0.4265207066490967, + 0.5137158534125632, + 0.6188244801492356, + 0.8547900390683324, + 0.5738588441830528, + 0.5868315187841466, + 0.605802020069569, + 0.5386869139978392, + 0.46549378002880726, + 0.7950564216675078, + 0.5979369618541293, + 0.5146111579194452, + 0.5710181192565312, + 0.6651517947050741, + 0.4921305628399319, + 0.5536483747780887, + 0.5160569848677441, + 0.5886098978179202, + 0.5678795579759912, + 0.6417859667789715, + 0.5750206243446846, + 0.6687057652231064, + 0.6889464223433626, + 0.7158948730243448, + 0.6524464210080366, + 0.48475624046603977, + 0.6293262018624814, + 0.6658548524055581, + 0.8068961801845745, + 0.46461559250469164, + 0.35628466513933904, + 0.6189367882427985, + 0.5025333511247599, + 0.667616663150839, + 0.631349293132061, + 0.6403335237363508, + 0.5851515791321271, + 0.5055588445956727, + 0.793676221909261, + 0.645971791805284, + 0.7711209596670969, + 0.4174255819776402, + 0.3932520004007395, + 0.4266884838956257, + 0.3796050783457108, + 0.9037284068779775, + 0.3513917535586817, + 0.9093583806686976, + 0.7807861088076247, + 0.49021465066908326, + 0.3805190508219696, + 0.3791957454169874, + 0.3904804151899842, + 0.6759510649617696, + 0.7809751159734425, + 0.770823354037984, + 0.6898502264155224, + 0.5892557828089173, + 0.5069044487221452, + 0.7016773636750404, + 0.6422354472735501, + 0.5450275109016883, + 0.6676705714922324, + 0.4824350004291121, + 0.5176755573508003, + 0.4294974722624069, + 0.8407728916802275, + 0.4926893893172545, + 0.5010044138647292, + 0.6804241156595126, + 0.7190776237291157, + 0.5483946687850312, + 0.37009877625959536, + 0.8850122957110442, + 0.4039243381249462, + 0.79639440128537, + 0.5293714641806666, + 0.6343203789335246, + 0.7560212688738474, + 0.47559234426792263, + 0.5907151170613896, + 0.6346476065362907, + 0.837695363151032, + 0.4426018719959861, + 0.6392758586138363, + 0.6010989589383561, + 0.7330261747201311, + 0.9036146660035137, + 0.46825980853442506, + 0.6170602715876213, + 0.5147222554126243, + 0.6767345698044107, + 0.7464803917207077, + 0.5842453239448101, + 0.4828023360635903, + 0.8801417582408275, + 0.8926283661735084, + 0.7643833907617555, + 0.8479095094387161, + 0.8575327782362981, + 0.848638113204014, + 0.8614587995369626, + 0.3922190002740208, + 0.41673690128527263, + 0.5399608868838922, + 0.4924750015867695, + 0.48646202119669735, + 0.6419971921869972, + 0.8753026150942003, + 0.8578135591564238, + 0.7115121023480722, + 0.840385273727762, + 0.8386493985436501, + 0.869099314751206, + 0.9149576199763884, + 0.5762103206569436, + 0.6710883266549696, + 0.7226519076504994, + 0.6929902773567945, + 0.8387855237027221, + 0.6246454316939338, + 0.920083121998481, + 0.8924566645206816, + 0.8812390339019867, + 0.9090550649451858, + 0.8990144597604816, + 0.887385928432215, + 0.8434410363510186, + 0.8686098764290984, + 0.9179984810690347, + 0.49484699769388124, + 0.9333462833520599, + 0.8883983453000962, + 0.8317207992557811, + 0.6440237684277023, + 0.8793916762006365, + 0.5598673338014449, + 0.7598121491766031, + 0.8820231348448214, + 0.6417709148474054, + 0.6312220836199673, + 0.72795053335605, + 0.5562850938422291, + 0.5894405546989149, + 0.7532536741406569, + 0.8956155206802322, + 0.779946022949342, + 0.8302363243113065, + 0.668027122874451, + 0.8774683443285541, + 0.8187841477864096, + 0.8483922777824399, + 0.9132906452401837, + 0.8377377524090386, + 0.908064593860337, + 0.8786809521128885, + 0.5672850841490568, + 0.8857389276571707, + 0.6129513185809292, + 0.5842869265215195, + 0.8910386125184524, + 0.9121107219616545, + 0.8912949503446019, + 0.660892360878431, + 0.6762563874533434, + 0.6927580894044757, + 0.8076742599658069, + 0.9045758843993464, + 0.748795975133535, + 0.9226019531946594, + 0.9277276023539566, + 0.6430458820926931, + 0.36475977954584776, + 0.556632373239221, + 0.9065560668091917, + 0.4999997745943332, + 0.6596373204330898, + 0.8937201665403991, + 0.8407226461198201, + 0.6848757473607495, + 0.8741489362571643, + 0.9022108443669905, + 0.8977778373584597, + 0.8883958994817469, + 0.8714940904471123, + 0.8995208983508398, + 0.8680577556830416, + 0.9384775760220211, + 0.7928463646313805, + 0.8425918386300055, + 0.91574392325603, + 0.6058257385487943, + 0.6142413241259267, + 0.6140994664891084, + 0.5848651695952762, + 0.9152787820645514, + 0.813981641621157, + 0.8564623862531221, + 0.8118672638381667, + 0.7799340715295605 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5435828510944237, + "valid_tracks": 12, + "per_track_pearson": [ + 0.515634210673939, + 0.561388550457527, + 0.5300860107738047, + 0.5557767354325277, + 0.5541099796146611, + 0.5793803842320339, + 0.49955720646237844, + 0.5537610347752414, + 0.5080782065296052, + 0.5475689366898105, + 0.5228515794014851, + 0.5948013780900699 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.3311190648549341, + "valid_tracks": 546, + "per_track_pearson": [ + 0.33647203432533773, + 0.366717747360364, + 0.3003478898497487, + 0.19837732312536735, + 0.11707123761470142, + 0.3158022017218262, + 0.21552814192151126, + 0.21139295214783005, + 0.3338118595470543, + 0.31766162427731015, + 0.35954294194807757, + 0.2146618558726287, + 0.22001244939580622, + 0.17840043741263392, + 0.18425297028659518, + 0.2311615929193379, + 0.2842619781481134, + 0.38725441164781316, + 0.36555410957599027, + 0.36873066748735245, + 0.29635784279514726, + 0.29929873847676264, + 0.44016675812806877, + 0.24934640035742597, + 0.30162723559007193, + 0.3527075641231144, + 0.3223462001607115, + 0.35843704797578724, + 0.3736963906217095, + 0.3834499056846389, + 0.2410754690580609, + 0.39719122594035755, + 0.2524516667179195, + 0.22038516876816786, + 0.40553495930712685, + 0.345632080412472, + 0.36763980339735813, + 0.42948740513105726, + 0.3671747900689612, + 0.40417368839123924, + 0.4485530376140363, + 0.37729021019081854, + 0.3542344736620949, + 0.30420975682799695, + 0.2140730269167994, + 0.28750185551227053, + 0.31174110663972665, + 0.3489949168233257, + 0.1908363488760957, + 0.24233002784546895, + 0.3865011084767751, + 0.3828851952437243, + 0.38533801397837203, + 0.2780156881252853, + 0.44153033536098235, + 0.43407572326440935, + 0.26207692766040364, + 0.44308192959729326, + 0.31157478531409283, + 0.41956789907241804, + 0.40873483030656593, + 0.2747293752155298, + 0.3546744466367625, + 0.29640791110919507, + 0.37080508689259845, + 0.42597480957567546, + 0.4239023082021117, + 0.2525191069195323, + 0.2014378110883882, + 0.17571858951038155, + 0.3311457399740184, + 0.3843947961899568, + 0.35060375171166835, + 0.3298720680187693, + 0.3751590400909563, + 0.30815595164992243, + 0.3594184386823786, + 0.34650661865958726, + 0.4064247829095722, + 0.39260916454570866, + 0.34409725788633244, + 0.2020116023810401, + 0.3539565592518423, + 0.4460958835630636, + 0.2013887017052782, + 0.3570200642027457, + 0.32340093544234677, + 0.436231282697774, + 0.29261593876582104, + 0.37686868167204723, + 0.3035662680367856, + 0.35291404619966177, + 0.3624397211247865, + 0.1840921163163058, + 0.3864671999211813, + 0.32260581231478896, + 0.3698515926385014, + 0.3649322492863903, + 0.3360506412714726, + 0.21939569782380608, + 0.3491136519109757, + 0.35904773208725876, + 0.38394149805745087, + 0.44179251064555847, + 0.40874866083118, + 0.39102389691848194, + 0.28850266181408246, + 0.39255691615458127, + 0.30450898436686663, + 0.3076855687910697, + 0.3224849769799006, + 0.32572316336114354, + 0.35101917008083033, + 0.3486692378774781, + 0.28942594696041635, + 0.3677932379793297, + 0.34682526451471035, + 0.35753808397755993, + 0.3437998363418472, + 0.37995940168297754, + 0.39244859046633884, + 0.3694271077778825, + 0.2673229759233357, + 0.31159397362915936, + 0.33042264647682507, + 0.37712869487700346, + 0.3539644504803977, + 0.3856873782412017, + 0.45229687858112216, + 0.3882867033071104, + 0.3987932446031154, + 0.30762571401680544, + 0.4003839602061018, + 0.4278911538613907, + 0.4482547748449648, + 0.33787020035303883, + 0.43084845290710844, + 0.3424601869911949, + 0.1865224071081026, + 0.3911914465885213, + 0.4062779582325426, + 0.3499405206582047, + 0.39536096804573984, + 0.30728799783961497, + 0.32921415340526905, + 0.35070284131459095, + 0.34356346768912677, + 0.45878267503206177, + 0.3322248339223, + 0.3234340479916167, + 0.40935661333431056, + 0.31061277152950906, + 0.36909036930562544, + 0.36929322827679256, + 0.3431956923547449, + 0.326886417075603, + 0.4314946752108095, + 0.4569442345646041, + 0.3187470200699581, + 0.29524480461967456, + 0.253779690200918, + 0.32616373594346304, + 0.3277689544920707, + 0.36990848612687754, + 0.3239814368655733, + 0.30286814291536246, + 0.3473414238372809, + 0.2666768109261824, + 0.21484594362992843, + 0.38077341908383716, + 0.3462343492083496, + 0.2935207464090567, + 0.3491866363020384, + 0.44976964541843667, + 0.356941481810425, + 0.21958466454404982, + 0.24195288858302588, + 0.3162745459366111, + 0.3982056626752824, + 0.4059601839880045, + 0.3918706315680148, + 0.40824989917178994, + 0.252481667474574, + 0.3917611738822692, + 0.3433685473788056, + 0.27897305807739087, + 0.33343698346670286, + 0.240149557601019, + 0.4480198972460866, + 0.3774671721933618, + 0.30830512178462716, + 0.2366360905186888, + 0.23118736087224878, + 0.33776282117232814, + 0.35159140264949107, + 0.2972524044694655, + 0.34616844794223944, + 0.36440753135176485, + 0.26518432865791963, + 0.34274410790678306, + 0.1785114710405268, + 0.18034869995490435, + 0.16890039073532848, + 0.16000536387382125, + 0.22283079581765058, + 0.3383327058486729, + 0.34131659820222576, + 0.2908375884382654, + 0.16220178789440567, + 0.31214565665796906, + 0.400638935730811, + 0.4132228854282478, + 0.4212280984038601, + 0.3755957467996728, + 0.36729536146074265, + 0.29246159913452835, + 0.40686667949314614, + 0.40984274512353136, + 0.3680828956722534, + 0.4043353947255786, + 0.3713709799446647, + 0.28632325304252043, + 0.38182195014977566, + 0.36857116228443443, + 0.45556903128589143, + 0.4199242951759754, + 0.35364909262122673, + 0.34937193543868655, + 0.3847531022859315, + 0.4315350083818351, + 0.28104063866371803, + 0.2989530872357476, + 0.4234261728847895, + 0.33443336848212946, + 0.3094872935727591, + 0.3267016797258912, + 0.43854578192644106, + 0.28665130106896836, + 0.3790978454144638, + 0.36946788092383653, + 0.17628662980011506, + 0.28214732988720914, + 0.2306031947982607, + 0.378748222819334, + 0.32429073704988814, + 0.3649912328871843, + 0.1630616726185073, + 0.4039840786003996, + 0.36099375922752064, + 0.3876848819077594, + 0.38707427468610334, + 0.2604883308393611, + 0.43272013788969954, + 0.19888247906632564, + 0.3842488123437131, + 0.30750870389065355, + 0.4395846990702061, + 0.39999067139756955, + 0.42062557986250104, + 0.4608589313950819, + 0.33790953941520896, + 0.36580452362938365, + 0.17416418271835452, + 0.3822624788785042, + 0.39866797462196857, + 0.283707063220993, + 0.18272112661265505, + 0.18776560162190545, + 0.2181368791755102, + 0.34255593850163796, + 0.41800123443661524, + 0.21749892106630778, + 0.16702496745335874, + 0.3178033119437312, + 0.3729132932211994, + 0.32946644726657726, + 0.22972475070843357, + 0.10924550092096291, + 0.3470130918706404, + 0.22989556677666156, + 0.2529029052663214, + 0.3449340453766312, + 0.30133070062722184, + 0.3700320373040531, + 0.21556181223694101, + 0.2337054578733567, + 0.212946271253341, + 0.20301607075024752, + 0.21878497171624497, + 0.2792472483319673, + 0.3701562084606283, + 0.3567302830917937, + 0.34234484628599077, + 0.2822574416035593, + 0.2919540465003739, + 0.42215496146215836, + 0.2286326883333625, + 0.26630204116677625, + 0.37642617608186174, + 0.31664597598103594, + 0.3506726306383987, + 0.3597398794434814, + 0.36396130188914666, + 0.22953794246221199, + 0.3740155355498196, + 0.23912339218138387, + 0.22069165724214163, + 0.37455814162102635, + 0.3602089826904924, + 0.40558846046872693, + 0.4161150978128663, + 0.43714307641847694, + 0.3887275901343365, + 0.435760114881239, + 0.36307546160188175, + 0.34718071944850465, + 0.3370905760242367, + 0.2297227325103808, + 0.3105870194811988, + 0.3511097143038836, + 0.3945630385922606, + 0.19608906098987144, + 0.2544245694718371, + 0.33834768256260417, + 0.3756749840860332, + 0.358685430296239, + 0.26933854939109125, + 0.429458478142489, + 0.41090147642783154, + 0.26536049468613715, + 0.4330013105749097, + 0.2967382551631803, + 0.4139882570745054, + 0.3894939024461398, + 0.26941522652218614, + 0.3552765612045657, + 0.2961172209076456, + 0.3751582368081231, + 0.41225647694916095, + 0.4012664760331989, + 0.24921031589959372, + 0.20441048721635205, + 0.17103197658431862, + 0.3253322968026811, + 0.37387381053093055, + 0.3360178717211092, + 0.31033578274777135, + 0.37173792533501865, + 0.2925403914262698, + 0.3333143041933732, + 0.31981487675181214, + 0.38186124386872267, + 0.37752961463122403, + 0.33509029848308136, + 0.20389357543125333, + 0.3508468945432347, + 0.42696754912884444, + 0.2085747966808416, + 0.35304611633961724, + 0.3507722259127045, + 0.42752065270679523, + 0.26920876902486934, + 0.3763920858146356, + 0.2920384302378838, + 0.34269908885852873, + 0.3601524162271922, + 0.18702356761470498, + 0.3741756197397767, + 0.3051590601383354, + 0.335367253628652, + 0.3687065015345055, + 0.3220815935074053, + 0.2100319737677839, + 0.3431712506650859, + 0.3584270327178056, + 0.3658909791432265, + 0.41697261598399965, + 0.38790881500089464, + 0.3934027717295858, + 0.27206729335252466, + 0.35289663044224284, + 0.28432891680977485, + 0.2941797850037899, + 0.3049475168619447, + 0.31284118368635055, + 0.34522105141879034, + 0.34737023591197586, + 0.27032924029806055, + 0.34881834295430997, + 0.31746905878189413, + 0.33876561075486344, + 0.3227260537562867, + 0.3658387522788523, + 0.3652277405004154, + 0.3532669811465407, + 0.26246327059071706, + 0.3098898911081147, + 0.3298847333996706, + 0.37617015046124086, + 0.33407659088921265, + 0.3827308231436429, + 0.44057410044823675, + 0.4061527485678602, + 0.37496059857923353, + 0.2841580141986023, + 0.35506046326866736, + 0.38478664203191604, + 0.42081365059667447, + 0.3355129089463252, + 0.4157977501813581, + 0.33061997334320226, + 0.2134540019389937, + 0.3787323258750179, + 0.3761888140688715, + 0.32970104234787134, + 0.39123689155545355, + 0.29178153188857026, + 0.32331835541089365, + 0.33670891584906204, + 0.3338673256221076, + 0.43706234787882964, + 0.358048824928273, + 0.3150126455714511, + 0.4024072520171002, + 0.30568695294827086, + 0.3478342057485102, + 0.3722704073284593, + 0.3332732241270767, + 0.33399770957787156, + 0.40964698615132356, + 0.44715416045092066, + 0.3139105596921759, + 0.3210423230414501, + 0.2633792879409595, + 0.30390199290303727, + 0.33318391817413034, + 0.3785104703321815, + 0.35453443926576506, + 0.30173179864920996, + 0.3513124702080989, + 0.29708496688459757, + 0.19615504605666512, + 0.4165645507179049, + 0.3465815970678566, + 0.2929748656916228, + 0.3518838663139232, + 0.4440958505715685, + 0.3403589571362803, + 0.19152252797379543, + 0.24042397906739266, + 0.3184787670893784, + 0.3687169526463995, + 0.40703476525356697, + 0.38588933818520166, + 0.4069740319046219, + 0.26321112064964275, + 0.38754608825560033, + 0.3498284608374721, + 0.3065432221007251, + 0.3387373769163224, + 0.25453920730554247, + 0.43254444090035854, + 0.37646766302598056, + 0.3133298678118126, + 0.26052124174717073, + 0.24540913470967265, + 0.297421547235469, + 0.3698877667791694, + 0.31882073093026303, + 0.35851450453662514, + 0.3624032948868072, + 0.2781753478992282, + 0.3405808357285812, + 0.1722832532535305, + 0.1519047960548983, + 0.13848781169144847, + 0.13342548895173498, + 0.19984782094993433, + 0.3548116513668635, + 0.3219089167557874, + 0.2896982737176637, + 0.15022637627994506, + 0.3265735368899341, + 0.4097262368853435, + 0.42826286078437026, + 0.44042522373538834, + 0.39514087442148393, + 0.39010293461525897, + 0.30602299418823703, + 0.4169209267339456, + 0.4134173311242639, + 0.36732625339526015, + 0.4078501727086663, + 0.3781612683415898, + 0.30841758710042083, + 0.40227177082843113, + 0.3562421949052453, + 0.47865440806580595, + 0.4171982734086462, + 0.38244543213316895, + 0.35077278141365503, + 0.37058517317364637, + 0.3901674123596956, + 0.23893934359694238, + 0.2386709269922005, + 0.43425190497966454, + 0.3437063288176398, + 0.32110227520051077, + 0.3254553751132837, + 0.43449425673813763, + 0.3075058057906934, + 0.3564357101926843, + 0.3484768671493379, + 0.15763252686257515, + 0.2807609333248172, + 0.21667939074227383, + 0.38864445314821594, + 0.2961987772552803, + 0.368492958088091, + 0.16666677377396583, + 0.3952116352184791, + 0.37002067885046197, + 0.3888030493902383, + 0.3947237605448931, + 0.2822688705314372, + 0.4787343985078435, + 0.18892733668515552, + 0.37301103055034723, + 0.26460434965752855, + 0.42740587848529515, + 0.4116932265729174, + 0.44696327034913347, + 0.4503280748294781, + 0.35001070001879775, + 0.3741948643504419, + 0.15396461902383068, + 0.3658576583065152, + 0.38213780182620355, + 0.30699210218570905, + 0.1645655222334007, + 0.16696910797312675, + 0.19607686265429985, + 0.34649127549693265, + 0.40902710387114827, + 0.18677567054624716, + 0.1404786635444965 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.6244303726349023, + "valid_tracks": 613, + "per_track_pearson": [ + 0.6138661060185109, + 0.6930177219392627, + 0.5954696008715418, + 0.6465623880601183, + 0.6804002894099331, + 0.47150387900591884, + 0.6051802212946772, + 0.6791873494675903, + 0.6660591952833764, + 0.689659933777327, + 0.6260239431378244, + 0.6051863301706129, + 0.5203392518501396, + 0.6782268283743018, + 0.649622311028827, + 0.6614117780842339, + 0.7051245918450805, + 0.6098303824057927, + 0.6129811232907842, + 0.6006450773297232, + 0.5951902990091141, + 0.6686368794430422, + 0.6227813730738483, + 0.5994189028596393, + 0.4944040172047915, + 0.6837798123960008, + 0.5490776385317252, + 0.624266152760426, + 0.6034672683114184, + 0.6210969412715063, + 0.6352047323653734, + 0.6644214283334162, + 0.6563378016081828, + 0.5516822440038952, + 0.6447076077248304, + 0.5092944316890091, + 0.634196110588549, + 0.6366738955346173, + 0.5828692470612389, + 0.6292126600720743, + 0.6875313561472138, + 0.5821647021946182, + 0.6718210423717204, + 0.6893304296385608, + 0.6189422063061747, + 0.533073008613668, + 0.5883172072304986, + 0.6597931268390456, + 0.6337574198834512, + 0.657678297836182, + 0.6485082935337805, + 0.6683609147017641, + 0.564098982987875, + 0.6189538007714168, + 0.49181858715236304, + 0.6171368019571364, + 0.6067827913727994, + 0.6770890454660856, + 0.7107925835552839, + 0.7265250702254407, + 0.7143245597621299, + 0.6462271286503574, + 0.6728970494556029, + 0.589573361710286, + 0.6632756420729503, + 0.6601582265470108, + 0.6843129038666543, + 0.6852813574246487, + 0.6632140573603448, + 0.7088721708814076, + 0.6730083797386719, + 0.6772612362051504, + 0.6420736169434845, + 0.6622607749286876, + 0.6536138882580539, + 0.6464270232952609, + 0.6748573057718833, + 0.6832507568490495, + 0.6585538135890088, + 0.7107948834216993, + 0.666287992845792, + 0.591621766483939, + 0.7268655603350127, + 0.7012638397936851, + 0.6128126982416463, + 0.6442300410031747, + 0.7076217870610968, + 0.6299354904870131, + 0.6783286263899487, + 0.6446508172546238, + 0.5488131946231801, + 0.686152894918307, + 0.7048872133232974, + 0.6861867815667362, + 0.6894058058648902, + 0.6971819661507089, + 0.7142636970243259, + 0.696154059521946, + 0.6459660715508507, + 0.5632066728348143, + 0.5472224762599313, + 0.5635673515188712, + 0.6125511315642207, + 0.6662606178742722, + 0.5904305439875065, + 0.6300963755605908, + 0.6745781871807242, + 0.6744309033575681, + 0.6762059428800071, + 0.6241936535767103, + 0.651106958473235, + 0.6650071222542144, + 0.6613377798916017, + 0.6254038496085496, + 0.6487771970722571, + 0.6781685919732741, + 0.6799357861839268, + 0.6663634298361775, + 0.6187804267924542, + 0.6777831844607339, + 0.6846841952135365, + 0.5950364203076204, + 0.6002610416699553, + 0.6241727127163712, + 0.7048179133918786, + 0.5370586790075655, + 0.5529975806122311, + 0.5458307751191587, + 0.5583660267713699, + 0.5895754187202952, + 0.6833729361951835, + 0.6178447979307463, + 0.6432430516796387, + 0.5977449461440081, + 0.6388358243773872, + 0.6654992306414905, + 0.6551424086166973, + 0.6777847365218899, + 0.6273363868861301, + 0.6729153906713541, + 0.6038895531797362, + 0.6103737044635252, + 0.6943018546918677, + 0.6443373023621176, + 0.5597613330442323, + 0.5807551636234993, + 0.580501624962762, + 0.5998910161433021, + 0.5525304096717841, + 0.627838537248802, + 0.5778132465456179, + 0.61355220271202, + 0.6415976265517217, + 0.5476340637940844, + 0.6133207032161142, + 0.5927498402672451, + 0.6267922005952284, + 0.6224515566878747, + 0.6154291164771761, + 0.6692010048571105, + 0.6784896566000529, + 0.6720127972043358, + 0.6547537341248246, + 0.5727088233288409, + 0.5689723134554094, + 0.5729517564025507, + 0.5764238416230879, + 0.6012778245735403, + 0.616124129616273, + 0.6196975128596639, + 0.712704897286713, + 0.5948340861327148, + 0.7168401403049308, + 0.6215425857201472, + 0.5693372019223895, + 0.5993292242780082, + 0.5789650721608747, + 0.6548065598411441, + 0.5433875461646253, + 0.5171283940573513, + 0.5214249539336052, + 0.6405449430763017, + 0.568573724026578, + 0.6560005291395032, + 0.5310480486576797, + 0.6167146187195389, + 0.6278531243326165, + 0.43829414858194554, + 0.5601637534618178, + 0.6029326332849549, + 0.6901252878736708, + 0.6124147127119359, + 0.6094084017926744, + 0.571389712217934, + 0.6582236478231668, + 0.6063737470599786, + 0.6257435986453324, + 0.4092225947254779, + 0.6100249795131064, + 0.617292209014415, + 0.6186732433441162, + 0.6867513698402855, + 0.6822741353529211, + 0.6654345587168878, + 0.649051290053565, + 0.6542371464625515, + 0.6600114575842976, + 0.3856150692338456, + 0.6227543758008414, + 0.6284214763970967, + 0.6333913477038938, + 0.6109537983305054, + 0.6731262774634972, + 0.6772258728826639, + 0.5654163795824205, + 0.6501959656197257, + 0.6518926138900708, + 0.5806532502463075, + 0.7034401434104561, + 0.6994030716697488, + 0.6805675489268836, + 0.7347318126797148, + 0.6440872919051239, + 0.644283025646013, + 0.635028611921307, + 0.6260941316699994, + 0.6409583812021546, + 0.6287845068230066, + 0.5410420971889025, + 0.7064396372418472, + 0.6018458953689128, + 0.6587764254731783, + 0.6425945198285207, + 0.5451774972973867, + 0.7057270126221482, + 0.5073516868083471, + 0.48754573508281435, + 0.6543368042873304, + 0.680440938235824, + 0.6533360719342194, + 0.6605413939283079, + 0.6709363069831082, + 0.6692470416486008, + 0.7249215693194163, + 0.34109085563451125, + 0.6166736187070255, + 0.6414818485454651, + 0.6528560975825154, + 0.5881493479709672, + 0.5918662258009041, + 0.5674298125151801, + 0.5980236360970169, + 0.5748037720436738, + 0.5825370658297535, + 0.6569567359646271, + 0.6299527833647517, + 0.5598929250017218, + 0.6716118814174526, + 0.6231216861880036, + 0.6514792140878216, + 0.5933311607669003, + 0.5889896344574616, + 0.6987395045716099, + 0.7041334067818721, + 0.6502445801913486, + 0.7050455134178781, + 0.6886734466524411, + 0.5510668847631836, + 0.6293136711265225, + 0.6289826677554415, + 0.5022642700029877, + 0.6079167909047088, + 0.6874186110080241, + 0.6996441592990824, + 0.699777968644135, + 0.6851304845941505, + 0.44611688498654817, + 0.5672900665102034, + 0.6731593523248158, + 0.6388159974253265, + 0.6991104692379199, + 0.6175474075114236, + 0.5786909794677365, + 0.531709231821917, + 0.6454584809637427, + 0.6154904808003768, + 0.6324115312837917, + 0.6920498484622152, + 0.6768121124768371, + 0.7030007956392065, + 0.60880070194766, + 0.6204419255780418, + 0.6766720414375674, + 0.6138634007246944, + 0.5626398904954748, + 0.499333161417562, + 0.6605267827301498, + 0.7058330789994542, + 0.7041777301449976, + 0.6340511723286991, + 0.645082047351192, + 0.5974052009827316, + 0.6752449308669981, + 0.6011353330208828, + 0.6118049083838825, + 0.681029296354222, + 0.6343137640743628, + 0.6942719696514097, + 0.7005511129473279, + 0.6505715655186616, + 0.6774193614264227, + 0.6802402763132128, + 0.6372802098783348, + 0.6442795223857772, + 0.612171686108092, + 0.5741203369165504, + 0.5875269402365481, + 0.6185579772053019, + 0.6754987194238468, + 0.6247372835052819, + 0.6770587814358094, + 0.7010512617445502, + 0.6897861661808155, + 0.5256372861010856, + 0.6002929933694549, + 0.474907186119553, + 0.6161190152579447, + 0.6084108208669211, + 0.6423435622644817, + 0.6836400285904526, + 0.6589040636543564, + 0.6865648157320383, + 0.6783905540895185, + 0.7008718877460717, + 0.5739045125760828, + 0.6818524549964463, + 0.680521523329023, + 0.7229172911234553, + 0.7185328850712236, + 0.6969164130821269, + 0.7178518435804655, + 0.6792669843644044, + 0.6786306063569164, + 0.6213005246061393, + 0.7070958352757089, + 0.681418108851329, + 0.6591299008649477, + 0.6832068703341543, + 0.7286408940754326, + 0.5848531668808739, + 0.6625608111510111, + 0.6337372525707045, + 0.5880571507818209, + 0.7237833605929042, + 0.704480683909926, + 0.633330873578232, + 0.6461156356981771, + 0.6474355414424873, + 0.6199878674828602, + 0.6911424633647157, + 0.5955275132242199, + 0.5838249026084127, + 0.6646930746639972, + 0.6468368711709416, + 0.6518619814264104, + 0.6208130472092461, + 0.6395073811164923, + 0.6592781374569946, + 0.6463083531364812, + 0.6062806942919416, + 0.5550277266048328, + 0.49781199148886623, + 0.517540052803777, + 0.5363992308887222, + 0.6963741436479932, + 0.5737185897542643, + 0.585586509870165, + 0.7209571504239346, + 0.6419918796616143, + 0.6707593905192588, + 0.6001694329662246, + 0.6444185795216985, + 0.6140418698205996, + 0.6139963265092351, + 0.6262971544115026, + 0.736675549725783, + 0.6831872407883061, + 0.6875736786608446, + 0.6266597430073737, + 0.5858911414093451, + 0.6365913203038238, + 0.7114217912792213, + 0.5715386697685612, + 0.5848147113734393, + 0.6001441629724698, + 0.7064229838678117, + 0.5189014223943745, + 0.535097317993886, + 0.5190983256703408, + 0.5177321056529577, + 0.6086267375263464, + 0.707935124769784, + 0.5790024106594501, + 0.6243153228253526, + 0.6941658118732842, + 0.6622170010592847, + 0.6899944535239847, + 0.6279664816044869, + 0.6899812298116458, + 0.6197148727429652, + 0.6648233184206412, + 0.5675196222438509, + 0.6301959189378585, + 0.6827306726843121, + 0.6229095554673475, + 0.5472090823118537, + 0.5536049292116944, + 0.6130051384930678, + 0.6400684590760601, + 0.6528369812616397, + 0.5802178851148372, + 0.5641153953450943, + 0.6255256555094239, + 0.6988918271143094, + 0.49370726045433183, + 0.5764921255604409, + 0.595491413111974, + 0.6862733041856875, + 0.5905761705668904, + 0.6719022424911401, + 0.6194584829899111, + 0.6712783055477022, + 0.6614011914132834, + 0.6640604717202836, + 0.5012006802466622, + 0.5000614235912112, + 0.5032631371927018, + 0.5025912920407752, + 0.5812791019552622, + 0.6154218941869594, + 0.622006058011203, + 0.7579290439033443, + 0.6096779250720948, + 0.7330638058926514, + 0.6100660534570693, + 0.5824281340651475, + 0.5835120473715797, + 0.5791547972157605, + 0.6852601618273159, + 0.49967538943420103, + 0.519709413991861, + 0.49144684902101915, + 0.6586021238903683, + 0.540434126983651, + 0.6086314738188274, + 0.5247539424807821, + 0.61631168389185, + 0.6585342406309329, + 0.46376446067236743, + 0.5714529832702313, + 0.5489547221534802, + 0.6580759944673342, + 0.6393896668748865, + 0.5616753687748184, + 0.5413200160943785, + 0.6669615570863279, + 0.6313992238424719, + 0.6307265513574188, + 0.3921321629165733, + 0.6618388731393872, + 0.6671276732396939, + 0.5882105376612466, + 0.7405243045938316, + 0.6592445819252105, + 0.6671415785500774, + 0.6670512908224864, + 0.6512729736990885, + 0.6574496229709196, + 0.31629833442732275, + 0.6544718847985636, + 0.6119923987833376, + 0.703982021310814, + 0.6811936503734867, + 0.7578432129528936, + 0.7576713935580498, + 0.6063297802780921, + 0.6921852612588045, + 0.7290440379175764, + 0.617815380860105, + 0.7750231642576467, + 0.6712889611801618, + 0.6167064230001095, + 0.7034280751155412, + 0.598352637935805, + 0.5904591826605651, + 0.5897450156223119, + 0.5487106900724957, + 0.5538979371367104, + 0.5664104282391174, + 0.48866223921152463, + 0.7349676921188609, + 0.5327876135395381, + 0.6363556185205687, + 0.718785728486877, + 0.5410832975567486, + 0.7233832893256937, + 0.47994095718395613, + 0.4850749483404679, + 0.6614499878894389, + 0.654352208774441, + 0.6033951104826948, + 0.6153104135143168, + 0.7003913051355385, + 0.7115527779645997, + 0.786844818600963, + 0.3452716883251259, + 0.6199413747160293, + 0.6335533232106281, + 0.6577368798292615, + 0.655528953230144, + 0.5930262733382926, + 0.5794095022282885, + 0.6028346293096521, + 0.6026934321732362, + 0.5969501586979509, + 0.6606663989187317, + 0.5980225448319693, + 0.5175487287835762, + 0.6835647801224982, + 0.6324691446524998, + 0.7235978284336637, + 0.5372978798799065, + 0.4972684466233173, + 0.6522198363090831, + 0.6602999083243007, + 0.6434560741147423, + 0.7145189688322958, + 0.7285734066692858, + 0.5072645599998415, + 0.6465631572229998, + 0.6495635738074658, + 0.4798694409645188, + 0.6662884945087961, + 0.3785448698700684, + 0.5558824854784841, + 0.6200564871442581, + 0.5621227036432165, + 0.629263234656093, + 0.5940322504319755, + 0.6454438816161057, + 0.6594438493130349, + 0.5726287259379004, + 0.5313501579118098, + 0.5371677986382574, + 0.5355052446692738, + 0.5254118297829672, + 0.524848847336121, + 0.536659848134644, + 0.5318889323941365, + 0.5051454575646493, + 0.48424490896811034, + 0.6342149945779947, + 0.6084416403419893, + 0.6132835860915085, + 0.6111434713480182, + 0.62335479424737, + 0.6134637195598228, + null, + 0.6276658860479515, + 0.676787070988735, + null, + 0.5227610383808681, + 0.6734115453194295, + 0.5193055877914728, + 0.655666506238634, + 0.6718106743886787, + 0.6704020489557482, + 0.6677163376387851, + 0.6625586544708824, + 0.6603555384395171, + 0.6467241173363933, + 0.6094733416130332, + 0.6004037992919767, + 0.5489819705409258, + 0.64305687026536, + 0.5181961537398024, + 0.6397271061986554, + 0.6191287182765589, + null, + 0.6313041562545546, + null, + null, + 0.6209319630481656, + null, + 0.6364553278424744, + 0.616960497185942, + null, + 0.6570314003820741, + null, + null, + null, + null, + null, + null, + null, + 0.6378507330641107, + null, + null, + null, + null, + 0.5542843588512608, + null, + 0.6343048382988534, + null, + 0.6276271555545423, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6567647925854134, + 0.48758972025076974, + 0.6147728094671326, + 0.6411875487580312, + 0.5606615730990954, + 0.6643060069096576, + null, + 0.6519469885734965, + null, + 0.6412688650671947, + null, + null, + 0.6570489952674267, + 0.6725862862052153, + 0.5971718939715746, + 0.5838643826361, + null, + null, + 0.6053825491056554, + null, + null, + 0.6367669766494846, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6440343884970265, + 0.630101900424525, + 0.646902904475795, + 0.6138445749749555, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.46795539486877774, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.25118978017869353, + 0.38113174687350243, + 0.3227125598191613, + 0.49211227924072126, + 0.4367984700754323, + 0.32437393339049003, + 0.4280633818217412, + 0.4984058638489107, + 0.4154215232767666, + 0.42948190773970707, + 0.4140456753184019, + 0.4272945130909759, + 0.3029040952990566, + 0.5011277334686673, + 0.417885320913752, + 0.42655980962587803, + 0.4940747040450959, + 0.3925008102054322, + 0.5250918384592627, + 0.41977808757149254, + 0.4176538487455324, + 0.36378464701398894, + 0.33688427091856676, + 0.5456524916109217, + 0.38834651681312243, + 0.4416265405405866, + 0.3612139005601612, + 0.3947526222686668, + 0.4296389175531929, + 0.4519094588503999, + 0.4551879203098345, + 0.4358087028596906, + 0.5070289841099123, + 0.3068021836878092, + 0.6799407078684785, + 0.29732275249261325, + 0.26585258101984655, + 0.1967010054515135, + 0.35232855120862494, + 0.2396436977777803, + 0.29609765826006584, + 0.480140380347918, + 0.3271169121918408, + 0.42705811135286126, + 0.3143845511351545, + 0.4159437716083261, + 0.3193324058412479, + 0.23899278878788194, + 0.48807973352634915, + 0.4607433574912864, + 0.4046610528121719, + 0.3785619681966723, + 0.39924197057224503, + 0.43717647903187745, + 0.4298046381121292, + 0.45030764221363206, + 0.37208247382938536, + 0.5294199464617073, + 0.5433320041006647, + 0.5439993537126953, + 0.3378551162439633, + 0.22082271153165395, + 0.41535561783320407, + 0.4417809327777704, + 0.291191169687533, + 0.37348585900654346, + 0.3658398028733702, + 0.4955880807958292, + 0.5150946381460371, + 0.4626226418477742, + 0.38477092195771356, + 0.7242093657250044, + 0.3778786099529715, + 0.40884310952045316, + 0.4119413817298992, + 0.44933107317423104, + 0.3357551941854278, + 0.6731506301688521, + 0.43625481930626603, + 0.46220662657826156, + 0.4163796195552662, + 0.4366788290328523, + 0.5022178297765637, + 0.4677246506543357, + 0.3932732886123851, + 0.34202056457568103, + 0.46939519818466907, + 0.4866868238876319, + 0.809549846542506, + 0.4989892815835038, + 0.41572965407033485, + 0.363497391661567, + 0.4832414452385307, + 0.41080717653371507, + 0.45079699548395236, + 0.5696913793278136, + 0.35736951368500675, + 0.4135384274359377, + 0.4217569384119408, + 0.3559282536701557, + 0.2573842111234239, + 0.39116505898841786, + 0.3512276901577198, + 0.3347452048333838, + 0.3832610982057611, + 0.275930605924714, + 0.3442451202508135, + 0.3730283247246097, + 0.3914900797738037, + 0.45050396643067386, + 0.277119413756675, + 0.3889889378071014, + 0.3627685339046681, + 0.4164578295186676, + 0.46223133715526, + 0.4559773157229641, + 0.43084289022272876, + 0.44130207615962114, + 0.415320053923129, + 0.4718749503579664, + 0.4225274022960408, + 0.456363273245499, + 0.4415256670923882, + 0.36110180542873505, + 0.41981389792024, + 0.37920604867915975, + 0.40989771497536176, + 0.3829106724015707, + 0.41344844981235734, + 0.5472338847839134, + 0.4638970033204258, + 0.38768002760403397, + 0.4523356440380072, + 0.6524170332685986, + 0.38677886466343653, + 0.44713095600288816, + 0.4500220136728067, + 0.34035394694820686, + 0.402154912444072, + 0.38143147777663333, + 0.4138893398936152, + 0.45681371122479386, + 0.5797206682150325, + 0.3952110222537317, + 0.4622505294502852, + 0.524536557019049, + 0.47676139348539776, + 0.34124454048541114, + 0.4609164817928388, + 0.483452803462423, + 0.4661270097510939, + 0.39842617830854177, + 0.5009955644591151, + 0.514588540690812, + 0.45249786547600374, + 0.45239644998874895, + 0.3657788296896693, + 0.41997695319449496, + 0.4191877813198083, + 0.47470403747017786, + 0.47924450894577675, + 0.40771447630375607, + 0.4384232154805425, + 0.4510175460491902, + 0.47137931479648126, + 0.4176368954126759, + 0.41169053785831783, + 0.4985563572450049, + 0.3462407582664675, + 0.4670721070828148, + 0.4438480897541539, + 0.39087242502637315, + 0.5095194292320329, + 0.40346503199419853, + 0.365137107193063, + 0.46100656804444007, + 0.3321933435994371, + 0.48508684233556326, + 0.4951045756795659, + 0.37556925959881554, + 0.47577278132091855, + 0.4619617614262186, + 0.37984062729389334, + 0.45774037908321946, + 0.3896381367337995, + 0.47685491099471194, + 0.47870530614687834, + 0.3474602954739281, + 0.37820236204698987, + 0.4534724116562142, + 0.3844910864805485, + 0.4115037238486428, + 0.4703448963425017, + 0.5129879399133153, + 0.4020582547782162, + 0.3897723025597451, + 0.4005514738730398, + 0.3971525561246043, + 0.30998961686162235, + 0.5093714060007668, + 0.4150658855099109, + 0.38738044131573024, + 0.4300421641903936, + 0.3782683330553544, + 0.48738237667619044, + 0.42264008856231217, + 0.4945567816171727, + 0.41716900051062483, + 0.36879111213873633, + 0.539547887039488, + 0.41097812947955636, + 0.4758479193224963, + 0.4397666975915909, + 0.45632706922656935, + 0.8637279592778804, + 0.4967341907444062, + 0.4185955370521353, + 0.4581340001795218, + 0.4752530720180025, + 0.5564866676728868, + 0.6221803743000591, + 0.6354967487461979, + 0.45069550451634255, + 0.40851291802577655, + 0.5013414603590124, + 0.7396218115774417, + 0.3764712214682702, + 0.4023471839494706, + 0.5569599019641096, + 0.5317941238649907, + 0.5928911863329551, + 0.4685218249052001, + 0.34641325993055355, + 0.4210132396536013, + 0.6092722986836294, + 0.4664679019442509, + 0.3764574684569384, + 0.6213560002782612, + 0.5631568833850568, + 0.46618827583019606, + 0.5060163800428594, + 0.7176875414140086, + 0.4987784853916039, + 0.5252553413047942, + 0.4594525451362277, + 0.7680209547071001, + 0.34919959447345955, + 0.6476277504134791, + 0.42231630636267975, + 0.5576182997902565, + 0.4537235792637127, + 0.3971677865392122, + 0.7076910091104953, + 0.5263126492701148, + 0.3576791067189677, + 0.3908536789577802, + 0.6546390797685733, + 0.40357536920903225, + 0.4555778140327085, + 0.5076216378911808, + 0.40452192694778, + 0.5037095705522514, + 0.5035243072741273, + 0.44232796245514755, + 0.4028579995172852, + 0.4845653082297999, + 0.3800728674587939, + 0.3622415096068134, + 0.6374494697285523, + 0.39812408013840106, + 0.46562007493934615, + 0.5804129440015393, + 0.4105132199759673, + 0.6658259894938201, + 0.37919073784259155, + 0.3819967241247482, + 0.567133879095504, + 0.5067987727477338, + 0.5894448459863573, + 0.7346866942288267, + 0.6778268380973792, + 0.4277415045405546, + 0.48124502393915985, + 0.5750987377904546, + 0.3743317472666869, + 0.36459090851067555, + 0.5747717592692864, + 0.5440494535722723, + 0.7425875372924557, + 0.5003013529140182, + 0.3863435101920759, + 0.4771962561616889, + 0.38900279962413403, + 0.4856590718883821, + 0.7124742173128449, + 0.6358316695330544, + 0.547968291127666, + 0.5456914844414759, + 0.5096746544499839, + 0.5254321555921455, + 0.7463826737814913, + 0.324022949712566, + 0.4611045141037176, + 0.33991096481370753, + 0.6185970593279932, + 0.4004387961363062, + 0.7420974780909615, + 0.44383019657864886, + 0.44755147967751, + 0.7027862186076642, + 0.5283066693528903, + 0.35489426266432306, + 0.5767295285404697, + 0.42473839717856243, + 0.5197690065740137, + 0.5068811217272576, + 0.36137529824555054, + 0.513512979920759, + 0.41868280282721665, + 0.4695784893628133, + 0.42766740859141, + 0.3741652907535216, + 0.43447886537943686, + 0.5402424043883428, + 0.5989237024282696, + 0.6719784060326088, + 0.44378549704021486, + 0.32959082267544576, + 0.4744442283853938, + 0.4828186073017358, + 0.44320055386864327, + 0.49468320383672193, + 0.500677177271751, + 0.4001054784768604, + 0.35188258006591827, + 0.5273668378042985, + 0.7687995217455843, + 0.5387429722642854, + 0.6142951339561913, + 0.5994485448330339, + 0.7744763552244588, + 0.5777766279862958, + 0.3761806994913127, + 0.5983945029272234, + 0.4063019384235334, + 0.5499455824220657, + 0.49412420112185024, + 0.407922145024382, + 0.5114405564790566, + 0.7886552177425987, + 0.25692488521300183, + 0.4823768602785325, + 0.48620663823275134, + 0.6327531119616177, + 0.3926845533939884, + 0.3361906736593726, + 0.4489954867477494, + 0.7279492223474594, + 0.4095510349404707, + 0.45945959987183826, + 0.5180136330322062, + 0.47073235336699926, + 0.44509690117154604, + 0.5143397194470628, + 0.6237398558423397, + 0.47508275535600414, + 0.42795154637117816, + 0.563808302255994, + 0.46813722273173336, + 0.5816058315591232, + 0.5660578548916114, + 0.5429920180511637, + 0.5650824768218374, + 0.3569292987374163, + 0.3793466509184595, + 0.5757538661279804, + 0.3410693221661213, + 0.703268539954774, + 0.3412491110029597, + 0.5743686855755736, + 0.5995301065885034, + 0.6930465428658439, + 0.32557787400859506, + 0.3906598898489738, + 0.3633233854551616, + 0.4120636636854797, + 0.344430953305054, + 0.32754971541438266, + 0.6304101025054597, + 0.4061809771878571, + 0.6874637621287641, + 0.5930262161587477, + 0.4723777851545536, + 0.6004664600081425, + 0.3949561504203452, + 0.5607136096619807, + 0.5456654130194286, + 0.4877338146642793, + 0.4018245635796583, + 0.47910092030581714, + 0.5181163572691155, + 0.5429720922087538, + 0.4085264807317872, + 0.3725996024916025, + 0.6388095404381712, + 0.49036159793484096, + 0.664100917838536, + 0.6546979237574918, + 0.5334708567436522, + 0.3530739384722827, + 0.3857476046265258, + 0.38464997271711654, + 0.5120401959531882, + 0.7504481169058632, + 0.48460390338401677, + 0.39715706815991886, + 0.7478517289935861, + 0.461329815404044, + 0.6315167876903521, + 0.542216014006115, + 0.4488929509366741, + 0.5085595277489244, + 0.4706759745029832, + 0.48407938786427435, + 0.5852833259230429, + 0.5543119998234592, + 0.5435469964968339, + 0.4623681954619001, + 0.4832057728810578, + 0.5310731309575664, + 0.5037350313965657, + 0.7716341915706486, + 0.5003015255737406, + 0.500787188177963, + 0.45158211854787333, + 0.3237382573595165, + 0.6687342871094413, + 0.42824991527003076, + 0.3653024399923125, + 0.5285703386315143, + 0.47186049614448816, + 0.365028025195694, + 0.49264538298956073, + 0.4534425190432579, + 0.7825715146478677, + 0.3273506423951879, + 0.4151927520272856, + 0.3736907562873399, + 0.4428168912621954, + 0.41282788484650174, + 0.33332621417535324, + 0.43495423913820247, + 0.2932011610115102, + 0.4362701897277966, + 0.7471266035611813, + 0.3444409841690055, + 0.4190368364358143, + 0.513577351549244, + 0.4318448194174206, + 0.4176700788421782, + 0.4902563956379932, + 0.36475628320332537, + 0.5319541310438424, + 0.5432036564652476, + 0.27714486044393033, + 0.6443738813812065, + 0.6962335766678642, + 0.5186006747281755, + 0.49687316453980357, + 0.5500605545631421, + 0.4419607345951289, + 0.46548137783265503, + 0.5719210661341572, + 0.7845223765613261, + 0.41569406280903104, + 0.615637110268836, + 0.36428147072473527, + 0.47089959603158094, + 0.49792521949455704, + 0.5000301440759065, + 0.41098826145745454, + 0.3528856520940313, + 0.3572091502387886, + 0.4058178486600178, + 0.5849266117233863, + 0.4762878283328274, + 0.35013378761220404, + 0.4179113527852313, + 0.473358036827271, + 0.32262128604141904, + 0.557602306682132, + 0.5114133260615019, + 0.435686710436768, + 0.46330718043110697, + 0.34304671132340314, + 0.43819057205329154, + 0.5170340645490447, + 0.5948934357755447, + 0.4388839745369406, + 0.7847849000561952, + 0.3805775787402527, + 0.5635723279530793, + 0.7440411517740236, + 0.6398892707190599, + 0.3289245785037749, + 0.5166115281629287, + 0.428937516815861, + 0.3704787003192415, + 0.3903207234112226, + 0.3999934863408083, + 0.4794879313170601, + 0.5058838227624528, + 0.38729480031147545, + 0.6232521736947877, + 0.7711185499375169, + 0.44515240807476075, + 0.4600624502595521, + 0.44903672502820147, + 0.5150129787708498, + 0.5297821640194035, + 0.49772957188291617, + 0.5120743350847314, + 0.3764539407922101, + 0.6685563021207986, + 0.495417634637233, + 0.5624072014180639, + 0.7196038079956126, + 0.3414398082129085, + 0.599552193811074, + 0.3975900071026353, + 0.4713541264279429, + 0.3554516138961331, + 0.5275252372805462, + 0.507632272567, + 0.3513672632814671, + 0.44957752658458144, + 0.4165183802042175, + 0.5447092968607471, + 0.6012848591209027, + 0.47492690005837906, + 0.47541634713037156, + 0.4257948117297227, + 0.7460085474304491, + 0.5050522069664382, + 0.5422686395218158, + 0.4771692659230766, + 0.47365957310225915, + 0.6212531137899995, + 0.3651682230244818, + 0.5774569149484408, + 0.39196310796241707, + 0.5578613430896295, + 0.43769320223012137, + 0.3657442414292397, + 0.37976125306042424, + 0.48447230152661347, + 0.38427199812121554, + 0.5174596293611964, + 0.7390094130385124, + 0.7301532946956656, + 0.4182670026805856, + 0.375498604129927, + 0.7072336818079917, + 0.45934985688329155, + 0.4038120854647133, + 0.5337852141091418, + 0.3702828351269437, + 0.6057892742311743, + 0.4408009166473994, + 0.5754335822606544, + 0.5946729204628244, + 0.46390374732985296, + 0.5220695527587925, + 0.6615705649543733, + 0.5312835785355884, + 0.3801347281291143, + 0.4370261949358302, + 0.3245910734466031, + 0.5488533477723392, + 0.3633590906779593, + 0.5288342680479352, + 0.5109152549835699, + 0.40685398746673157, + 0.7263503963091923, + 0.3465346584469115, + 0.7320092417555539, + 0.46094719397534184, + 0.5486866310895875, + 0.748695524123673, + 0.49164031144427195, + 0.560651495180338, + 0.420109256566749, + 0.7089123365520175, + 0.43166628526865675, + 0.351885560091965, + 0.5733906635607051, + 0.5324932578435593, + 0.647835158932022, + 0.5580004714348624, + 0.41572777996136784, + 0.4950263649659531, + 0.3193940862780011, + 0.47379286193023373, + 0.5740592048625854, + 0.5238551439388393, + 0.37311947442218363, + 0.5521760753784941, + 0.5935219452619658, + 0.5789688972902298, + 0.4514417706709936, + 0.4122074180877855, + 0.5177923599303015, + 0.4735270357657373, + 0.5653885035456607, + 0.5017116318327585, + 0.3789026503177188, + 0.4227589583044864, + 0.5213056754029717, + 0.4132222982625026, + 0.7416873515333013, + 0.5307986978502276, + 0.46519992742788774, + 0.5082309959649334, + 0.4147969852380951, + 0.45665734355819326, + 0.5487102229457105, + 0.5599218704918947, + 0.7153037994021368, + 0.5514153989268479, + 0.5823328824303324, + 0.5394506779549796, + 0.5061901842197922, + 0.585153934801475, + 0.41889715460737, + 0.6448150485945251, + 0.603318228153638, + 0.7288931268235778, + 0.5311613985634782, + 0.5543482419712501, + 0.5481799993282835, + 0.4525957686706852, + 0.5444611995537535, + 0.36539715310538623, + 0.6886902555488377, + 0.4964713182784986, + 0.7412426135643225, + 0.7654622512247989, + 0.5007945765569406, + 0.46464268352179927, + 0.5114826505547986, + 0.3772664839912837, + 0.5142104863360526, + 0.7631626228728768, + 0.49871511307205046, + 0.5276765111114056, + 0.5934508246320026, + 0.5315064274412998, + 0.3702473553635246, + 0.6201409857431283, + 0.44563843448894147, + 0.40716674496449184, + 0.7384256950605913, + 0.42971542531844475, + 0.45095603086263003, + 0.41631107881928797, + 0.49464860429984425, + 0.5701381401078351, + 0.6384277549636133, + 0.6853303681646147, + 0.6296753684761313, + 0.5963720321503504, + 0.47330271666647244, + 0.6154349494631561, + 0.571338111375976, + 0.7839639776956102, + 0.7475596836030955, + 0.38275362353471887, + 0.3989719740738351, + 0.3756797155911901, + 0.5311939433224837, + 0.42904861904919367, + 0.7924719195187311, + 0.39421410449009514, + 0.4336602919718639, + 0.5843887438901626, + 0.49748862660608173, + 0.5765242349539605, + 0.6334846894976414, + 0.48207558708534803, + 0.6238553391380635, + 0.3967757238980609, + 0.6933154823736494, + 0.7851142446499297, + 0.7691898681992413, + 0.5950073086583959, + 0.44866380492811325, + 0.5698292511655494, + 0.7480931266320825, + 0.6978357683280726, + 0.634393665558957, + 0.5035520666125692, + 0.47791368874740314, + 0.5047919228679626, + 0.7412612026792725, + 0.43947506659606045, + 0.4512230365119392, + 0.3392835172614785, + 0.6096831250820259, + 0.33568211347120436, + 0.4334567688951452, + 0.6789975534147685, + 0.5500319540527638, + 0.6914660776685274, + 0.5579969689245845, + 0.5284788186373183, + 0.6747729771108055, + 0.6420522588988684, + 0.575287789366499, + 0.3979212041957251, + 0.6543945886583237, + 0.5337972833110597, + 0.4208276143189121, + 0.3726780219382023, + 0.49598485871266856, + 0.6460530426187542, + 0.5302713077080233, + 0.6230510153268408, + 0.585950118205823, + 0.4395711586585627, + 0.4254977448804821, + 0.6958061941276069, + 0.516316761636694, + 0.62111982288415, + 0.7299780917606665, + 0.42919934142166144, + 0.40070353430100036, + 0.39731101260670226, + 0.3942322204575966, + 0.7197404158992744, + 0.7414758561220792, + 0.5733210419443048, + 0.43520535160132545, + 0.48498780700910243, + 0.4066651032887996, + 0.7867566622064215, + 0.3685733296758829, + 0.49930286191924467, + 0.35741278689313893, + 0.5242590220184472, + 0.3938080701967444, + 0.5766285612147672, + 0.5022976804989853, + 0.371758410046806, + 0.7053474851825302, + 0.5228769208826476, + 0.4442416574203871, + 0.3608731120971732, + 0.37645125322897033, + 0.4272344740575935, + 0.43936106126515045, + 0.3930156785317471, + 0.38753664575383645, + 0.422450032792279, + 0.41449876173316025, + 0.43432203382874535, + 0.35707114405760826, + 0.46479371829275995, + 0.22615368526217827, + 0.3067067175920132, + 0.38013506969790006, + 0.39745466263884094, + 0.4676743560870999, + 0.46529583847424777, + 0.390850560179329, + 0.4125707051831724, + 0.38005883502154486, + 0.6871193929510275, + 0.3785812132407774, + 0.34616409609182797, + 0.7908135262402353, + 0.3727190710234272, + 0.45821597616600457, + 0.38154529477947635, + 0.45488947320005396, + 0.3221104683101513, + 0.441227742877532, + 0.36371563095876464, + 0.5053643965309994, + 0.28776197472875736, + 0.4203999040083465, + 0.40725404236076684, + 0.5230512772566286, + 0.4330871536193988, + 0.47812637725237167, + 0.31488195997683166, + 0.49181858481501317, + 0.434180993841136, + 0.39023494090903255, + 0.4788537982038594, + 0.41817463709280817, + 0.4852989572792358, + 0.42955263737413796, + 0.4081936481010237, + 0.8112642911699388, + 0.35110075653017386, + 0.49194057889574994, + 0.3719604572458258, + 0.4407381206874734, + 0.3552814953124076, + 0.42092431037223393, + 0.46735877346516685, + 0.5715427726702963, + 0.3543036911696074, + 0.3985899231679954, + 0.3248074454677438, + 0.4163566737583365, + 0.48070998780672486, + 0.46767778040431063, + 0.49375264220977777, + 0.3869207529140911, + 0.37999723201155355, + 0.47612069000675866, + 0.4457426930672085, + 0.4582590404096696, + 0.34311817422829294, + 0.3321394572136143, + 0.5026753519849793, + 0.4636991330116166, + 0.43387297882224124, + 0.4063300835518039, + 0.42846302135818626, + 0.8375389419463941, + 0.34126845289018326, + 0.381770511453294, + 0.3827567746666962, + 0.3579484241807397, + 0.4021453199561752, + 0.39993003812099875, + 0.35179464096334845, + 0.3702383270455824, + 0.3643310658077908, + 0.4833062910352439, + 0.45270340383074553, + 0.4592818977731709, + 0.48511945183816213, + 0.4780908473640033, + 0.4039552589721496, + 0.573980276462792, + 0.44801726404608366, + 0.4775767697942888, + 0.5124181357907394, + 0.4690799897585095, + 0.42324888289446744, + 0.33491665794723313, + 0.4270149743248006, + 0.541286899448377, + 0.4973984647520792, + 0.36677371292827615, + 0.39212111278961903, + 0.43111786545908043, + 0.48472364732859474, + 0.3754120049886864, + 0.4193784983527115, + 0.5077242916378002, + 0.5590622673035892, + 0.3675492388968619, + 0.44218225115883375, + 0.3728086540094728, + 0.3770156908581199, + 0.37725420832663714, + 0.7958174542917187, + 0.4280394980046975, + 0.45871074698991166, + 0.37125763395536565, + 0.39531409317825045, + 0.4192870987936184, + 0.7108416853012011, + 0.4651813608767038, + 0.4398791326904243, + 0.46348404039625996, + 0.43599725966964975, + 0.4650888355287464, + 0.47577002080697733, + 0.5097864875429349, + 0.3096188745535241, + 0.38509337680190137, + 0.46472846014833097, + 0.4659283853391972, + 0.4320251063962298, + 0.429460938433091, + 0.42665114891778566, + 0.3639239020853718, + 0.40734729770192557, + 0.421230462348806, + 0.4208994019207403, + 0.6210330018243799, + 0.3978261636340589, + 0.43310127609889815, + 0.3569149515968152, + 0.5969935480401246, + 0.41771849274509393, + 0.38875818123423117, + 0.45168112656124493, + 0.4712210381661492, + 0.45930029377088966, + 0.3784454322435628, + 0.5665790453495889, + 0.49125815236289666, + 0.36685073339473456, + 0.4558406468469023, + 0.4648304596026836, + 0.4929192273790479, + 0.4136227276860236, + 0.47811801843077656, + 0.36799035303303895, + 0.6679069321696137, + 0.40830944415620696, + 0.4677474786081115, + 0.48619048945471954, + 0.4323712957351243, + 0.43124451762183136, + 0.45405906801276913, + 0.3525405005704419, + 0.4157868121137854, + 0.6101735415469298, + 0.4054988044678586, + 0.3922364684683717, + 0.5302390781397041, + 0.44035508067574364, + 0.42694383099648825, + 0.6365715469239674, + 0.42814200706504113, + 0.4965750146745851, + 0.4453404769884101, + 0.4220083459368042, + 0.5130062708586035, + 0.4142506064311888, + 0.5060173100079667, + 0.4326232218658111, + 0.49429290751322363, + 0.3856868227382492, + 0.4864587794574397, + 0.44301854950439473, + 0.3691267145243099, + 0.34353961387152165, + 0.4065578643886166, + 0.44970529880587423, + 0.5647042113839316, + 0.4770887574774493, + 0.4651828777970056, + 0.5044432933141075, + 0.4108657647589439, + 0.4037132728638251, + 0.4974667864858987, + 0.38420292808954587, + 0.42257521819944815, + 0.5049003873240951, + 0.4175903708283103, + 0.5125878823035308, + 0.513737408057173, + 0.41950717448293195, + 0.42155949834858125, + 0.5322449250768365, + 0.43313886269033686, + 0.45418647583891397, + 0.3561825975572239, + 0.5140287185000018, + 0.612725507326115, + 0.7977779754943867, + 0.44544312383662166, + 0.4235572499333386, + 0.44317076760877666, + 0.4018788496752601, + 0.3780304443916022, + 0.39006194678696726, + 0.478221582921657, + 0.4695987145824781, + 0.432007029053967, + 0.46223611433600237, + 0.4976104735060435, + 0.5158245743054813, + 0.387382682009611, + 0.3759904084237, + 0.47630226331290515, + 0.4301481966992762, + 0.40639122096765923, + 0.43772361950606575, + 0.4153227440738607, + 0.42173668319981017, + 0.3300294029476199, + 0.5088302268220896, + 0.4554779390299526, + 0.4110620218113451, + 0.41859731133910494, + 0.3305323785405173, + 0.46939770199825864, + 0.3138668333319866, + 0.5486129428198224, + 0.42405723893837544, + 0.36197258285186795, + 0.4665651740902779, + 0.437623096253965, + 0.3692586179561898, + 0.6171553192835025, + 0.5334284334962502, + 0.3598888970637934, + 0.5005213823375436, + 0.4354629775363723, + 0.45526922511165935, + 0.46897540247183783, + 0.32091959024734956, + 0.4105206626491256, + 0.5226092734256893, + 0.3933617911421322, + 0.459507550790535, + 0.44172087701549534, + 0.4758605074423652, + 0.4936705361253977, + 0.41495747494846213, + 0.5183698317767089, + 0.4778935280925492, + 0.41844224879048714, + 0.35884127077326183, + 0.4716472042993231, + 0.375866003811754, + 0.41970683560732186, + 0.4294037054348163, + 0.49214731544098816, + 0.5486561284474256, + 0.47462328033893764, + 0.43445996017572425, + 0.48648505597260217, + 0.4584661620327292, + 0.4512476611275021, + 0.53451470247238, + 0.4174617420308278, + 0.3431374659149613, + 0.5085475846737153, + 0.4400239608843301, + 0.46744986839360886, + 0.45697118301222905, + 0.45449551944697836, + 0.5137284953479713, + 0.38384633120856226, + 0.4473865175888959, + 0.368957546527744, + 0.37517690272995374, + 0.36624992072628076, + 0.44129833177300404, + 0.37932291944997093, + 0.454279483184852, + 0.5373458874627304, + 0.45587970668783234, + 0.49667744088151, + 0.7808538158657187, + 0.5691254752908211, + 0.4630727015712861, + 0.4273773038629264, + 0.47432749542106206, + 0.4756473459490583, + 0.3591925095039998, + 0.5843942141061481, + 0.4891547534981131, + 0.4689081461917686, + 0.48359535524129443, + 0.4168284239426825, + 0.5135246329040839, + 0.5237369749163895, + 0.4270908101565382, + 0.5512073432433227, + 0.5168786480124489, + 0.40169184724874235, + 0.45677967649032203, + 0.5310412096290404, + 0.43353112354328854, + 0.4624989538730011, + 0.36947117215334074, + 0.5693013602139173, + 0.5307922253966104, + 0.8257871354096622, + 0.543879876163993, + 0.4628641146071092, + 0.2944117369494174, + 0.48418907513283993, + 0.44121068003535663, + 0.6168110964369079, + 0.3619100186693959, + 0.5642785672445387, + 0.35704808684392064, + 0.395096215806715, + 0.3929457212420447, + 0.37247359401627816, + 0.4622341974808165, + 0.3810815597825375, + 0.4420890844547341, + 0.460410748713516, + 0.4308290092979176, + 0.3846430009238419, + 0.3797014479118439, + 0.4308306404771292, + 0.3728153511840638, + 0.5595985447506557, + 0.44054693552415597, + 0.5195824046504537, + 0.4374388011493708, + 0.39596676393421365, + 0.5697305626775094, + 0.49181598084009187, + 0.3907963286983207, + 0.603264365816346, + 0.3316565117038769, + 0.4657774554846492, + 0.37931538889369276, + 0.4013725480705744, + 0.34681997851047425, + 0.4127408639746814, + 0.5063476477699392, + 0.3672503056673642, + 0.699537356134443, + 0.4671942759521941, + 0.39583114341316755, + 0.42760816592792367, + 0.3943486301311703, + 0.6079381901486091, + 0.33521485659397393, + 0.4043576192449296, + 0.4030237665093663, + 0.4542537046135698, + 0.3818821236283159, + 0.4329844864747994, + 0.33058075681301274, + 0.4730994988256061, + 0.7700180188158856, + 0.4980088934799381, + 0.3983031825021102, + 0.4185150634864411, + 0.4467384307044327, + 0.3570034664334981, + 0.3902743199401977, + 0.46409979548514635, + 0.755293705616967, + 0.566172421081612, + 0.4683135928552161, + 0.4084056185739177, + 0.38711500493167833, + 0.426593539452954, + 0.3513789671362342, + 0.4835628739696582, + 0.5396296424743012, + 0.40724927572111114, + 0.41064396437136796, + 0.397443759193241, + 0.5133108993208132, + 0.37383370549727246, + 0.5555971706445062, + 0.48484125686135193, + 0.3142457779784299, + 0.42344171585304985, + 0.4453869329173789, + 0.5374071459727233, + 0.545316293123965, + 0.25328104642345894, + 0.7295257044857717, + 0.7154455306326356, + 0.21795445747677547, + 0.3443291354382151, + 0.44195026890171507, + 0.47680223553051015, + 0.31944389155054925, + 0.423606913845335, + 0.474416153826106, + 0.48398295736212515, + 0.4955065307718215, + 0.3914890590058952, + 0.4392599465378928, + 0.37567896307900983, + 0.3352257499066308, + 0.4516908898836603, + 0.4045152706422603, + 0.39951246999618056, + 0.5224554937136922, + 0.4485412764999491, + 0.5424660761357082, + 0.45591583256176155, + 0.6155432090885283, + 0.4421057718381067, + 0.4851913432356139, + 0.5106713153548754, + 0.46871848112657194, + 0.3504177955875545, + 0.6366277529675288, + 0.633231767088762, + 0.46460119887985185, + 0.3887990237951707, + 0.412478041252424, + 0.5442657929529156, + 0.46334612760795174, + 0.513828408839768, + 0.5113417323337273, + 0.49007909056759724, + 0.45543813388976057, + 0.39253149632942, + 0.4220524006293015, + 0.578867416257785, + 0.5044050055545398, + 0.4821724527954705, + 0.5049696040940496, + 0.41901611540919564, + 0.27881055004489097, + 0.35511685977646723, + 0.3995315957360121, + 0.5436608937785506, + 0.4807684054761373, + 0.3698071286306747, + 0.4872653040133099, + 0.3861439425669155, + 0.34672963741929597, + 0.4122489870567238, + 0.4288603930331519, + 0.4010836108261352, + 0.45513571378424955, + 0.6357767795447263, + 0.40143518031807907, + 0.46389360096535187, + 0.5782731070834892, + 0.4124405963753293, + 0.5202581504476486, + 0.6079933263798943, + 0.4707511324849509, + 0.501504070703562, + 0.4941455350490864, + 0.45649930498883895, + 0.34369409188930944, + 0.3778137662633023, + 0.506036183832999, + 0.5228071317342203, + 0.49581051757872496, + 0.3860017851974387, + 0.4844109733265989, + 0.5008325195642096, + 0.5277242349092881, + 0.41759162462315735, + 0.4784887827770537, + 0.39555389989871803, + 0.49029621231900006, + 0.456219227220029, + 0.43174199133119473, + 0.3901230120625093, + 0.47602579055227456, + 0.48441937646878097, + 0.3156315076331582, + 0.40463770289520395, + 0.449406244740796, + 0.4241172684137232, + 0.48058794191097054, + 0.506805139911475, + 0.5118511922842136, + 0.47121088517591897, + 0.5116836356947798, + 0.41146647807014314, + 0.4808794857047577, + 0.4835586584000059, + 0.4444074698926776, + 0.4560420359901179, + 0.4672180286032998, + 0.4613062304381268, + 0.47947660780270107, + 0.4889679963648684, + 0.7283257796708131, + 0.554098996754731, + 0.5141470756120601, + 0.5454903692459201, + 0.5273423293485832, + 0.5055250384500036, + 0.5476818642232074, + 0.37623445525925137, + 0.41835868760580835, + 0.48121476783421285, + 0.3584561204907256, + 0.4627289524024399, + 0.510670065271397, + 0.45123157145180254, + 0.45594156188626445, + 0.48095166238859716, + 0.3442681534341848, + 0.5579230545835213, + 0.5011040578010478, + 0.4012945389990261, + 0.45290441180080504, + 0.4013172784705204, + 0.3715271586716578, + 0.4740573069151097, + 0.46500085782658346, + 0.4423899952561199, + 0.47684757957366347, + 0.41109404234343344, + 0.4984367231513571, + 0.38223107306678694, + 0.4043589251624292, + 0.44450121127378206, + 0.5076664358979465, + 0.5006569510022103, + 0.4821504575696586, + 0.4226113192368969, + 0.3648274868540793, + 0.3668932234770358, + 0.4595642021724491, + 0.42379341517071123, + 0.4633810850141802, + 0.49621188122142157, + 0.4450056759422844, + 0.42131838776139485, + 0.37195450564224164, + 0.42565435296715975, + 0.5438777760373152, + 0.4224015236658332, + 0.2162418869649797, + 0.3750958813409628, + 0.45700213491090175, + 0.32243072586071264, + 0.31231254720663737, + 0.4756448097538589, + 0.4736572821002955, + 0.34412616596467516, + 0.3220469368340443, + 0.41008998557020293, + 0.4024670760817422, + 0.5468355078165265, + 0.3253012805008963, + 0.3327411264784688, + 0.35548838549284173, + 0.37452615765960207, + 0.28559068923757913, + 0.34814230645558625, + 0.32740646811386404, + 0.36207603277989836, + 0.33559732791688596, + 0.4509846307698918, + 0.3625029002526006, + 0.406122496322407, + 0.40794933477168027, + 0.49723857982829944, + 0.3519141065692232, + 0.41370151893979745, + 0.4067813443807664, + 0.41626936121383096, + 0.4109744746434814, + 0.41878147842814956, + 0.40688169741095803, + 0.6429238693764054, + 0.35290503172654836, + 0.45887824008869277, + 0.5874577157931149, + 0.5022140394662109, + 0.37507155868725917, + 0.4784703583052765, + 0.42906974929761676, + 0.47183509215941966, + 0.4836250514119097, + 0.3706464036136919, + 0.5871788502790182, + 0.4406778890967688, + 0.43734758993598627, + 0.3725465210784546, + 0.4451167943740144, + 0.40573815243760664, + 0.4713017638031781, + 0.5037123626652809, + 0.4608726277220753, + 0.47902824977686614, + 0.5785993398979972, + 0.45436446379313905, + 0.4208155350569639, + 0.5180119194451444, + 0.5863335513896532, + 0.5394649091439363, + 0.4549838050751281, + 0.5855345338672574, + 0.57255195163683, + 0.39368663838500667, + 0.40564554746003123, + 0.4541450863342631, + 0.35683777974188646, + 0.6781821727526235, + 0.4961486226411291, + 0.43961964963096484, + 0.3646152448495143, + 0.47559428914913027, + 0.4112038087289693, + 0.5296659348316233, + 0.448616529627306, + 0.47545134644025877, + 0.5434092166257024, + 0.35296282143012003, + 0.4319434961039945, + 0.5626207028357944, + 0.3607437834065698, + 0.4431116674161649, + 0.3400056948768645, + 0.38562745161621736, + 0.41456192292130356, + 0.41542660198952047, + 0.4866644155219115, + 0.359943414016939, + 0.3407369906276564, + 0.45026891601194347, + 0.4127348698214912, + 0.387749933457892, + 0.46492179690606394, + 0.475620279987119, + 0.44077144082197944, + 0.5390127958594443, + 0.5708641129786548, + 0.4762438480961486, + 0.37178315320791405, + 0.4903315573665303, + 0.4538674709338802, + 0.40751242782145575, + 0.44820812123849896, + 0.34076602168800274, + 0.34230045893012057, + 0.3212720458958636, + 0.3807893304045168, + 0.44406247647317754, + 0.6575000231451874, + 0.4464572299898121, + 0.4371838348428266, + 0.3085348195887191, + 0.3673032766459204, + 0.32829825451970696, + 0.41869036998570025, + 0.3788760690506581, + 0.4015810395540193, + 0.3655071181023847, + 0.4296424512297824, + 0.4038670378115535, + 0.43181537627705907, + 0.42066196283770646, + 0.38287280972689486, + 0.5086337494103264, + 0.5037281695770167, + 0.39356888527889927, + 0.4393404355514511, + 0.45140180279170994, + 0.3983518620293105, + 0.37138127680785404, + 0.5188166632802355, + 0.4572168388985427, + 0.5882793007109036, + 0.46065527747145374, + 0.45250457763768936, + 0.4537102903553867, + 0.4609535217980841, + 0.35811653433114, + 0.6239491307491255, + 0.41762138896779794, + 0.23930368261558213, + 0.4865783215665186, + 0.5957006904540069, + 0.3823385044797636, + 0.4918124720016193, + 0.764389490928117, + 0.4596230505232914, + 0.5350345202036161, + 0.32096515071847276, + 0.5003249458636503, + 0.35698397886539046, + 0.38174069057341886, + 0.3606486750113469, + 0.314844001856918, + 0.457169506568796, + 0.26367192997770605, + 0.41382679505829, + 0.4640973427294754, + 0.44620941376762085, + 0.41218754813719716, + 0.39773744423380464, + 0.3670934152640393, + 0.46758719493065354, + 0.5311784114032423, + 0.5520856512079889, + 0.3669079270396069, + 0.4619202411790997, + 0.4318937238793245, + 0.7389647268314972, + 0.42636283404020475, + 0.5665581708588234, + 0.7407090956866685, + 0.7198401265618429, + 0.47913006737328295, + 0.40206004597934325, + 0.35180003328769455, + 0.46582667081070267, + 0.5107269898886068, + 0.5000363879931102, + 0.587591715982186, + 0.4333464058071365, + 0.5594312863044796, + 0.47615893624394207, + 0.4062116341225534, + 0.7071303021708338, + 0.6866940319297825, + 0.5029076452065213, + 0.42776008120538195, + 0.4823896313115114, + 0.431385926329203, + 0.4652550960043118, + 0.650306724631731, + 0.6749355765270338, + 0.4380801133216157, + 0.4481780519493837, + 0.4476573176061011, + 0.3987064075348238, + 0.5041571893231412, + 0.37275994809909074, + 0.5428208943610765, + 0.5378124112759052, + 0.37650701928930946, + 0.46294484229167315, + 0.4730942861318929, + 0.3917248178561165, + 0.545872023241415, + 0.5248159929111543, + 0.43986587544051176, + 0.3943731051588734, + 0.5022230108255097, + 0.4652417755124886, + 0.37797468886103824, + 0.36080982750901025, + 0.30156846163166146, + 0.26565000214560724, + 0.3180216787945674, + 0.5018718510083158, + 0.5535293601013661, + 0.41211334675207656, + 0.3289226609111089, + 0.41648895600203373, + 0.49996104346059916, + 0.49681853000768517, + 0.5283618803867793, + 0.41957967044535716, + 0.5052904966588235, + 0.5563989241359341, + 0.4266586941746012, + 0.5864632382774976, + 0.5429275402477928, + 0.4637703239793457, + 0.4440147394457679, + 0.39506790194776453, + 0.2658457929267822, + 0.3085774082794335, + 0.3004063926962166, + 0.5629902640510486, + 0.47800322261704414, + 0.44142208238801767, + 0.5601361051552166, + 0.5819479048922978, + 0.3406142240007583, + 0.48869333645774016, + 0.3981689221127634, + 0.7230159029207652, + 0.451564692264321, + 0.4668078836001433, + 0.3755970341467343, + 0.46397979332587574, + 0.5057124692766665, + 0.45177828182604207, + 0.42302505218382686, + 0.5080475793328946, + 0.4660222478190411, + 0.5794831318883548, + 0.4342737383364944, + 0.508457184765728, + 0.4600115974899536, + 0.44493290552477005, + 0.4274527182093526, + 0.4912019895414581, + 0.451966084127419, + 0.505762480989696, + 0.38350092169971567, + 0.3627153713083333, + 0.35794413928388524, + 0.4348346597129973, + 0.4785714439406932, + 0.4175068683161135, + 0.48662646182179076, + 0.5120474863096316, + 0.4660948502982715, + 0.4413576792521195, + 0.45381085188222187, + 0.27824769756164586, + 0.4485118186625153, + 0.32286254455884367, + 0.4683272277886997, + 0.24468180868611747, + 0.6755892534567862, + 0.33798304813093466, + 0.5138268585581446, + 0.5972101275634848, + 0.5193840945728834, + 0.28408681746921677, + 0.4813639570762298, + 0.3843473632515483, + 0.7833114352263886, + 0.30213043462891626, + 0.5614306291361516, + 0.5297932908603946, + 0.5563984655796987, + 0.6424456131793446, + 0.4947351636240715, + 0.6522389711494895, + 0.48347807189408853, + 0.4367725829708629, + 0.3839851696916464, + 0.4806759941914702, + 0.3375320015969123, + 0.7262587563818346 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.49952563027719993, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.578538721188377, + 0.7654290189492409, + 0.7397003780233555, + 0.8222324273630752, + 0.7470351105580934, + 0.6913009209229668, + 0.7409851084168972, + 0.6841727838868342, + 0.4381051357473947, + 0.7245670878956528, + 0.3894665733192949, + 0.5207839198651784, + 0.42584489573120504, + 0.3935971661452734, + 0.4349185906840617, + 0.35330150743397337, + 0.40598474184471744, + 0.5970469688534678, + 0.5875392617526347, + 0.48276380304215855, + 0.36212139142819144, + 0.6938213494356191, + 0.4750981454497306, + 0.6108562073458517, + 0.3931823849412393, + 0.34947485217312235, + 0.45632113580589617, + 0.5619354726245496, + 0.6599239112032653, + 0.5172078520935496, + 0.33898923463168557, + 0.43931341462801504, + 0.43929095724300166, + 0.390594443329581, + 0.49929779582859324, + 0.38236652607649785, + 0.49714455939991303, + 0.5632410435750632, + 0.6252074333911636, + 0.6340323111786521, + 0.794789546768183, + 0.7861707624860087, + 0.7798024383846246, + 0.6687134810125326, + 0.48519128689340973, + 0.5550951965605037, + 0.7861672782316852, + 0.43897944226950386, + 0.42302615775615504, + 0.5637246507317369, + 0.46898627562072576, + 0.7817776070775464, + 0.7979414321581682, + 0.484217307532755, + 0.49642759980507306, + 0.423129530420005, + 0.4129769449738518, + 0.46137227087498733, + 0.45851565664867, + 0.5840280206102512, + 0.5426226097142703, + 0.5098487074760366, + 0.517667162538877, + 0.5596477087950391, + 0.6602863864414094, + 0.4570930078461109, + 0.4181984932088993, + 0.6290900912743185, + 0.4577867285563586, + 0.5345905824948819, + 0.556213096134135, + 0.5379567716456202, + 0.5080057679364729, + 0.5680955153270174, + 0.47853598134869313, + 0.5797873614444625, + 0.3643380839065549, + 0.3965361491574062, + 0.45364726985209075, + 0.3638761848056033, + 0.48073257639554634, + 0.6023114559234191, + 0.4377586257594319, + 0.4206799296666763, + 0.4222967296294871, + 0.5145815271328009, + 0.500861028100815, + 0.4740683678056109, + 0.39575948567592556, + 0.4250597031435654, + 0.38010608329909895, + 0.3844079723883448, + 0.5529187578604006, + 0.5755852397624189, + 0.4512500323750102, + 0.6151679631861467, + 0.40717420050714026, + 0.5894378911954352, + 0.43850014612080224, + 0.5031227166948998, + 0.4886161953754774, + 0.5593813841393461, + 0.33922371604498186, + 0.5009798102008367, + 0.37391373363658986, + 0.3894511249044956, + 0.5206730591206167, + 0.5516683006221682, + 0.37749400447315085, + 0.8156823465070202, + 0.8096729698783813, + 0.8143481512708097, + 0.8330156127843762, + 0.5918350493718684, + 0.7096790157376942, + 0.7269806829281923, + 0.7640543654873554, + 0.7163359200364436, + 0.5918824712431799, + 0.6678678497314754, + 0.5785112822696006, + 0.7791451653675908, + 0.6709064576873923, + 0.452280903772715, + 0.3605897265992551, + 0.3358088566771559, + 0.3685781429828538, + 0.5846824365888108, + 0.45440719517990535, + 0.32058270753978074, + 0.5020258351646115, + 0.46554469492190587, + 0.5176962096853972, + 0.4213301618680753, + 0.44651898714166155, + 0.5011318603635487, + 0.49719723853042946, + 0.38884916935200464, + 0.42384771454671394, + 0.42870083938599385, + 0.3919580542406973, + 0.3407117143629108, + 0.45893009087490905, + 0.615753982497998, + 0.33405414048718995, + 0.5805485880782114, + 0.38355666581528874, + 0.509903492621251, + 0.4595019913031752, + 0.582085828245024, + 0.34724174628507426, + 0.5190775335959579, + 0.3587032424398194, + 0.38167110282490135, + 0.4427790605386613, + 0.37916590755229906, + 0.30012107349527306, + 0.2199806292686649, + 0.39135937442601904, + 0.3903182429313879, + 0.5607565044713552, + 0.5469242122598781, + 0.6228502728038556, + 0.5544276472595739, + 0.4048942230993317, + 0.5183593259602273, + 0.5657710205598434, + 0.5551602260261497, + 0.4614096545730412, + 0.5062097431326819, + 0.45324465080410115, + 0.32582837219633043, + 0.5101810555770331, + 0.4791949846897567, + 0.30466348357568546, + 0.3248938458345705, + 0.5320178078447511, + 0.5994841964417829, + 0.5891567118264246, + 0.5133435871634288, + 0.35372883790498943, + 0.5844089267261006, + 0.5312212057519994, + 0.5484978969585633, + 0.3364532588023987, + 0.5853707574880718, + 0.49431960655474255, + 0.5491998815589731, + 0.28377105137931014, + 0.46883427676474276, + 0.3713601785068443, + 0.30098679443306214, + 0.5619729144355734, + 0.47382440698337597, + 0.5880060684984846, + 0.4129372025600677, + 0.6013231151886087, + 0.3926288355587594, + 0.4874714463999933, + 0.424632364496034, + 0.3624508388878651, + 0.29875222805362, + 0.40501982029129024, + 0.4057737137087176, + 0.6489521198016416, + 0.3858278224592211, + 0.323571405244583, + 0.4930685586689391, + 0.5902421464241037, + 0.43961807946290477, + 0.49099670532446515, + 0.467470074506483, + 0.3758874746936432, + 0.4159584019812502, + 0.40296936027974983, + 0.5553806784335309, + 0.636270877214137, + 0.6368651519512065, + 0.5981551118981309, + 0.36919286513473154, + 0.5425771068983185, + 0.4890716633663641, + 0.30928254517237513, + 0.37121033276564874, + 0.40256467994111816, + 0.24576174962130876, + 0.5684988109594133, + 0.35961576500501824, + 0.35948596283073236, + 0.5571610931025894, + 0.3166772397258258, + 0.29978608909518506, + 0.553266086572038, + 0.43786351098584636, + 0.3512328245242802, + 0.6482834394475984, + 0.5717845006248384, + 0.3322815497877948, + 0.3416620923311665, + 0.5036438566470656, + 0.34936194437223317, + 0.4789207131124715, + 0.342820602594329, + 0.5218453769277155, + 0.4777263574052488, + 0.419043316213983, + 0.38401790450586126, + 0.3571661593324329, + 0.4801750990752835, + 0.3757266002215909, + 0.4653681722830728, + 0.35706307901434475, + 0.4265090834429738, + 0.3839408921826764, + 0.39370939903160274, + 0.4748669737227596, + 0.4366021035040873, + 0.4081561898609279, + 0.46359731939448334, + 0.46632927005303554, + 0.3567283571603582, + 0.5678880644395335, + 0.356003772747292, + 0.3892579308821497, + 0.42128448442844124, + 0.34866958949890514, + 0.3868682664148868, + 0.5378999961326533, + 0.3019295211138756, + 0.37766217188221635, + 0.36667178357554925, + 0.37986972504049277, + 0.3954805501377703, + 0.4312719422741659, + 0.4062157608796118, + 0.5108246685941156, + 0.31799591367835217, + 0.48229623511546876, + 0.5291545870836469, + 0.3368183035897647, + 0.3340195149524129, + 0.6391124842657143, + 0.507751858817803, + 0.30775853993420854, + 0.5772751456682685, + 0.5344401828960075, + 0.5058696322629231, + 0.509280222414469, + 0.5888216729912235, + 0.5585489188078689, + 0.57859027517703, + 0.5647527363938973, + 0.5666756239337881, + 0.452159067342826, + 0.5052653223251051, + 0.5750018598100103, + 0.5144767911857381, + 0.48907015374826623, + 0.6635771931002037, + 0.35520212259212225, + 0.6146208375254395, + 0.6759949995645128, + 0.731270706737463, + 0.7053024412372904, + 0.5439551920470738, + 0.5811667885724174, + 0.44252197662761084, + 0.7028714758161344, + 0.5077886174695012, + 0.5218429038597524, + 0.36116084997497044, + 0.5427095471158505, + 0.4573573522705394, + 0.4663598638780988, + 0.5840843326820703, + 0.3765993011576485, + 0.45348670154937554, + 0.3524600866229092, + 0.4773364277245711, + 0.4125428131259687, + 0.36865249710140086, + 0.4148160969332912, + 0.3062847395826643, + 0.37117196762873106, + 0.4067541603993141, + 0.41514907267070544, + 0.40239971366360994, + 0.41178527989768093, + 0.6231625977029986, + 0.5239817321246075, + 0.48235142600569586, + 0.5511566450145389, + 0.5095649937708265, + 0.3541826564989165, + 0.3337884065465128, + 0.6096729701012163, + 0.43561596599800856, + 0.3669848815963507, + 0.48074258477818144, + 0.5970391191078392, + 0.6267211679153417, + 0.47499729682742897, + 0.3840345493790459, + 0.5102028970457113, + 0.40468164888911506, + 0.374594225460475, + 0.5297451662162656, + 0.376119523919047, + 0.8280144245471628, + 0.5363507107742942, + 0.4373536297448437, + 0.4358082311163454, + 0.4700786713574479, + 0.467479114708788, + 0.43250216075671455, + 0.5442438481846957, + 0.4026589822310992, + 0.5712199092645234, + 0.46902289476488157, + 0.5396603412980806, + 0.5902832742708215, + 0.48392094587675877, + 0.5036752876050644, + 0.42232385945075535, + 0.44489319404595534, + 0.6297099645367613, + 0.599733994186557, + 0.45392839308926003, + 0.6211875278615414, + 0.49989487808158756, + 0.591476876121593, + 0.45982964911386726, + 0.4990302174388379, + 0.4517013829639535, + 0.5914589432777981, + 0.6070232687582383, + 0.646357373668637, + 0.5738574680585604, + 0.5675384609089328, + 0.5569655155988803, + 0.5199120993013344, + 0.4342210769568533, + 0.4273412066196002, + 0.3627950632654954, + 0.41925940250669846, + 0.3526536248552073, + 0.3831859251034962, + 0.4045805360333148, + 0.5539954098342834, + 0.38892619506628673, + 0.3640539740291115, + 0.42290785437720135, + 0.5926341526650506, + 0.6608634791516929, + 0.6202806251142429, + 0.6032278955923622, + 0.44232430581904014, + 0.6195966191902315, + 0.48701622891573665, + 0.6545345293559456, + 0.5038131618374351, + 0.6109964415500362, + 0.4771985527569671, + 0.5893452161142015, + 0.6175782856745792, + 0.5277702026125787, + 0.5188518899996625, + 0.5585997340454839, + 0.6218158334481686, + 0.5710235837551176, + 0.4498504155314778, + 0.4297567619907284, + 0.37796399780772827, + 0.40828862513069136, + 0.5573967093723818, + 0.6016730391253934, + 0.3519517316249566, + 0.6497078254988825, + 0.6280262386010653, + 0.513437464249181, + 0.43506174162987965, + 0.5044855220499419, + 0.5985189338989695, + 0.5602072249012089, + 0.5128478690234983, + 0.8625198905612327, + 0.5016435191731923, + 0.605597765922286, + 0.557993345480568, + 0.5918169541721648, + 0.49992430901327567, + 0.432502195794395, + 0.6314701513291615, + 0.49369401224247605, + 0.4541541373025884, + 0.4847835550729848, + 0.4805439593950697, + 0.4891081541119882, + 0.4525226395184636, + 0.48397849899879425, + 0.470658116558312, + 0.49665368869613535, + 0.3419628337913563, + 0.5103075589589198, + 0.35110446067112006, + 0.42378393736991643, + 0.3113035663875956, + 0.514230902406903, + 0.4082352823210592, + 0.3540051823093128, + 0.5813468301115461, + 0.4188996142641752, + 0.3131947019739569, + 0.5741260946894889, + 0.49631926585855507, + 0.4995332180254129, + 0.5531838361247621, + 0.4389836374846103, + 0.44610179602914357, + 0.6628231321177418, + 0.44796833828751503, + 0.5654087251650146, + 0.45812945199865096, + 0.3285371759401212, + 0.30644524422827796, + 0.4275421787174644, + 0.5271728120251117, + 0.6237123462108852, + 0.6125627563787969, + 0.5090840254909982, + 0.5709806408096959, + 0.5022017292262199, + 0.5339706903785957, + 0.46881614753080786, + 0.44956057019898604, + 0.46141592554899175, + 0.4441656514507448, + 0.4998578802886569, + 0.4347010349757886, + 0.47380662255513534, + 0.5570779508889765, + 0.49574555109561846, + 0.47709449400060555, + 0.40135616074670744, + 0.4399137207480934, + 0.5319220480673545, + 0.5855759316338295, + 0.5723355598331246, + 0.5861532513557087, + 0.6052788595119115, + 0.6360710780424339, + 0.6776300116594849, + 0.6600596619836475, + 0.4959141216760406, + 0.542115243648053, + 0.5174774571601009, + 0.5621321081171109, + 0.4340559280699498, + 0.5339259736458629, + 0.4989651644016573, + 0.592757325856514, + 0.424942246288514, + 0.6005122718024203, + 0.6630356258059059, + 0.6331036534920587, + 0.5833809951335885, + 0.6564804324253561, + 0.558064726195354, + 0.48494869338096047, + 0.4099577869475449, + 0.4650937055072678, + 0.365871889400658, + 0.42317022550800065, + 0.5790492173423093, + 0.6221926355543291, + 0.5033861882071715, + 0.4098205880643248, + 0.4233837173254242, + 0.5311071613622524, + 0.5245852568181082, + 0.4297311321321094, + 0.5441119533076695, + 0.522739421465534, + 0.4031944944968105, + 0.45531585253997087, + 0.4214275104355645, + 0.49817424444794833, + 0.4800646542846176, + 0.37235525585841484, + 0.4423436395810555, + 0.5040719563443736, + 0.4293247505085052, + 0.42020196974969665, + 0.45708810241061437, + 0.4160846295128148, + 0.40650915559664386, + 0.4726888517312069, + 0.5760005855008892, + 0.4549917457892014, + 0.3255778963495031, + 0.6060329295970867, + 0.4274220960900864, + 0.5854496896405891, + 0.5737220665600848, + 0.4113668412770472, + 0.543330791568768, + 0.6155354672148068, + 0.5305981595039503, + 0.6776779675106562, + 0.5041457916200117, + 0.5600203399460714, + 0.533995701329572, + 0.5190435830481841, + 0.5663310137935701, + 0.527743611100828, + 0.38801014069190737, + 0.508197056079351, + 0.549072160108507, + 0.45771943181164804, + 0.3447092205174374, + 0.4895182611166676, + 0.5076409906351215, + 0.49140331199271325, + 0.6655871274938182, + 0.5712811656470476, + 0.5776463084655128, + 0.5923855111341542, + 0.6599055840939256, + 0.5808651146910669, + 0.6367973054628384, + 0.6486464567991931, + 0.6367187875744693, + 0.4809275233753022, + 0.5019833357461949, + 0.45605036668671295, + 0.3672859421498605, + 0.32945656886563507, + 0.3972519199110917, + 0.42593812482063736, + 0.37402592675702045, + 0.34193273409875163, + 0.5253784106553497, + 0.3350438558074394, + 0.491396557277164, + 0.3766415386486297, + 0.5600545227609324, + 0.6206630769126078, + 0.5189039139666639, + 0.3681363310378048, + 0.4168973524372892, + 0.5893577954780314, + 0.37193945930000133, + 0.5204155245568589, + 0.34320252967937287, + 0.8373205596180285, + 0.5579103170632403, + 0.45178349730441725, + 0.6365393881133151, + 0.5942037583414976, + 0.5199836642154662, + 0.45699875869547074, + 0.5193969845997164, + 0.5968459679582943, + 0.4730102179506297, + 0.6116999159509829, + 0.5688788340292126, + 0.33514189229559405, + 0.5312648422772039, + 0.3387332035740316, + 0.4081877620473107, + 0.34976866570622794, + 0.5471236971868109, + 0.45554406974505196, + 0.4022587217907936, + 0.5562656689375495, + 0.5914177657891431, + 0.414027219017612, + 0.5174886280080632, + 0.45613237372016385, + 0.49281983413362607, + 0.5219890838062551, + 0.49546888665671696, + 0.37579686682475877, + 0.36851517597319156, + 0.6109824445941877, + 0.4462327870038123, + 0.40885845761501394, + 0.547740302573002, + 0.4373744345154329, + 0.47786479620153693, + 0.3602031642086037, + 0.40384450732419275, + 0.4531825864217844, + 0.5090261879687339, + 0.5131027391170436, + 0.38811513019957006, + 0.5109146802441757, + 0.34162863637024615, + 0.5374221365932967, + 0.48601721613462345, + 0.5188862670797914, + 0.5292673081987388, + 0.4377454671434778, + 0.4932378792212246, + 0.4129191836190685, + 0.534644716243807, + 0.44560519216075434, + 0.4852195523977536, + 0.5029674011010747, + 0.3859400159924193, + 0.5354761229664957, + 0.2977455480902526, + 0.673452961086372, + 0.6759119510012354, + 0.5741807240949641, + 0.5645308727095412, + 0.6187990581768948, + 0.6150224805702541, + 0.30431580120199847, + 0.42450865306290625, + 0.3854561128944991, + 0.43289980385217564, + 0.37055600051530013, + 0.3744679895837144, + 0.43051163366161743, + 0.5016368504528342, + 0.48565285731730307, + 0.3889873506887826, + 0.4115156148974007, + 0.3425499166431061, + 0.48016615761692344, + 0.2692825165048827, + 0.4508902496426971, + 0.5056821145440762, + 0.5144316003808007, + 0.5323862092139349, + 0.4546357540896908, + 0.5628529844834899, + 0.31811928117956173, + 0.5473293119346274, + 0.48103966741256965, + 0.4287393965481406, + 0.40988038762491363, + 0.44556200446966365, + 0.4540111348506457, + 0.4638663985120855, + 0.5040944298945126, + 0.5278847790775718, + 0.5662690228536986, + 0.38016149907754, + 0.41639520855858847, + 0.4905888011126314, + 0.5500753531216303, + 0.5391205913669096, + 0.5459803155236823, + 0.3734977805887114, + 0.6022313741388247, + 0.49431152471400197, + 0.482466026810636, + 0.6212696748033977, + 0.44776012628442646, + 0.3808527776009348, + 0.33928782003102254, + 0.4302903143557056, + 0.47630601444369536, + 0.4261321973232885, + 0.4385119303338991, + 0.450748820632179, + 0.5689671870398348, + 0.6192364599813516, + 0.6705445022686292, + 0.564236501195413, + 0.5797632910984688, + 0.5387840003709571, + 0.6075242088152888, + 0.5485127923797826, + 0.6208836260749284, + 0.3939873086402483, + 0.5089035989965869, + 0.6332900797259937, + 0.6804083707848296, + 0.2292532515379103, + 0.42761413934799053, + 0.44310380847832187, + 0.539208544165976, + 0.5314938863592212, + 0.5573696744615567, + 0.4210052242610474, + 0.535959134320471, + 0.7548801613653415, + 0.6244695730076122, + 0.7908607872141238, + 0.8050182869291392, + 0.6686600023113314, + 0.7661124811495552, + 0.7729512277340257, + 0.42706795980750323, + 0.4407914919147921, + 0.45307637605765794, + 0.5517848976758296, + 0.3743113818104405, + 0.5100383961564725, + 0.4029814393565699, + 0.5088208637267261, + 0.445782344731003, + 0.4265982921073605, + 0.3496010050418584, + 0.4552028262671103, + 0.4444033720679451, + 0.47807597244603667, + 0.45736524672139117, + 0.3520937272016615, + 0.37263033437433213, + 0.35966293732056276, + 0.3779592637458175, + 0.291198151823309, + 0.37708872864188053, + 0.3351332515956522, + 0.5558351062843935, + 0.5973146371401697, + 0.3980695852892903, + 0.34035412586598224, + 0.2897116412895092, + 0.5443516372222907, + 0.6470458780183238, + 0.6514536304502427, + 0.4000917410553069, + 0.39109173288969434, + 0.2854614210356805, + 0.5243507483076826, + 0.3522317149838892, + 0.41726873950635957, + 0.507236970251298, + 0.5778420792227456, + 0.4936995356093282, + 0.3419553883097752, + 0.48956589340619383, + 0.5496333886447885, + 0.539213413500713, + 0.4867199927262129, + 0.5354223710992683, + 0.4409252637798408, + 0.535855011848116, + 0.5116193170209034, + 0.5756057862913251, + 0.3804262956532132, + 0.4997685933958898, + 0.6858915922050721, + 0.4712638016277108, + 0.6852547378912306, + 0.41329693712992277, + 0.5643854853504354, + 0.7218147703772587, + 0.5875805881549936, + 0.4257936224294507, + 0.40187423065982003, + 0.3714838683606552, + 0.34629745198900874, + 0.4912972713207554, + 0.6204199084333297, + 0.5004679029300916, + 0.4379356372735628, + 0.3648637154673811, + 0.5464823060140227, + 0.5365986840332717, + 0.5765623116009303, + 0.6092521959511853, + 0.5618908448372029, + 0.5253736415168526, + 0.441509644250655, + 0.5949053827847725, + 0.33061774751875184, + 0.4402222922783345, + 0.4984281252313996, + 0.44275058099497344, + 0.6492744050670372, + 0.6539245333917268, + 0.621600077788666, + 0.6822993393033524, + 0.7578554850380117, + 0.614406987342164, + 0.7671293328674763, + 0.5427620449259681, + 0.48604810135088, + 0.4691162725140513, + 0.6114978673809652, + 0.3961540734444779, + 0.5425021962578809, + 0.5580666423335725, + 0.5475565238846323, + 0.5118866205326601, + 0.47309283125850987, + 0.4807085982038437, + 0.5280357491283941, + 0.44213728489060145, + 0.5759208021836264, + 0.5504980896879373, + 0.5663683916262064, + 0.6071965831559182, + 0.6598201953638145, + 0.5913852726950246, + 0.46497903964851134, + 0.6169585935471936, + 0.598539475791906, + 0.3876626194658933, + 0.4708460364820189, + 0.44597941711954026, + 0.4449922231654865, + 0.4286119834837518, + 0.4559786072748719, + 0.5137597192839511, + 0.5034333784855083, + 0.518026404407137, + 0.49948909687661, + 0.3290909352100291, + 0.48277519362676297, + 0.5677604350390614, + 0.5149327879713016, + 0.6192925300180291, + 0.6495327261259195, + 0.749387544848536, + 0.3723127655662722, + 0.38754312783857775, + 0.5406501348298066, + 0.43355490147038267, + 0.3157248850166654, + 0.3809610197197682, + 0.3626043306820786, + 0.6918279821416623, + 0.6439660050278797, + 0.6708667435138563, + 0.38351980186054174, + 0.43169076259842903, + 0.350751788658536, + 0.5975125374446759, + 0.5063431551749432, + 0.2996615510671507, + 0.3527353462987677, + 0.5013237353157315, + 0.4577497401183574, + 0.5249859909122432, + 0.5358803659199007, + 0.4529278366202262, + 0.4438485870318159, + 0.33166605094316937, + 0.6988569830925105, + 0.5791367273991119, + 0.7079430812931451, + 0.5953241165246509, + 0.32669098806445135, + 0.5494844368884251, + 0.5649942188475627, + 0.5643717625055945, + 0.5178349648538847, + 0.4263001806641861, + 0.5330292968541143, + 0.5806948103028013, + 0.5754542777630381, + 0.5183666407645984, + 0.3995817776118384, + 0.5122884908513197, + 0.41729461680927277, + 0.4906532018893196, + 0.6514851058465383, + 0.5990253261823588, + 0.5994330503957367, + 0.4780812269409179, + 0.5063132608000764, + 0.5051506962454299, + 0.46161317277453723, + 0.5331231825455152, + 0.45029546410617166, + 0.31741447904130127, + 0.4925657275660655, + 0.5784827897978622, + 0.3026726778181563, + 0.7686812835286552, + 0.4674989193998399, + 0.6164775369409143, + 0.7411705419978128, + 0.6893941662788692, + 0.41416817394615185, + 0.8128250341340195, + 0.5103739223579131, + 0.6816979655327168, + 0.42019143895940186, + 0.6013676225740512, + 0.6050521695509454, + 0.687194833898831, + 0.6420828590721095, + 0.7049081988064765, + 0.6286636647978037, + 0.44820715428157426, + 0.6031672558411566, + 0.5651643834711899, + 0.4876482160624087, + 0.7094037957848208, + 0.39210441684999014, + 0.6205958085142849, + 0.6253953286810936, + 0.6974632978125372, + 0.6459626159478501, + 0.7111924070117798, + 0.5906472927669588, + 0.436980090506386, + 0.5524886957025954, + 0.5635428090354838, + 0.5500052113940835, + 0.44177205607703657, + 0.40548631225549103, + 0.4844299539354013, + 0.5636178676742031, + 0.576051591564266, + 0.45280713358250757, + 0.5920037989422496, + 0.6035039377829277, + 0.46018587250523574, + 0.5851295320923277, + 0.5759393026252735, + 0.5506025152529386, + 0.48783826392734886, + 0.43157223574928044, + 0.462685840530328, + 0.5282969119060826, + 0.45637661681932906, + 0.3831234421301542, + 0.4831953056596377, + 0.3974823949083444, + 0.5242475440827729, + 0.4523662801885704, + 0.38177089699080224, + 0.39092624610484145, + 0.4083775343478076, + 0.5154681764870712, + 0.4931690403567912, + 0.39726768013749175, + 0.3605985059202975, + 0.4460738266768073, + 0.5977053249793037, + 0.6665603046248336, + 0.3502120105920056, + 0.5676961108106464, + 0.6159787795328137, + 0.6017383529791991, + 0.6041403907116774, + 0.4556722879659651, + 0.565346594081152, + 0.44039834695665947, + 0.6113946576640539, + 0.4821937418885274, + 0.6615371001603451, + 0.5960215798895875, + 0.6041767376281696, + 0.35050344406201245, + 0.3792647057662234, + 0.4340761756995462, + 0.5268270239237574, + 0.5045880054726388, + 0.5902175939702679, + 0.3722401615488986, + 0.6141443902481276, + 0.5436627998888103, + 0.583091877052696, + 0.5765761022085436, + 0.5471862534033833, + 0.4609478609203456, + 0.5746739230024949, + 0.5989280397584578, + 0.5834063448056138, + 0.5654696307666983, + 0.4732139368393684, + 0.3770100885913523, + 0.40255045706028625, + 0.4067480907641094, + 0.27921713395530673, + 0.42487906520914204, + 0.5524337204679285, + 0.467137250780215, + 0.4535480664549711, + 0.48194347323597225, + 0.4390521664495503, + 0.543616777151467, + 0.5741473487816051, + 0.3917718014235439, + 0.5144357949299107, + 0.5988472443940336, + 0.4012430854626426, + 0.4423007910726203, + 0.5316200332736636, + 0.5085238585056336, + 0.6217129035495238, + 0.38591802360621075, + 0.5582929112055841, + 0.40211853363554334, + 0.6630921587893807, + 0.6101657639907919, + 0.6662975731538854, + 0.34570311447616703, + 0.36654896315474783, + 0.3456343652603453, + 0.32290259386335385, + 0.3459789957256652, + 0.5027720940544017, + 0.5419997463271429, + 0.3163071844657145, + 0.44408822666868325, + 0.46359578008229496, + 0.49210418295699915, + 0.5673139291981485, + 0.5351472757882623, + 0.6747506877210946, + 0.3405972565210461, + 0.4784184205491086, + 0.48558063040068566, + 0.4442762256375959, + 0.4628735528455794, + 0.5691331006632175, + 0.5269938252993948, + 0.32805538040072496, + 0.56717435393831, + 0.49546755183654273, + 0.4667951635621884, + 0.6163835886878335, + 0.47141341537245335, + 0.3929773106299579, + 0.4244882070353424, + 0.4793904563890797, + 0.4738403426923261, + 0.4402751595530464, + 0.47212098290737375, + 0.628822038369218, + 0.48789579219273155, + 0.6416392841366766, + 0.5445022328699349, + 0.5348394304648812, + 0.6008148500487304, + 0.5570866507656929, + 0.5465058791205634, + 0.5320401386935796, + 0.6597682147850443, + 0.630625336737284, + 0.5854965273248836, + 0.6062607152965857, + 0.7515337722463613, + 0.4912050627790117, + 0.4885762685915614, + 0.3780028291617372, + 0.5592603519203235, + 0.5905022689806285, + 0.5524882107330453, + 0.4850131286380525, + 0.5723334423872574, + 0.39969367882813256, + 0.5146508582735269, + 0.4380064235528819, + 0.32331524026174263, + 0.4972744423767499, + 0.47968467898215034, + 0.4907513561702739, + 0.4929555417671933, + 0.36927769590694565, + 0.6407799720962719, + 0.6767791384191268, + 0.552170609908821, + 0.5933938292901674, + 0.6608289007753008, + 0.546129561851908, + 0.34778240718926867, + 0.45605304307598155, + 0.42970193984882044, + 0.4055776531934233, + 0.4390212787887316, + 0.46978460958248425, + 0.5201669986765037, + 0.48184649362413534, + 0.3555967017450167 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.39941893156349584, + "valid_tracks": 4, + "per_track_pearson": [ + 0.4333569748621104, + 0.36155937596426657, + 0.44141753876499973, + 0.3613418366626067 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.5915210285106998, + "valid_tracks": 626, + "per_track_pearson": [ + 0.6010955306827028, + 0.5883039439326824, + 0.5953496030799891, + 0.5915048012955123, + 0.6075298746010199, + 0.596672301337434, + 0.5221280104768494, + 0.5644409045445528, + 0.5998806895506311, + 0.5945061572527484, + 0.5961782222872277, + 0.5987311166346533, + 0.5699951612825169, + 0.5773628662170731, + 0.6025254141909048, + 0.5953823482319183, + 0.5843261891268539, + 0.5813562064818709, + 0.6080797611441427, + 0.6033820682261943, + 0.6199405641673622, + 0.6092414897906558, + 0.6074561051743278, + 0.5754682616847744, + 0.5784017241200676, + 0.6005447078541214, + 0.6009793519186967, + 0.5812414533388534, + 0.5913610932757339, + 0.5932790451320095, + 0.5967184095709109, + 0.6016380495290967, + 0.5829531068935343, + 0.612473424129506, + 0.5818185184582264, + 0.6144167773115761, + 0.567871516961494, + 0.5934001177549063, + 0.5851232462290308, + 0.608938091185922, + 0.610294376568126, + 0.5806497432917697, + 0.561773889233494, + 0.5830574615837292, + 0.6019950881941871, + 0.5726753950122367, + 0.5942627271630729, + 0.5980625371779416, + 0.5964283799232872, + 0.590099881528493, + 0.5720906572392969, + 0.5684771939674917, + 0.5905772009055694, + 0.5979162668019888, + 0.5744436646930463, + 0.5864119805479387, + 0.5968413540153962, + 0.5962280373623489, + 0.5999951245909187, + 0.5969991623288324, + 0.5838978771778192, + 0.5921754209458174, + 0.5998439822379309, + 0.5899705415037678, + 0.585267636155527, + 0.5809835601249499, + 0.595114846655751, + 0.5981777361827364, + 0.5795568698523195, + 0.5895965897693763, + 0.5993539780742712, + 0.5941008560394913, + 0.5961654876429813, + 0.5877934410879755, + 0.6053796363994278, + 0.6071249430757967, + 0.5883795630647688, + 0.5935996800404437, + 0.5816390855636673, + 0.5822841424150473, + 0.6181921385833334, + 0.60381600298096, + 0.5976116995678775, + 0.5818666779111215, + 0.585174462240502, + 0.5881649899025875, + 0.5967285320232741, + 0.5977535942108745, + 0.5986773645478464, + 0.595521249559126, + 0.5853672435378431, + 0.5881120284414563, + 0.5940308772248598, + 0.5915809392454021, + 0.6003672157182036, + 0.6130759632190698, + 0.5941080583104978, + 0.5966280865553841, + 0.6018923573238953, + 0.596711058506255, + 0.5801548396624268, + 0.5870153245292072, + 0.6045195118778267, + 0.5859867486306678, + 0.5972560800497426, + 0.5804281588464812, + 0.5697400087681033, + 0.5967902325907852, + 0.5947781151079298, + 0.5907622413984746, + 0.5850375929952688, + 0.5896466982433297, + 0.6042951629738337, + 0.5917960011378475, + 0.594651096213599, + 0.5972041852259772, + 0.5781606714761521, + 0.5809521634141681, + 0.5773102568236169, + 0.5991591153469615, + 0.6076011691392509, + 0.5906099057203619, + 0.5982443416126069, + 0.5883640035155248, + 0.5804149982439217, + null, + 0.6008438760374586, + 0.5837081201913291, + 0.5813713010751443, + 0.5993945097140592, + 0.5970015591826049, + 0.5869971181044905, + 0.5939098117422227, + 0.586725161846079, + 0.5877383814941153, + 0.5904149342672578, + null, + 0.6003478428815182, + 0.6078529366513774, + 0.5681558392239756, + 0.5819329268347333, + 0.5927582842307285, + 0.5815204008903445, + 0.60502608643392, + 0.6037391038644141, + 0.5898770060863724, + 0.5818425222450634, + 0.6004027794600674, + 0.5874835621944674, + 0.5886607277973008, + 0.5935484088161884, + 0.5979068592269391, + 0.5873049283983444, + 0.5951189535469085, + 0.5989798050079475, + 0.5926891228511358, + 0.5877309815422505, + 0.5963636019565904, + 0.5983743099300967, + 0.5924585673358661, + 0.5908567911197949, + 0.5969807826654124, + 0.6074423816362229, + 0.6074735964068139, + 0.5954926227694989, + 0.599278983712573, + 0.6010969289364685, + 0.5737916685088014, + 0.5653086520037671, + 0.5803279998890125, + 0.5972137733936376, + 0.5943955337816541, + 0.5904822159425066, + 0.5952209315069448, + 0.5973084819866961, + 0.5592400017499932, + 0.59290957840899, + 0.598207701937047, + 0.579954881228819, + 0.5910284518445487, + 0.5948929108368664, + 0.5937551327473127, + 0.5754892723468485, + 0.5981879374836145, + 0.5853765668272718, + 0.5611889672594848, + 0.6077134878870085, + 0.5952198194997336, + 0.6067041590081619, + 0.5814976861321097, + 0.6034744024124626, + 0.581330999187774, + 0.604878470330128, + 0.6016662985304999, + 0.5977935011891008, + 0.6049177287687832, + 0.5994189103245052, + 0.5727901626306681, + 0.6085411325585555, + 0.6073242838725432, + null, + 0.6031265430284293, + 0.6092499420622403, + 0.6017227731428693, + 0.6048559051650931, + 0.6087850071252274, + 0.6057175486600462, + 0.5966535703317384, + null, + null, + 0.5990199626870893, + 0.6066952220778424, + null, + 0.5854166685731678, + 0.5634765521085869, + 0.5831134923162874, + 0.6048992861911999, + 0.6079009217096629, + 0.612231731277558, + 0.5997176023201907, + null, + 0.6105348671922572, + 0.5907976993594429, + null, + 0.5435976058120312, + null, + 0.5733111104146392, + 0.5914064741783536, + 0.6027616668917303, + 0.6064675729551844, + null, + 0.5891101660632545, + 0.5918921751193685, + 0.5976736872427826, + null, + 0.6004776230107451, + null, + 0.5913577323455801, + null, + 0.5775991222797239, + 0.5913847724122334, + null, + 0.5874775795595952, + 0.5962593631599731, + null, + null, + 0.6045120529994643, + 0.6027102509652423, + 0.5984779688836085, + 0.6045611692649173, + null, + 0.5999092498581314, + null, + 0.6035714627187492, + 0.620585762137873, + null, + 0.49949587522927347, + 0.5988321575171336, + 0.600121016304357, + null, + 0.5943199825887434, + 0.6112187498994951, + null, + 0.6091847482844926, + 0.6070394153864985, + 0.6044335185598319, + null, + null, + null, + null, + 0.6139660401016243, + null, + null, + 0.5989775171632973, + 0.5983125728110603, + 0.6085691132167039, + null, + 0.607193765704603, + 0.607106992897358, + null, + null, + 0.6122664813330153, + 0.5973031321726384, + 0.5986312495454507, + 0.6058784916696052, + 0.6132660080129757, + 0.6084853433717806, + 0.5875625065312835, + 0.6082588489061955, + 0.6004452130382933, + 0.6003257490255304, + 0.6000539511936629, + 0.6117898197672194, + null, + 0.5849009275431631, + 0.6013238804877887, + 0.6009949394990342, + 0.5977729593008666, + 0.5766838575678739, + 0.5611396639795329, + 0.6062622128403147, + 0.5915071848162313, + 0.5968469360185196, + 0.5990638224610713, + 0.6010614667904165, + 0.5996662391071059, + null, + 0.5917614737383768, + 0.6167397974673636, + 0.6138499959414032, + null, + 0.6032811923497524, + 0.5190379025089846, + null, + 0.61130224294229, + null, + 0.5907388071859617, + 0.6008029397761616, + 0.6007587064245132, + 0.5852470790843981, + 0.5992148883373106, + 0.5980714444980817, + 0.5862456896984488, + null, + null, + 0.5996514185234371, + 0.5828466542549827, + null, + 0.6024302519759532, + null, + 0.6027692939919459, + 0.5974448744284222, + 0.5963084579818417, + 0.596617429964899, + null, + null, + null, + null, + 0.6103452915542443, + null, + 0.6018348845392925, + 0.605939743710574, + null, + null, + 0.598767890744132, + 0.6108936179949711, + 0.6052730915783227, + null, + 0.5924284321923174, + 0.5920158901583236, + 0.5972984365672022, + null, + 0.610181302544689, + null, + null, + 0.5986254997547639, + null, + 0.6041724329890185, + null, + null, + 0.593136020173561, + 0.6017507034567798, + 0.5948406412365487, + 0.5958957770438196, + 0.5889570580529467, + null, + 0.5856216979430972, + 0.5916589622355147, + 0.5845311927924727, + 0.5693009763833838, + 0.5648589807802781, + 0.5986757024263062, + 0.6022319025191387, + 0.526938096353288, + 0.6003872152417707, + 0.5789731678487378, + 0.5983278011326164, + 0.5873764421231716, + 0.5927845725435881, + 0.5914721546616313, + 0.6081509402518054, + 0.5887138320208021, + 0.5842641920874428, + 0.5877842887232391, + 0.5852295601209864, + 0.5875007005093367, + 0.5838627984617336, + 0.607524203575224, + 0.6291474499689241, + 0.5923737163242048, + 0.5159802686464507, + 0.6033342727793833, + 0.6008580863263225, + 0.5714662368165208, + 0.5932948285431834, + 0.6005991379299755, + 0.5907540620639901, + 0.5988199687658994, + 0.557777774313995, + 0.5233818590482525, + 0.5888258368314802, + 0.5075577758925777, + 0.5924617037929276, + 0.4852648223912857, + 0.5857408842344052, + 0.5883614155305457, + 0.6134332721354832, + 0.6222953717029297, + 0.5484080239030114, + 0.5524203018980843, + 0.5820800427800528, + 0.5887095028415983, + 0.5378875354650805, + 0.5649477568056863, + 0.5568070953405483, + 0.5790760520865146, + 0.5493739217092481, + 0.5659602254696179, + 0.521242240132764, + 0.5799595530941557, + 0.573514702339363, + 0.5306084298050309, + 0.5826707487975459, + 0.6079418596800733, + 0.5856507882431303, + 0.5816827332382998, + 0.5964253196864655, + 0.5670066055834101, + 0.5869482368035494, + 0.5872400301619296, + 0.6083971854617466, + 0.5900256177375534, + 0.6030840894714459, + 0.6006723043534588, + 0.6213155111952173, + 0.6122126931232347, + 0.5934542436886718, + 0.585116159332789, + 0.5800354521314145, + 0.5937170568786745, + 0.5964856941957442, + 0.587664664308125, + 0.5979076397710131, + 0.5934822093771465, + 0.573250602918825, + 0.5719167594929508, + 0.5878182212893008, + 0.5851082713936752, + 0.5728313908398828, + 0.5878592435112832, + 0.6115507084598576, + 0.5965103542711335, + 0.5782279017110148, + 0.5749014199359055, + 0.5866701981162323, + 0.5896741299868483, + 0.5802246448214183, + 0.5906094277442343, + 0.590628667435685, + 0.5969365177594924, + 0.5994239339578309, + 0.6064421315311738, + 0.5969761145104328, + 0.5965195648924598, + 0.5840827440664516, + 0.584450467903877, + 0.6034942855385105, + 0.5940063163691954, + 0.6058868864809162, + 0.5874311512038508, + 0.5985745384214592, + 0.6219175031892783, + 0.5941103538911029, + 0.578736288725029, + 0.5820497825588915, + 0.5953680840648288, + 0.5934725671163532, + 0.5889680289542043, + 0.582631461024061, + 0.5803502549221535, + 0.5836727011973567, + 0.5946786699041738, + 0.6021880868236231, + 0.5926591658131323, + 0.6007872015466722, + 0.5905505896884247, + 0.602047257123774, + 0.5870040450784927, + 0.6216890807410776, + 0.5915725121426815, + 0.5855028268015977, + 0.6136037177757977, + null, + 0.6019119524326295, + 0.5751815447390539, + 0.6000342779909509, + 0.6048883121886416, + 0.617489415512648, + 0.6004553718814192, + 0.602680050884179, + 0.6260960434508266, + 0.5956038068589993, + 0.6074951675463535, + null, + 0.5748247993046742, + 0.6116825828471277, + 0.622864723552569, + 0.6116013986468006, + 0.6198886274576065, + 0.6106013306530076, + 0.5856612160538395, + 0.610844056988873, + 0.6261389443938588, + 0.6321091302603016, + 0.5855362566245552, + 0.6309480719471968, + 0.6203106662585485, + 0.6263257858160094, + 0.6322092987454153, + 0.5990777775840783, + 0.5781007583623113, + 0.549713194458382, + 0.5510827005172169, + 0.5499505489341313, + 0.6222736643604134, + 0.6014160237043785, + 0.6106938865484419, + 0.6138784427394559, + 0.6110854126716917, + 0.6015926369596032, + 0.617114561966545, + 0.6083333553214882, + 0.6088570414092201, + 0.5907845265017794, + 0.579122692763056, + 0.5991531263149698, + 0.5981715508443716, + 0.5967186542163289, + 0.6119108415236838, + 0.6083677729777505, + 0.6052601322400932, + 0.5997961778803271, + 0.6036856607559413, + 0.5855314020756445, + 0.5931781962613629, + 0.576819334324183, + 0.5917533640598202, + 0.6005261991500589, + 0.6073648161276768, + 0.5754280459302676, + 0.5930662043665591, + 0.6018447471178443, + 0.6111003686390762, + 0.5922922946085865, + 0.582643622612855, + 0.609760230726385, + 0.5962081356569248, + 0.6055643850515239, + 0.6145777666930865, + 0.5660894568205201, + 0.6102162966269222, + 0.6075665221505554, + 0.6050125060599546, + 0.612091032912958, + 0.5802705450936881, + 0.5775983016561642, + 0.5934085546247346, + null, + 0.561124746848904, + 0.6247886194684982, + 0.5710288875155343, + 0.6255365848805093, + 0.5944952905022014, + 0.5815300228997078, + 0.5721838139969506, + null, + null, + 0.5916015358364498, + 0.6197266393021014, + null, + 0.5684650410443605, + 0.5492875966083084, + 0.5665118154984383, + 0.5913505553445016, + 0.6082172074273411, + 0.6003002572498505, + 0.5666961463777719, + null, + 0.5788682507262668, + 0.5861662897032829, + null, + 0.5487352640338997, + null, + 0.563268462064371, + 0.571647750063627, + 0.5716841582601357, + 0.5883713535382538, + null, + 0.5734043431278253, + 0.5644117475880387, + 0.6024313587707062, + null, + 0.5603529713323564, + null, + 0.5920165226899746, + null, + 0.5783740248910578, + 0.5825713610330984, + null, + 0.5777148859186625, + 0.561751857166987, + null, + null, + 0.5621333601426224, + 0.6244285719337428, + 0.5796078927496203, + 0.5832600321554411, + null, + 0.5926551332553551, + null, + 0.5867608285646346, + 0.6145888701624532, + null, + 0.46774093207232575, + 0.6002290329069669, + 0.5667919972420172, + null, + 0.6108536917107369, + 0.619293405983927, + null, + 0.5752649123528226, + 0.5553468171393178, + 0.5754250785680463, + null, + null, + null, + null, + 0.5906262613983801, + null, + null, + 0.6044792530372715, + 0.6001871456826391, + 0.5536911672578436, + null, + 0.5647997245431994, + 0.5706962278219841, + null, + null, + 0.5993184350502214, + 0.59449974044719, + 0.5947289730465667, + 0.6091662964708316, + 0.6170600805186127, + 0.622510112536907, + 0.5847553832342088, + 0.60585598629966, + 0.5898742820644063, + 0.6055654895662462, + 0.607326427977279, + 0.6067217712815053, + null, + 0.588155105175371, + 0.5895288029968173, + 0.5674261365240829, + 0.5812211741575286, + 0.5913092029198711, + 0.5673643771325368, + 0.5861930586497817, + 0.5618011541206531, + 0.5572491595120013, + 0.6028617765270567, + 0.6059556762954382, + 0.6013011621418024, + null, + 0.6022214257975949, + 0.578130645027166, + 0.5901838666954231, + null, + 0.6037078044581741, + 0.5221795680568749, + null, + 0.6110566088331716, + null, + 0.5599497615403574, + 0.607072985660565, + 0.5981545244105709, + 0.5778045678501736, + 0.582420278028233, + 0.5960627587958267, + 0.5942894647034525, + null, + null, + 0.5741416235729712, + 0.567423663988913, + null, + 0.588214381917572, + null, + 0.6018983162098482, + 0.5872236872010631, + 0.6124949677823701, + 0.5674188683422853, + null, + null, + null, + null, + 0.6232012389118184, + null, + 0.5912916551570114, + 0.5956151994426235, + null, + null, + 0.6159006501319091, + 0.6050488509846863, + 0.6022827218420062, + null, + 0.6115097766024895, + 0.6142036732689568, + 0.5804135159744953, + null, + 0.6027176415555425, + null, + null, + 0.6199629719965905, + null, + 0.5927029537340204, + null, + null, + 0.5896496041238571, + 0.5562285099273386, + 0.5584194996649233, + 0.557620849917694, + 0.5634514817830406, + null, + 0.5665794172292761 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978, + 0.4556436391090978 + ] + }, + "junctions": { + "mean_pearson": 0.38073057497615626, + "valid_tracks": 626, + "per_track_pearson": [ + 0.47249576471204957, + 0.3804534831420561, + 0.4125920328305197, + 0.3931589644892049, + 0.4939980763150037, + 0.5049141395744583, + 0.11778660156879028, + 0.32128180224805036, + 0.4061003615163488, + 0.33152820847132114, + 0.4590624396846516, + 0.39763297774987905, + 0.32168163976030356, + 0.4189765571523736, + 0.3601204137365165, + 0.3630245861407091, + 0.3584054853143759, + 0.3289030405405996, + 0.3965445939507814, + 0.45560087459933585, + 0.3519238857584378, + 0.4128508645400656, + 0.3127906958650049, + 0.4178831233626408, + 0.38973652053864555, + 0.2974953049839243, + 0.3914920014393729, + 0.3454288280792883, + 0.36074316382995086, + 0.3931511233153048, + 0.43601258906885376, + 0.47251856088532274, + 0.42287404261672196, + 0.45216122698200567, + 0.45031548254566706, + 0.39946767106115866, + 0.4218856122326274, + 0.38380067139858415, + 0.5041712308428563, + 0.34867346302255114, + 0.4968125427726537, + 0.39910648962122725, + 0.3878908896777736, + 0.4259794039902948, + 0.39403212070937144, + 0.36291283376800937, + 0.3227174517473769, + 0.39276758329664546, + 0.38014027850252047, + 0.3468487160372996, + 0.39227196335761544, + 0.3338439631695621, + 0.37161741661432884, + 0.3943122623805186, + 0.35514121971035306, + 0.40467688200487434, + 0.3490103390117654, + 0.2690788441037133, + 0.2903725842085234, + 0.4069458708285091, + 0.25913340709033017, + 0.26066757795490175, + 0.28996506536977507, + 0.29733989224362267, + 0.40314495687012075, + 0.3876371428479666, + 0.40460312652731184, + 0.3830164406618389, + 0.4235718679770916, + 0.32579197821403544, + 0.292025649083841, + 0.36973020588156297, + 0.2889405689381344, + 0.28154787189492575, + 0.41331649608490806, + 0.3400467381646737, + 0.32448445973863554, + 0.3009807368829338, + 0.3207362509718026, + 0.4024572405370697, + 0.3979003219597802, + 0.3199748908097807, + 0.34318731412697934, + 0.47752328485100665, + 0.4474116794843748, + 0.29325555687767796, + 0.3267497011744861, + 0.3384927910837309, + 0.28611417119204197, + 0.33495968249761143, + 0.35047021184132293, + 0.4167919310965039, + 0.29310270974453956, + 0.14781208793585718, + 0.38642439099890996, + 0.2541417014492137, + 0.4020508393704652, + 0.4323238041876071, + 0.42687570523797685, + 0.3788282496264661, + 0.34446067503949557, + 0.49368779324151096, + 0.3796626313843952, + 0.29179905710010745, + 0.3319620511910838, + 0.3008355519312383, + 0.3360214549602803, + 0.4143302898250261, + 0.3095817677371089, + 0.33490647977498306, + 0.36761482259545836, + 0.3852850210784768, + 0.4029680094790495, + 0.3730377828241898, + 0.3807133070576126, + 0.4094381219095719, + 0.2940342133411478, + 0.41002442858756016, + 0.3874382945492834, + 0.32565033380021247, + 0.2952309510752068, + 0.3705769828749817, + 0.40235548418078454, + 0.29647778709373857, + 0.3894867706908405, + null, + 0.3451531768460327, + 0.35961025368957394, + 0.20153754634112575, + 0.3107044578076872, + 0.3992269794468132, + 0.2445382894666262, + 0.40630738033166497, + 0.43521712599031603, + 0.36570578959700895, + 0.4722191451523516, + null, + 0.3580717856619318, + 0.3833719113482879, + 0.3845107323119867, + 0.44910318354452455, + 0.42987209083251765, + 0.39573795469213613, + 0.3258195647573144, + 0.3604552530649694, + 0.16250826305927532, + 0.4028003807403598, + 0.3726674458335008, + 0.31577268943127595, + 0.28847839548206033, + 0.4619523064701733, + 0.4041195194624574, + 0.27786258609140957, + 0.4135402682842967, + 0.5097366889124755, + 0.5029593398924141, + 0.3951106160219494, + 0.25385617377194364, + 0.4093153015027895, + 0.20482518511030873, + 0.4766370679508008, + 0.3384679804160361, + 0.42060257852891597, + 0.39004316433321917, + 0.377380134619108, + 0.36052851340887043, + 0.48006420822225926, + 0.2457986792810887, + 0.33989096500321864, + 0.3496956855331637, + 0.3695314382126624, + 0.34704212148245206, + 0.47750257772999516, + 0.25486433571191464, + 0.31218268298313956, + 0.454958606810105, + 0.2351616255527238, + 0.297035374969491, + 0.2620003014871772, + 0.4218689061263763, + 0.2846271758266087, + 0.313940956275101, + 0.431602526951409, + 0.47080699992269615, + 0.3162323634892368, + 0.38354419594589867, + 0.44096934497344037, + 0.46974413540240945, + 0.4850291561844149, + 0.35883822564285234, + 0.4244023215133789, + 0.36001727611235435, + 0.3659773001150593, + 0.48004055080586916, + 0.5280230474571389, + 0.5017171852785615, + 0.38757490954144147, + 0.31899078917444174, + 0.36978228201138164, + 0.33534153090370683, + null, + 0.3774102527826586, + 0.5209350995936676, + 0.4927656778460089, + 0.4338766521028879, + 0.30596933604320015, + 0.3675840134796064, + 0.32133711696292144, + null, + null, + 0.30511507345444616, + 0.49759412743443476, + null, + 0.35171316801654784, + 0.466116931570673, + 0.2635552151436324, + 0.33662268356074343, + 0.4389151039085497, + 0.43411418797995927, + 0.2921633736970451, + null, + 0.4421153479426744, + 0.4670589685377646, + null, + 0.3785283002351359, + null, + 0.38352927667619374, + 0.3754945519392478, + 0.2629905238192148, + 0.36860338331345394, + null, + 0.22156028130887584, + 0.2962091971296612, + 0.42788057162721155, + null, + 0.37000494449033794, + null, + 0.4090482252634367, + null, + 0.22520678533516766, + 0.40292274043675186, + null, + 0.3433578035631227, + 0.43787884078436595, + null, + null, + 0.3369747199195629, + 0.40775456369804236, + 0.3361955788201197, + 0.33621280338776766, + null, + 0.5198323251432763, + null, + 0.4221946230128621, + 0.40112632566252476, + null, + 0.4053404749409796, + 0.485914774766645, + 0.468623186912727, + null, + 0.4264325937713783, + 0.38851106958305975, + null, + 0.5451146849838532, + 0.5389102310956593, + 0.4694476773660036, + null, + null, + null, + null, + 0.35553222702541015, + null, + null, + 0.32757518757362564, + 0.29436025118451914, + 0.49648016658273536, + null, + 0.4539371105309725, + 0.43886789263713155, + null, + null, + 0.4058844408957724, + 0.42702214289707247, + 0.3458782037281479, + 0.4067827164674437, + 0.4618717256313174, + 0.44702467351580105, + 0.412294245548645, + 0.4135042810502604, + 0.427748984605986, + 0.31399502083766584, + 0.4079033919433276, + 0.51947858023374, + null, + 0.3816663527742719, + 0.5020051172391315, + 0.25805132601433534, + 0.316444379828652, + 0.4808883588881353, + 0.3803602037162248, + 0.38091665113289874, + 0.3336102863874223, + 0.42735481437678685, + 0.31459996500882653, + 0.4761368792598981, + 0.4324447595278486, + null, + 0.3984115897661295, + 0.3154759409528882, + 0.34587491254178143, + null, + 0.33390292135581867, + 0.32051304838474515, + null, + 0.5250851733976577, + null, + 0.3044850489144695, + 0.40802175697765203, + 0.37383885985605475, + 0.34912184318350753, + 0.3509523122096735, + 0.40420865042354753, + 0.366286020798226, + null, + null, + 0.372660997941695, + 0.3970513433443928, + null, + 0.43219876010106584, + null, + 0.38615610938869654, + 0.4021394759594362, + 0.3352851200853258, + 0.46855462703967243, + null, + null, + null, + null, + 0.3132366903073749, + null, + 0.37870407795228106, + 0.3827986002671564, + null, + null, + 0.4551403673450039, + 0.2907066748481925, + 0.3201752076934387, + null, + 0.554492197475378, + 0.40379782689032145, + 0.3898690161788722, + null, + 0.4407375945931713, + null, + null, + 0.3991160949352991, + null, + 0.27624432468639226, + null, + null, + 0.3307357876202282, + 0.4014433673046863, + 0.38320052298535545, + 0.4205082992397791, + 0.39260036639365303, + null, + 0.3717645795672527, + 0.4020374013869657, + 0.32988730962417856, + 0.6033381484205094, + 0.3516469939651021, + 0.5197305222439752, + 0.38588424438187074, + 0.3243095744139476, + 0.23731044141854502, + 0.37941184650589105, + 0.4198225480716052, + 0.286407504349404, + 0.3963011116132511, + 0.3769628200987957, + 0.3012644146982844, + 0.31796402416354297, + 0.3290089077823806, + 0.341672014765599, + 0.5091700890873532, + 0.5053131831468287, + 0.45371310054599046, + 0.4517785321982978, + 0.4729733660373895, + 0.34523417230215947, + 0.5678161138002308, + 0.25669202614499914, + 0.34083949278698955, + 0.46253254460235094, + 0.44062986994842734, + 0.38229687430240744, + 0.26269893297287455, + 0.33247203550465804, + 0.43938642884946283, + 0.3871725806289786, + 0.5370040759364213, + 0.19835741326040313, + 0.39557989166134, + 0.166963394351671, + 0.38349641721635375, + 0.5451706744095738, + 0.41231250874999686, + 0.42583388516699466, + 0.261876793054183, + 0.20091417096633812, + 0.2158684922926145, + 0.3586258179216833, + 0.2881398556364621, + 0.2620268748610143, + 0.43140364063799835, + 0.2192555056830657, + 0.33079315619922395, + 0.20304649129054397, + 0.2945426550455022, + 0.2092278022150431, + 0.2641826886093745, + 0.1928296730189476, + 0.3447792655120745, + 0.39531430308065657, + 0.34757671780888344, + 0.32299256449311664, + 0.3151997603376525, + 0.4077510194136566, + 0.3032275234718844, + 0.4193891679136518, + 0.34223666626657356, + 0.3427384349236539, + 0.25646580058442026, + 0.414814318643023, + 0.36748954549547713, + 0.24759070994969457, + 0.31250131527990177, + 0.3718760852055529, + 0.34160770222433046, + 0.3646907245286125, + 0.4345470814569034, + 0.3562020072149472, + 0.37924238827953116, + 0.39906771968354615, + 0.5056735966057516, + 0.3824550843748304, + 0.4052633626488631, + 0.3568676149144446, + 0.4149639749059343, + 0.4171617336086302, + 0.3368235897923769, + 0.35822851604259787, + 0.3532804361763743, + 0.468513025376736, + 0.4335636586534371, + 0.3389421701188712, + 0.3741335334282921, + 0.44048971369824097, + 0.380878592854023, + 0.38000940835665814, + 0.3490923063138893, + 0.3100674634626429, + 0.2919232533203213, + 0.4192129305705116, + 0.34951655883209176, + 0.45594862052118096, + 0.49568727079579566, + 0.2083419511586019, + 0.3388462416235523, + 0.3432494758706562, + 0.3698332902260078, + 0.423548007784129, + 0.3664195172307818, + 0.26987313142659464, + 0.411997961078304, + 0.4202313490787487, + 0.4404677687308341, + 0.3033526341698675, + 0.4436990850830639, + 0.312043404946257, + 0.39088370796691396, + 0.42231513804858273, + 0.3048176232886953, + 0.472134411956742, + 0.23359975553063242, + 0.3378776844917392, + 0.2833207222469079, + 0.5058269961392977, + 0.35166111190431876, + 0.3211857538122175, + 0.34804206308046187, + 0.49132921986506833, + null, + 0.388352102213147, + 0.4920912217378222, + 0.2861079725123144, + 0.47687361052218435, + 0.41654592528062545, + 0.2728070529252381, + 0.27580008453212207, + 0.4108711498544307, + 0.3806441252390214, + 0.2757240787765485, + null, + 0.3193250150282022, + 0.27423723815055334, + 0.2281455287615597, + 0.24743645527576166, + 0.3799588637689429, + 0.3097061617656676, + 0.3243741046788142, + 0.416508466063891, + 0.19457336572558961, + 0.24616378112989296, + 0.34409041937812196, + 0.339879001432065, + 0.3337451966186773, + 0.4056255423845542, + 0.5108666262871543, + 0.36115981982200457, + 0.36144771236616585, + 0.536737775971566, + 0.4716068776233024, + 0.3518029072425456, + 0.35923204850104057, + 0.3439496076212578, + 0.2565164241643998, + 0.3002569101175269, + 0.2826054756974131, + 0.5042837910368622, + 0.5376200146502831, + 0.34127591942828595, + 0.3070158747238659, + 0.46569972147379934, + 0.42111592395447683, + 0.345209422535474, + 0.4624879638794813, + 0.3271645416501974, + 0.33000615208684625, + 0.36100070936798884, + 0.2627518891127525, + 0.3004205574638779, + 0.35897788512267115, + 0.40195337678451454, + 0.31628131629869766, + 0.2908140336118943, + 0.38410047256039703, + 0.3073873908174893, + 0.269096820564568, + 0.5221284779433326, + 0.5060862610944984, + 0.4120362411705893, + 0.5424443853010893, + 0.5406364298990942, + 0.26709183224278454, + 0.4856250013187814, + 0.36168050078479835, + 0.5586065006002935, + 0.39569842980967707, + 0.3223644865903564, + 0.2716855664109379, + 0.3439070513844299, + 0.305447151264794, + 0.3208188817797219, + 0.4461025813238949, + 0.4244891713075824, + 0.5619652732717804, + null, + 0.3891791228123781, + 0.4247008465575635, + 0.5483109581282056, + 0.5281372798925074, + 0.45045293573257905, + 0.259245346383498, + 0.38606649342082533, + null, + null, + 0.4470817171680728, + 0.515965565348797, + null, + 0.21816421720016343, + 0.48041239030974325, + 0.19278332956370609, + 0.3438975966963034, + 0.34078447592691635, + 0.3467076480000501, + 0.3613323523302025, + null, + 0.4577236391006628, + 0.36682489675290897, + null, + 0.42904408974682784, + null, + 0.5579543781018584, + 0.39321421490410785, + 0.2883527615666325, + 0.32600976558765316, + null, + 0.418566506816533, + 0.33905815795292793, + 0.4809625485110587, + null, + 0.3978260708535296, + null, + 0.36897952888748303, + null, + 0.38983534943478576, + 0.44066085783814474, + null, + 0.4078453957953824, + 0.4538295799111559, + null, + null, + 0.43114052322490476, + 0.3757603951460994, + 0.22490726223224303, + 0.34679258329131396, + null, + 0.47555565873981864, + null, + 0.3223855150093525, + 0.4507109587600338, + null, + 0.23099391884965417, + 0.507794451851246, + 0.45171992557520624, + null, + 0.40341753444941275, + 0.5113293660954198, + null, + 0.45109082280957596, + 0.4897047340995893, + 0.46597338473574224, + null, + null, + null, + null, + 0.46958993484104544, + null, + null, + 0.49203657176354854, + 0.46471940587636834, + 0.5073766491148556, + null, + 0.33813995469301794, + 0.5044429312755567, + null, + null, + 0.35143566087588823, + 0.13975837181968206, + 0.4251569729920797, + 0.24859340012084294, + 0.3253028879742107, + 0.441468613420678, + 0.1956156123071636, + 0.34376043922138666, + 0.2100409201611079, + 0.3763526715285669, + 0.4233549728672382, + 0.34006949520752844, + null, + 0.3266504212888747, + 0.5050204393846878, + 0.5054910116740255, + 0.46311754625826235, + 0.2257483403724718, + 0.44667036260420034, + 0.412017093781433, + 0.4088399520464439, + 0.4543081576518673, + 0.40545899030804405, + 0.4594624522898452, + 0.3973833191285646, + null, + 0.39652849645073235, + 0.4768420490123773, + 0.4250004863523722, + null, + 0.5361061993825486, + 0.4012405425965828, + null, + 0.4243781141868821, + null, + 0.5147572105319128, + 0.4850434940852725, + 0.32438781577902587, + 0.4724280313943023, + 0.4786864047076655, + 0.5219522729630344, + 0.5391327747698785, + null, + null, + 0.4732890301346835, + 0.5785084472775968, + null, + 0.4050703329536045, + null, + 0.43555219788374056, + 0.44397999246960335, + 0.46672468269260314, + 0.45940720872230106, + null, + null, + null, + null, + 0.3691687713703965, + null, + 0.41351585283313197, + 0.572975146849416, + null, + null, + 0.47845143209992347, + 0.24799654366798932, + 0.3605552562842057, + null, + 0.3858791431840947, + 0.38826223310417746, + 0.38767663966025523, + null, + 0.3702239549118026, + null, + null, + 0.39623672591975273, + null, + 0.29748605759973046, + null, + null, + 0.4269785437906329, + 0.5244646642026195, + 0.5669597984979632, + 0.5842123189246041, + 0.4727609114431373, + null, + 0.3754432467923525 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.7284768211920529, + 0.7828947368421053, + 0.7330960854092526, + 0.7215189873417721, + 0.7715231788079471, + 0.7423728813559322, + 0.721830985915493, + 0.7361111111111112, + 0.7532051282051282, + 0.7269503546099291, + 0.7474747474747475, + 0.7016949152542373, + 0.7175324675324676, + 0.75, + 0.7766666666666666, + 0.7467105263157895, + 0.7377622377622378, + 0.7435897435897436, + 0.7484076433121019, + 0.7516339869281046, + 0.7086330935251799, + 0.7451612903225806, + 0.698051948051948, + 0.7509293680297398, + 0.7153024911032029, + 0.7532894736842105, + 0.782051282051282, + 0.7627118644067796, + 0.7533783783783784, + 0.7517241379310344, + 0.708904109589041, + 0.7313915857605178, + 0.746268656716418, + 0.7659574468085106, + 0.7289377289377289, + 0.722397476340694, + 0.7099236641221374, + 0.7460815047021944, + 0.6923076923076923, + 0.7483221476510067, + 0.7303754266211604, + 0.7835051546391752, + 0.7404580152671756, + 0.7481481481481481, + 0.7311475409836066, + 0.7466216216216216, + 0.7453874538745388, + 0.7168458781362007, + 0.7056737588652482, + 0.7687074829931972, + 0.718978102189781, + 0.7573529411764706, + 0.7159090909090909, + 0.7482993197278912, + 0.7425373134328358, + 0.6938775510204082, + 0.7093425605536332, + 0.7748344370860927, + 0.7363013698630136, + 0.721830985915493, + 0.7034482758620689, + 0.7639344262295082, + 0.7483443708609272, + 0.7508532423208191, + 0.7315436241610739, + 0.7689873417721519, + 0.7195945945945946, + 0.7483221476510067, + 0.7167235494880546, + 0.7777777777777778, + 0.7508896797153025, + 0.7258064516129032, + 0.7474747474747475, + 0.7615894039735099, + 0.7526881720430108, + 0.7728813559322034, + 0.7684887459807074, + 0.7424749163879598, + 0.7517241379310344, + 0.7789115646258503, + 0.7384105960264901, + 0.759493670886076, + 0.773972602739726, + 0.7734627831715211, + 0.7649122807017544, + 0.7647058823529411, + 0.7636986301369864, + 0.7516339869281046, + 0.7571428571428571, + 0.7658227848101266, + 0.7491289198606271, + 0.7697594501718213, + 0.7719298245614035, + 0.7448979591836735, + 0.7432432432432432, + 0.7291666666666666, + 0.7728813559322034, + 0.71280276816609, + 0.7615894039735099, + 0.7531645569620253, + 0.7166666666666667, + 0.7672131147540984, + 0.7615658362989324, + 0.7763578274760383, + 0.7440273037542662, + 0.7733333333333333, + 0.7116104868913857, + 0.7859531772575251, + 0.7574750830564784, + 0.75, + 0.7178571428571429, + 0.7583892617449665, + 0.7662337662337663, + 0.735593220338983, + 0.7482993197278912, + 0.7567567567567568, + 0.7275862068965517, + 0.7432432432432432, + 0.7328519855595668, + 0.7679180887372014, + 0.7214285714285714, + 0.7559322033898305, + 0.7389705882352942, + 0.7598684210526315, + 0.7165991902834008, + 0.9919354838709677, + 0.7750865051903114, + 0.7456445993031359, + 0.75, + 0.7574626865671642, + 0.7318840579710145, + 0.7724137931034483, + 0.7456445993031359, + 0.7105263157894737, + 0.7578947368421053, + 0.7391304347826086, + 0.9758064516129032, + 0.7389830508474576, + 0.7126865671641791, + 0.7598684210526315, + 0.7664233576642335, + 0.7014925373134329, + 0.7316176470588235, + 0.7344827586206897, + 0.7368421052631579, + 0.7213740458015268, + 0.7125984251968503, + 0.71875, + 0.6691176470588235, + 0.7455830388692579, + 0.7111111111111111, + 0.7198581560283688, + 0.7106227106227107, + 0.762987012987013, + 0.7551724137931034, + 0.7491408934707904, + 0.7272727272727273, + 0.7906137184115524, + 0.7279151943462897, + 0.7226277372262774, + 0.7083333333333334, + 0.7395833333333334, + 0.7648902821316614, + 0.7142857142857143, + 0.7283018867924528, + 0.7517985611510791, + 0.7474048442906575, + 0.7748344370860927, + 0.690566037735849, + 0.7333333333333333, + 0.7326007326007326, + 0.7133105802047781, + 0.7443609022556391, + 0.7636986301369864, + 0.75, + 0.7587548638132295, + 0.7355072463768116, + 0.7317073170731707, + 0.6911764705882353, + 0.7272727272727273, + 0.7386759581881533, + 0.7425373134328358, + 0.7121771217712177, + 0.7025089605734767, + 0.7676056338028169, + 0.7859778597785978, + 0.7363013698630136, + 0.7373737373737373, + 0.7204301075268817, + 0.7967213114754098, + 0.7256944444444444, + 0.7830882352941176, + 0.7214285714285714, + 0.6830985915492958, + 0.7003484320557491, + 0.6784452296819788, + 0.6727941176470589, + 0.7291666666666666, + 0.738562091503268, + 0.75, + 0.9689922480620154, + 0.7598684210526315, + 0.7186440677966102, + 0.6976744186046512, + 0.7364864864864865, + 0.7556270096463023, + 0.734982332155477, + 0.6973684210526315, + 1.0, + 0.9621212121212122, + 0.7722772277227723, + 0.7532467532467533, + 0.9846153846153847, + 0.7137931034482758, + 0.6153846153846154, + 0.6970802919708029, + 0.7064846416382252, + 0.7284768211920529, + 0.7310344827586207, + 0.7114093959731543, + 0.975609756097561, + 0.7515527950310559, + 0.738831615120275, + 0.9924812030075187, + 0.6484375, + 0.9929577464788732, + 0.6763636363636364, + 0.7368421052631579, + 0.7183098591549296, + 0.7306397306397306, + 0.9782608695652174, + 0.7028985507246377, + 0.704225352112676, + 0.7263157894736842, + 0.968503937007874, + 0.6734006734006734, + 0.9846153846153847, + 0.7272727272727273, + 1.0, + 0.73, + 0.7248322147651006, + 0.9854014598540146, + 0.7137546468401487, + 0.7442622950819672, + 0.9612403100775194, + 0.9765625, + 0.7375886524822695, + 0.7549019607843137, + 0.7254901960784313, + 0.7003367003367004, + 0.984, + 0.7208480565371025, + 0.9612403100775194, + 0.7305194805194806, + 0.697594501718213, + 0.9791666666666666, + 0.6884615384615385, + 0.7041800643086816, + 0.7090909090909091, + 0.9782608695652174, + 0.7, + 0.7391304347826086, + 0.9752066115702479, + 0.7442622950819672, + 0.6982456140350877, + 0.7322033898305085, + 0.9626865671641791, + 0.9922480620155039, + 0.9689922480620154, + 0.9833333333333333, + 0.7275747508305648, + 0.9787234042553191, + 0.9763779527559056, + 0.7414965986394558, + 0.6985815602836879, + 0.7147435897435898, + 0.9916666666666667, + 0.6996587030716723, + 0.7191780821917808, + 1.0, + 0.9923076923076923, + 0.7560975609756098, + 0.6956521739130435, + 0.725752508361204, + 0.722972972972973, + 0.7303754266211604, + 0.6972789115646258, + 0.7133105802047781, + 0.7058823529411765, + 0.6847457627118644, + 0.7153024911032029, + 0.767515923566879, + 0.7232472324723247, + 0.9925373134328358, + 0.7333333333333333, + 0.706081081081081, + 0.7, + 0.7306397306397306, + 0.7414965986394558, + 0.7291666666666666, + 0.7157894736842105, + 0.6931034482758621, + 0.7311475409836066, + 0.706081081081081, + 0.7065217391304348, + 0.7098976109215017, + 0.991869918699187, + 0.7491961414790996, + 0.7331081081081081, + 0.7532051282051282, + 1.0, + 0.7220447284345048, + 0.6774193548387096, + 0.9752066115702479, + 0.7285223367697594, + 0.9850746268656716, + 0.7467948717948718, + 0.71875, + 0.735973597359736, + 0.6811594202898551, + 0.7244897959183674, + 0.708904109589041, + 0.7202797202797203, + 0.9858156028368794, + 1.0, + 0.7156862745098039, + 0.6870748299319728, + 0.9776119402985075, + 0.7111913357400722, + 0.9923076923076923, + 0.7242524916943521, + 0.7214765100671141, + 0.6654545454545454, + 0.7275747508305648, + 0.9714285714285714, + 0.9921875, + 0.968, + 0.9761904761904762, + 0.6986301369863014, + 0.9781021897810219, + 0.7162629757785467, + 0.7714285714285715, + 0.9826086956521739, + 0.9858156028368794, + 0.7323943661971831, + 0.7142857142857143, + 0.7185185185185186, + 0.992, + 0.7363013698630136, + 0.7171717171717171, + 0.6843971631205674, + 0.9838709677419355, + 0.7802547770700637, + 1.0, + 0.9767441860465116, + 0.726962457337884, + 0.9754098360655737, + 0.7216828478964401, + 0.9763779527559056, + 0.9844961240310077, + 0.6833976833976834, + 0.7272727272727273, + 0.738831615120275, + 0.711340206185567, + 0.7007042253521126, + 1.0, + 0.7185430463576159, + 0.7218045112781954, + 0.7298245614035088, + 0.75, + 0.75, + 0.8164556962025317, + 0.6868327402135231, + 0.7142857142857143, + 0.6415094339622641, + 0.7337883959044369, + 0.7230215827338129, + 0.740484429065744, + 0.6895306859205776, + 0.7294520547945206, + 0.726027397260274, + 0.724907063197026, + 0.6996336996336996, + 0.6901408450704225, + 0.6851851851851852, + 0.7285714285714285, + 0.7235494880546075, + 0.7132867132867133, + 0.7279151943462897, + 0.7792642140468228, + 0.7481203007518797, + 0.6964285714285714, + 0.7490909090909091, + 0.7384105960264901, + 0.7700348432055749, + 0.7347670250896058, + 0.6679841897233202, + 0.7649122807017544, + 0.7536764705882353, + 0.7137096774193549, + 0.7553191489361702, + 0.7289377289377289, + 0.7730496453900709, + 0.7051792828685259, + 0.6879432624113475, + 0.7062937062937062, + 0.7230215827338129, + 0.6654929577464789, + 0.781021897810219, + 0.7491166077738516, + 0.6980392156862745, + 0.7353951890034365, + 0.7446043165467626, + 0.7218045112781954, + 0.732824427480916, + 0.720754716981132, + 0.7115384615384616, + 0.7269076305220884, + 0.7434944237918215, + 0.7045454545454546, + 0.7651245551601423, + 0.7105263157894737, + 0.7211895910780669, + 0.7805755395683454, + 0.7491039426523297, + 0.701067615658363, + 0.6884057971014492, + 0.7456445993031359, + 0.7528089887640449, + 0.7747440273037542, + 0.73992673992674, + 0.7364620938628159, + 0.7352941176470589, + 0.7035714285714286, + 0.6563706563706564, + 0.7224334600760456, + 0.6989247311827957, + 0.7372262773722628, + 0.6884057971014492, + 0.7762711864406779, + 0.7025089605734767, + 0.7240143369175627, + 0.7282608695652174, + 0.7103448275862069, + 0.7208480565371025, + 0.7302158273381295, + 0.7283018867924528, + 0.7474048442906575, + 0.7191780821917808, + 0.6958174904942965, + 0.7074074074074074, + 0.7304964539007093, + 0.7352941176470589, + 0.7442748091603053, + 0.7604166666666666, + 0.6861313868613139, + 0.7216117216117216, + 0.7526132404181185, + 0.7117437722419929, + 0.7206896551724138, + 0.7323420074349443, + 0.6771929824561403, + 0.7025089605734767, + 0.7024221453287197, + 0.6914498141263941, + 0.7226027397260274, + 0.7132616487455197, + 0.685823754789272, + 0.6970802919708029, + 0.7463768115942029, + 0.7535714285714286, + 0.7355072463768116, + 0.7395833333333334, + 0.7245283018867924, + 0.7444444444444445, + 0.712280701754386, + 0.7153024911032029, + 0.704225352112676, + 0.7340823970037453, + 0.7417218543046358, + 0.7204301075268817, + 0.7311827956989247, + 0.7076923076923077, + 0.6996466431095406, + 0.6846153846153846, + 0.691699604743083, + 0.6863468634686347, + 0.7263157894736842, + 0.7096774193548387, + 0.706959706959707, + 0.7491166077738516, + 0.6802973977695167, + 1.0, + 0.75177304964539, + 0.7357142857142858, + 0.7588652482269503, + 0.7154150197628458, + 0.7436823104693141, + 0.7552447552447552, + 0.7314487632508834, + 0.7030075187969925, + 0.759581881533101, + 0.6870229007633588, + 0.9813084112149533, + 0.7560975609756098, + 0.6801470588235294, + 0.8041958041958042, + 0.8222996515679443, + 0.6988847583643123, + 0.7056737588652482, + 0.6993006993006993, + 0.6808510638297872, + 0.6654275092936803, + 0.7052238805970149, + 0.6816479400749064, + 0.6787003610108303, + 0.706959706959707, + 0.6461538461538462, + 0.6605166051660517, + 0.7240143369175627, + 0.7416974169741697, + 0.7433962264150943, + 0.7655172413793103, + 0.7472527472527473, + 0.7593984962406015, + 0.7162162162162162, + 0.7407407407407407, + 0.6920289855072463, + 0.684981684981685, + 0.7734627831715211, + 0.6993006993006993, + 0.7078651685393258, + 0.7047244094488189, + 0.7054794520547946, + 0.7634408602150538, + 0.6489795918367347, + 0.6840148698884758, + 0.7391304347826086, + 0.6777777777777778, + 0.7170542635658915, + 0.7385159010600707, + 0.7269372693726938, + 0.7195121951219512, + 0.723404255319149, + 0.7321428571428571, + 0.7272727272727273, + 0.7226277372262774, + 0.7212543554006968, + 0.7323420074349443, + 0.6727941176470589, + 0.6923076923076923, + 0.7640845070422535, + 0.7407407407407407, + 0.6951672862453532, + 0.7194244604316546, + 0.6690140845070423, + 0.7607142857142857, + 0.7457044673539519, + 0.7269372693726938, + 0.7420494699646644, + 0.6678966789667896, + 0.6738351254480287, + 0.6762589928057554, + 0.6584507042253521, + 0.676923076923077, + 0.7903225806451613, + 0.7542662116040956, + 0.9914529914529915, + 0.75177304964539, + 0.7575757575757576, + 0.6716417910447762, + 0.7142857142857143, + 0.7773972602739726, + 0.7414965986394558, + 0.7031802120141343, + 0.9920634920634921, + 0.9666666666666667, + 0.8119122257053292, + 0.7817589576547231, + 0.944, + 0.7157894736842105, + 0.6678200692041523, + 0.676923076923077, + 0.7303754266211604, + 0.7177700348432056, + 0.7117437722419929, + 0.7314487632508834, + 0.953125, + 0.7419354838709677, + 0.6925925925925925, + 0.9923076923076923, + 0.6380597014925373, + 1.0, + 0.6779661016949152, + 0.7473309608540926, + 0.7567567567567568, + 0.6912280701754386, + 0.9481481481481482, + 0.6886446886446886, + 0.7087719298245614, + 0.6875, + 0.9922480620155039, + 0.701067615658363, + 0.9908256880733946, + 0.6855123674911661, + 0.9523809523809523, + 0.7153558052434457, + 0.7090909090909091, + 1.0, + 0.6909090909090909, + 0.7419354838709677, + 0.9851851851851852, + 0.968, + 0.7586206896551724, + 0.7420494699646644, + 0.7631578947368421, + 0.734006734006734, + 0.9830508474576272, + 0.6770428015564203, + 1.0, + 0.7056737588652482, + 0.7201492537313433, + 0.9761904761904762, + 0.6383763837638377, + 0.7062937062937062, + 0.6654275092936803, + 0.9844961240310077, + 0.7101449275362319, + 0.7429577464788732, + 0.9831932773109243, + 0.7235494880546075, + 0.7190635451505016, + 0.7419354838709677, + 0.9779411764705882, + 0.9838709677419355, + 0.9920634920634921, + 0.9819819819819819, + 0.7508650519031141, + 0.9908256880733946, + 0.981651376146789, + 0.7304964539007093, + 0.7517006802721088, + 0.7533783783783784, + 1.0, + 0.7377622377622378, + 0.7317880794701986, + 0.9915966386554622, + 0.952, + 0.7335640138408305, + 0.7594501718213058, + 0.7163636363636363, + 0.7557755775577558, + 0.7403508771929824, + 0.718978102189781, + 0.6691176470588235, + 0.7517006802721088, + 0.6867924528301886, + 0.7094339622641509, + 0.7661016949152543, + 0.7056737588652482, + 1.0, + 0.7708333333333334, + 0.7168458781362007, + 0.7517006802721088, + 0.7333333333333333, + 0.6958041958041958, + 0.7097902097902098, + 0.7123287671232876, + 0.7289377289377289, + 0.7348993288590604, + 0.7243816254416962, + 0.7185185185185186, + 0.6945454545454546, + 0.9606299212598425, + 0.7392857142857143, + 0.7357142857142858, + 0.7833333333333333, + 0.983739837398374, + 0.7370242214532872, + 0.7190635451505016, + 0.9672131147540983, + 0.7245901639344262, + 0.9918032786885246, + 0.7574750830564784, + 0.7366548042704626, + 0.7601351351351351, + 0.6862745098039216, + 0.7431506849315068, + 0.6865671641791045, + 0.6615969581749049, + 0.9754098360655737, + 0.983739837398374, + 0.7288732394366197, + 0.6774193548387096, + 1.0, + 0.7046263345195729, + 1.0, + 0.7201365187713311, + 0.698961937716263, + 0.6888111888111889, + 0.7205882352941176, + 0.9708029197080292, + 0.9921259842519685, + 1.0, + 0.9906542056074766, + 0.7147887323943662, + 0.991304347826087, + 0.6923076923076923, + 0.7912457912457912, + 0.959349593495935, + 0.9923076923076923, + 0.7311827956989247, + 0.7298245614035088, + 0.7071428571428572, + 0.9915966386554622, + 0.7347670250896058, + 0.7317073170731707, + 0.7235494880546075, + 0.9906542056074766, + 0.7727272727272727, + 0.983739837398374, + 0.9917355371900827, + 0.7454545454545455, + 0.9833333333333333, + 0.7056603773584905, + 0.9724770642201835, + 0.992, + 0.6826568265682657, + 0.7508650519031141, + 0.7664473684210527, + 0.7412587412587412, + 0.735593220338983, + 0.9747899159663865, + 0.7220338983050848 + ] + }, + "contact_maps": { + "mean_pearson": 0.33420325689713787, + "valid_tracks": 28, + "per_track_pearson": [ + 0.3226864374003948, + 0.35829024970845097, + 0.3750131538668185, + 0.30964071997943043, + 0.2431726365078143, + 0.2830069084720374, + 0.3410250520650591, + 0.2599904399621589, + 0.4079841225725386, + 0.37577862724317557, + 0.30752561239287496, + 0.36283536737665123, + 0.3644439860423661, + 0.2832297683020212, + 0.3196577286012534, + 0.44858594727691525, + 0.09702272962400005, + 0.36065179295343064, + 0.35628476537129994, + 0.35871767998056825, + 0.30602158611867003, + 0.30973346914671157, + 0.3116504804832239, + 0.37298552536677543, + 0.3777007654107906, + 0.3876221467359219, + 0.3544349094363562, + 0.40199858472215155 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.49334943479928534, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..576cf097a4a42e0d44f5a62b518ba944d342f95f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S029/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4258169009370233, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5559431880860453, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.10341363733998217, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7243319344187241, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.5740876938615527, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.5819396607928249, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.26566523021852473, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.4296821336600264, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.1528028572549477, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22418102178488727, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.4634272042775439, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.18307647819150158, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.22499318783259625, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/README.md new file mode 100644 index 0000000000000000000000000000000000000000..96c816010891750d4d1532eeebd588192215bfb0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/README.md @@ -0,0 +1,99 @@ +# SeqFunc small8 with a fixed six-checkpoint parameter average with transcript-direction junction decoding with calibrated splice probabilities + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. One fixed equal parameter average ofsixsavedstatesfromourunchangedsmallscale13trajectoryat324000,325000,326000,327000,328000,and328160trainingexamples. Thefinalstatecompletes8full41020intervalFOLD_0permutations/82040batch4updates,seed17,variance-balancedobjective,original14epochcosineLRschedule. Theexisting1000-examplecheckpointcadenceisnotchanged. Firstfivecapturesuseoneopenprogressinode;thelastusesimmutableepoch-08.pt. Missedsnapshotsfailclosed;noalternativestatesorweightsareselected. AccumulatefloatingtensorsinFP64,castonce;nonfloatingbuffersmustmatch. 13290504learnedparameters,0additionalSGD,oneaveragedmodelsameinferencecostservesbothentries. Rawspliceoutputsandoriginalsparsedecodersretained. Optimizer/RNGomitted:averagedcheckpointisinference-only,notatrainingresumestate. ThisfollowupismotivatedbySWA30complexTraitGymdevelopmentgainbutdiffersinrecipe,batchsize,andaveragingdensity;transferisunknown. Snapshotstepsanduniformweightsarefixedbeforecapture,notfittedtodevelopmentlabels. Comparewithunchangedrawsmall8,alreadycoveredbyoriginalcheckpointmonitor;noextra controltraining. Native52/full200windows2697queriesall13/native624andallthree4hruntimeestimatesprecedepromotion. Noexternalpretrainedparameters,testlabels,orlabel-derivedinferencecache. + +This checkpoint records **328,160 training examples in 82,040 optimizer updates**, 8 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +1231.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. The complete reports below determine this checkpoint's track and variant tradeoffs. + +This fixed-link follow-up uses the frozen SWA68 average of six parameter states ending at the complete eighth training pass. It applies the original four training-only offsets without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. Both entry points use the same link. All200 validation windows and all13 development metrics are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. A publication additionally requires native replay and three sampled runtime estimates below four hours. + +## Measured validation results + +Family-mean Pearson: **0.496924** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.590946 | +| dnase | 0.667164 | +| procap | 0.545689 | +| cage | 0.332173 | +| rna_seq | 0.625238 | +| chip_tf | 0.470676 | +| chip_histone | 0.501773 | +| splice_sites | 0.399322 | +| splice_usage | 0.598374 | +| junctions | 0.392829 | +| contact_maps | 0.341983 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.4495 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5458 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.0453 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7366 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.5682 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.5625 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2524 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4341 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1458 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2232 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4798 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2275 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2256 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..01240d8fb5c7349ba5e6e9a381900037dba944d5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/swa68/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..69cb497f74729d562ce1352606f4fcfd3c0ba56d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/package_manifest.json @@ -0,0 +1,72 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af", + "files": { + "README.md": "2025c334082224ff28e416cd1234c424cb94f13522bda6d5345ed9cf6ca303a5", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/evaluate_swa68_checkpoint.py": "db858ffdeabfb5e7cfffb8a9132152a094814a2bc837c528fa1cdaf85fba4c05", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_history/run_swa68_study.py": "fcb73022e4f4eeb1882dbf6872e8ed1906c783e6114931fdc071b3628d293464", + "training_history/snapshot_average.py": "ae2c8bd6c082e9e3bed0c51623a5ff9a65c24364400a1b77fb8a4f44ba5d0f04", + "training_history/swa68-averaging-report.json": "e8585369a2da7afa4639a3acf1e11aa461bdc9c96cbe12829425685a42b15bf5", + "training_history/swa68-captures.json": "09f9a7db7610dc3e064b3423c2ddf73e30304f3d60597e5ddaf12315875589f2", + "training_history/swa68-plan.json": "c882cec098a1cb458d11af94c522394b085d6bd965e271c72f30a761b9a1ce49", + "training_history/swa68-reused-averaging-mechanical-check.json": "2dd7fdebee69ccbf99e6ae8d5dff7612e8a31ab0678950555125a83a306914f3", + "training_history/test_snapshot_average.py": "902061aabe9817ed2348dafea2bdb480ab1cf8cc6e36aa0506825cc109321f3d", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "a748946a87a48dde7f00db96286165d8633732aa3bb31619ccb60d11262aae56", + "training_assets_manifest.json": "9c476e8e32171654beeac521d30aa07400cb6c94fa2439f37f10fb2b78149da0", + "decoding_policy.json": "11e570ff39ee4699cec72b775c4ed61094e4087078c609674af7161652a9d02a", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/fixed-swa105-master-plan.json": "476385cb3bbeb89ea43a7340e33196672e64266c1ec8c43bb7294a17e66cb638", + "training_history/fixed-swa105-resolved-parents.json": "d7b237225d1fae09495dec363fb17be275705fcfc8054171e2c55877bac2302c", + "training_history/fixed-swa105-parent-check.json": "2934a03c5e0b9877a149012a92a56041cfe19c650cae910bf5d3c0050fd08562", + "training_history/fixed_swa105.py": "38d7aa4cbe9a8140cd6386d5d30a71f5ac3502e85088e3f934af8db81fc4e2e9", + "training_history/run_fixed_swa105.py": "17c0f28061c51a3bf73fc7467f6d6928d9e061e843e174a368e3346d88fe9fc2", + "validation_tracks.json": "eb33597560cf0437cfdca24bcb93c4ee208a74a1507c911f23fe518daaccb749", + "validation_variants.json": "30f3b46b448cf83eb08b09fc6acd580e1bc92161da34ee945976476af9e90282" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..163a3d2327596bb12f95712396ea75230aa799f5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.py @@ -0,0 +1,102 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + trunk = model._trunk(dna) + group = root.create_group(f"perturbations/p{index:06d}") + features = None + for family in row.families: + if family == "splice_sites": + logits = model._splice_site_logits(trunk)[:, target_start:target_start + config.target_length] + values = splice_output_scores(logits[:, row.splice_start:row.splice_end]) + else: + features = model._dense_features(trunk) if features is None else features + values = model.dense_heads[family](features)[:, row.dense_bin_start:row.dense_bin_end] + # float32: one variant's effect is often below float16 resolution (about 0.004 at + # asinh values of 4-8), so A and B would round to the same numbers. + values = values.float().cpu().numpy() + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..c4be95252c0f0b42bed5a0f6067e2503133c2fa2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/predict.py @@ -0,0 +1,134 @@ +#!/usr/bin/env python3 +"""Stream the single benchmark model into the sharded prediction contract.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import torch +import zarr + +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES, PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from splice_output import splice_output_scores + + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample["example_id"] + dna = sample["dna"][None].to(device) + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + group = root.create_group(f"examples/{example_id}") + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy().astype(np.float16), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy().astype(np.float16), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy().astype(np.float16), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy().astype(np.float16), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + ids.append(example_id) + print(json.dumps({"event": "predict", "example": len(ids), "total": len(dataset)}), flush=True) + root.attrs.update({"schema": PREDICTION_SCHEMA, "example_ids": ids}) + (output_root / "example_ids.json").write_text(json.dumps(ids) + "\n") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--split", default="test", choices=("valid", "test"), + help="manifest to predict; the verifier uses test, local validation uses valid") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/cached_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/cached_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3677c1e8518ea97db0dd6759f3d636f5fff27c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/cached_read_store.py @@ -0,0 +1,71 @@ +"""Optional local byte cache for immutable training Zarr stores. + +Source storage stays read-only. Cached entries contain a key fingerprint, length, +and SHA256 of the original compressed bytes; every hit verifies all three. Writes +replace a private temporary inode atomically. A missing/corrupt/unwritable cache +falls back to the original store, so this layer changes neither values nor keys. +Only training loaders may opt into it; it is not an inference asset. +""" +import hashlib +import os +from pathlib import Path +import struct +import tempfile + +from .direct_read_store import ParallelDirectReadDirectoryStore + +MAGIC = b'SFBC0001' +HEADER = struct.Struct('<8sQ32s32s') + + +class CachedReadDirectoryStore(ParallelDirectReadDirectoryStore): + def __init__(self, path, *, cache_root, namespace, io_workers=4): + super().__init__(path, io_workers=io_workers) + if not namespace: + raise ValueError('An immutable input namespace is required') + self.cache_root = Path(cache_root) + self.namespace = namespace + self.store_key = hashlib.sha256((namespace+'\0'+os.path.abspath(path)).encode()).digest() + self.statistics = {'hit': 0, 'miss': 0, 'corrupt': 0, 'write_failure': 0} + + def cache_location(self, key): + key = self._normalize_key(key) + identity = hashlib.sha256(self.store_key+key.encode()).digest() + name = identity.hex() + return self.cache_root/name[:2]/name[2:], identity + + def __getitem__(self, key): + key = self._normalize_key(key) + destination, identity = self.cache_location(key) + try: + blob = destination.read_bytes() + if len(blob) >= HEADER.size: + magic, size, cached_key, checksum = HEADER.unpack_from(blob) + payload = blob[HEADER.size:] + if (magic == MAGIC and size == len(payload) and cached_key == identity + and hashlib.sha256(payload).digest() == checksum): + self.statistics['hit'] += 1 + return payload + self.statistics['corrupt'] += 1 + except OSError: + pass + self.statistics['miss'] += 1 + payload = super().__getitem__(key) + temporary = None + try: + destination.parent.mkdir(parents=True, exist_ok=True) + with tempfile.NamedTemporaryFile(prefix='.pending-', dir=destination.parent, delete=False) as stream: + temporary = stream.name + stream.write(HEADER.pack(MAGIC, len(payload), identity, hashlib.sha256(payload).digest())) + stream.write(payload) + os.replace(temporary, destination) + temporary = None + except OSError: + self.statistics['write_failure'] += 1 + finally: + if temporary is not None: + try: + os.unlink(temporary) + except OSError: + pass + return payload diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..275046a7b3696c54f023056a4392913bd14b095c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_assets_manifest.json @@ -0,0 +1,29 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/evaluate_swa68_checkpoint.py": "db858ffdeabfb5e7cfffb8a9132152a094814a2bc837c528fa1cdaf85fba4c05", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_history/run_swa68_study.py": "fcb73022e4f4eeb1882dbf6872e8ed1906c783e6114931fdc071b3628d293464", + "training_history/snapshot_average.py": "ae2c8bd6c082e9e3bed0c51623a5ff9a65c24364400a1b77fb8a4f44ba5d0f04", + "training_history/swa68-averaging-report.json": "e8585369a2da7afa4639a3acf1e11aa461bdc9c96cbe12829425685a42b15bf5", + "training_history/swa68-captures.json": "09f9a7db7610dc3e064b3423c2ddf73e30304f3d60597e5ddaf12315875589f2", + "training_history/swa68-plan.json": "c882cec098a1cb458d11af94c522394b085d6bd965e271c72f30a761b9a1ce49", + "training_history/swa68-reused-averaging-mechanical-check.json": "2dd7fdebee69ccbf99e6ae8d5dff7612e8a31ab0678950555125a83a306914f3", + "training_history/test_snapshot_average.py": "902061aabe9817ed2348dafea2bdb480ab1cf8cc6e36aa0506825cc109321f3d" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-integration_check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-integration_check.json new file mode 100644 index 0000000000000000000000000000000000000000..1377348921feb42310e66b133df8ebee9f0d591c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-integration_check.json @@ -0,0 +1,166 @@ +{ + "created_utc": "2026-10-25T03:48:07.725319+00:00", + "preparation_sha256": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "source_sha256": "b0921bfd3d3dc47f3cb82debc0b492d8291693c40e0cef028764c713c1257e85", + "original_full256_report_sha256": "e6620ff13b00a36857dad22fa0c9d1ad5c2b266eca70773650d280998ba20005", + "examples": 16, + "worker_processes": 4, + "ordered_indices": [ + 38509, + 39126, + 38926, + 17432, + 6551, + 22599, + 24749, + 10790, + 7022, + 30789, + 17340, + 40258, + 26222, + 22674, + 23658, + 17167 + ], + "all_returned_training_ids_and_tensor_bytes_exact": true, + "reference_and_cache_both_match_original_report": true, + "public_inference_and_validation_cache_disabled_despite_environment": true, + "default_cache_disabled": true, + "models_and_trainers_unchanged": true, + "rows": { + "reference": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ], + "cached": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ] + }, + "no_active_source_or_training_changes": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/evaluate_swa68_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/evaluate_swa68_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..7733af836e3cabcf46aa244a61d5e8fa65295a8d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/evaluate_swa68_checkpoint.py @@ -0,0 +1,156 @@ +"""Stage and replay-check one immutable swa68 checkpoint or its matched control.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import math +import os +from pathlib import Path +import subprocess +import sys +import time + +import torch +from verify_package import inference_signature, verify_package +from vendor_training_assets import verify_training_assets + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan', type=Path) + parser.add_argument('role', choices=['candidate']) + args = parser.parse_args() + plan = json.loads(args.plan.read_text()) + here, study = Path(__file__).resolve().parent, args.plan.parent + spec = plan['evaluation'][args.role] + checkpoint, source, output = map(Path, [spec['checkpoint'], spec['source'], spec['output']]) + output.mkdir(parents=True, exist_ok=True) + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(2) + for name, expected in plan['helper_sha256'].items(): + assert digest(here/name) == expected, name + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') + environment.pop('SEQFUNC_TRAIN_BYTECACHE', None) + + def status(phase, **extra): + row = {'phase': phase, 'role': args.role, 'utc': datetime.now(timezone.utc).isoformat(), **extra} + temporary = output/'pipeline-state.tmp' + temporary.write_text(json.dumps(row, indent=2)+'\n') + temporary.replace(output/'pipeline-state.json') + print(json.dumps(row), flush=True) + + def run(phase, *command): + status(phase) + with (output/(phase+'.log')).open('a') as stream: + subprocess.run([sys.executable, *map(str, command)], env=environment, + stdout=stream, stderr=subprocess.STDOUT, check=True) + + # Prevent duplicated evaluators after manual/operator recovery. + with (output/'controller.lock').open('a') as controller_lock: + fcntl.flock(controller_lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + status('waiting_for_immutable_checkpoint') + while not checkpoint.exists(): + time.sleep(5) + state = torch.load(checkpoint, map_location='cpu', weights_only=True) + assert state['global_step'] == 328160 and state['optimizer_steps'] == 82040 + assert state['epoch'] == 8 and state['offset'] == 0 + assert state['config']['local_context_blocks'] == 6 + assert state.get('weights_only_average') is True and 'optimizer' not in state + del state + final_digest = digest(checkpoint) + + def staged(path): + if not path.exists(): + return False + try: + verify_package(path) + verify_training_assets(source, path) + assert json.loads((path/'package_manifest.json').read_text())['source_checkpoint_sha256'] == final_digest + return True + except (OSError, ValueError, KeyError, AssertionError): + path.rename(path.with_name(path.name+'-incomplete-'+str(time.time_ns()))) + return False + + raw, package = output/'raw-package', output/'package' + if not staged(raw): + run('readme', here/'make_run_readme.py', checkpoint, output/'README.package.md', + '--name', spec['name'], '--lineage', spec['lineage']) + run('package', source/'make_package.py', checkpoint, raw, '--readme', output/'README.package.md') + run('training-assets', here/'vendor_training_assets.py', source, raw) + if not staged(package): + run('orientation', here/'make_orientation_candidate.py', raw, package) + signature = inference_signature(package) + native = study/('native-'+args.role) + with Path(plan['evaluation_lock']).open('a') as lock: + status('waiting_for_native_and_evaluation_slot') + fcntl.flock(lock, fcntl.LOCK_EX) + if not (native/'report.json').exists(): + run('native', here/'check_native_recovery.py', package, native) + report = json.loads((native/'report.json').read_text()) + assert report['package_inference_signature'] == signature + assert report['all_arrays_bit_identical'] and report['native_vendored_entrypoints_passed'] + assert report['partial_perturbation_recovery_passed'] + assert report['arrays'] == {'tracks': 30, 'perturbations': 22} + if not (output/'evaluation.json').exists(): + run('evaluate', here/'evaluate_package.py', package, output, + '--resume-partial-tracks', '--resume-partial-perturbations') + assert json.loads((output/'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output/'valid-score/report.json').read_text()) + variants = json.loads((output/'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + control = Path(plan['control']) + status('waiting_for_complete_fixed_control') + while not (control/'evaluation.json').exists(): + time.sleep(5) + before = json.loads((control/'valid-score/report.json').read_text()) + before_variants = json.loads((control/'valid-variant-score/report.json').read_text()) + assert before['examples'] == 200 and before['invalid_reason'] is None + assert len(before['families']) == 11 and len(before_variants) == 13 + for name,value in variants.items(): + for key in ['rows','perturbations','metric','anchors']: + assert value[key] == before_variants[name][key],(name,key) + first = json.loads((package/'training_provenance.json').read_text()) + second = json.loads((control/'package/training_provenance.json').read_text()) + for key in ['config','global_step','optimizer_steps','epoch','offset','examples','objective', + 'seed','schedule_epochs','family_loss_scales','loss_scale_provenance']: + assert first[key] == second[key],key + comparison = dict(created_utc=datetime.now(timezone.utc).isoformat(), + fixed_control=str(control),same_training_exposure_recipe_and_config=True, + same_variant_eligibility_and_anchors=True, + reward_delta=tracks['overall_mean_family_pearson']-before['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-before['families'][k]['mean_pearson'] for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-before_variants[k]['model'] for k,v in variants.items()}) + (study/'matched-comparison.json').write_text(json.dumps(comparison,indent=2)+'\n') + latency = study/'native-latency' + latency.mkdir(exist_ok=True) + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency/'plan.json').exists(): + assert json.loads((latency/'plan.json').read_text()) == timing_plan + else: + (latency/'plan.json').write_text(json.dumps(timing_plan, indent=2)+'\n') + run('native-latency', here/'measure_native_latency.py', latency/'plan.json') + timing = json.loads((latency/'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + keys = ['development_mixture_seconds', 'slowest_observed_stratum_mean_seconds', 'pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in keys) + passed = max(estimates[k] for k in keys) <= 14400 + (study/'runtime-gate.json').write_text(json.dumps(dict(passed=passed, estimates=estimates, + report_sha256=digest(latency/'report.json'), + scope='All three predeclared shared-load sample estimates must fit4h; not guaranteed bounds.'), indent=2)+'\n') + if passed: + run('publish', here/'promote_if_better.py', package, output) + status('complete' if passed else 'complete_accuracy_runtime_gate_failed', + checkpoint_sha256=final_digest, runtime_passed=passed) + + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-master-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-master-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..44aed05ac2fa17c42a6ca24fd0ebc799c73888f9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-master-plan.json @@ -0,0 +1,521 @@ +{ + "created_utc": "2026-10-26T00:31:08.603070+00:00", + "members": { + "averaged-small8": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/swa68", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/calib105-swa68", + "study": "/workspace/research/studies/fixed_swa105" + } + }, + "capacities": { + "small": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 13290504 + } + }, + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "calibration_reference": "/workspace/research/experiments/calib57/package/splice_calibration.json", + "helper_sha256": { + "/run-state/workspace/research/seqfunc/check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "/run-state/workspace/research/seqfunc/check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "/run-state/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "/run-state/workspace/research/seqfunc/diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "/run-state/workspace/research/seqfunc/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "/run-state/workspace/research/seqfunc/evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "/run-state/workspace/research/seqfunc/finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "/run-state/workspace/research/seqfunc/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "/run-state/workspace/research/seqfunc/make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "/run-state/workspace/research/seqfunc/materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "/run-state/workspace/research/seqfunc/measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "/run-state/workspace/research/seqfunc/promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "/run-state/workspace/research/seqfunc/publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "/run-state/workspace/research/seqfunc/resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "/run-state/workspace/research/seqfunc/resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "/run-state/workspace/research/seqfunc/run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "/run-state/workspace/research/seqfunc/run_fixed_link_followup69.py": "3234a647219aae3df72afee9e5d9ed671d9db8b08adafff58bbffcbebcafb40d", + "/run-state/workspace/research/seqfunc/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "/run-state/workspace/research/seqfunc/verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "/run-state/workspace/research/seqfunc/fixed_swa105.py": "38d7aa4cbe9a8140cd6386d5d30a71f5ac3502e85088e3f934af8db81fc4e2e9", + "/run-state/workspace/research/seqfunc/run_fixed_swa105.py": "17c0f28061c51a3bf73fc7467f6d6928d9e061e843e174a368e3346d88fe9fc2", + "/run-state/workspace/research/seqfunc/prepare_fixed_swa105.py": "1a33480f654a3f306a9fc9684974b2f5286360337775ae5a12d2f2b26e026983" + }, + "evidence_sha256": { + "/workspace/research/studies/fixed_link69/plan.json": "45541635f549af8ce6c3db3fc7c857720970a5ee35e9de3c6a57f41dbe1f56bc", + "/workspace/research/studies/fixed_link61/plan.json": "100e4da9bf4faa715e901a0aa928dab65fc19f60970312e3e0eb548fb5b70860", + "/workspace/research/studies/fixed_link57/complete-summary.json": "5bba2a23162a77a523777f19de0be819d44cb44246c8c5e2f4bef913f5ee7a87", + "/workspace/research/studies/scale13/study-cached-io.json": "88aa6d95ef93776ed3551d68d98b1a31d726aa4b0270edbfc16b3f16ceeb937f", + "/workspace/research/studies/scale13/loss_scales.json": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42", + "/workspace/data/public/train.parquet": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "/workspace/research/experiments/calib57/package/splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5", + "/workspace/research/experiments/scale13-large-epoch04/package/training_provenance.json": "6589e14349cbd5419c814aa3d292c81f5328f1441bc75bc3ea7cae2b114dcdac", + "/workspace/research/studies/fixed_swa105/admission.json": "9b3b6249098088521c9780f766e97250030a2e0cdcbfade62a9091707b2ba1bc", + "/workspace/research/studies/late_fixed_links81/plan.json": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "/workspace/research/studies/fixed_swa105/preparation.json": "e4a0b6e8fbe737fd79f00b70404f133197f5d4d67bd1bd9af3d986b33f2318f2", + "/workspace/research/studies/fixed_swa105/run_fixed_swa105.py.diff": "80604535f270bfced5c4d16a491aef8b6ea46900d6f434b7d8b232723ff02289", + "/workspace/research/studies/swa68/plan.json": "c882cec098a1cb458d11af94c522394b085d6bd965e271c72f30a761b9a1ce49", + "/workspace/research/studies/swa68/averaging-report.json": "e8585369a2da7afa4639a3acf1e11aa461bdc9c96cbe12829425685a42b15bf5", + "/workspace/research/studies/swa68/captures.json": "09f9a7db7610dc3e064b3423c2ddf73e30304f3d60597e5ddaf12315875589f2", + "/workspace/research/studies/swa68/matched-comparison.json": "0dda9b57329d6643b5ffddf3e86563fc3b5820a71c44a2b86ee34f07809391f5", + "/workspace/research/studies/swa68/runtime-gate.json": "216435d26923fcf926ab28358c0cade7be54600a9f434235b0ddfddbe33b96f7", + "/workspace/research/studies/swa68/native-candidate/report.json": "2d9264b22785b7901770878b779a0b0983283f7efada60db70d3c0a9acad2eef", + "/workspace/research/studies/swa68/native-latency/report.json": "5e28008ad413279398d91d3267ff83adac0461436f519eb3a7397b19d387ecfb", + "/workspace/research/studies/swa68/diagnostics/plan.json": "24feb959f9b1dbd0bda056f664e33e3777700b6d0acccf048c03805c384b4a2f", + "/workspace/research/experiments/swa68/evaluation.json": "81e0600d2d438d3fc0db72a406563fdb523d0309335f390efead372c2b68b994", + "/workspace/research/experiments/swa68/evaluation-package.json": "b132c36bfcc26411c921f2bf4a2d68b5d4d3c97f51dd271466598f599d5a5c9e", + "/workspace/research/experiments/swa68/valid-score/report.json": "c7292832eae9eba365792fe3ded06a02e0667856783f952ad1c4e9f1d422e7ab", + "/workspace/research/experiments/swa68/valid-variant-score/report.json": "e31488471849ca68ad95663072134f9406da0ff3ddae6b6561d7aebd980fc66b", + "/workspace/research/experiments/swa68/package/training_provenance.json": "a748946a87a48dde7f00db96286165d8633732aa3bb31619ccb60d11262aae56", + "/workspace/research/experiments/scale13-small-epoch08/evaluation.json": "048a4f8e805c8476b4240977d2e6c6e800dd8a509ec1183f099e9726cc51cd11", + "/workspace/research/experiments/scale13-small-epoch08/evaluation-package.json": "84a37d41701a015643c86a120d0c8b1794af718fce1478824f8aaf0798411cf5", + "/workspace/research/experiments/scale13-small-epoch08/valid-score/report.json": "2baaccea669daefc1fde4adec4477eeb813e1e17fc5d64ed3dc1f8b54dff676f", + "/workspace/research/experiments/scale13-small-epoch08/valid-variant-score/report.json": "e204ff744b560b4df16ad6c114205480864faffac1a5e22c519f4a1960638d76", + "/workspace/research/experiments/scale13-small-epoch08/package/training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01", + "/workspace/research/studies/swa68/diagnostics/tracks-complete.json": "1bdbf01b898abe8a6a504cbb8e83caac86b264fffa0de488f8f922163d9131dd", + "/workspace/research/analysis/swa68-vs-small8-tracks.json": "4bac48c6fbcbd958e67e88bdec38722600578e4d5b3bc37db385c9de9672e7be", + "/workspace/research/studies/swa68/diagnostics/dense-variants-complete.json": "21ad69650d54524c99624aeec56855739b33d29d1b910ecb505f90fedda71927", + "/workspace/research/analysis/swa68-vs-small8-dense-variants.json": "321a34b65e0bb4894ac661733aa9de12ca333b92a5db4b8077cccf436e169c88", + "/workspace/research/studies/swa68/diagnostics/remaining-variants-complete.json": "96b80e839e6664f652bf0c418dee288e10f1bb5f52ea6fa46a461c5a3897aacf", + "/workspace/research/analysis/swa68-vs-small8-remaining-variants.json": "f804a2cd28475fe97c942f4393189397e9a09622bcfad7c1dfda7c5effacbe9f" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Research-only prepared-calib105-swa68 backing must remain while referenced. Never an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "native_latency_template": { + "created_utc": "2026-10-25T09:54:57.944795+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.", + "scope": "Own16-window/96-query native replay and all624 arrays for fixedSWA105; three shared-load4h estimates required before any publication.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "The rawlarge2 parent timing is contextual only. Require fresh candidate replay and allthree estimates <=14400seconds; no estimate is a guaranteed upper bound." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "bounded_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "native_timeout_seconds": 1800, + "materialization_timeout_seconds": 1800, + "parent_minimum_reward_gain": 1e-06, + "publication_minimum_reward_gain": 1e-06, + "hypothesis": "Apply the original four training-prior offsets to the sole frozen SWA68 average. No new averaging, fitting, neural updates, orientation, sparse selection, or interpolation change.", + "admission": "Separate scientific/resource review after complete SWA68 full/native/runtime and all11/all13 paired diagnostics. Require the exact average to improve rawsmall8 reward by >1e-6. No alternate parent or weighting search.", + "validation": "Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.", + "diagnostics": "Original paired tracks and remaining-seven metrics versus rawSWA68, with eight unchanged official reports checked exactly. A separately frozen all11/all13 paired plan compares with fixed81-small8, isolating averaging under the same output link. No fitted variant constants.", + "nonimprovement_policy": "A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.", + "limitations": [ + "The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.", + "The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.", + "Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.", + "No test labels, variant-label training, external parameters or inference caches are used." + ], + "average_checkpoint_sha256": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-parent-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-parent-check.json new file mode 100644 index 0000000000000000000000000000000000000000..d06776b1d72ce26c927eb0023d96e1be09855bac --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-parent-check.json @@ -0,0 +1,96 @@ +{ + "created_utc": "2026-10-26T00:31:08.603070+00:00", + "plan_sha256": "476385cb3bbeb89ea43a7340e33196672e64266c1ec8c43bb7294a17e66cb638", + "actual_completed_parent_checks": [ + { + "experiment": "/workspace/research/experiments/swa68", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.46591372099876976, + "sha256": { + "/workspace/research/experiments/swa68/evaluation.json": "81e0600d2d438d3fc0db72a406563fdb523d0309335f390efead372c2b68b994", + "/workspace/research/experiments/swa68/evaluation-package.json": "b132c36bfcc26411c921f2bf4a2d68b5d4d3c97f51dd271466598f599d5a5c9e", + "/workspace/research/experiments/swa68/valid-score/report.json": "c7292832eae9eba365792fe3ded06a02e0667856783f952ad1c4e9f1d422e7ab", + "/workspace/research/experiments/swa68/valid-variant-score/report.json": "e31488471849ca68ad95663072134f9406da0ff3ddae6b6561d7aebd980fc66b", + "/workspace/research/experiments/swa68/package/training_provenance.json": "a748946a87a48dde7f00db96286165d8633732aa3bb31619ccb60d11262aae56" + }, + "average_provenance": { + "kind": "fixed_checkpoint_parameter_average", + "snapshots": [ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c" + } + ], + "equal_weights": true, + "accumulation_dtype": "float64", + "cast_once_to_original_tensor_dtype": true, + "nonfloating_buffers_identical": true, + "optimizer_and_RNG_excluded_not_a_training_resume_checkpoint": true, + "additional_training_updates": 0, + "development_labels_fit_weights": false + }, + "paired_phase_output_sha256": { + "tracks": "4bac48c6fbcbd958e67e88bdec38722600578e4d5b3bc37db385c9de9672e7be", + "dense-variants": "321a34b65e0bb4894ac661733aa9de12ca333b92a5db4b8077cccf436e169c88", + "remaining-variants": "f804a2cd28475fe97c942f4393189397e9a09622bcfad7c1dfda7c5effacbe9f" + }, + "native_runtime_report_sha256": "5e28008ad413279398d91d3267ff83adac0461436f519eb3a7397b19d387ecfb", + "average_checkpoint_sha256": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af" + }, + { + "experiment": "/workspace/research/experiments/scale13-small-epoch08", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4622727224599102, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch08/evaluation.json": "048a4f8e805c8476b4240977d2e6c6e800dd8a509ec1183f099e9726cc51cd11", + "/workspace/research/experiments/scale13-small-epoch08/evaluation-package.json": "84a37d41701a015643c86a120d0c8b1794af718fce1478824f8aaf0798411cf5", + "/workspace/research/experiments/scale13-small-epoch08/valid-score/report.json": "2baaccea669daefc1fde4adec4477eeb813e1e17fc5d64ed3dc1f8b54dff676f", + "/workspace/research/experiments/scale13-small-epoch08/valid-variant-score/report.json": "e204ff744b560b4df16ad6c114205480864faffac1a5e22c519f4a1960638d76", + "/workspace/research/experiments/scale13-small-epoch08/package/training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01" + } + } + ], + "gpu_used": false, + "packages_changed": false, + "no_new_predictions_or_parameter_fitting": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-resolved-parents.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-resolved-parents.json new file mode 100644 index 0000000000000000000000000000000000000000..ed0ac1cc5fb636d8a66a8203d7758f4f6452cbd5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed-swa105-resolved-parents.json @@ -0,0 +1,101 @@ +{ + "master_plan_sha256": "476385cb3bbeb89ea43a7340e33196672e64266c1ec8c43bb7294a17e66cb638", + "member": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/swa68", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/calib105-swa68", + "study": "/workspace/research/studies/fixed_swa105" + }, + "receipts": { + "parent": { + "experiment": "/workspace/research/experiments/swa68", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.46591372099876976, + "sha256": { + "/workspace/research/experiments/swa68/evaluation.json": "81e0600d2d438d3fc0db72a406563fdb523d0309335f390efead372c2b68b994", + "/workspace/research/experiments/swa68/evaluation-package.json": "b132c36bfcc26411c921f2bf4a2d68b5d4d3c97f51dd271466598f599d5a5c9e", + "/workspace/research/experiments/swa68/valid-score/report.json": "c7292832eae9eba365792fe3ded06a02e0667856783f952ad1c4e9f1d422e7ab", + "/workspace/research/experiments/swa68/valid-variant-score/report.json": "e31488471849ca68ad95663072134f9406da0ff3ddae6b6561d7aebd980fc66b", + "/workspace/research/experiments/swa68/package/training_provenance.json": "a748946a87a48dde7f00db96286165d8633732aa3bb31619ccb60d11262aae56" + }, + "average_provenance": { + "kind": "fixed_checkpoint_parameter_average", + "snapshots": [ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c" + } + ], + "equal_weights": true, + "accumulation_dtype": "float64", + "cast_once_to_original_tensor_dtype": true, + "nonfloating_buffers_identical": true, + "optimizer_and_RNG_excluded_not_a_training_resume_checkpoint": true, + "additional_training_updates": 0, + "development_labels_fit_weights": false + }, + "paired_phase_output_sha256": { + "tracks": "4bac48c6fbcbd958e67e88bdec38722600578e4d5b3bc37db385c9de9672e7be", + "dense-variants": "321a34b65e0bb4894ac661733aa9de12ca333b92a5db4b8077cccf436e169c88", + "remaining-variants": "f804a2cd28475fe97c942f4393189397e9a09622bcfad7c1dfda7c5effacbe9f" + }, + "native_runtime_report_sha256": "5e28008ad413279398d91d3267ff83adac0461436f519eb3a7397b19d387ecfb", + "average_checkpoint_sha256": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af" + }, + "previous": { + "experiment": "/workspace/research/experiments/scale13-small-epoch08", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4622727224599102, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch08/evaluation.json": "048a4f8e805c8476b4240977d2e6c6e800dd8a509ec1183f099e9726cc51cd11", + "/workspace/research/experiments/scale13-small-epoch08/evaluation-package.json": "84a37d41701a015643c86a120d0c8b1794af718fce1478824f8aaf0798411cf5", + "/workspace/research/experiments/scale13-small-epoch08/valid-score/report.json": "2baaccea669daefc1fde4adec4477eeb813e1e17fc5d64ed3dc1f8b54dff676f", + "/workspace/research/experiments/scale13-small-epoch08/valid-variant-score/report.json": "e204ff744b560b4df16ad6c114205480864faffac1a5e22c519f4a1960638d76", + "/workspace/research/experiments/scale13-small-epoch08/package/training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01" + } + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed_swa105.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed_swa105.py new file mode 100644 index 0000000000000000000000000000000000000000..90dfc1cd0c2d83ab63ffb9d5590f33e81e78c0f8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/fixed_swa105.py @@ -0,0 +1,89 @@ +"""Bind the existing fixed-link pipeline to the sole six-state SWA68 parent.""" +import json +import math +from pathlib import Path + +from late_fixed_links81 import ( + ROOT, HERE, digest, write_json, + parent_ready as original_parent_ready, + parent_receipt as original_parent_receipt, +) + +STUDY = ROOT / 'studies/fixed_swa105' +PARENT = ROOT / 'experiments/swa68' +CONTROL = ROOT / 'experiments/scale13-small-epoch08' + + +def read(path): + return json.loads(Path(path).read_text()) + + +def parent_ready(path): + path = Path(path) + assert path.resolve() in {PARENT.resolve(), CONTROL.resolve()} + if not original_parent_ready(path): + return False + if path.resolve() == CONTROL.resolve(): + return True + prerequisites = [ + ROOT / 'studies/swa68/matched-comparison.json', + ROOT / 'studies/swa68/runtime-gate.json', + ROOT / 'studies/swa68/native-latency/report.json', + *[ROOT / 'studies/swa68/diagnostics' / (name+'-complete.json') + for name in ['tracks', 'dense-variants', 'remaining-variants']], + ] + return all(p.exists() for p in prerequisites) + + +def parent_receipt(path, capacity, epoch, plan): + path = Path(path) + assert parent_ready(path) + assert capacity == 'small' and epoch == 8 + receipt = original_parent_receipt(path, capacity, epoch, plan) + if path.resolve() == CONTROL.resolve(): + return receipt + assert path.resolve() == PARENT.resolve() + parent_study = ROOT / 'studies/swa68' + average = read(parent_study/'averaging-report.json') + assert average['output_checkpoint_sha256'] == plan['average_checkpoint_sha256'] + assert digest(path/'last.pt') == plan['average_checkpoint_sha256'] + assert read(path/'package/package_manifest.json')['source_checkpoint_sha256'] == plan['average_checkpoint_sha256'] + assert [row['step'] for row in average['snapshots']] == [324000,325000,326000,327000,328000,328160] + assert average['equal_weights'] and average['accumulation_dtype'] == 'float64' + assert average['cast_once_to_original_tensor_dtype'] and average['nonfloating_buffers_identical'] + assert average['optimizer_and_RNG_excluded_not_a_training_resume_checkpoint'] + assert average['additional_training_updates'] == 0 and not average['development_labels_fit_weights'] + intervention = read(path/'package/training_provenance.json')['intervention'] + assert intervention['kind'] == 'fixed_checkpoint_parameter_average' + assert all(average[key] == value for key, value in intervention.items()) + comparison = read(parent_study/'matched-comparison.json') + assert Path(comparison['fixed_control']).resolve() == CONTROL.resolve() + assert comparison['same_training_exposure_recipe_and_config'] + assert comparison['same_variant_eligibility_and_anchors'] + assert comparison['reward_delta'] > plan['parent_minimum_reward_gain'] + native = read(parent_study/'native-candidate/report.json') + assert native['package_inference_signature'] == receipt['inference_signature'] + assert native['all_arrays_bit_identical'] and native['native_vendored_entrypoints_passed'] + assert native['partial_perturbation_recovery_passed'] + assert native['arrays'] == {'tracks':30,'perturbations':22} + latency_path = parent_study/'native-latency/report.json' + latency = read(latency_path) + assert latency['package_inference_signature'] == receipt['inference_signature'] + assert latency['all_arrays_bit_exact'] + gate = read(parent_study/'runtime-gate.json') + assert gate['passed'] and gate['report_sha256'] == digest(latency_path) + estimates = latency['extrapolated_seconds'] + names = ['development_mixture_seconds', 'slowest_observed_stratum_mean_seconds', 'pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) and estimates[k] <= 14400 for k in names) + paired = parent_study/'diagnostics' + phases = read(paired/'plan.json')['phases'] + assert [phase['name'] for phase in phases] == ['tracks','dense-variants','remaining-variants'] + phase_hashes = {} + for phase in phases: + marker = read(paired/(phase['name']+'-complete.json')) + assert marker['output_sha256'] == digest(phase['output']) + phase_hashes[phase['name']] = marker['output_sha256'] + receipt.update(average_provenance=intervention, paired_phase_output_sha256=phase_hashes, + native_runtime_report_sha256=digest(latency_path), + average_checkpoint_sha256=plan['average_checkpoint_sha256']) + return receipt diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_fixed_swa105.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_fixed_swa105.py new file mode 100644 index 0000000000000000000000000000000000000000..42bd20a40c32e826364d63cd1feb6eceb0de983c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_fixed_swa105.py @@ -0,0 +1,268 @@ +"""Apply the existing fixed link to the single completed six-state SWA68 average.""" +import argparse +from datetime import datetime, timezone +import fcntl +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from check_splice_link_reuse import check_link_only +from fixed_swa105 import ROOT, STUDY, HERE, digest, write_json, parent_ready, parent_receipt +from verify_package import inference_signature, verify_package + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('member') + args = parser.parse_args() + plan_path = STUDY / 'plan.json' + plan = json.loads(plan_path.read_text()) + member = plan['members'][args.member] + study, parent, previous, output = [Path(member[k]) for k in ['study','parent','previous_raw','output']] + lock = (study / 'controller.lock').open('a') + fcntl.flock(lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + complete = study / 'complete.json' + if complete.exists(): + assert json.loads(complete.read_text())['master_plan_sha256'] == digest(plan_path) + print(json.dumps(dict(phase='already_complete_no_rerun', member=args.member)), flush=True) + return + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + assert digest(plan['training_prior']) == plan['training_prior_sha256'] + audit = json.loads((STUDY / 'parent-check.json').read_text()) + assert audit['plan_sha256'] == digest(plan_path) and len(audit['actual_completed_parent_checks']) == 2 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2', + PYTHONNOUSERSITE='1', PYTHONDONTWRITEBYTECODE='1') + for key in ['PYTHONPATH','PYTHONHOME','SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def status(phase, **extra): + record = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), member=args.member, **extra) + write_json(study / 'state.json', record) + print(json.dumps(record), flush=True) + + def finish(phase, **extra): + write_json(complete, dict(phase=phase, master_plan_sha256=digest(plan_path), + created_utc=datetime.now(timezone.utc).isoformat(), **extra)) + status(phase, **extra) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study / f'{phase}.log').open('a') as stream: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=stream, stderr=subprocess.STDOUT, start_new_session=True) + try: + code = process.wait(timeout=timeout) + assert code == 0, (phase, code) + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try:process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL); process.wait() + + def wait_memory(): + while int(subprocess.check_output(['nvidia-smi','--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB'] * 1024: + time.sleep(5) + + status('waiting_for_complete_declared_raw_parents') + while not (parent_ready(parent) and parent_ready(previous)): + time.sleep(10) + receipts = dict(parent=parent_receipt(parent, member['capacity'], member['epoch'], plan), + previous=parent_receipt(previous, member['capacity'], member['previous_epoch'], plan)) + resolved_path = study / 'resolved-parents.json' + resolved = dict(master_plan_sha256=digest(plan_path), member=member, receipts=receipts) + if resolved_path.exists():assert json.loads(resolved_path.read_text()) == resolved + else:write_json(resolved_path, resolved) + admitted = receipts['parent']['reward'] > receipts['previous']['reward'] + plan['parent_minimum_reward_gain'] + gate = dict(admitted=admitted, parent_reward=receipts['parent']['reward'], + previous_raw_reward=receipts['previous']['reward'], minimum_gain=plan['parent_minimum_reward_gain']) + if not (study / 'parent-gate.json').exists():write_json(study / 'parent-gate.json', gate) + else:assert json.loads((study / 'parent-gate.json').read_text()) == gate + if not admitted: + finish('rejected_at_parent_gain_gate', **gate) + return + output.mkdir(exist_ok=True) + package = output / 'package' + construction_path = study / 'construction-check.json' + if not construction_path.exists(): + if package.exists(): + try:verify_package(package);check_link_only(parent / 'package', package) + except (OSError,ValueError,KeyError,AssertionError): + package.rename(package.with_name('package-incomplete-' + str(time.time_ns()))) + if not package.exists(): + execute('build', [HERE / 'make_splice_probability_candidate.py', parent / 'package', package, + '--training-prior', plan['training_prior']]) + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] + assert json.loads((package / 'package_manifest.json').read_text())['parameters'] == plan['capacities'][member['capacity']]['parameters'] + before = (parent / 'package/predict.py').read_text() + marker = 'from reference.alphagenome_50m.model import CompactModel, ModelConfig\n' + old = 'data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16),' + new = 'data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16),' + assert before.count(marker) == before.count(old) == 1 + expected = before.replace(marker, marker + 'from splice_output import splice_output_scores\n').replace(old, new) + assert (package / 'predict.py').read_text() == expected + for name in ['infer.sh','perturb.sh']: + assert digest(package / name) == digest(parent / 'package' / name) + assert json.loads((package / 'splice_calibration.json').read_text()) == json.loads(Path(plan['calibration_reference']).read_text()) + history = package / 'training_history'; history.mkdir(exist_ok=True) + additions = {'fixed-swa105-master-plan.json': plan_path, + 'fixed-swa105-resolved-parents.json': resolved_path, + 'fixed-swa105-parent-check.json': STUDY / 'parent-check.json', + 'fixed_swa105.py': HERE / 'fixed_swa105.py', + 'run_fixed_swa105.py': Path(__file__)} + for name, source in additions.items(): + (history / name).write_bytes(source.read_bytes()) + readme = package / 'README.md' + body = readme.read_text().replace( + 'In preliminary\ndiagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed.', + "The complete reports below determine this checkpoint's track and variant tradeoffs.") + note = ('\nThis fixed-link follow-up uses the frozen SWA68 average of six parameter states ' + 'ending at the complete eighth training pass. It applies the original four training-only offsets ' + 'without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. ' + 'Both entry points use the same link. All200 validation windows and all13 development metrics ' + 'are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. ' + 'A publication additionally requires native replay and three sampled runtime estimates below four hours.\n') + if note.strip() not in body:body += note + readme.write_text(body) + manifest = json.loads((package / 'package_manifest.json').read_text()) + for name in ['README.md', *['training_history/' + name for name in additions]]: + manifest['files'][name] = digest(package / name) + write_json(package / 'package_manifest.json', manifest) + verification = verify_package(package) + write_json(construction_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), resolved_parents_sha256=digest(resolved_path), + package_inference_signature=inference_signature(package), exact_link_only_checks=checks, + parameters=manifest['parameters'], network_updates=0, prior_refitted=False, package_verified=True)) + construction = json.loads(construction_path.read_text()) + assert construction['master_plan_sha256'] == digest(plan_path) + assert construction['resolved_parents_sha256'] == digest(resolved_path) + verify_package(package) + signature = inference_signature(package) + assert construction['package_inference_signature'] == signature + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] and checks['candidate'] == signature + native = study / 'native-recovery' + if not (native / 'report.json').exists() or not (output / 'valid-variants').exists(): + status('waiting_for_bounded_native_and_derivation_slot') + with Path(plan['bounded_lock']).open('a') as probe_lock: + fcntl.flock(probe_lock, fcntl.LOCK_EX);wait_memory() + if not (native / 'report.json').exists(): + execute('native', [HERE / 'check_native_recovery.py', package, native], plan['native_timeout_seconds']) + if not (output / 'valid-variants').exists(): + materialization = dict(plan['materialization'], candidates=[dict(parent=str(parent), candidate=str(output), + parent_inference_signature=checks['parent'], candidate_inference_signature=signature)]) + path = study / 'materialization-plan.json' + if path.exists():assert json.loads(path.read_text()) == materialization + else:write_json(path, materialization) + execute('materialize', [HERE / 'materialize_splice_link_outputs_v2.py', path], plan['materialization_timeout_seconds']) + proof = json.loads((native / 'report.json').read_text()) + assert proof['package_inference_signature'] == signature and proof['all_arrays_bit_identical'] + assert proof['native_vendored_entrypoints_passed'] and proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks':30,'perturbations':22} + derivation = json.loads((output / 'valid-variants/perturb-link-derivation.json').read_text()) + assert derivation['checks']['candidate'] == signature and derivation['checks']['parent'] == checks['parent'] + assert derivation['native_arrays_bit_exact'] == 36 and derivation['queries'] == 2697 + if not (output / 'evaluation.json').exists(): + status('waiting_for_full_evaluation_slot') + with Path(plan['evaluation_lock']).open('a') as evaluation_lock: + fcntl.flock(evaluation_lock, fcntl.LOCK_EX);wait_memory() + execute('evaluate', [HERE / 'evaluate_package.py', package, output, + '--resume-partial-tracks','--resume-partial-perturbations']) + assert json.loads((output / 'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output / 'valid-score/report.json').read_text()) + variants = json.loads((output / 'valid-variant-score/report.json').read_text()) + raw_tracks = json.loads((parent / 'valid-score/report.json').read_text()) + raw_variants = json.loads((parent / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None and len(variants) == 13 + affected = {'sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'} + for family, report in tracks['families'].items(): + if family != 'splice_sites':assert report == raw_tracks['families'][family],family + for name, report in variants.items(): + assert math.isfinite(report['model']) + for field in ['rows','perturbations','metric','anchors']:assert report[field] == raw_variants[name][field],(name,field) + if name not in affected:assert report == raw_variants[name],name + comparison_path = study / 'matched-comparison.json' + if not comparison_path.exists(): + write_json(comparison_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), reward=tracks['overall_mean_family_pearson'], + reward_delta=tracks['overall_mean_family_pearson']-raw_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-raw_tracks['families'][k]['mean_pearson'] for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-raw_variants[k]['model'] for k,v in variants.items()}, + unaffected_family_reports_exact=10, unaffected_variant_reports_exact=8, package_inference_signature=signature)) + decision_path = study / 'publication-decision.json' + if not decision_path.exists(): + current = json.loads(Path('/workspace/submission/validation_tracks.json').read_text())['overall_mean_family_pearson'] + eligible = tracks['overall_mean_family_pearson'] > current + plan['publication_minimum_reward_gain'] + runtime_passed = None + if eligible: + latency = study / 'native-latency';latency.mkdir(exist_ok=True) + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency / 'plan.json').exists():assert json.loads((latency / 'plan.json').read_text()) == timing_plan + else:write_json(latency / 'plan.json', timing_plan) + if not (latency / 'report.json').exists(): + execute('native-latency', [HERE / 'measure_native_latency.py', latency / 'plan.json']) + timing = json.loads((latency / 'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + names = ['development_mixture_seconds','slowest_observed_stratum_mean_seconds','pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in names) + runtime_passed = max(estimates[k] for k in names) <= 14400 + write_json(study / 'runtime-gate.json', dict(passed=runtime_passed, estimates=estimates, + report_sha256=digest(latency / 'report.json'), scope='Shared-load extrapolations, not guaranteed upper bounds.')) + if runtime_passed:execute('publish', [HERE / 'promote_if_better.py', package, output]) + write_json(decision_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward=tracks['overall_mean_family_pearson'], current_reward_at_review=current, + eligible_at_review=eligible, runtime_passed=runtime_passed, + publication='Atomic promoter rechecks the current strongest package; see publish.log if run.')) + for phase, helper in [('tracks','diagnose_track_uncertainty.py'), + ('remaining-variants','diagnose_remaining_variant_uncertainty.py')]: + report_path = study / f'paired-{phase}.json' + marker_path = study / f'paired-{phase}-complete.json' + if marker_path.exists(): + assert json.loads(marker_path.read_text())['output_sha256'] == digest(report_path) + continue + def diagnostic_complete(): + try: + report = json.loads(report_path.read_text()) + assert Path(report['candidate']).resolve() == output.resolve() + assert Path(report['control']).resolve() == parent.resolve() + assert report['replicates'] == 2000 and report['seed'] == 1729 + assert report['diagnostic_only_no_parameter_fitting'] + leaf = 'valid-score/report.json' if phase == 'tracks' else 'valid-variant-score/report.json' + assert report['report_sha256'] == {'candidate':digest(output / leaf),'control':digest(parent / leaf)} + if phase == 'tracks': + assert len(report['point_deltas']) == 12 and 'reward' in report['point_deltas'] + assert set(report['bootstrap']) == {'chromosome','chromosome_10Mb_blocks'} + for result in report['bootstrap'].values(): + assert set(result['deltas']) == set(report['point_deltas']) + else: + assert set(report['evaluations']) == affected | {'caqtl','bqtl_spi1'} + for item in report['evaluations'].values(): + assert set(item['bootstrap']) == {'variant_id','genomic_block'} + for result in item['bootstrap'].values(): + assert result['reference_replicates_checked'] >= 32 + assert result['maximum_absolute_metric_error'] < 1e-10 + return True + except (FileNotFoundError,json.JSONDecodeError,AssertionError,KeyError): + return False + if not diagnostic_complete(): + execute('paired-' + phase, [HERE / helper, output, parent, report_path, + '--replicates','2000','--seed','1729']) + assert diagnostic_complete(), ('incomplete diagnostic', phase) + write_json(marker_path, dict(output_sha256=digest(report_path), master_plan_sha256=digest(plan_path))) + finish('complete', reward=tracks['overall_mean_family_pearson'], + matched_comparison_sha256=digest(comparison_path), publication_decision=json.loads(decision_path.read_text()), + diagnostics='All11families/sevenvariants, with eight unaffected official variant reports exact; not anall13bootstrap.') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_swa68_study.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_swa68_study.py new file mode 100644 index 0000000000000000000000000000000000000000..7c762ca8e6a2db80e782e20ddb9f1139fcf0e43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/run_swa68_study.py @@ -0,0 +1,144 @@ +"""Capture six fixed existing small8 checkpoints, average once, and evaluate one model.""" +import argparse +from datetime import datetime, timezone +import fcntl +import gc +import json +import os +from pathlib import Path +import shutil +import signal +import subprocess +import sys +import time + +import torch +from snapshot_average import average, capture, digest + + +def atomic_json(path, value): + temporary = path.with_suffix('.tmp') + temporary.write_text(json.dumps(value, indent=2)+'\n') + temporary.replace(path) + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan', type=Path) + args = parser.parse_args() + plan = json.loads(args.plan.read_text()) + study, here = args.plan.parent, Path(__file__).resolve().parent + with (study/'controller.lock').open('a') as controller_lock: + fcntl.flock(controller_lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + run(plan, args.plan, study, here) + + +def run(plan, plan_path, study, here): + for name, expected in plan['helper_sha256'].items(): + assert digest(here/name) == expected, name + source, output, training = map(Path,[plan['source'],plan['output'],plan['training_directory']]) + for name, expected in plan['source_sha256'].items(): + assert digest(source/name) == expected, name + for path, expected in plan['evidence_sha256'].items(): + assert digest(path) == expected, path + torch.set_num_threads(1) + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + output.mkdir(parents=True,exist_ok=True) + environment = dict(os.environ,OPENBLAS_NUM_THREADS='1',OMP_NUM_THREADS='2',MKL_NUM_THREADS='2') + environment.pop('SEQFUNC_TRAIN_BYTECACHE',None) + steps = plan['snapshot_steps'] + assert steps == [324000,325000,326000,327000,328000,328160] + assert steps[-1] == 8*41020 and all(step % 1000 == 0 for step in steps[:-1]) + history = source/'training_history' + history.mkdir(exist_ok=True) + for name in ['snapshot_average.py','run_swa68_study.py','evaluate_swa68_checkpoint.py']: + shutil.copy2(here/name,history/name) + shutil.copy2(plan_path,history/'swa68-plan.json') + + def status(phase, **values): + record = dict(phase=phase,utc=datetime.now(timezone.utc).isoformat(),**values) + atomic_json(study/'state.json',record) + print(json.dumps(record),flush=True) + + records = [] + for step in steps: + destination = study/'snapshots'/f'step-{step:06d}.pt' + original = training/('epoch-08.pt' if step == steps[-1] else 'progress.pt') + status('waiting_for_exact_snapshot',step=step,captured=len(records),source=str(original)) + seen = None + while True: + try: + info = original.stat() + signature = (info.st_ino,info.st_mtime_ns,info.st_size) + except FileNotFoundError: + signature = None + if destination.exists() or signature != seen: + record = capture(original,destination,step) + seen = signature + if record is not None: + break + time.sleep(2) + state = torch.load(destination,map_location='cpu',weights_only=True) + assert state['global_step'] == step and state['optimizer_steps'] == step//4 + assert state['examples'] == 41020 and state['schedule_epochs'] == 14 and state['seed'] == 17 + assert state['objective'] == 'variance_balanced' + assert state['epoch']*41020+state['offset'] == step + assert state['config'] == plan['model_config'] + assert state['family_loss_scales'] == plan['family_loss_scales'] + for key,value in plan['training_arguments'].items(): + assert state['training_arguments'][key] == value,(step,key) + del state + gc.collect() + records.append(record) + atomic_json(study/'captures.json',records) + status('snapshot_captured',**record,captured=len(records)) + + status('averaging_six_fixed_snapshots') + state, intervention = average([record['path'] for record in records],steps) + assert state['global_step'] == 328160 and state['optimizer_steps'] == 82040 + assert state['epoch'] == 8 and state['offset'] == 0 + assert state['weights_only_average'] and 'optimizer' not in state and 'rng_cpu' not in state + parameters = sum(value.numel() for value in state['model'].values()) + assert parameters == plan['parameters'] == 13290504 + checkpoint = output/'last.pt' + if checkpoint.exists(): + old = torch.load(checkpoint,map_location='cpu',weights_only=True) + assert old['intervention'] == intervention + assert old['model'].keys() == state['model'].keys() + assert all(torch.equal(old['model'][key],value) for key,value in state['model'].items()) + del old + else: + temporary = checkpoint.with_suffix('.tmp') + torch.save(state,temporary) + temporary.replace(checkpoint) + del state + gc.collect() + report = dict(created_utc=datetime.now(timezone.utc).isoformat(),**intervention, + output_checkpoint_sha256=digest(checkpoint),parameters=parameters, + same_trajectory_as_control=plan['control'],final_snapshot_source='immutable epoch-08.pt', + no_training_schedule_or_checkpoint_cadence_change=True) + atomic_json(study/'averaging-report.json',report) + shutil.copy2(study/'averaging-report.json',history/'swa68-averaging-report.json') + shutil.copy2(study/'captures.json',history/'swa68-captures.json') + status('candidate_evaluation') + with (study/'evaluate.log').open('a') as log: + process = subprocess.Popen([sys.executable,str(here/'evaluate_swa68_checkpoint.py'), + str(plan_path),'candidate'],env=environment,stdout=log,stderr=subprocess.STDOUT,start_new_session=True) + try: + code = process.wait() + if code: + raise RuntimeError(f'evaluation failed with exit status {code}') + except BaseException: + if process.poll() is None: + os.killpg(process.pid,signal.SIGTERM) + try: + process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid,signal.SIGKILL) + process.wait() + raise + status('complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/snapshot_average.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/snapshot_average.py new file mode 100644 index 0000000000000000000000000000000000000000..7c47f7e1c6212203ac9712a051b05af0773db5c7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/snapshot_average.py @@ -0,0 +1,103 @@ +"""Capture exact saved checkpoints and average a fixed list of model states.""" +from copy import deepcopy +import hashlib +from pathlib import Path +import shutil + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def capture(source, destination, expected_step): + """Return capture metadata, or None while the source has not reached the step. + + A single open inode supplies the inspected state and the copied bytes. A missed + step is an error, never an implicit change to the prespecified averaging window. + Existing destination files support recovery after either atomic rename. + """ + source, destination = Path(source), Path(destination) + if destination.exists(): + state = torch.load(destination, map_location='cpu', weights_only=True) + assert state['global_step'] == expected_step, 'saved capture has wrong step' + return {'step': expected_step, 'sha256': digest(destination), + 'path': str(destination), 'source': str(source)} + try: + stream = source.open('rb') + except FileNotFoundError: + return None + with stream: + state = torch.load(stream, map_location='cpu', weights_only=True) + step = state['global_step'] + if step < expected_step: + return None + if step > expected_step: + raise ValueError(f'Missed fixed snapshot {expected_step}; observed {step}') + del state + stream.seek(0) + destination.parent.mkdir(parents=True, exist_ok=True) + temporary = destination.with_suffix('.tmp') + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8 * 1024 * 1024) + temporary.replace(destination) + return {'step': expected_step, 'sha256': digest(destination), + 'path': str(destination), 'source': str(source)} + + +def average(paths, expected_steps): + """Equal arithmetic mean in FP64, cast once to each original floating dtype.""" + assert len(paths) == len(expected_steps) > 1 + assert list(expected_steps) == sorted(set(expected_steps)) + sums, constants, template, records, final = {}, {}, None, [], None + invariant_fields = ('config', 'examples', 'schedule_epochs', 'seed', 'objective', + 'objective_parameters') + for path, step in zip(map(Path, paths), expected_steps): + state = torch.load(path, map_location='cpu', weights_only=True) + assert state['global_step'] == step + invariants = {key: state.get(key) for key in invariant_fields} + model = state['model'] + if template is None: + template = deepcopy(invariants) + layout = {key: (value.shape, value.dtype) for key, value in model.items()} + else: + assert invariants == template, 'snapshot training recipe changed' + assert model.keys() == layout.keys(), 'snapshot model keys changed' + for key, value in model.items(): + assert (value.shape, value.dtype) == layout[key], key + if value.is_floating_point(): + assert torch.isfinite(value).all(), f'nonfinite tensor: {key}' + if key not in sums: + sums[key] = value.double().clone() + else: + sums[key].add_(value.double()) + else: + assert not value.is_complex(), 'complex model tensors are unsupported' + if key not in constants: + constants[key] = value.clone() + else: + assert torch.equal(constants[key], value), f'nonfloating buffer changed: {key}' + records.append({'step': step, 'sha256': digest(path)}) + final = state + mean = {key: ((sums[key] / len(paths)).to(layout[key][1]) if key in sums else constants[key]) + for key in layout} + excluded = {'optimizer', 'scaler', 'rng_cpu', 'rng_cuda'} + output = {key: value for key, value in final.items() if key not in excluded} + output['model'] = mean + intervention = { + 'kind': 'fixed_checkpoint_parameter_average', + 'snapshots': records, 'equal_weights': True, 'accumulation_dtype': 'float64', + 'cast_once_to_original_tensor_dtype': True, + 'nonfloating_buffers_identical': True, + 'optimizer_and_RNG_excluded_not_a_training_resume_checkpoint': True, + 'additional_training_updates': 0, 'development_labels_fit_weights': False, + } + history = list(output.get('interventions', [])) + if output.get('intervention') and output['intervention'] not in history: + history.append(output['intervention']) + history.append(intervention) + output.update(intervention=intervention, interventions=history, + weights_only_average=True) + return output, intervention diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-averaging-report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-averaging-report.json new file mode 100644 index 0000000000000000000000000000000000000000..a2c0cbff4dcbf5ca1e4dd22c65cbad72903e603d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-averaging-report.json @@ -0,0 +1,42 @@ +{ + "created_utc": "2026-10-25T18:27:47.015926+00:00", + "kind": "fixed_checkpoint_parameter_average", + "snapshots": [ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c" + } + ], + "equal_weights": true, + "accumulation_dtype": "float64", + "cast_once_to_original_tensor_dtype": true, + "nonfloating_buffers_identical": true, + "optimizer_and_RNG_excluded_not_a_training_resume_checkpoint": true, + "additional_training_updates": 0, + "development_labels_fit_weights": false, + "output_checkpoint_sha256": "4be1f73ee80bd804a186deab5245ef0c7f4c2fd5b33acc070ed880e2e95d72af", + "parameters": 13290504, + "same_trajectory_as_control": "/workspace/research/experiments/scale13-small-epoch08", + "final_snapshot_source": "immutable epoch-08.pt", + "no_training_schedule_or_checkpoint_cadence_change": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-captures.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-captures.json new file mode 100644 index 0000000000000000000000000000000000000000..c7dbc2e9926917ef907d5e9b7d7d961cb240449c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-captures.json @@ -0,0 +1,38 @@ +[ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae", + "path": "/workspace/research/studies/swa68/snapshots/step-324000.pt", + "source": "/workspace/research/experiments/scale13-small/progress.pt" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2", + "path": "/workspace/research/studies/swa68/snapshots/step-325000.pt", + "source": "/workspace/research/experiments/scale13-small/progress.pt" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7", + "path": "/workspace/research/studies/swa68/snapshots/step-326000.pt", + "source": "/workspace/research/experiments/scale13-small/progress.pt" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e", + "path": "/workspace/research/studies/swa68/snapshots/step-327000.pt", + "source": "/workspace/research/experiments/scale13-small/progress.pt" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5", + "path": "/workspace/research/studies/swa68/snapshots/step-328000.pt", + "source": "/workspace/research/experiments/scale13-small/progress.pt" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c", + "path": "/workspace/research/studies/swa68/snapshots/step-328160.pt", + "source": "/workspace/research/experiments/scale13-small/epoch-08.pt" + } +] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..5123c3cf7084e5a0aef27b26dfff21ebd7c59d1f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-plan.json @@ -0,0 +1,512 @@ +{ + "schema": "seqfunc-fixed-small8-tail-average/v1", + "created_utc": "2026-10-25T13:32:10.317098+00:00", + "source": "/workspace/research/studies/swa68/source", + "source_parent": "/workspace/research/studies/loader_io/cached-integration/source", + "source_sha256": { + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "benchmark_batches.py": "9b843960109d2929fd30a993bcba5b0f7b8703aff5bf9bfeca93d6ade1ca7791", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "afb19fb051b895efcc0460b344bcc298a6ce5c0a3aa322dec4b963d5373b7735", + "make_package.py": "af5a3d3671c54784f345be9b942b121aaf07ac554670409caf068c211a1fa142", + "package.sh": "c1450052f2c28e2449f38be0523e4f22bb2f68ab839f754f9113c250ff0dcce7", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "perturb.sh": "3c61de823019e8f8232223a07447f2925a71454edecfcab918df36c6d0321543", + "predict.py": "2725592456c55c188b22c376f0cb249e9f46b4ae1370d2226df7143da358a8b2", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "provisional_large.yaml": "7391e29bd79f80f4edfeee8efa9a61ffb95ed7fda2e1ff4a6a48bdba21c0a82a", + "provisional_large_local64.yaml": "4616441df1c88ea6a9f911a4226fea581c2c610e90ba583a34705e961b06d735", + "publish.sh": "95e3c079aa5a1c4c7a996e76cc3234f3777f1f764f8827fc97a96c873a8705e3", + "publish_atomic.py": "63adee6090f9f15aeafca369235f68de33b95f89ac468c76194446b207c31b41", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tests/conftest.py": "43f45639b77003e29387eeab62bc62a1acfc778a88aaa71d760104936d6f3be9", + "reference/alphagenome_50m/tests/test_bridge.py": "564470de114618d3b3d6fcc7a6dc4dd8be90abcc833c5407ae5dd436dfe9fbc0", + "reference/alphagenome_50m/tests/test_contract.py": "6a1a788a621ca6d29853d5dd6e4c4682c0bf8781978eed6944056037d83579f4", + "reference/alphagenome_50m/tests/test_local.py": "31eb2f113979a965894cab62001e35b93180d404f76dd80d9b1219e461434b01", + "reference/alphagenome_50m/tests/test_metrics.py": "8dc6f1a596740b3f234c8e2b2bb7b6cd5a2f5d4544f0fdf85764d54113aea1cc", + "reference/alphagenome_50m/tests/test_model.py": "c744dbf1f0ff946924b0e335d6cef2f8df7f050753e03ce6b883a9ca9330796c", + "reference/alphagenome_50m/tests/test_uniform_splice.py": "e3a67005441db9e16f04f535863d05e1dbe2321e03adeee0ba4cd3c00ece57d4", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "test_batched_training.py": "682992ae3b80524693f37bf7127e10acb238e328957071d13c617ead63df723f", + "test_bounded_residuals.py": "1532db4c09aa4bb88f21ddafc1e529b7b8d1c3afbc39efe737dfac7ca6a1a284", + "test_context_extension.py": "9ea0497f991d0f5bb722e6db7148fdc21d72c4067ecdfe718436e3a14cace455", + "test_training.py": "de749aed5f8e3ec3948f874c8d5564a503461638aad5d82a96dc07228c45c8e9", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497" + }, + "training_directory": "/workspace/research/experiments/scale13-small", + "control": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/swa68", + "snapshot_steps": [ + 324000, + 325000, + 326000, + 327000, + 328000, + 328160 + ], + "weights": [ + 0.16666666666666666, + 0.16666666666666666, + 0.16666666666666666, + 0.16666666666666666, + 0.16666666666666666, + 0.16666666666666666 + ], + "parameters": 13290504, + "additional_training_updates": 0, + "model_config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "training_arguments": { + "batch_size": 4, + "epochs": 14, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "save_every": 1000 + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "helper_sha256": { + "run_swa68_study.py": "fcb73022e4f4eeb1882dbf6872e8ed1906c783e6114931fdc071b3628d293464", + "snapshot_average.py": "ae2c8bd6c082e9e3bed0c51623a5ff9a65c24364400a1b77fb8a4f44ba5d0f04", + "evaluate_swa68_checkpoint.py": "db858ffdeabfb5e7cfffb8a9132152a094814a2bc837c528fa1cdaf85fba4c05", + "make_run_readme.py": "49940541d605d2b601ce2bc53f96c3164f4d629a2a982288b095381c6418cbb8", + "vendor_training_assets.py": "0789127e2ed86143ca0bb195cfa76a09d13d931cafd88e0387431f06fa80216e", + "make_orientation_candidate.py": "3ec3e59fee38a5d7107c957cfc578d612f92645453c3c507e8d437f2a7d4e708", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363" + }, + "evidence_sha256": { + "/workspace/research/studies/swa68/preflight.json": "f24753bb354d0aa1261768202c159910a4d0713e1925e3693d9057de62495354", + "/workspace/research/studies/swa30/mechanical-check.json": "2dd7fdebee69ccbf99e6ae8d5dff7612e8a31ab0678950555125a83a306914f3", + "/workspace/research/studies/swa30/complete-raw-comparison.json": "36fdf6b85726f048627d7653ddb602039b6b3bf5f3b6038c21b9673b238e809c", + "/workspace/research/studies/swa30/diagnostics/raw/uncertainty-summary.json": "1bf3ec62b8b5fcbddb5186a624bac54f18d203ac6a9d0c469ff272126037e4a5", + "/workspace/research/studies/swa30/study.json": "0ab208202620b512d90fe7456a13313e111420a35966775c51de127659c443bd", + "/workspace/research/studies/scale13/study-cached-io.json": "88aa6d95ef93776ed3551d68d98b1a31d726aa4b0270edbfc16b3f16ceeb937f" + }, + "evaluation": { + "candidate": { + "checkpoint": "/workspace/research/experiments/swa68/last.pt", + "source": "/workspace/research/studies/swa68/source", + "output": "/workspace/research/experiments/swa68", + "name": "SeqFunc small8 with a fixed six-checkpoint parameter average", + "lineage": "One fixed equal parameter average ofsixsavedstatesfromourunchangedsmallscale13trajectoryat324000,325000,326000,327000,328000,and328160trainingexamples. Thefinalstatecompletes8full41020intervalFOLD_0permutations/82040batch4updates,seed17,variance-balancedobjective,original14epochcosineLRschedule. Theexisting1000-examplecheckpointcadenceisnotchanged. Firstfivecapturesuseoneopenprogressinode;thelastusesimmutableepoch-08.pt. Missedsnapshotsfailclosed;noalternativestatesorweightsareselected. AccumulatefloatingtensorsinFP64,castonce;nonfloatingbuffersmustmatch. 13290504learnedparameters,0additionalSGD,oneaveragedmodelsameinferencecostservesbothentries. Rawspliceoutputsandoriginalsparsedecodersretained. Optimizer/RNGomitted:averagedcheckpointisinference-only,notatrainingresumestate. ThisfollowupismotivatedbySWA30complexTraitGymdevelopmentgainbutdiffersinrecipe,batchsize,andaveragingdensity;transferisunknown. Snapshotstepsanduniformweightsarefixedbeforecapture,notfittedtodevelopmentlabels. Comparewithunchangedrawsmall8,alreadycoveredbyoriginalcheckpointmonitor;noextra controltraining. Native52/full200windows2697queriesall13/native624andallthree4hruntimeestimatesprecedepromotion. Noexternalpretrainedparameters,testlabels,orlabel-derivedinferencecache." + } + }, + "native_latency_template": { + "created_utc": "2026-10-25T13:32:10.317098+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "16GiB schedulingguard forone13.29Msmallmodel; noorientationensemble oradditionaltraining. Samearchitectureasmaintrajectory.", + "scope": "Native16window96query624arrayreplay ofonefixedsix-checkpointparametermeanafterfull200/2697/13validation,allthree4hruntimeestimatesbeforepromotion.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "Requireallthreeestimates<=14400s;thisisshared-loadsampling,notaruntimeupperbound." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "hardware": { + "gpus": 1, + "gpu": "H10080GB", + "cpus": 22, + "ram_GiB_request": 224 + }, + "recovery": "StartthiswatcherBEFOREresumingmaintrainingifapproaching324000. Exclusivecontrollerlock;exactcapturehelperfromSWA30,existingatomiccapturesreusedandhashed. Ifanyscheduledprogresssnapshotismissed,failclosed;neverreplaceit. Finalstateusesimmutableepoch08eveniflast.ptadvances. Averagedlast.ptisNOTatrainingresumestate. No furthercalibrationcandidateispredeclared.", + "selection": "Solefixedstrongersmall8trajectory,6equalcheckpointscoveringlast4160examples. Cadenceandwindowchosenmechanicallybeforecapture; noalternativewindoworweightsearch. Raw-onlyfollowup;oldercandidatepoolsunchanged.", + "caveat": "One-seedexploratorytransferfromolderSWA30recipe;trackrewardgainthereuncertainandonlycomplexTraitGymCIpositive. NoMendelianoruniversalvariantbenefitassumed. Fullcomparisonsrequired; runtimeestimatesnotguaranteedbounds." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-reused-averaging-mechanical-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-reused-averaging-mechanical-check.json new file mode 100644 index 0000000000000000000000000000000000000000..0bd0c493433f363c00b96976e968f8a5edc83803 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/swa68-reused-averaging-mechanical-check.json @@ -0,0 +1,15 @@ +{ + "status": "passed", + "helper_sha256": "ae2c8bd6c082e9e3bed0c51623a5ff9a65c24364400a1b77fb8a4f44ba5d0f04", + "test_sha256": "902061aabe9817ed2348dafea2bdb480ab1cf8cc6e36aa0506825cc109321f3d", + "tests": [ + "missing/not-yet-reached snapshot waits", + "single open inode survives atomic source replacement", + "capture retry is byte identical", + "missed/wrong steps rejected", + "FP64 accumulation and one cast match analytical mean with cancellation", + "optimizer/RNG excluded", + "fixed recipe/nonfloating buffer/nonfinite/order guards" + ], + "additional_training_or_GPU_compute": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/test_snapshot_average.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/test_snapshot_average.py new file mode 100644 index 0000000000000000000000000000000000000000..421860567edc8fb3089deec43d8d364734c0b825 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_history/test_snapshot_average.py @@ -0,0 +1,80 @@ +"""Mechanical checks for exact capture, averaging and interrupted recovery.""" +from pathlib import Path +import tempfile +from unittest.mock import patch + +import torch + +from snapshot_average import average, capture, digest + + +def state(step, values, **extra): + return {'global_step': step, 'config': {'width': 2}, 'examples': 41020, + 'schedule_epochs': 4, 'seed': 17, 'objective': 'balanced_splice', + 'model': {'weight': torch.tensor(values, dtype=torch.float32), + 'constant': torch.tensor([9], dtype=torch.int64)}, + 'optimizer': {'state': 123}, 'rng_cpu': torch.arange(5), **extra} + + +def expect_failure(function, exception=AssertionError): + try: + function() + except exception: + return + raise AssertionError('Expected rejection did not occur') + + +def main(): + with tempfile.TemporaryDirectory(prefix='snapshot-average-fixture-') as temporary: + root = Path(temporary) + source, captured = root / 'progress.pt', root / 'captured.pt' + assert capture(source, captured, 100) is None + torch.save(state(50, [1, 2]), source) + assert capture(source, captured, 100) is None + torch.save(state(100, [3, 7]), source) + expected_hash = digest(source) + original_load = torch.load + + def replace_after_load(stream, **kwargs): + result = original_load(stream, **kwargs) + replacement = source.with_suffix('.next') + torch.save(state(200, [8, 9]), replacement) + replacement.replace(source) + return result + + with patch('snapshot_average.torch.load', side_effect=replace_after_load): + record = capture(source, captured, 100) + assert record['sha256'] == expected_hash == digest(captured) + assert capture(source, captured, 100) == record + expect_failure(lambda: capture(source, root / 'missed.pt', 100), ValueError) + expect_failure(lambda: capture(source, captured, 200)) + + paths, steps = [], [100, 200, 240] + arrays = [[1e8, 3], [1, -2], [-1e8, 8]] + for step, values in zip(steps, arrays): + path = root / f'{step}.pt' + torch.save(state(step, values), path) + paths.append(path) + result, intervention = average(paths, steps) + expected = torch.tensor(arrays, dtype=torch.float64).mean(0).float() + assert torch.equal(result['model']['weight'], expected) + assert result['global_step'] == 240 + assert 'optimizer' not in result and 'rng_cpu' not in result + assert torch.equal(result['model']['constant'], torch.tensor([9])) + again, again_intervention = average(paths, steps) + assert again_intervention == intervention + assert torch.equal(again['model']['weight'], result['model']['weight']) + torch.save(state(240, [2, 3], objective='rna_profile'), paths[-1]) + expect_failure(lambda: average(paths, steps)) + changed = state(240, [2, 3]) + changed['model']['constant'][0] = 10 + torch.save(changed, paths[-1]) + expect_failure(lambda: average(paths, steps)) + torch.save(state(240, [float('nan'), 3]), paths[-1]) + expect_failure(lambda: average(paths, steps)) + expect_failure(lambda: average(paths, [100, 240, 200])) + print('Exact inode capture, retry, missed-step rejection, FP64 mean, recipe/buffer/nonfinite guards passed.') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..9f45a4ef656aa077dce103c513e8509d0eb173d8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/training_provenance.json @@ -0,0 +1,193 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 8, + "offset": 0, + "global_step": 328160, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 73897.99778079404, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "intervention": { + "kind": "fixed_checkpoint_parameter_average", + "snapshots": [ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c" + } + ], + "equal_weights": true, + "accumulation_dtype": "float64", + "cast_once_to_original_tensor_dtype": true, + "nonfloating_buffers_identical": true, + "optimizer_and_RNG_excluded_not_a_training_resume_checkpoint": true, + "additional_training_updates": 0, + "development_labels_fit_weights": false + }, + "interventions": [ + { + "kind": "fixed_checkpoint_parameter_average", + "snapshots": [ + { + "step": 324000, + "sha256": "37695f91b7d8a37846df5372e27807353fd49d11c788f5a4f877ca840accdcae" + }, + { + "step": 325000, + "sha256": "66afb07c5bf176a5c2915fb82109f9beb93f918b822f4bbc71c346492654b4c2" + }, + { + "step": 326000, + "sha256": "661c84abd884afff183f269a8d14635e4854e49526e0c6e84abc4b915ba156b7" + }, + { + "step": 327000, + "sha256": "8a06e2fad1c3235ae4c8b8ba4a6934907e436af513d00fbfebac1c8a8775a42e" + }, + { + "step": 328000, + "sha256": "40c3aa3e6e2cf135f32670a443a236024ab28d9c2884c26ed7bb30216e46f4e5" + }, + { + "step": 328160, + "sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c" + } + ], + "equal_weights": true, + "accumulation_dtype": "float64", + "cast_once_to_original_tensor_dtype": true, + "nonfloating_buffers_identical": true, + "optimizer_and_RNG_excluded_not_a_training_resume_checkpoint": true, + "additional_training_updates": 0, + "development_labels_fit_weights": false + } + ], + "optimizer_steps": 82040, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..db3b14be4a89606b3170257326102ff6f549eccd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5909460463201712, + "valid_tracks": 167, + "per_track_pearson": [ + 0.7110539840385793, + 0.7346622653945708, + 0.4630974191305131, + 0.6587310881834911, + 0.4712654680281801, + 0.47161044693849685, + 0.49255300634037846, + 0.5756237421871789, + 0.590601922610321, + 0.6564972861636392, + 0.5902001793673592, + 0.5682760629486349, + 0.40181746135863566, + 0.5319234501305731, + 0.6939668747284352, + 0.5167332255151724, + 0.5665194731671939, + 0.5358863247291561, + 0.5702200395818412, + 0.5539608374273263, + 0.5221006235677889, + 0.4843797761407928, + 0.5619321992258494, + 0.5373106850022399, + 0.5367672022706237, + 0.5157939215800286, + 0.4938390984469013, + 0.5152734284355842, + 0.5255882002184615, + 0.5857152954367086, + 0.5033634804976237, + 0.540727372692966, + 0.49636635512816035, + 0.4142764806619685, + 0.3844069833299901, + 0.5998743818197315, + 0.5365682924476776, + 0.645529310775157, + 0.5266864335396517, + 0.5181636515155384, + 0.5972477876772195, + 0.5993735314453349, + 0.5150743015598616, + 0.4916580640430457, + 0.5633171644388285, + 0.5778841118914027, + 0.44744064536276484, + 0.5010681683601655, + 0.49523701977413176, + 0.5386890537998005, + 0.5099537087072948, + 0.47565056424775887, + 0.573728871882647, + 0.6108835484151283, + 0.4702245336067935, + 0.6937275166350286, + 0.8130176982693537, + 0.561460221328401, + 0.6042337311691801, + 0.6516670035965557, + 0.7758816898363711, + 0.5128486677387497, + 0.7002960479116217, + 0.6871711940555847, + 0.646466200788803, + 0.5215094616853477, + 0.5838522065886246, + 0.6503870578824794, + 0.8023484790716566, + 0.7214254268897868, + 0.5379605694880417, + 0.4912294453139966, + 0.46438402439073, + 0.5530805932636279, + 0.5146869243911271, + 0.5022096523795726, + 0.509406753238259, + 0.5514249810411569, + 0.5977822112819006, + 0.5055062912019557, + 0.5874418171714343, + 0.6110803306988181, + 0.5259325470797374, + 0.4993153299088408, + 0.5343638720575142, + 0.5258296982453212, + 0.5041474667347003, + 0.4947817172828721, + 0.46401492963020136, + 0.5181584539183199, + 0.5227358709036223, + 0.5179906209834026, + 0.49812008668867275, + 0.5177332281398536, + 0.579626561430115, + 0.40601648681855057, + 0.45674580821132355, + 0.44694895438097304, + 0.5048852797192213, + 0.520944955126509, + 0.4709635395596874, + 0.47422291321209, + 0.5191807210271355, + 0.4788613637326866, + 0.5585244470098697, + 0.5655751698065808, + 0.5528718224760147, + 0.534844768264645, + 0.5569554175965816, + 0.5668318972755393, + 0.620745853673369, + 0.5423972684464045, + 0.5359485409215047, + 0.4552656896111854, + 0.48052441756090697, + 0.4533994071973378, + 0.4924933085122429, + 0.4964500737433951, + 0.5113721974924159, + 0.4891665348840111, + 0.6673624911804935, + 0.68152343891645, + 0.7470620478243805, + 0.7391673526166065, + 0.7291254459729221, + 0.5878142197374792, + 0.8667985083803131, + 0.7347313577353052, + 0.7742473970310896, + 0.6930516154358456, + 0.7371698599677398, + 0.5748587716295751, + 0.6013703011864492, + 0.5746302674062708, + 0.7217206184078143, + 0.7547481060484083, + 0.7058720428639427, + 0.5520113763495672, + 0.43616311611563663, + 0.8240174213058867, + 0.7948738942507163, + 0.8148022038255804, + 0.8089600637641134, + 0.5873522684928774, + 0.8770825471694358, + 0.7884721564657098, + 0.7529854932175901, + 0.7416364617026271, + 0.8472528451198754, + 0.8250638550725221, + 0.6308780488728354, + 0.6345938154837107, + 0.7342871135572722, + 0.7625232226346519, + 0.7781466496178444, + 0.6174494463686775, + 0.7533422508032733, + 0.6071059599394084, + 0.6091710345329352, + 0.6994834249142561, + 0.6101184910015727, + 0.7871363261415525, + 0.7894164535404288, + 0.6585511903782028, + 0.6055393130472229, + 0.601262064025145, + 0.752389162929551 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.667163659594539, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5805086567195046, + 0.9317381699200643, + 0.7995113547100109, + 0.5382195817565406, + 0.5487679073654811, + 0.5616774880084474, + 0.9451423667664935, + 0.7373639539504133, + 0.47033476388119055, + 0.3672966723654966, + 0.8968182247090651, + 0.7238312937823337, + 0.4471083185567402, + 0.5383654734021491, + 0.5176736843336043, + 0.8211968632177253, + 0.8046488392498273, + 0.8572887984508126, + 0.8753810571924348, + 0.9034168860760806, + 0.7168903407437868, + 0.6407731146878404, + 0.9135621071611675, + 0.5491648316659871, + 0.5658412012618992, + 0.7906763807382846, + 0.47291869581699253, + 0.6884338106119615, + 0.5443382528812469, + 0.5727504206240378, + 0.6529941250480132, + 0.4889787944719328, + 0.850776501529245, + 0.6363814457393868, + 0.889007823271356, + 0.8970498909541832, + 0.4618658715414032, + 0.8375127275703945, + 0.8830555977947324, + 0.8702249484333647, + 0.8389392705793745, + 0.6794965961191834, + 0.7152471812363614, + 0.6547577516859735, + 0.5524181500498992, + 0.6493847401073074, + 0.7442185006843861, + 0.516902702601993, + 0.6177244710245701, + 0.4235670248691181, + 0.4045793226934098, + 0.5870714363452201, + 0.5263323214330725, + 0.8361031590920955, + 0.4599257818993389, + 0.4289735027300682, + 0.4835866095702587, + 0.5355934384704394, + 0.49999767362438996, + 0.5404176986057444, + 0.47654720162772624, + 0.6331395529506755, + 0.5232741915665757, + 0.5116076202526332, + 0.5409645359854099, + 0.5465211333551747, + 0.5770556054566087, + 0.5069909453930904, + 0.4081438561792192, + 0.45153069880712726, + 0.5263611012791093, + 0.6736761809309862, + 0.532116585077053, + 0.6561179150225751, + 0.38696123856737225, + 0.8229001132565021, + 0.9030292215528684, + 0.689490693871538, + 0.46713103604041667, + 0.4825907905494919, + 0.5642214927836431, + 0.5388189319706499, + 0.47946841558041153, + 0.5159501830756683, + 0.6012785002206429, + 0.5220616775567375, + 0.6368869185345603, + 0.5290899439523629, + 0.6585495892941136, + 0.5227435942948173, + 0.6536690995598328, + 0.42912204759045336, + 0.742338547981007, + 0.49948125368987123, + 0.7332613934569978, + 0.5293963680338335, + 0.7739589169879457, + 0.7436030763662229, + 0.396670855657727, + 0.4042575130519902, + 0.5002678124257561, + 0.3962914129903994, + 0.3794669808137656, + 0.7521361109700905, + 0.7248652026031234, + 0.42852718005480817, + 0.5154784246039685, + 0.6164881836785016, + 0.8561784667931559, + 0.5814103709171994, + 0.5857978811476955, + 0.6074065375860911, + 0.5393435582584082, + 0.4657239582786866, + 0.796801269798762, + 0.5995727453890788, + 0.518343044780911, + 0.5760182579298241, + 0.6675186044615184, + 0.49437815780720085, + 0.5535315846305674, + 0.5179825593141719, + 0.5889866330729024, + 0.573779372432034, + 0.6437761612261337, + 0.5762576711626625, + 0.6752877430851925, + 0.689424268036482, + 0.7159012776495609, + 0.658337749436345, + 0.4865922585456887, + 0.6318261307242984, + 0.6675957203183539, + 0.8096643449639133, + 0.46520662331914625, + 0.3579067235744918, + 0.6209192035226664, + 0.50378491197032, + 0.6690184250282989, + 0.6324650910624242, + 0.6422053024435307, + 0.5865592843363977, + 0.5066182402086968, + 0.7948870616864269, + 0.6490743220695727, + 0.7712325941730258, + 0.42062560190593085, + 0.3920472588986656, + 0.42786432879454783, + 0.38161606218723165, + 0.9059265361053236, + 0.35189685787259434, + 0.9107852732945166, + 0.7821463018170177, + 0.4919718706254488, + 0.38178338762708924, + 0.37892228622029017, + 0.39232513019193943, + 0.6785744650580994, + 0.7822098213115598, + 0.7732773119637208, + 0.6917273908598739, + 0.5897861327961741, + 0.5075365875073856, + 0.7029434217547633, + 0.6440286830381783, + 0.5484848814972243, + 0.6675823183713047, + 0.48424508437323577, + 0.5185857695584253, + 0.4319389954529395, + 0.8420392049744775, + 0.49583430832240316, + 0.5032648216358224, + 0.6845755294142482, + 0.7208965474028237, + 0.5501845126547035, + 0.3720106934118903, + 0.885150561629717, + 0.40520067642665797, + 0.7981190680797496, + 0.5325398538021465, + 0.6357117343064935, + 0.7572181025591255, + 0.47757192625182243, + 0.5960705465301884, + 0.6387907103654505, + 0.8395287827898094, + 0.4444936850055379, + 0.6421801426812588, + 0.6078580416540443, + 0.7378642150840288, + 0.9048107279192434, + 0.470394255545981, + 0.6195230080298287, + 0.5173236483506364, + 0.6796742086856057, + 0.7487439216488286, + 0.5877046696369741, + 0.48523595318798896, + 0.8814954558285635, + 0.8938197936885605, + 0.7652406008402144, + 0.8487088051306513, + 0.8583286199023024, + 0.8495419181045522, + 0.8629549145247986, + 0.39463477289547455, + 0.418989860331301, + 0.543138854368094, + 0.4947666036214584, + 0.4886236680189766, + 0.6444242247467198, + 0.8764967675452645, + 0.8587865466729079, + 0.7127002114923261, + 0.8417489963481004, + 0.8397823628108528, + 0.8700689791046741, + 0.9156140557075209, + 0.5780745308491341, + 0.675574340653391, + 0.7266028097875132, + 0.6975111187707943, + 0.839983450606481, + 0.6288820767418797, + 0.9210444889348673, + 0.8933965969009398, + 0.8818196912752031, + 0.9101573618857797, + 0.9000450591223288, + 0.8889649047496355, + 0.8444356868556838, + 0.8694670041919561, + 0.9189062484274692, + 0.4966070750855829, + 0.934597204472324, + 0.8892431834999086, + 0.8332599314883575, + 0.6461055838428089, + 0.8802111283879359, + 0.5617109705005007, + 0.7609789754144665, + 0.8826895720133741, + 0.6450473938826582, + 0.6331353816644796, + 0.7298304789688839, + 0.5593511376471868, + 0.5929410247220246, + 0.7557214505598974, + 0.8963445393812746, + 0.7817468578158314, + 0.8319060897999908, + 0.6698240476192792, + 0.8784137499114921, + 0.8196555553580276, + 0.8497202555836626, + 0.9143883067146582, + 0.8383745474111216, + 0.9087213171083673, + 0.8798111133986473, + 0.5719579677204047, + 0.8864141610859317, + 0.6195031819027915, + 0.5855917066591437, + 0.89188225439076, + 0.9129019866037823, + 0.8921177474168095, + 0.6654312904322366, + 0.6783197597391482, + 0.6945069169219692, + 0.8094290083930834, + 0.9056849589290569, + 0.7510740908003816, + 0.9237853020344613, + 0.9286896191057873, + 0.6446751265607479, + 0.3664836071788958, + 0.558623661447953, + 0.9074232742722483, + 0.5017608835979086, + 0.6615918990310257, + 0.8948887086532777, + 0.8417931066488052, + 0.6870701641462847, + 0.874928388381081, + 0.9032260825120204, + 0.8985335884699522, + 0.8893297686859183, + 0.8728216749411215, + 0.9006310512448613, + 0.8692748250505312, + 0.9392904905394207, + 0.7934838412386872, + 0.8442659858890585, + 0.9161393442246606, + 0.6079636878102553, + 0.6163053368117404, + 0.6164431583112204, + 0.5866991905597924, + 0.9161587037400787, + 0.8152033665518632, + 0.8573997146585113, + 0.8131783660065705, + 0.7810866292463798 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5456893620345681, + "valid_tracks": 12, + "per_track_pearson": [ + 0.5165329319170904, + 0.5628705047693143, + 0.5330290554995643, + 0.5562716649271179, + 0.5528845528660249, + 0.5802728980504692, + 0.5030319691372853, + 0.5571799685494934, + 0.5101955399295778, + 0.5507439807983417, + 0.5254329114611395, + 0.5998263665093976 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.3321729198998934, + "valid_tracks": 546, + "per_track_pearson": [ + 0.3399461154187062, + 0.3662935815490913, + 0.29884378524561717, + 0.19825032948228405, + 0.11726785866797752, + 0.31674052130601643, + 0.21747539419229198, + 0.21306384681151028, + 0.3351978927684121, + 0.3214074386970494, + 0.3595921414711201, + 0.21570773825384515, + 0.21970340374036718, + 0.17915615524326187, + 0.18691403993449332, + 0.23218352389044736, + 0.2855981347520439, + 0.38831579661934, + 0.3655189241739984, + 0.3695673966059422, + 0.297050597657248, + 0.30030952865986543, + 0.4414827164222243, + 0.25002341450354315, + 0.303056279169438, + 0.3533707774482419, + 0.3232949595930284, + 0.360873535009134, + 0.3762898724896822, + 0.3854945990375278, + 0.24157903850018472, + 0.39833851084990984, + 0.2535423325256741, + 0.22340391927163486, + 0.40647163165802247, + 0.3471543622075453, + 0.36716081820407204, + 0.43077256213284715, + 0.3677722481986355, + 0.4062087958104609, + 0.45052530381054345, + 0.3797912956172182, + 0.35496783095762924, + 0.30291880768786267, + 0.21627434119834077, + 0.2876653189301527, + 0.31131244028724964, + 0.3503008077324379, + 0.19047086254962273, + 0.2420173238835883, + 0.3843509762252701, + 0.3856936588844381, + 0.38754001980926284, + 0.2795914708820876, + 0.44229453816685954, + 0.43593918680549726, + 0.2635093814662868, + 0.4442877632533057, + 0.31407250011114873, + 0.4205360051307567, + 0.41107726099921293, + 0.2760068333109572, + 0.35586395252524755, + 0.29765104353566946, + 0.3728028080475872, + 0.42684182801629295, + 0.4254359350475103, + 0.25253848058973144, + 0.2034603935740102, + 0.17614602759172335, + 0.33208996338028746, + 0.3867204783479795, + 0.350236796504569, + 0.3308134046402251, + 0.37681864146076177, + 0.3094613332527258, + 0.35965872378418284, + 0.347126986852348, + 0.4078395586151209, + 0.3922102281594686, + 0.34606452972222573, + 0.2023981628488306, + 0.3558764654912561, + 0.4467530710122345, + 0.2026329675980975, + 0.3587552754033206, + 0.3261544807189043, + 0.4376492309461503, + 0.2931702008813515, + 0.3786310228725676, + 0.3042388981724857, + 0.35386838978479757, + 0.3638518800850647, + 0.1849197327420256, + 0.3874509873330154, + 0.3233069712749726, + 0.3695909038801176, + 0.3658390232856435, + 0.3379396958230054, + 0.21954366763177613, + 0.35000729766874417, + 0.3609861423179412, + 0.38575033488170235, + 0.44340274059147944, + 0.40955642799745456, + 0.3927363645418409, + 0.2890060063272086, + 0.3937883283584779, + 0.30597428299562407, + 0.3097944630864265, + 0.3244816680446596, + 0.3281356462075792, + 0.3528514207499914, + 0.349396912739392, + 0.290380079952459, + 0.3687038234785256, + 0.34778829098472946, + 0.35846028516608597, + 0.34531396766860517, + 0.38140167117100027, + 0.39378581052602357, + 0.37135148761135534, + 0.2681137872824555, + 0.31250445982003006, + 0.3324965762261284, + 0.37866227780042144, + 0.35495431391714777, + 0.3869480778729229, + 0.4532952817683232, + 0.3898805122136137, + 0.39888658149010525, + 0.3078289781895221, + 0.40045012196498686, + 0.4271421150023691, + 0.448743685563046, + 0.3408090292782027, + 0.43184956435500066, + 0.34279908678385257, + 0.18763954865428856, + 0.39223993673158514, + 0.4077640553921461, + 0.3511814134207901, + 0.39550388511375817, + 0.30966311987753414, + 0.3301454865432697, + 0.35282278794837063, + 0.3450246040035794, + 0.4610510165514645, + 0.33199992471756584, + 0.32589192823081947, + 0.413163472247948, + 0.3125410288347815, + 0.37122910789876834, + 0.3739761960052485, + 0.3467514621346399, + 0.3298688625598404, + 0.43400131329127317, + 0.4602439339331147, + 0.32117373564464646, + 0.2990682889391001, + 0.2561092582352829, + 0.32694996904573453, + 0.32986226909815264, + 0.3692820790734387, + 0.3262486505919913, + 0.30265805438337123, + 0.34703119640580476, + 0.2681751135783402, + 0.21611490961993246, + 0.38048908425467143, + 0.3480582292297951, + 0.2946632108485206, + 0.34917083942640154, + 0.45221403586884396, + 0.3587824192354044, + 0.2204740872259359, + 0.2418535321921416, + 0.31521331318366463, + 0.3981370378689418, + 0.4059934687523473, + 0.3929647532666466, + 0.40967015600692774, + 0.2512075945763124, + 0.3919143361360167, + 0.3426380299057948, + 0.27917556790863557, + 0.33305933665467385, + 0.24036351679983492, + 0.44877221124980043, + 0.3767420853519833, + 0.3082348288692841, + 0.23640078795162248, + 0.2304290546570552, + 0.33943961943388334, + 0.351817100916599, + 0.297313632742367, + 0.3460227818563455, + 0.3637140625063568, + 0.2690030941221967, + 0.34413292258534545, + 0.17953980393064695, + 0.18140786902336767, + 0.16992917102264185, + 0.16127283746794635, + 0.22347083614170507, + 0.33909503303704003, + 0.3403683808756706, + 0.2915667882862228, + 0.16328773922920414, + 0.31281853457332076, + 0.40050798910685, + 0.4079716316978909, + 0.41796270361809007, + 0.3740413676938207, + 0.3654484998265498, + 0.29305546817769745, + 0.4056966946532182, + 0.4102133448659795, + 0.3688184617055518, + 0.40498300510521407, + 0.37098696312756824, + 0.2876623974616672, + 0.3797938767781724, + 0.36904274686600175, + 0.45059244177087554, + 0.4199118384030132, + 0.3554734028097261, + 0.3494465787016276, + 0.3837165701799881, + 0.43047883610935817, + 0.27997293894611774, + 0.29810101075582185, + 0.4233466563504676, + 0.33574096046804475, + 0.3111059822467487, + 0.32677967983533357, + 0.43910103486548, + 0.2870661997006651, + 0.3796032535946685, + 0.37076231332068094, + 0.1772042787842966, + 0.2797738771886263, + 0.23105859660399447, + 0.3788446619259041, + 0.32358138468257014, + 0.364940600920153, + 0.1629847890614974, + 0.40370973930562987, + 0.3608955575570936, + 0.3878434962611444, + 0.3873344423913084, + 0.2622699376812528, + 0.43478161199202386, + 0.19950960147644348, + 0.38510482939315155, + 0.30867163053214014, + 0.4402939473137136, + 0.3993957088443782, + 0.41919168520047867, + 0.4618831380753048, + 0.3372815970892719, + 0.3654808550963825, + 0.17534338031710642, + 0.3854006608145292, + 0.39912210397169895, + 0.28387892419761346, + 0.18348475204193593, + 0.18899243441200725, + 0.21927226972242286, + 0.3427650459411655, + 0.41837600259, + 0.21878994356982773, + 0.1680901247020731, + 0.3230064928560321, + 0.3705981110031295, + 0.33082341011037253, + 0.23197885057980716, + 0.11001250810704374, + 0.34828767879183375, + 0.23101493529557937, + 0.25229119392099975, + 0.345425029449382, + 0.3034382649116849, + 0.3719149898590435, + 0.2155996701460514, + 0.2348260354758261, + 0.21287296256565746, + 0.20370361417756147, + 0.2198582014393431, + 0.2803554570094809, + 0.3729043168443512, + 0.35863514364145904, + 0.3422966030067407, + 0.2835568731820201, + 0.292619812477169, + 0.4232307185803054, + 0.2296364636281134, + 0.26897253401820054, + 0.37801094368131744, + 0.31690344981508917, + 0.3522581104270195, + 0.35999291505699504, + 0.3654018753776476, + 0.23152505182841998, + 0.37332101132343026, + 0.24118734902621647, + 0.22237792372213167, + 0.3745479933924082, + 0.3600808320884278, + 0.406386291278187, + 0.4166989535597019, + 0.4349087955512887, + 0.39030798124264615, + 0.43839259198532277, + 0.3632289689156423, + 0.3486124937592527, + 0.3378610325721699, + 0.22875609323913118, + 0.3115529128117831, + 0.3502752012059727, + 0.3966129379714917, + 0.19749057450067592, + 0.25440868684667145, + 0.3394016454399528, + 0.37617436119238346, + 0.36001518144250794, + 0.270347993866913, + 0.42904218676673356, + 0.41102295474989303, + 0.2682098326505745, + 0.4339833469819343, + 0.29536733566435075, + 0.4172548863824033, + 0.3899968135211386, + 0.27010174155785727, + 0.3556285570012433, + 0.2977645818969772, + 0.3762079680407844, + 0.4137618375631272, + 0.40035656070177666, + 0.24974772852723998, + 0.204886070440869, + 0.17163811521705802, + 0.32453696475531735, + 0.3751165377025025, + 0.33619929449415553, + 0.31152184691272333, + 0.37234488060615173, + 0.29258039476844677, + 0.33380917526228293, + 0.3214904618565704, + 0.38303470018349006, + 0.3795755832117629, + 0.3351245365577782, + 0.2051033306809364, + 0.3524755486858867, + 0.4288934995628596, + 0.20968243914485257, + 0.3531937624924803, + 0.3519217357623334, + 0.42874472884116993, + 0.2690724595004033, + 0.3774488603099479, + 0.29300583925714324, + 0.34333008606653603, + 0.36304515438184154, + 0.18850614029453094, + 0.37620589167228713, + 0.3071907755505401, + 0.3369437964465604, + 0.36907057765094636, + 0.3227776949000684, + 0.21110606516490857, + 0.3456711307078514, + 0.3606943070785114, + 0.36782937110412117, + 0.4171272580943159, + 0.38966755454977214, + 0.39335975957483504, + 0.27171201979243953, + 0.35323202676658966, + 0.2851518178172045, + 0.29566289105362537, + 0.30545130542533294, + 0.31449944628449306, + 0.3487657589238648, + 0.34938362313813254, + 0.2708520886422824, + 0.3501295936613625, + 0.31833004265058756, + 0.33928324091286394, + 0.3224993415501448, + 0.3671096484750194, + 0.3649918673998824, + 0.3541421935915492, + 0.26690841926119985, + 0.3119021564993163, + 0.332058779290539, + 0.37767231720382377, + 0.33458076046528823, + 0.3855745881104028, + 0.4414798897941478, + 0.4081176522579043, + 0.37530188889863786, + 0.2825846952917093, + 0.35531396097796275, + 0.3864844016057192, + 0.42226665580506945, + 0.3362786254996387, + 0.4172370353344605, + 0.3312189941431185, + 0.21273467475512609, + 0.38131985757965975, + 0.3770636480746733, + 0.3314205483842659, + 0.39389790563828064, + 0.29301474786431864, + 0.3250249837590402, + 0.33766449274516225, + 0.3347124796873272, + 0.4388161450750578, + 0.35560675484932147, + 0.3159273911547446, + 0.4046893312961, + 0.30587590579327123, + 0.35117235203552843, + 0.37587075568742945, + 0.33411422854155504, + 0.3368325531927558, + 0.4115279503222357, + 0.4494175075570501, + 0.31437037253636035, + 0.32179727266635955, + 0.2664842095312009, + 0.3044830514989348, + 0.3321602957781904, + 0.379618010007656, + 0.35692666402207857, + 0.29853883036105017, + 0.3512258898451102, + 0.29664108864624433, + 0.19832909869899726, + 0.4165880163611426, + 0.34739711162585385, + 0.2943180251723927, + 0.3473085438534698, + 0.44612687466182527, + 0.34120119409541105, + 0.19363602452498171, + 0.24187988206635325, + 0.3181053510744167, + 0.3659722662580244, + 0.41177774505800074, + 0.38815488923324726, + 0.41199256209157514, + 0.25633727961468666, + 0.38773034568172526, + 0.3520610449148436, + 0.30847350215519037, + 0.33717153886282114, + 0.2537256678392448, + 0.43439214693692024, + 0.3744163765931213, + 0.31390322851197205, + 0.26252110221764746, + 0.2354501362229873, + 0.29090822610256956, + 0.36969987844173713, + 0.31944136376918764, + 0.35904671918817405, + 0.3620025941921331, + 0.278614128089441, + 0.3389305116758315, + 0.17176860553426657, + 0.14817236594276498, + 0.13826194924155746, + 0.13137135645022205, + 0.19380231693297695, + 0.35299318454761275, + 0.31615616926244744, + 0.28885663942812817, + 0.1511833414025921, + 0.32685939128066266, + 0.4181425667623562, + 0.4384314973781357, + 0.4511465966317709, + 0.4010420339891663, + 0.3957509344766095, + 0.30995443904819786, + 0.4255679469036966, + 0.4191979626521193, + 0.371838348268744, + 0.41363483634931325, + 0.3853782318290764, + 0.31363390846490113, + 0.413806633128116, + 0.3580927869793003, + 0.492937229949474, + 0.4246197351149134, + 0.38983067537359156, + 0.3565966729701443, + 0.365968311265951, + 0.39189279013018663, + 0.2330019154246853, + 0.23496534762590016, + 0.435990474709011, + 0.34252194273262765, + 0.32235903142197764, + 0.32363148825274707, + 0.43542866105413186, + 0.3079154798570214, + 0.35577300210033613, + 0.3490851794305536, + 0.15775832001676096, + 0.27269716197366006, + 0.21565329720023205, + 0.39588986263450204, + 0.2951734365360623, + 0.3741281104313618, + 0.16527354349932125, + 0.40028210508396084, + 0.3744029001618471, + 0.3901247174700928, + 0.3946232360007794, + 0.282929337502048, + 0.48125886755705855, + 0.19066752929087327, + 0.3741283578056491, + 0.26608580882546573, + 0.4324589446648046, + 0.42069165834752464, + 0.45772989724125235, + 0.45416137962575487, + 0.3579395851251276, + 0.372224185853941, + 0.15530182389784916, + 0.36774973706433, + 0.38344925254516404, + 0.30913401936182283, + 0.16418479644633324, + 0.16634115022886364, + 0.19754103637318735, + 0.3515938144546474, + 0.4159928158647926, + 0.18805763617226018, + 0.14176699076149424 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.6252383460321363, + "valid_tracks": 613, + "per_track_pearson": [ + 0.6141401210466474, + 0.6900422526971038, + 0.5968095038054948, + 0.6472081948677849, + 0.6754035638525588, + 0.4716934070802244, + 0.6045091084868711, + 0.6744689815964408, + 0.6612545272778392, + 0.6873833040718348, + 0.6223042981810479, + 0.6038805103192373, + 0.518460552342386, + 0.6792439822526951, + 0.6483134419397143, + 0.6624406498866235, + 0.7077964183449027, + 0.6126777423466434, + 0.6187355067320184, + 0.6021778232950856, + 0.5967264141503218, + 0.6661445998287058, + 0.6231979424561405, + 0.5997919020170018, + 0.4903062588214625, + 0.6842312574459215, + 0.5538315561879779, + 0.6253584460706059, + 0.6018675350825571, + 0.6193872352780188, + 0.6341461602205781, + 0.659306942561178, + 0.6559437780700297, + 0.5533156123727679, + 0.6445270767069359, + 0.5101465710510213, + 0.635529273211051, + 0.6377813772515033, + 0.5831535266716603, + 0.6285419898973105, + 0.6868790548197461, + 0.5787790593489774, + 0.670225855294101, + 0.6872889067531408, + 0.6194013004621232, + 0.5306818856235508, + 0.588761377967426, + 0.6609390935345217, + 0.6340222306291293, + 0.6583299325333803, + 0.6493008025093093, + 0.6688650281221993, + 0.5668426587783909, + 0.6206045100996581, + 0.48799050366922303, + 0.615879106586778, + 0.6062777230185611, + 0.6745613884971756, + 0.7092288258415986, + 0.7235526695279539, + 0.7123160992207157, + 0.6466450547730707, + 0.672098833577512, + 0.5909296989162183, + 0.6622594520240129, + 0.6566603804468873, + 0.6825567953807736, + 0.6863888322546001, + 0.6645892573341184, + 0.7101756350808248, + 0.6739242225758342, + 0.6757776429922674, + 0.6393659559206775, + 0.6592783564765514, + 0.6511359660559753, + 0.6440079885015357, + 0.6733659264696668, + 0.6818267740439274, + 0.6542908839900966, + 0.7083280165506491, + 0.6614591031076594, + 0.589913566236559, + 0.7272864050082632, + 0.7001427175119813, + 0.612711700768664, + 0.6477421732864682, + 0.703060784853374, + 0.631111377756263, + 0.6793791919751498, + 0.6445903327188285, + 0.5473377221617107, + 0.6852001482804656, + 0.7014794256022295, + 0.6819142992987708, + 0.6848430230502117, + 0.6929215344691546, + 0.7109247174976102, + 0.6925595866703843, + 0.647355466510064, + 0.5617976289925485, + 0.5442844640313033, + 0.5638520011357844, + 0.6085670538038829, + 0.6641943754384442, + 0.5904190857621727, + 0.6307576320141821, + 0.6771186542201163, + 0.6735850826159319, + 0.6775728891105788, + 0.6249121505073709, + 0.6540614796793628, + 0.663887094926505, + 0.661217823378522, + 0.6207550053456948, + 0.6489238174713222, + 0.6802600146396459, + 0.6827512088959204, + 0.6661895760490905, + 0.620224206132776, + 0.6782895543123598, + 0.686546208882453, + 0.594633708157416, + 0.601797837007129, + 0.6238941963726583, + 0.705052133332656, + 0.5389696172091671, + 0.5563567286521977, + 0.5460789998563664, + 0.5594656758514188, + 0.5918752268314394, + 0.6830756422188571, + 0.6173702715143143, + 0.6440079130151087, + 0.6004374295002531, + 0.641235082676102, + 0.6658176456054544, + 0.6545973744470398, + 0.6779208969499673, + 0.6280995659746293, + 0.6738311251424629, + 0.6015048269073915, + 0.6078215485224882, + 0.6946790058028203, + 0.6431483515911761, + 0.563934546874846, + 0.580697922476159, + 0.5812180578865888, + 0.5979418109631884, + 0.5579894437521039, + 0.6262741645469617, + 0.5772071786032212, + 0.6138776312446433, + 0.6453432232396057, + 0.5478571966957373, + 0.6142179905544949, + 0.592642087613058, + 0.629403360247734, + 0.6236689914893799, + 0.6158761551245521, + 0.6673808639586524, + 0.6790628481448867, + 0.6737621484702923, + 0.6578960975770747, + 0.5697698529797245, + 0.566013076784615, + 0.5700234724890337, + 0.5729569778212525, + 0.6023990503935067, + 0.6176426237792892, + 0.6181832852368397, + 0.7064003511662774, + 0.5955134428835446, + 0.714532593700939, + 0.6226821846494325, + 0.5711100445113209, + 0.5989015412869791, + 0.5810989525057262, + 0.6600950651206346, + 0.5455226407945502, + 0.5170377796377109, + 0.5191542539339684, + 0.6412214945750051, + 0.567363803358834, + 0.6555653487204398, + 0.5313934457392199, + 0.6168073801308878, + 0.6288786358940237, + 0.4358177638406754, + 0.5603319318166753, + 0.6000723537134094, + 0.6900502476650028, + 0.610330262773124, + 0.612612823668678, + 0.5734595962904155, + 0.6522342714036332, + 0.6066565975423911, + 0.6268710813132293, + 0.40971137925154777, + 0.6079492087812326, + 0.6197768473161058, + 0.6177150505685367, + 0.6798967059814159, + 0.6820141718786257, + 0.6652079381519589, + 0.6455463847465189, + 0.6494277023116329, + 0.6577835122241419, + 0.38036317181290596, + 0.6222398149372627, + 0.6255948419101427, + 0.619308019252175, + 0.5981708505150999, + 0.659115224863428, + 0.6679558256972864, + 0.5657811532987237, + 0.6496410624297994, + 0.63677749323831, + 0.5781440128132169, + 0.6923041397665417, + 0.6995763190716972, + 0.679618078019965, + 0.7326884504129082, + 0.6412533374490856, + 0.6412937095237604, + 0.6319418246794846, + 0.6242201419634874, + 0.6376931225157034, + 0.6265401048993159, + 0.5395838452130008, + 0.700126377350106, + 0.5991985917428921, + 0.6572745773924421, + 0.6459569018526067, + 0.545572201784774, + 0.7088813171399263, + 0.5079195342375593, + 0.4863514529012465, + 0.655833572102997, + 0.681306544324906, + 0.6519971012717681, + 0.6606922474139849, + 0.6732486523738663, + 0.6652867741360836, + 0.7151277260089118, + 0.33831105258180955, + 0.6181236229761908, + 0.6406499849157965, + 0.6527233802410118, + 0.5904661384687103, + 0.5920621363609873, + 0.5688001944747193, + 0.5991364299990314, + 0.5722991427170057, + 0.5798019563453122, + 0.6582499998967232, + 0.6280498558608366, + 0.5565131059834, + 0.6737843377696973, + 0.6241530616324832, + 0.653536453418017, + 0.5939898447952321, + 0.5903831690286787, + 0.6994249275001388, + 0.7048524025117314, + 0.651970292488144, + 0.6994850194232328, + 0.6898372946482388, + 0.5512493383562432, + 0.6290755927817003, + 0.630746187087148, + 0.5021210514005289, + 0.6129594316545977, + 0.6927543054623674, + 0.7013321241245937, + 0.69995488977978, + 0.687558962737684, + 0.44843307799523474, + 0.571077147566656, + 0.6748257382998121, + 0.6421020998691498, + 0.7015632419429098, + 0.6117396972456024, + 0.5734920146668055, + 0.5331480628291502, + 0.6486403374089096, + 0.6200508164898019, + 0.6353752923839587, + 0.6920664909670575, + 0.6781635843870693, + 0.7073164953965669, + 0.6102154229523097, + 0.6222262680983354, + 0.6764834563755536, + 0.6182719793688904, + 0.5619564318896422, + 0.48554867895739895, + 0.6635915955490856, + 0.7123495254890727, + 0.7051264848140192, + 0.6323764378198558, + 0.6453740187116737, + 0.5994928514480963, + 0.6718520830277324, + 0.5976901788044651, + 0.6122942196515184, + 0.6850879471602336, + 0.6333836898784955, + 0.6959113036514449, + 0.7010547394149562, + 0.6506617344426268, + 0.6789089822440175, + 0.678852070389125, + 0.6406284074781513, + 0.6447399596157235, + 0.6143723823674997, + 0.5756718846793923, + 0.5937781654401519, + 0.6136149378207392, + 0.6751919424447659, + 0.6266252760591517, + 0.6719727831327932, + 0.7048565175936213, + 0.6890838000090448, + 0.5291198876150799, + 0.6018953771506352, + 0.4619674821373559, + 0.6158793774267349, + 0.6113012508907001, + 0.6410899879633676, + 0.6855259288627865, + 0.6617399446876775, + 0.6890855707376977, + 0.6880199998773954, + 0.7039804772229852, + 0.5749784713693313, + 0.6874810589810957, + 0.6793384164523258, + 0.7273401980277372, + 0.7223469011985897, + 0.7037390745984695, + 0.7226917148465486, + 0.6849266961901538, + 0.6829637687091482, + 0.6237513237780726, + 0.7108686763698256, + 0.6849862957982887, + 0.6654534366428233, + 0.6911326237854976, + 0.7303122718482766, + 0.5831533184218243, + 0.6633434059605866, + 0.6334428849814487, + 0.5920900761620552, + 0.7266177338070252, + 0.7069754961089885, + 0.6360817818215403, + 0.6546427497044455, + 0.648095388541162, + 0.6216187794144513, + 0.6925848996989564, + 0.5965897652235165, + 0.5840493710922351, + 0.6689958802369018, + 0.6475038187418605, + 0.6510543478723354, + 0.6200296312957131, + 0.640123840959557, + 0.659593762415433, + 0.6462920224200221, + 0.612731244771515, + 0.5590878658343192, + 0.49654998769737435, + 0.5173614648168833, + 0.5366821812562733, + 0.7001227104503395, + 0.5757578234502547, + 0.58956587320525, + 0.7254370501136216, + 0.6428770097991616, + 0.6694985827158231, + 0.6001142829989025, + 0.6439945357202862, + 0.6144630900010748, + 0.6161785343687686, + 0.6319515336680354, + 0.7398502245439788, + 0.6837859732217324, + 0.6870699963330584, + 0.6293504250188623, + 0.5868193042274589, + 0.6388291778924119, + 0.7130050548903787, + 0.570732418813114, + 0.5880094329515932, + 0.6001980222788315, + 0.7061898034704369, + 0.5228864099533985, + 0.5388331972393058, + 0.5205233343576765, + 0.5171681220395122, + 0.6080148618103212, + 0.7074643897262094, + 0.5792064376134756, + 0.6265058606903966, + 0.6988802217378298, + 0.6615760652038756, + 0.687070044282369, + 0.6291366500657194, + 0.6922827219401989, + 0.6211672637717823, + 0.6671479467801444, + 0.5688162914630711, + 0.6309011926957744, + 0.6845212359880828, + 0.6251515322824118, + 0.5503576279062053, + 0.5533819742358506, + 0.614056616783153, + 0.6388390683958527, + 0.6600618314405027, + 0.5834650159923328, + 0.5659730412973998, + 0.6277898280304706, + 0.6990259446123858, + 0.49062826176711194, + 0.5775783375629682, + 0.5992987566927481, + 0.6880634928445892, + 0.594367471070334, + 0.6715816414926818, + 0.6185915989867437, + 0.6724893598552707, + 0.6633540020369928, + 0.6726841673950638, + 0.4971952848377851, + 0.4974302964712736, + 0.4997336908454851, + 0.49952010256333224, + 0.5851063859295667, + 0.6180011102236256, + 0.6259722369377905, + 0.7595019191949798, + 0.6135785140595534, + 0.7376562615953153, + 0.6129614919370296, + 0.5837570600062729, + 0.5870273707941583, + 0.5819252663713628, + 0.6944915022265786, + 0.5007737563201403, + 0.5213512699814237, + 0.49176610139174903, + 0.6658576904052016, + 0.5408062041909447, + 0.6170093678762992, + 0.5252075263757681, + 0.618806475643726, + 0.6617435779271119, + 0.4659548769621979, + 0.5720626557324789, + 0.5493264997575799, + 0.6560719422158368, + 0.6498181108092159, + 0.5657390041326449, + 0.5450627836249984, + 0.6741813170296805, + 0.6347156286269734, + 0.631742287202057, + 0.39340869000338013, + 0.6689384448463911, + 0.6640973397018584, + 0.5891495238481478, + 0.7489003802417193, + 0.6603739601411532, + 0.6704573464735473, + 0.6748079682057508, + 0.657435635250956, + 0.6580735338729249, + 0.31485539857122885, + 0.6624827718215935, + 0.6128438810906383, + 0.729365877733194, + 0.7021527626083317, + 0.7783445230213973, + 0.7706054438716935, + 0.608249383037824, + 0.6954453980150934, + 0.7522152489840532, + 0.626277192126862, + 0.7925427410304928, + 0.669963170267481, + 0.6154312009044302, + 0.7151411621257896, + 0.5999594279531781, + 0.5909692147941785, + 0.5885191963405035, + 0.5489156745773008, + 0.5508337914580013, + 0.5675303957307801, + 0.4848570752814633, + 0.7368607279558583, + 0.5310192110849729, + 0.6368594830107305, + 0.7279049033379305, + 0.5412143908743137, + 0.7286969858427225, + 0.4796235553167318, + 0.48631460394163367, + 0.6666212701668469, + 0.6570841127396994, + 0.6056097873321744, + 0.6179707024847615, + 0.7049693197707817, + 0.7237413860217152, + 0.7985710480927253, + 0.3489272912713409, + 0.6237688688317787, + 0.6357560387675274, + 0.6693706020108401, + 0.6568199085766396, + 0.5917646174293714, + 0.5810587495654392, + 0.602436328583481, + 0.6077094851048553, + 0.6004396751145683, + 0.6635171742428874, + 0.5912323303310596, + 0.5152149093311603, + 0.681246705274761, + 0.6344526964745744, + 0.7251240593953547, + 0.5367379743088804, + 0.49344150698804007, + 0.6531364609922311, + 0.663087349055801, + 0.6456836490492722, + 0.7384728375917191, + 0.7307438381780137, + 0.5048861113410099, + 0.648689582970148, + 0.6471632909137465, + 0.4812864447559763, + 0.6717570857720956, + 0.3827799583975387, + 0.5545665047898872, + 0.6169194023207353, + 0.5626922758336782, + 0.6298764357958991, + 0.5944439045108607, + 0.6448825135502859, + 0.6591660545180282, + 0.5708220043412708, + 0.5318993677662608, + 0.5368902076154427, + 0.5341238378601438, + 0.5242317996876745, + 0.5248634353209507, + 0.5355930868420347, + 0.5308985072029776, + 0.5041566714612694, + 0.4813242945757439, + 0.634574433451911, + 0.6099966368261108, + 0.6142359540222226, + 0.6136108763439666, + 0.626236497033806, + 0.6165927048849111, + null, + 0.6296655836380222, + 0.6786090722452871, + null, + 0.5242197016871415, + 0.673733321870681, + 0.5210863442923405, + 0.6561618086129185, + 0.6720722055416395, + 0.6721687642446502, + 0.6683689194636846, + 0.6641282310933363, + 0.6633145434216855, + 0.6502820187688158, + 0.6125516878289398, + 0.6010815368647312, + 0.5498194040286798, + 0.6452334858475978, + 0.5178159761739394, + 0.6417650411023451, + 0.6210026915062253, + null, + 0.6313776752365027, + null, + null, + 0.624323855723473, + null, + 0.6384135679971165, + 0.6162597476762761, + null, + 0.6594684450458003, + null, + null, + null, + null, + null, + null, + null, + 0.6400674648292304, + null, + null, + null, + null, + 0.5549209478935677, + null, + 0.633840958540276, + null, + 0.6271812367832673, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6585727237672073, + 0.4904185337664877, + 0.6148301985111625, + 0.6436117130498012, + 0.564070209821972, + 0.6676387827831116, + null, + 0.6530040791576596, + null, + 0.6412272730374218, + null, + null, + 0.659575579209427, + 0.675078714581031, + 0.597646712109856, + 0.5836394670330376, + null, + null, + 0.6081927826681119, + null, + null, + 0.6388057389036781, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6461750443871503, + 0.6315993354378848, + 0.6490747822528834, + 0.6153528433932863, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4706757407826344, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.2520491000028878, + 0.38816569197011397, + 0.3253230645913208, + 0.49421592039229323, + 0.4386696908590364, + 0.3272144350983886, + 0.42948889496032266, + 0.5022157105956215, + 0.4169348539986208, + 0.434036940513175, + 0.4143536076463046, + 0.43061402297454915, + 0.3042680322008531, + 0.5065691164364543, + 0.4235328882414876, + 0.43075583815144614, + 0.49871345761271413, + 0.39665659770931166, + 0.5290378936052044, + 0.4270785550360502, + 0.42324016526423003, + 0.36568542195576875, + 0.33916481873567395, + 0.5505711870740047, + 0.3951699733265064, + 0.44764028348992513, + 0.3688957891706988, + 0.3971038404533408, + 0.4355681933462405, + 0.4541588131421155, + 0.4608585127770516, + 0.43965388663785654, + 0.5095042195640155, + 0.30680652373646455, + 0.6815726748963146, + 0.29846861323729684, + 0.26722454033492216, + 0.19715296076906555, + 0.3523715642090409, + 0.2416887855057758, + 0.2965453444105483, + 0.4816352403581345, + 0.32796274724924596, + 0.4290265898462881, + 0.315439114964985, + 0.41635173360995165, + 0.3199840816458938, + 0.2404373316388146, + 0.48773770239360803, + 0.4605738433454729, + 0.4062377988667265, + 0.3799824452051575, + 0.40000272698808836, + 0.4397604113244686, + 0.43083331237931327, + 0.4507755176975788, + 0.3733995605687173, + 0.5329980245797048, + 0.5488119055866305, + 0.5494165088188606, + 0.34274250411436047, + 0.2227772970526926, + 0.4158605292875298, + 0.4431200457431709, + 0.29152555449738715, + 0.3737708116936225, + 0.3677661279523206, + 0.4942417679509094, + 0.5129699481083879, + 0.4603230728182746, + 0.38307229945095755, + 0.7243574593115089, + 0.3758878142863948, + 0.4048020141277823, + 0.40551547655070985, + 0.4480189017146753, + 0.3312542093725064, + 0.6731083675150759, + 0.4342782509042554, + 0.46031658127821184, + 0.41538765975011, + 0.43440446500352253, + 0.5006149994524423, + 0.4653645630067004, + 0.38798773131410447, + 0.3381679440341264, + 0.46809954322709885, + 0.48801048350153875, + 0.8092122132645412, + 0.4982887863797126, + 0.4121634871472176, + 0.3613788953678276, + 0.48485812200389533, + 0.4076824524824917, + 0.44887369682481953, + 0.5674444449005837, + 0.3558403345804424, + 0.40980696772208225, + 0.4179862631509346, + 0.35128959832067136, + 0.2618531281905254, + 0.3935850965889992, + 0.35172638764303155, + 0.3362369373878232, + 0.383678972193054, + 0.27615711123112074, + 0.3454693868437632, + 0.3707292493906803, + 0.39127913105989687, + 0.45382343038296813, + 0.277021288782223, + 0.3899186047890437, + 0.36210169773498424, + 0.41620491686812683, + 0.4624174751285416, + 0.45703063139609074, + 0.43043505252825, + 0.4413068408944699, + 0.41739983949980486, + 0.47202296684563355, + 0.4232309340718763, + 0.45672190252627043, + 0.4406735248822512, + 0.3616936723335499, + 0.41798232127709345, + 0.37816270179785383, + 0.41140121693522774, + 0.3856909749190341, + 0.41609304362501537, + 0.5479328277610526, + 0.46331803129160276, + 0.3885918713254055, + 0.45273152149203666, + 0.6501526435894504, + 0.3864692218619694, + 0.4482650560302946, + 0.45210201214265344, + 0.3379185312006884, + 0.4028668487139604, + 0.38252706948831355, + 0.41422961829385974, + 0.45786940424853556, + 0.5799871806326578, + 0.393880052634779, + 0.4600456387341839, + 0.5254876117241033, + 0.47701472683643636, + 0.3420830747455901, + 0.45865681732174857, + 0.4852337504794319, + 0.46908500334923126, + 0.3941276738281103, + 0.5005382857044345, + 0.5132457232858056, + 0.4534923367310135, + 0.45148787128488543, + 0.3668729737109226, + 0.4185115466791558, + 0.4210066964643764, + 0.47308255688491124, + 0.47711664023623707, + 0.4070291125349318, + 0.4377638461951634, + 0.45103621350723205, + 0.47115919048786875, + 0.41925716218505643, + 0.4134651684536786, + 0.49935012609257773, + 0.3472667301830813, + 0.4701271419052787, + 0.4454324488757026, + 0.3907258661612914, + 0.5106354599600575, + 0.4037835671655529, + 0.3651746475183979, + 0.45959369183323123, + 0.332564927841183, + 0.4848584761797479, + 0.49684421381483895, + 0.3763143006325939, + 0.47798539780035576, + 0.463960015702508, + 0.3837380477148264, + 0.4572630051999995, + 0.38982765476573883, + 0.4753892767287984, + 0.47928422978871665, + 0.34742682564670535, + 0.3748042467480817, + 0.4534219561530579, + 0.38790068958394425, + 0.41348728661957085, + 0.4685002727631702, + 0.5154107443646263, + 0.40218676171681383, + 0.3903283625181954, + 0.4011258351503712, + 0.39869983645298773, + 0.3108599622462036, + 0.5068696241932382, + 0.41362095265086724, + 0.38985561298025745, + 0.43210863571446734, + 0.3776609956131816, + 0.48700472915780374, + 0.42420255046935174, + 0.49486837259191535, + 0.4148212242540697, + 0.36820414764493153, + 0.5385983596845952, + 0.41382472013005767, + 0.4817019194990419, + 0.4484913201245712, + 0.4630730618820223, + 0.8637521204317854, + 0.5032076286504206, + 0.4236799237197798, + 0.46152847214503, + 0.47923980806187877, + 0.5591678460919083, + 0.6242895893397364, + 0.6389618938223474, + 0.4562146272508114, + 0.41282542056068994, + 0.5047663137979058, + 0.7411089503565689, + 0.38216551343498395, + 0.4094055070163429, + 0.5594206767237666, + 0.5348054541903396, + 0.5967673526364633, + 0.47390455609156723, + 0.3572236483395497, + 0.4278454385959478, + 0.6122717303214548, + 0.4718086607053905, + 0.3867143681581999, + 0.6237111117129129, + 0.5653428769951487, + 0.47025036313224006, + 0.5095491278450237, + 0.7192552574051535, + 0.5029221380866905, + 0.5312326269660329, + 0.46315620553784204, + 0.7689641913547962, + 0.3556646309423161, + 0.6496540879552827, + 0.4278957854693033, + 0.560352194731796, + 0.45825629606170676, + 0.399735101195259, + 0.7091365360158921, + 0.5293776932008769, + 0.3645095340256351, + 0.39305360926086935, + 0.6566043806305286, + 0.40989371983764394, + 0.45947351233736566, + 0.5102210175730132, + 0.4138872396993693, + 0.5079735510168011, + 0.5086592851055484, + 0.4479149121910301, + 0.40679788483173834, + 0.4870787132197713, + 0.3870492340573651, + 0.36881028842989466, + 0.6394004714137275, + 0.40284102886073647, + 0.4688893703121813, + 0.5852413177309871, + 0.41705257549007474, + 0.6674237635698914, + 0.38705491679804854, + 0.38853179295748613, + 0.5703963203239887, + 0.5102203774797605, + 0.5915763583002792, + 0.7360783471834043, + 0.6801203124000268, + 0.4334454749467087, + 0.48439075634452144, + 0.579178632143371, + 0.38100764024990685, + 0.3706305707927393, + 0.5773244545272357, + 0.5483030784056468, + 0.7452440702629167, + 0.505714171308271, + 0.3917586157592394, + 0.4815583071575267, + 0.394566553176098, + 0.48887529431950183, + 0.7142739141427908, + 0.6384824581404206, + 0.5555897697916459, + 0.5485001211916735, + 0.51425229604318, + 0.5279290669998289, + 0.7473609096612105, + 0.32836941409363574, + 0.46457243504272, + 0.35106585710383953, + 0.6212748089334287, + 0.40762548728700965, + 0.7430880971732094, + 0.4498500773301533, + 0.4530512658873135, + 0.7051950491429757, + 0.5314213920252067, + 0.36211568616315126, + 0.5793402997943214, + 0.4294315197147983, + 0.5223273800647096, + 0.5089436931751885, + 0.36640774097327294, + 0.5164161840513757, + 0.42445203182628527, + 0.4730122625526364, + 0.4349146051546802, + 0.37970531003896735, + 0.43928875122321753, + 0.5440318115485737, + 0.6016346612613158, + 0.6735310362810452, + 0.4485623668157171, + 0.33619110634493327, + 0.47952116367625497, + 0.4874369549889924, + 0.4462891904587682, + 0.4982638951951195, + 0.5062222728961293, + 0.4049315999419248, + 0.35589265878080484, + 0.5315613927413344, + 0.7699236921979237, + 0.5417161659547081, + 0.6168600641759723, + 0.6015130120984821, + 0.7750748334527899, + 0.5813264246059987, + 0.38257271192572545, + 0.6016345756498425, + 0.41240580584977965, + 0.5541804283363594, + 0.500255977982455, + 0.4133505311140226, + 0.5150681818240559, + 0.789864342853543, + 0.2594426901925617, + 0.4863662274804336, + 0.48972250430912, + 0.6349804463798051, + 0.39611001538818397, + 0.3423036709070228, + 0.45406732694536067, + 0.7293186243915708, + 0.4153311608316907, + 0.46378684958623145, + 0.5216015070698894, + 0.4748005267027878, + 0.44917831310753137, + 0.5184795735229609, + 0.6259559315761997, + 0.47846416341029274, + 0.43377325786714604, + 0.5672648906932721, + 0.4709620504465793, + 0.5841042925091133, + 0.5704642006122396, + 0.5463497119445918, + 0.5677519044900231, + 0.3641131615376556, + 0.3858098730041586, + 0.5810090729627571, + 0.34629120111495937, + 0.7044409814534112, + 0.34784946577361053, + 0.5768972219505772, + 0.6041780322101928, + 0.6947109057337103, + 0.3328058944458065, + 0.3982589087320228, + 0.3703554171159152, + 0.4197101838760995, + 0.3500584213357177, + 0.3337733177929682, + 0.6330548682030966, + 0.41265131465632837, + 0.6889392823707808, + 0.5960087508029237, + 0.4769126636421285, + 0.603907246074376, + 0.3989282264130852, + 0.5643747273574985, + 0.5498251131694604, + 0.49073777177111017, + 0.4064758002370514, + 0.4824570287487474, + 0.5223227722294139, + 0.5467810072297019, + 0.41387643735986485, + 0.37928808516495915, + 0.6417816777822142, + 0.4941111058273183, + 0.6659267734611948, + 0.6579072532271227, + 0.5375425422606787, + 0.3589566545160657, + 0.39210526945789675, + 0.39076352097619893, + 0.5244355141443529, + 0.7514811311322885, + 0.4876345248585844, + 0.40177763193380367, + 0.7492305426589099, + 0.46627144968962997, + 0.6340731695546062, + 0.5456167212465257, + 0.45503638435394395, + 0.5119926096737613, + 0.4760906449944476, + 0.4898017903878557, + 0.5872993892431372, + 0.5577422605608912, + 0.5463204789157251, + 0.4678044054607059, + 0.4876857461778285, + 0.5395439298618114, + 0.508031089548269, + 0.7725962751124642, + 0.5040157737369109, + 0.5053741774150964, + 0.4564559962081529, + 0.33010361997316007, + 0.6703254314744145, + 0.4325164570936449, + 0.37263660715693464, + 0.5325684034354798, + 0.47939978798064314, + 0.37188204422888915, + 0.49483326070931277, + 0.4583971099485452, + 0.7836787594263297, + 0.330140405644757, + 0.42053812833029164, + 0.3847786023568424, + 0.44661089488781547, + 0.41853014099704827, + 0.33902683905185443, + 0.4418405768655559, + 0.2991417034231582, + 0.4415010224787274, + 0.7479860591166899, + 0.35150782254820967, + 0.42331798317278096, + 0.5174746447222678, + 0.43602076253793715, + 0.4223454092141344, + 0.4938330163227774, + 0.3690408270818655, + 0.5358213721394381, + 0.5473493844610298, + 0.2807709989544317, + 0.6471074594307249, + 0.6984026366048264, + 0.5214789960479566, + 0.5004982600673401, + 0.5523560416574409, + 0.4460444546364964, + 0.4688512677773519, + 0.5744628984419565, + 0.7852229678435256, + 0.4212936819899944, + 0.6176999127259106, + 0.3720179690403288, + 0.47381634984086296, + 0.5020277573359385, + 0.5036822348499701, + 0.4179471904848583, + 0.36550998795253814, + 0.36333422252250686, + 0.4076401324838225, + 0.5878404253885963, + 0.4821685859668751, + 0.35579060156892606, + 0.42211187787491844, + 0.4764270111942688, + 0.32912510408492057, + 0.560827647883019, + 0.5168259222102805, + 0.4429622106135618, + 0.46738940011762536, + 0.35010835497000836, + 0.44384873591320495, + 0.5221754111127629, + 0.5973343673078689, + 0.4417111269442983, + 0.7857407469103319, + 0.3882487115662485, + 0.5660490007350324, + 0.7451326361819478, + 0.6419845735818148, + 0.33312453012344956, + 0.5208772642973656, + 0.43205567426827496, + 0.37711798922217765, + 0.3970044523073546, + 0.4098309074676467, + 0.48399378455312075, + 0.5087640145901958, + 0.3929770748068246, + 0.6258938582958136, + 0.7720371200124236, + 0.4494840585612193, + 0.46563253562435625, + 0.4540089699514879, + 0.5194947331662807, + 0.5346251126882081, + 0.5011992582349419, + 0.5160516987426232, + 0.3828256035837497, + 0.6711166290035804, + 0.49898832270707805, + 0.5656946655609666, + 0.7214832299558969, + 0.34836814262578836, + 0.6019057291535879, + 0.40152934547618996, + 0.4763421327956817, + 0.3590987251657742, + 0.5302762473169349, + 0.5113806180453677, + 0.3571257306217288, + 0.4559323728063391, + 0.4199345349479957, + 0.5458816176349348, + 0.6046572752589119, + 0.47977751618631953, + 0.4798426853502276, + 0.42938198802670685, + 0.7474551247784881, + 0.5191137940260188, + 0.5493359708620563, + 0.4815674569823648, + 0.47774085439214004, + 0.6237507169144562, + 0.37140980494619225, + 0.5801211417389646, + 0.39770641225859565, + 0.5618206669161674, + 0.4431734593265296, + 0.37253275820164583, + 0.38679333785807235, + 0.4883564959956462, + 0.38999742172695145, + 0.520746909087011, + 0.7406163245287459, + 0.7314055856177352, + 0.42424564949496246, + 0.38327093694931375, + 0.7088617180392911, + 0.4622951886641312, + 0.40800195069157325, + 0.5392246125056298, + 0.3748978779656206, + 0.6082857069869213, + 0.44373057370151525, + 0.57864962884029, + 0.5975874856679558, + 0.47059497124366045, + 0.5260862419221765, + 0.663749365410453, + 0.5347932245705018, + 0.38690903115932235, + 0.44382865704102437, + 0.33202071368447605, + 0.5522008424279082, + 0.3673980798530507, + 0.5322090396723994, + 0.5143472462543764, + 0.41312299023390875, + 0.7278996458615016, + 0.3520903379640269, + 0.7331321837762209, + 0.4658267425701477, + 0.5519907718087096, + 0.749866687529965, + 0.4959514055612725, + 0.5633124669269897, + 0.4323001436597135, + 0.7106476537360833, + 0.43751774884479966, + 0.35843462786023167, + 0.5763498003514028, + 0.5360267076383478, + 0.6499023194368247, + 0.5608443441270158, + 0.4213750719883446, + 0.49844273345781154, + 0.3253359277625617, + 0.47987234306410387, + 0.5766995612549511, + 0.5279947322547097, + 0.37840810652621376, + 0.5563541008113857, + 0.5965552679801981, + 0.58077412856486, + 0.4558615886444014, + 0.4209801730706929, + 0.5206825705107092, + 0.4774150006687795, + 0.5678034878832399, + 0.505564416571034, + 0.3851811808529075, + 0.4279456140939848, + 0.5245743258816044, + 0.4182709159082142, + 0.7432333747484325, + 0.534184109521902, + 0.46903863243102467, + 0.5115483428940247, + 0.4207746359852511, + 0.4605706879849159, + 0.5511550877280816, + 0.563028082481986, + 0.7167183829169896, + 0.5547938151053883, + 0.5849797388011382, + 0.542781444546588, + 0.5113920902860384, + 0.5874107212535742, + 0.42347521386874987, + 0.6467915091878546, + 0.6052382737801367, + 0.7302219972921665, + 0.5349214023341679, + 0.5575657063932297, + 0.5508839123440457, + 0.4623420414271857, + 0.5470703510251324, + 0.3718398671281094, + 0.6903548935974656, + 0.5016527972792165, + 0.7423431948198389, + 0.7662746727904894, + 0.5057053773095416, + 0.4697134036172873, + 0.5144690027173291, + 0.3831298285957236, + 0.5183499325007189, + 0.7640936679427679, + 0.5014180599111523, + 0.5316643422846672, + 0.5962380483668046, + 0.5361331872482616, + 0.37272272362141645, + 0.6221904936923556, + 0.4496825137208455, + 0.4126672471940819, + 0.7395188343747924, + 0.4342238938354788, + 0.45514264171022967, + 0.4217591181008678, + 0.4982216357268146, + 0.5728193146007284, + 0.6405610747664527, + 0.6871067461922141, + 0.6318676243883479, + 0.5986549385316927, + 0.4777134190399516, + 0.6177851988125663, + 0.5742828037758754, + 0.7848283719588198, + 0.7486490017838654, + 0.39065346541393325, + 0.40474206161065795, + 0.3822304169278742, + 0.5347425017174151, + 0.43252237190015014, + 0.7932526000200586, + 0.3990427016463419, + 0.4381991593386083, + 0.5875612259562659, + 0.5010601803482634, + 0.5803730647930239, + 0.6356372102437279, + 0.4869222102273504, + 0.6257516470958594, + 0.40138509792439614, + 0.6953809499327366, + 0.7858735883127879, + 0.7700508515387816, + 0.5972962825648351, + 0.45226842429899244, + 0.5724668288298492, + 0.749053273474471, + 0.6996483980609137, + 0.6367749498351143, + 0.5075988842854551, + 0.48295100714996725, + 0.5079458785388101, + 0.7423253690574925, + 0.4450239073659084, + 0.45711002640747517, + 0.344608809249973, + 0.6121001590586646, + 0.34195541966576426, + 0.43894185691800436, + 0.6804734724949383, + 0.5525288801692975, + 0.6933421840588059, + 0.5618853460464565, + 0.5329739538662934, + 0.676645908800573, + 0.6440288795069653, + 0.5778530688191998, + 0.40318200116019576, + 0.65672450257356, + 0.5366456092587012, + 0.4253894914881417, + 0.3781371642459797, + 0.49937818444344223, + 0.6483454986236835, + 0.5339440296344469, + 0.625480846648562, + 0.5886217537071281, + 0.4442842742769498, + 0.4298922294017208, + 0.6976683965425542, + 0.5193091581872196, + 0.6236102738785629, + 0.7312967158533021, + 0.43355720882699605, + 0.40631246869557697, + 0.40385950075978816, + 0.40065736756616926, + 0.7212195747400023, + 0.7424579110308603, + 0.5753726361011319, + 0.4391131664187093, + 0.49090474098096876, + 0.4119505788392672, + 0.7873620724794407, + 0.3741650442110935, + 0.5029119905304978, + 0.3627530225735976, + 0.5275106128651385, + 0.39937193587887687, + 0.5790105780435205, + 0.5067295774179827, + 0.37722634632905894, + 0.7073086741071292, + 0.5265184100853781, + 0.445046265652426, + 0.3615072107175691, + 0.3773934387253055, + 0.42784241740979767, + 0.4397371405159887, + 0.39397909470706877, + 0.38821264931925303, + 0.4231192064671775, + 0.4152578982595308, + 0.4353220693260522, + 0.35835086952736167, + 0.46563148920905556, + 0.2266267753554952, + 0.3046906123745441, + 0.3798580866591179, + 0.39757453114932767, + 0.4657039954359528, + 0.46550866196675184, + 0.39250140568870917, + 0.4130431475032768, + 0.3801704192522696, + 0.6874403495504638, + 0.3781172947775399, + 0.34764612043875265, + 0.7906144318909364, + 0.37349239756800795, + 0.4588481137227188, + 0.3814310258557244, + 0.4549344478332571, + 0.3219903276859232, + 0.442614642897041, + 0.364609580674233, + 0.5050823888461791, + 0.28749807608815015, + 0.42111720332146974, + 0.407506390736814, + 0.5233053078575233, + 0.43269133300100243, + 0.47873914675191215, + 0.31560749602330307, + 0.4916699150664612, + 0.432697217054818, + 0.39114128749698, + 0.4794736219750762, + 0.418804187527746, + 0.4849158425641037, + 0.4296605504427065, + 0.40672617405525435, + 0.8113754703911307, + 0.3503184439396404, + 0.4939522223763716, + 0.37283723815108927, + 0.44076918066115395, + 0.35507689169527507, + 0.4222569755444703, + 0.46758101263103125, + 0.5719046404314398, + 0.3542932431450543, + 0.3991259621332223, + 0.32170221179138075, + 0.4172180661026688, + 0.481655250606546, + 0.46769970705969377, + 0.49322316106991293, + 0.38680427644875154, + 0.3814390568825303, + 0.4761702034792936, + 0.4457177675056065, + 0.45885641383764225, + 0.34186083891859487, + 0.3326729868026831, + 0.5020063558033291, + 0.46399020446420186, + 0.4343823040150568, + 0.4059450226926785, + 0.427894402439585, + 0.8374838295250772, + 0.3417083384729276, + 0.37999428803293134, + 0.3825504499691671, + 0.3578988860637675, + 0.400424271555245, + 0.39825616369942685, + 0.3520876779397201, + 0.37374276113083704, + 0.3646164524592394, + 0.48358702275041215, + 0.4594307355169649, + 0.46231168465217365, + 0.48645137131402993, + 0.48392724313025137, + 0.40559356009343484, + 0.5757064778842906, + 0.4519475856166693, + 0.4838816532318047, + 0.5127837705505706, + 0.47138437493217145, + 0.42924403227657304, + 0.3379382817695525, + 0.43151884958852216, + 0.5452520528214877, + 0.4982457099507726, + 0.36979922271894106, + 0.401579283644417, + 0.4350471327240074, + 0.48698143482877326, + 0.37801763792855103, + 0.4238263481615301, + 0.5094701072550867, + 0.5617413555506219, + 0.3721093196757044, + 0.44309863824760604, + 0.3764770852751564, + 0.3821009817823987, + 0.381057092864255, + 0.7968407358773397, + 0.42939800424359637, + 0.4623565255377182, + 0.3755692259001637, + 0.3969100467443113, + 0.4219654624998992, + 0.7119785507102357, + 0.46759706788389876, + 0.4450244053929145, + 0.4646461662586402, + 0.4379284183268948, + 0.473625834246771, + 0.47950724565856345, + 0.5121340825212611, + 0.31259431176739993, + 0.38814818701065473, + 0.4666023910225303, + 0.4693821768980878, + 0.43370688362561804, + 0.43236095894845533, + 0.4292967149153029, + 0.36907510603335686, + 0.4098186249862657, + 0.4234949295107856, + 0.4267884255351401, + 0.6215720633115704, + 0.398927517922936, + 0.43509605051820843, + 0.36063973658297177, + 0.5998815907883923, + 0.4197255331216393, + 0.3926024160437411, + 0.45506399377716233, + 0.4788666292961533, + 0.4624454779800537, + 0.3827096279505009, + 0.5690036372682598, + 0.49465578284071104, + 0.37037385890828484, + 0.4595565367072315, + 0.46651006551964314, + 0.4992436082804506, + 0.4165355921399595, + 0.4799410682284425, + 0.37169139880219654, + 0.6689264362932809, + 0.4101497242993985, + 0.4692855631728043, + 0.4885338358150404, + 0.4367142725532693, + 0.4346236157145366, + 0.4595306143132018, + 0.35556563660668794, + 0.41822247202507307, + 0.6118155198182751, + 0.408932007002191, + 0.39722178777373707, + 0.5319176279503453, + 0.4417035031078961, + 0.4293759297277008, + 0.6382710342395668, + 0.43328397489582277, + 0.5011301612384815, + 0.4508126625698669, + 0.4263872168593225, + 0.5157031566803825, + 0.4192630902586541, + 0.5091783374131077, + 0.43459517913389745, + 0.49802899795263617, + 0.3883189351767568, + 0.49260488372696837, + 0.44677828670068714, + 0.3718684674240985, + 0.3439708514642843, + 0.41256657921666356, + 0.4515592831660534, + 0.5669962935523638, + 0.4789749374164988, + 0.4665302235138025, + 0.5081351717672505, + 0.41523144012541024, + 0.4098436348194143, + 0.499265191055134, + 0.3861542207289573, + 0.42422739063097387, + 0.5069481206679566, + 0.4215961092979952, + 0.5209971190018993, + 0.5149837981418688, + 0.4217921012442951, + 0.424764682422409, + 0.5339808765585888, + 0.4385811172873231, + 0.45604990954453395, + 0.3597642035112748, + 0.5161432142014613, + 0.6139600831283406, + 0.7986359255940748, + 0.4474574189093317, + 0.4318843809292611, + 0.44538364184998264, + 0.40672063150215715, + 0.3807641154280957, + 0.3921105992759605, + 0.48204888390482525, + 0.475821919665777, + 0.43757083694287763, + 0.4640930287418411, + 0.5052106714888644, + 0.5245747215629034, + 0.3907321141399973, + 0.3796985008555166, + 0.48051686649389097, + 0.4311974546520361, + 0.40818273513341913, + 0.44151941164489444, + 0.4176268106935625, + 0.42570783559890857, + 0.33282441583665134, + 0.5172205590409111, + 0.4570192979242123, + 0.4155009785884131, + 0.4210125440407639, + 0.3331290431607098, + 0.47686281808428355, + 0.3174059774592971, + 0.551051662541792, + 0.42957331602380466, + 0.36450539010679717, + 0.4727462792881134, + 0.4415509207958645, + 0.371589377759023, + 0.6186511294405307, + 0.5361244377799983, + 0.3617659797853431, + 0.5034671241492106, + 0.4373415096869756, + 0.46027140961343344, + 0.47287332483333605, + 0.3227624865357566, + 0.41501638549681014, + 0.5240191960412786, + 0.39658092575062404, + 0.46211931435711423, + 0.4446393317627672, + 0.4779010896662001, + 0.4944790434375182, + 0.4184420406411585, + 0.519803428375473, + 0.48036830475107073, + 0.42252684154263404, + 0.3608672311260713, + 0.47595757646337306, + 0.3800015217013082, + 0.42231474653978673, + 0.4305203077783671, + 0.49272629733325124, + 0.5508090578265905, + 0.4786507212130463, + 0.4373547304420923, + 0.49282713153136476, + 0.4612035737558045, + 0.4531368313916399, + 0.535297215235679, + 0.41910976760309626, + 0.34558414483625793, + 0.5097613177134542, + 0.44427723364398547, + 0.469049783872731, + 0.45805031646410754, + 0.45852638968728004, + 0.5154330372221407, + 0.3865516542932446, + 0.4532629610807464, + 0.37147736752497595, + 0.3765542845137112, + 0.36804769673280047, + 0.4448944101990096, + 0.3827687935592605, + 0.4574829106552676, + 0.5387276090039689, + 0.46042042783201864, + 0.4975646108103075, + 0.7816640355282076, + 0.5714824773815254, + 0.4659607222223328, + 0.4293929276102417, + 0.4762716206765573, + 0.48438808655571236, + 0.3633893988430039, + 0.5864369182889105, + 0.4911821500671511, + 0.47810761107225863, + 0.4856319726971609, + 0.4203816426353087, + 0.5161890182957952, + 0.5255757049225673, + 0.43078986762486915, + 0.5529808346556051, + 0.5211649708387139, + 0.40603818202215497, + 0.4581976397193777, + 0.5340012778297419, + 0.4397195727342638, + 0.4695630928415004, + 0.37351459855466507, + 0.569685468340681, + 0.5319857960543098, + 0.82647463069711, + 0.547249026003091, + 0.46629956013034174, + 0.2955655461789862, + 0.4858931403451097, + 0.44206866817352064, + 0.6183174232170054, + 0.36458104426547155, + 0.5673755469986623, + 0.36015460147151657, + 0.4052316441199282, + 0.39721567432612936, + 0.3758595862659608, + 0.4656060056043027, + 0.38380135044187746, + 0.44473393969714625, + 0.4623756027204347, + 0.4364049686487588, + 0.3892793073698674, + 0.38376940431358864, + 0.4335841919917166, + 0.3794656701169661, + 0.5640231320368967, + 0.4424890344723922, + 0.5210948792233676, + 0.43984363104937513, + 0.3982846695872726, + 0.5715501248519198, + 0.49402503794731384, + 0.3958855693539693, + 0.6046625095852853, + 0.3347563310715035, + 0.4705400147165961, + 0.3814700163567929, + 0.40763956041495925, + 0.3482458173175594, + 0.4165228342342812, + 0.5103151099988809, + 0.36920243021359156, + 0.7005816467742039, + 0.4679088047662265, + 0.39915711447681407, + 0.4322317488757377, + 0.3996860368546319, + 0.610216613536762, + 0.33786198599975514, + 0.4082307177212384, + 0.4046897465082435, + 0.4577301067949291, + 0.3868997450891011, + 0.43589481351466874, + 0.3330347083426378, + 0.4755722175423284, + 0.770628378002536, + 0.5041305750118623, + 0.4007113802668275, + 0.4219637517038835, + 0.4505523932146222, + 0.36211438797888595, + 0.3933248128857564, + 0.46717083305673834, + 0.7560392406027584, + 0.5676009275539772, + 0.47201549760046124, + 0.41023294409981376, + 0.3898851360901425, + 0.42921712431199444, + 0.35666948112834224, + 0.48870499877389867, + 0.5411702338680461, + 0.41093961573138743, + 0.40996810472378503, + 0.40066487752156604, + 0.5208137199679083, + 0.3753636144912715, + 0.5550375418521402, + 0.48434686638857866, + 0.31273672899151733, + 0.42313238652589547, + 0.4452395696416918, + 0.5377063604483697, + 0.5460977454198078, + 0.2555253320409322, + 0.7306821563198346, + 0.7157036904585022, + 0.21833656504757445, + 0.34781799934111757, + 0.4455825113732359, + 0.4784464231154406, + 0.3235599894417443, + 0.4275002927628274, + 0.47521807099496743, + 0.4847636426399063, + 0.4973722129771651, + 0.3932436866287377, + 0.44560043172937785, + 0.37690247480828737, + 0.33631802945868444, + 0.45247282197834815, + 0.40645040684718015, + 0.40136636209063825, + 0.5229708229162038, + 0.4527537884397553, + 0.5476031689920985, + 0.45666469211511945, + 0.616245442686294, + 0.4431630192008089, + 0.48854898843640954, + 0.5105788653202449, + 0.4685433132864529, + 0.35153411858813816, + 0.6374003018901543, + 0.6404721171222667, + 0.4664895425676161, + 0.3897726029992242, + 0.4154469203834774, + 0.5450967409618999, + 0.47090043767907525, + 0.5209106065461877, + 0.5197241413787573, + 0.49116963024110993, + 0.45607685873194814, + 0.392697574374559, + 0.4256097397903016, + 0.5819463392889754, + 0.504411967253657, + 0.4824558034560747, + 0.50489902628096, + 0.4217207350835114, + 0.2798428917692877, + 0.3613293798864728, + 0.40254089140556926, + 0.545464137745472, + 0.4809325974711193, + 0.37565419528678073, + 0.48954236008224433, + 0.3882342979630446, + 0.3472377021552677, + 0.41505851595354676, + 0.43167242144368356, + 0.4044028867389775, + 0.4552362445278267, + 0.6363828785930311, + 0.40637249188637187, + 0.46574682188515326, + 0.5791315769834043, + 0.41335191795977844, + 0.5211446659463429, + 0.6088233782859489, + 0.4712550921046709, + 0.5058207007874737, + 0.4984243553383555, + 0.4594050327164564, + 0.3476275171721762, + 0.37888959143212375, + 0.5060141599957442, + 0.5245100960049498, + 0.4958918987333602, + 0.39107251943741295, + 0.48595698632623735, + 0.5040566829342712, + 0.5277249306428629, + 0.41816576395416405, + 0.47928905846173897, + 0.3962964972447781, + 0.4922480518316861, + 0.4560208367164957, + 0.4325097351913308, + 0.3955206173006078, + 0.4765415987842545, + 0.4847502295993798, + 0.31896779096953015, + 0.4064444350903029, + 0.4501323737324498, + 0.42530930468707184, + 0.4870632083182242, + 0.507302568810994, + 0.5130871480456248, + 0.47256716041914315, + 0.5145205717280268, + 0.41485532020959714, + 0.48106101663561107, + 0.48888605459960266, + 0.4483211126430796, + 0.4578252604624064, + 0.46745267860934986, + 0.46389307896278686, + 0.4808263493223358, + 0.4899834511892897, + 0.7293698602038823, + 0.5555901886919316, + 0.5155971587997888, + 0.5466595686118008, + 0.528251413544456, + 0.5066352484209242, + 0.5482954548456738, + 0.37761881059771124, + 0.41815051908857914, + 0.48137384345561707, + 0.35857824143396233, + 0.4630385636454168, + 0.5108715121121558, + 0.4512356146457897, + 0.4562440536518725, + 0.4811656198957146, + 0.3443043811267559, + 0.5575378147836367, + 0.5012163834439695, + 0.4016319960417654, + 0.45262536460991076, + 0.40120312243904177, + 0.3713811187129494, + 0.4740427276046422, + 0.4650928798039991, + 0.44232548966906904, + 0.47699284818540394, + 0.41087981637800597, + 0.49847887825280024, + 0.3828377017506998, + 0.40515906221331477, + 0.44495716797158275, + 0.5084259071905908, + 0.5013963067363547, + 0.48249003080095504, + 0.4227118407063668, + 0.364493586578316, + 0.3675476250155202, + 0.45956305814901127, + 0.4245274564434477, + 0.46486962984388347, + 0.4960154548931238, + 0.4447027918058787, + 0.4213032442837759, + 0.37228465335101074, + 0.4254329466113707, + 0.5435177484421471, + 0.4225063841616029, + 0.21763878356176652, + 0.37676186442068144, + 0.45798693247567135, + 0.3239993447180655, + 0.31422646748612953, + 0.4767670277298127, + 0.47473995457589674, + 0.34470200242297494, + 0.32308824222275107, + 0.4149841251976519, + 0.40830490520612367, + 0.5495367582611367, + 0.3326105138830362, + 0.3404216691346271, + 0.36311701184746786, + 0.3826834100398507, + 0.2920484564512456, + 0.3561555161802412, + 0.33308333755094943, + 0.369576483740268, + 0.34113017668014456, + 0.4584737958193021, + 0.3690465438094015, + 0.4118983684524376, + 0.41370707514899746, + 0.4994401526385459, + 0.3577571565092487, + 0.41852924006465675, + 0.41326160073725954, + 0.4226090884269423, + 0.41770280941955845, + 0.42228451208492773, + 0.412149346123067, + 0.6460179032158821, + 0.35214023747134443, + 0.46044573594061033, + 0.5909661218787412, + 0.5049953844636051, + 0.37672598583376316, + 0.48067270412001534, + 0.4311599816571072, + 0.4742903348241109, + 0.48594191420806476, + 0.37169636051209726, + 0.5891581438659186, + 0.4432896353035879, + 0.4397047656464267, + 0.37558974876203294, + 0.4483156763992549, + 0.40926020781897093, + 0.473657379341848, + 0.5083202475570658, + 0.463105981931416, + 0.4813046578444483, + 0.5789939742620754, + 0.4564036868633311, + 0.4232579910936287, + 0.5194896352200149, + 0.5874694101576066, + 0.5415103979726026, + 0.45697130492659793, + 0.5877759195530681, + 0.5743240320961449, + 0.3960790236909989, + 0.4083757911642291, + 0.4575825462676757, + 0.3601430703375758, + 0.6823561918403933, + 0.497619015851824, + 0.4420876233835676, + 0.3667750579115848, + 0.47780083137565277, + 0.413193985353262, + 0.5320335645224391, + 0.45098724698960624, + 0.47730535909629956, + 0.5445127521700317, + 0.3554361051301604, + 0.43259531535290324, + 0.5638754140916199, + 0.36220931636015563, + 0.44453095731977804, + 0.3402515194808492, + 0.3869726839262943, + 0.41643622012471665, + 0.4169561744075561, + 0.4878414602948096, + 0.3610684724532265, + 0.3414438999505791, + 0.4509746198780002, + 0.41399698317771044, + 0.388222489233016, + 0.466259983890407, + 0.47885519543209015, + 0.44179685786076617, + 0.5405800011921268, + 0.5717011784935618, + 0.47649620203205645, + 0.3726379044749035, + 0.4914434380408523, + 0.4546002583629076, + 0.40840529160838845, + 0.4497195895582753, + 0.3421995437077257, + 0.3429603930915344, + 0.32295366654338936, + 0.38128800589325484, + 0.4464762391338083, + 0.658823164913915, + 0.44817912324492953, + 0.43884000830428616, + 0.3093935042075888, + 0.3679301356866176, + 0.3302708684960607, + 0.418960431716443, + 0.3794524171101294, + 0.4023721397801325, + 0.3660617660684115, + 0.4306371769806444, + 0.4032180926881452, + 0.43259884322291164, + 0.42438403120686985, + 0.38656570296003345, + 0.5113792448691409, + 0.5068008331091781, + 0.39751392240854666, + 0.4429074338007599, + 0.45518809856936976, + 0.401606424097166, + 0.3768979683682968, + 0.5215022058473794, + 0.4602110264258862, + 0.5900879276116767, + 0.46357713458430727, + 0.4559625784228489, + 0.45637379890638063, + 0.4637547966594226, + 0.3606801045159083, + 0.6248700597404612, + 0.41694257142770724, + 0.2400498620067809, + 0.4897852105115916, + 0.6081800683520308, + 0.38431966215653535, + 0.4962558838972588, + 0.7645526738511789, + 0.45980521404409214, + 0.537681834973764, + 0.3234728540250816, + 0.5031776879199239, + 0.3675259588973045, + 0.3864636212148963, + 0.3663962035954835, + 0.31826190565329854, + 0.45792686957787715, + 0.26492509930146435, + 0.44169437436890396, + 0.46503219696338083, + 0.44720981396117493, + 0.41324131171697037, + 0.3985126469116948, + 0.37104989930232196, + 0.4715919192047608, + 0.5326830510098522, + 0.5537240848725878, + 0.37326302305895015, + 0.4635866287401696, + 0.4349276375566228, + 0.7398344219103226, + 0.4296562173584076, + 0.5675911026911019, + 0.7417205130241153, + 0.7208393540821595, + 0.48299143796961436, + 0.40691508120525893, + 0.3572972937608556, + 0.4678849934711154, + 0.5123631847258243, + 0.5015011684922751, + 0.5889629169161767, + 0.43555105704369473, + 0.5606527695861586, + 0.4797708702689677, + 0.408641362229473, + 0.7079145575157896, + 0.6877550506086498, + 0.5064127411402816, + 0.42911411117130255, + 0.4842193401467037, + 0.432206305109045, + 0.46769837690058796, + 0.6512841686949049, + 0.6760369297908381, + 0.43930097255422407, + 0.4501114982417107, + 0.45012544204986815, + 0.4007592999561705, + 0.5038907477327086, + 0.3789172799115743, + 0.5437331379163247, + 0.5401554577359597, + 0.37727635421281097, + 0.4672022005750868, + 0.4771194949642523, + 0.39968279268115453, + 0.5477092334797292, + 0.5281437008422606, + 0.44060532548216613, + 0.3949967462277727, + 0.5033451138767954, + 0.46592096813087774, + 0.37985036165383557, + 0.36189883133478995, + 0.30164414409102547, + 0.26660042204968387, + 0.3191997972087975, + 0.5056953164715297, + 0.5572600129721055, + 0.41584989154338703, + 0.3315285404362821, + 0.420483407895616, + 0.5020867699905311, + 0.5006713182531171, + 0.5303448879170006, + 0.4223616781118762, + 0.5083934438789662, + 0.5593079559526575, + 0.4302053379439915, + 0.5886970411260106, + 0.5456571929953439, + 0.4656477128261886, + 0.44612183184199283, + 0.3966697091944415, + 0.26516265915819487, + 0.31282456476166565, + 0.30012493770508564, + 0.5646743892082987, + 0.47899059875692024, + 0.44159612242816926, + 0.5656771322961464, + 0.5874713604289872, + 0.34102240080469315, + 0.4942901635659883, + 0.39868615052589657, + 0.7244129726310754, + 0.4544524580382459, + 0.46714867806078125, + 0.378850452062721, + 0.46679593546416503, + 0.5084919612491859, + 0.4551626781939067, + 0.42573355545098485, + 0.5087811017993368, + 0.4686008190020391, + 0.5829047176524403, + 0.436516633720663, + 0.5104128424319778, + 0.4625271455969346, + 0.44709703672491047, + 0.42997658890035834, + 0.4913483521325594, + 0.45506599210622134, + 0.5085836442389563, + 0.3890885843138994, + 0.3672711177759517, + 0.36297919189830097, + 0.43894224018202205, + 0.4830700127645588, + 0.4211016924038533, + 0.48941562936772476, + 0.5139785788941226, + 0.47025458815175314, + 0.44191601254514296, + 0.45626798248806766, + 0.2811140710863932, + 0.45206987716296787, + 0.32453875804494176, + 0.4698599725518726, + 0.24530244203129478, + 0.6761807377902319, + 0.3424863363727305, + 0.5154453707127866, + 0.5994119178548908, + 0.5213803780040616, + 0.2853880261871529, + 0.48355036240345733, + 0.38568659799178606, + 0.7841258308805066, + 0.30281679610531037, + 0.5646624392354147, + 0.5313272025700055, + 0.5593864403578349, + 0.6431622419717202, + 0.49786062567560374, + 0.6567979257420747, + 0.4850892281995401, + 0.43906486839001035, + 0.3832605701345271, + 0.4825664585260537, + 0.3381011263457215, + 0.7279481901211879 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.5017726981280421, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.579407691212294, + 0.7663179374394278, + 0.7402105822638541, + 0.8228774521683243, + 0.7483014060763515, + 0.6940974431465642, + 0.7421960577478316, + 0.6854076327109414, + 0.43894688817033284, + 0.7252118034706522, + 0.3857621285623741, + 0.5219569509006132, + 0.42830677089976044, + 0.39844825398175826, + 0.43620995703760307, + 0.3555189494611304, + 0.40764511069453685, + 0.5980052889338927, + 0.5894900211576466, + 0.48426568329640934, + 0.3641494291646424, + 0.6940281996590535, + 0.4773252106791074, + 0.6110518118894892, + 0.39498996080109, + 0.35258210431058773, + 0.46461659852147147, + 0.569975510150701, + 0.6659987367364072, + 0.51946304505301, + 0.3415343965076517, + 0.4412055296008422, + 0.4421919707947232, + 0.3929980022665457, + 0.501687367136213, + 0.3849906705173343, + 0.4998172986893943, + 0.5663449341880598, + 0.6299437777518827, + 0.6348112243665726, + 0.7959417293563575, + 0.7869588636288551, + 0.7810025227451532, + 0.670366637248862, + 0.4939257951586584, + 0.5574553218157617, + 0.7874665752833145, + 0.44087011409167653, + 0.4221012508337841, + 0.5654729911580041, + 0.4681146987803859, + 0.782534904978894, + 0.7985159091465394, + 0.4927003497397151, + 0.4985571501939976, + 0.42425456459102967, + 0.4123686279806845, + 0.4628444585124859, + 0.4583621421761326, + 0.5859716721159516, + 0.5459738117994195, + 0.5118581359294887, + 0.5199855486362376, + 0.5598649554963211, + 0.6650230661741875, + 0.4591965972966209, + 0.41895734830007836, + 0.6326776196460433, + 0.4613154623757866, + 0.5408256363479023, + 0.5588750350768219, + 0.5398402049175569, + 0.5079549706811738, + 0.5721077671166556, + 0.48035921191077136, + 0.5840500246031692, + 0.36750571169802826, + 0.40062379921850505, + 0.4558461409796084, + 0.36569017128669057, + 0.4836904439384216, + 0.604423213103808, + 0.43854140058022173, + 0.42125275666368495, + 0.42360626994684597, + 0.5181523455380006, + 0.5035649161680785, + 0.47912616595755525, + 0.39684512204970734, + 0.42867384625574106, + 0.3844205751970656, + 0.3860327850278688, + 0.5551608919389646, + 0.5772616850367376, + 0.45463150058760626, + 0.6184956740456631, + 0.4086415870743169, + 0.5896199868590297, + 0.44240316697885335, + 0.5039708860361386, + 0.4904225025481021, + 0.5612324019864812, + 0.34081028024011384, + 0.5027278187709707, + 0.3751979838546388, + 0.39040847903739145, + 0.5211386115306521, + 0.5540050586640798, + 0.3781555516177154, + 0.8160418341980279, + 0.8097978216766772, + 0.8147105923209045, + 0.8332339264648513, + 0.5930252304127119, + 0.7104622973641348, + 0.7277427549898862, + 0.764706186458685, + 0.7174579810308587, + 0.5927341354457488, + 0.6682476317112175, + 0.5789620166955963, + 0.7794795713336313, + 0.6707877101364903, + 0.45550107034593246, + 0.36123263675545253, + 0.3367044182911315, + 0.37011136707135955, + 0.586125010235639, + 0.4553762862921279, + 0.32172252007665375, + 0.5040756669873148, + 0.4704681787179822, + 0.5196311504508965, + 0.4232846891093578, + 0.447002788897809, + 0.5072977359632007, + 0.49870960417629245, + 0.3900006194544635, + 0.42454375716858234, + 0.4290691012430924, + 0.393041304270293, + 0.3415502862427573, + 0.4622815165884229, + 0.6156935902419488, + 0.33583885142045644, + 0.5821432417778503, + 0.38579442153037774, + 0.5099843935872753, + 0.46194260721307195, + 0.5823070715859058, + 0.3488466588512576, + 0.5212917311647043, + 0.36000843935085003, + 0.38333864101713333, + 0.4461902531554341, + 0.38161712706869716, + 0.3020880033580577, + 0.22284704987184242, + 0.39427254783538834, + 0.391449368789995, + 0.562931170918438, + 0.5534805340467995, + 0.626603914640753, + 0.5573476069666095, + 0.4053740747667842, + 0.5233296319290895, + 0.5734030958025569, + 0.5624240099393507, + 0.46803609702004517, + 0.5088842157255032, + 0.45271016315891693, + 0.325353943322486, + 0.5103235206972816, + 0.48203564648961406, + 0.305258446866083, + 0.32418565967359464, + 0.5321332340783012, + 0.6018203253688041, + 0.5905699614201262, + 0.5161493730791843, + 0.3546066341867483, + 0.5868087554063763, + 0.5335458171518265, + 0.549611220625461, + 0.3375187818634816, + 0.5884112787059659, + 0.4961287349024857, + 0.5505585239540878, + 0.28438672167405166, + 0.4702552471925056, + 0.3713509948767254, + 0.3037856421627818, + 0.5648623423801803, + 0.475860475836961, + 0.5891617906095887, + 0.41611719415638476, + 0.6013198671873301, + 0.3945636604228143, + 0.4895732087557245, + 0.4256908267739295, + 0.3627208966905201, + 0.3005240884714175, + 0.40593303513613005, + 0.4088698311017752, + 0.649533690221617, + 0.38706981954222053, + 0.32438226118907526, + 0.4975434423483863, + 0.5932988405332803, + 0.44190176034163614, + 0.4906974814910279, + 0.46760852711720907, + 0.3808615484946021, + 0.4178098042098512, + 0.40434490078803326, + 0.5581127851459708, + 0.639316847301423, + 0.639688312144532, + 0.6012940169322857, + 0.3752832388984487, + 0.545399543539239, + 0.4896093306719744, + 0.30972772541703747, + 0.37305257276602816, + 0.40257677476055126, + 0.2502991683684282, + 0.569973963831564, + 0.3612567542849665, + 0.3597135661072972, + 0.5581934215928654, + 0.3182487112126788, + 0.3041830253457974, + 0.5557811754396411, + 0.43922417623413157, + 0.35217168370314456, + 0.6485776788644607, + 0.5732562475676986, + 0.33319384667482144, + 0.3416858831535508, + 0.5054758157492326, + 0.3504830830044306, + 0.4809708989215491, + 0.34319753538437914, + 0.5259471991763999, + 0.4785736540877554, + 0.42001608000202223, + 0.3851887132665842, + 0.35841028411113657, + 0.4813804146578302, + 0.37671564447099254, + 0.4682032969246292, + 0.3577226454434254, + 0.42735796968434636, + 0.38484449534324094, + 0.3945046220118798, + 0.4753881272210665, + 0.43856012873461486, + 0.409036302794865, + 0.46398706111479016, + 0.46785072053773485, + 0.35796074645508597, + 0.5696479016852702, + 0.3576452347016862, + 0.39004532646349194, + 0.42323026773235173, + 0.3508144916426645, + 0.38856828132176263, + 0.5411722905912085, + 0.3029322595209025, + 0.3781641858131475, + 0.3679986752351538, + 0.38009947881963585, + 0.3961363092399979, + 0.43222071283255953, + 0.4067406479789679, + 0.5127903042255104, + 0.3193126618914958, + 0.48378454699235296, + 0.531247633985913, + 0.3383353470059173, + 0.33541992638341805, + 0.6408262179038412, + 0.5090531943763984, + 0.3093952630757198, + 0.580662072260884, + 0.5385751535287442, + 0.508005422898391, + 0.5154664667277826, + 0.5926004683832615, + 0.5615004839190497, + 0.5805624099207283, + 0.5669493898020754, + 0.5730794026453744, + 0.45555218601362546, + 0.5118191208142893, + 0.5772979791346996, + 0.5176798686727748, + 0.4889234661353158, + 0.664228088414037, + 0.3619142558923381, + 0.6174140177013645, + 0.6770211908386364, + 0.7339885060081842, + 0.7079881525533809, + 0.5441747819490367, + 0.5858615213697862, + 0.4447771658489924, + 0.7038911885359451, + 0.5097855792741782, + 0.5249122896347932, + 0.36266932467702717, + 0.5458867785197723, + 0.4604364939151944, + 0.47049368649608325, + 0.587619246942701, + 0.37710629387598754, + 0.45566546820601955, + 0.3579026913462748, + 0.4796756749488557, + 0.4137797886436099, + 0.37008644978719607, + 0.416848337330374, + 0.3092909997209366, + 0.37392922130836553, + 0.40801226926527506, + 0.4158016396938478, + 0.4033533990246788, + 0.41340350572598417, + 0.6238523417846445, + 0.5242563643861782, + 0.48258407114696444, + 0.5477696768440016, + 0.511753116990343, + 0.3552959437826918, + 0.3348673701211063, + 0.6103601003443164, + 0.4343592193414352, + 0.36922812450480874, + 0.4844591920249836, + 0.6030705109450237, + 0.6317264874473828, + 0.4783541015949152, + 0.38688593616449984, + 0.514404781494469, + 0.4078616058490959, + 0.3780451299641126, + 0.5333970741954435, + 0.3810245755919593, + 0.8283178326016963, + 0.5393724610850661, + 0.44248188934964877, + 0.44135526024360233, + 0.47419082143009805, + 0.47121039408404713, + 0.43593881501685483, + 0.548787851759269, + 0.4095452473716543, + 0.5749605219994087, + 0.4728178638002635, + 0.5480182589314327, + 0.5924527705135015, + 0.4888361165010679, + 0.5080353237637287, + 0.42535941043559433, + 0.4454028109881567, + 0.6338449000359624, + 0.6027995737350248, + 0.4556463402810312, + 0.6246975213770025, + 0.49858353611650147, + 0.5939236866251068, + 0.4652932103430651, + 0.5045352053652876, + 0.4553711079071416, + 0.5919395919612157, + 0.6066852249349747, + 0.6482418719075755, + 0.5744800778748205, + 0.5679955587912092, + 0.5560973834147891, + 0.5217901238524068, + 0.43356683551338143, + 0.4300052898550615, + 0.36847422656529244, + 0.426457446966853, + 0.3595905316208877, + 0.3886945493094905, + 0.4094680786434701, + 0.5620771943003222, + 0.39391320174427036, + 0.3646647632005327, + 0.4244253449955205, + 0.5953049713688194, + 0.6639382830957934, + 0.6226934768676186, + 0.6053604908248432, + 0.4432184868894182, + 0.622358426506127, + 0.4884570089071388, + 0.6555880484781937, + 0.5057059514066461, + 0.6111755623019047, + 0.4785701742603107, + 0.59373034455102, + 0.6198267571424377, + 0.5287282925095022, + 0.5203598466083258, + 0.5599029756779282, + 0.6249418081050829, + 0.5728121896309907, + 0.45340733334032657, + 0.4303765003712234, + 0.3786789122514018, + 0.4087294968859652, + 0.559793108725241, + 0.6045244139119863, + 0.35469635448350434, + 0.6532351477416531, + 0.6313569002620327, + 0.5183009352851126, + 0.4380609826401246, + 0.5071990166914992, + 0.6022458627960431, + 0.5636583231982853, + 0.5137658900156508, + 0.8635282665373084, + 0.5055035756880893, + 0.6123563882452873, + 0.5645733657121915, + 0.5954118072211398, + 0.502272286260609, + 0.4347132726636049, + 0.6339666117189056, + 0.49479191738422795, + 0.4562629282144234, + 0.48562750194506205, + 0.4777292587137874, + 0.48914572354068586, + 0.4518911732063566, + 0.48452637837846824, + 0.4706703779720118, + 0.4987060110579835, + 0.34317554695687064, + 0.5118315714103987, + 0.35157428408069336, + 0.42455303934618743, + 0.31323826164074625, + 0.5159332709498098, + 0.4105598332122218, + 0.3547648057649474, + 0.5833953193693828, + 0.4208573501495926, + 0.31476417141737845, + 0.5742474812469395, + 0.496897303918993, + 0.49945503763181687, + 0.5555066041706264, + 0.4394065364083481, + 0.44640936661581826, + 0.6640507602964857, + 0.4484560208245506, + 0.5659236528364194, + 0.45990600411198473, + 0.32975173966697185, + 0.308190855460777, + 0.4299992061999034, + 0.5300552765106714, + 0.6279980452085657, + 0.6144523770415631, + 0.5122787576944173, + 0.572821074678804, + 0.5050408376798616, + 0.5376111340329798, + 0.4699189368010198, + 0.4488997516852856, + 0.4613950561135726, + 0.4404793242958727, + 0.4997386955837141, + 0.43428789204264323, + 0.4735112541821607, + 0.5585316219696526, + 0.4968298416129705, + 0.47817336895683665, + 0.39854999947695774, + 0.4417417448581609, + 0.5351782481536512, + 0.5872859567971442, + 0.5754439713113917, + 0.5872332851446624, + 0.6072273117264066, + 0.6382990823670803, + 0.678687223743429, + 0.6614906214267399, + 0.4983353351568945, + 0.5437409839277368, + 0.5213461027909521, + 0.5644656616112893, + 0.43593487600594544, + 0.535025806725168, + 0.5021965831644465, + 0.5944880827376849, + 0.4258283860083421, + 0.6036382763431846, + 0.6643485712273641, + 0.633566318432353, + 0.5838594892268658, + 0.6572904782333941, + 0.5585596101970528, + 0.48534539763769646, + 0.4115150610374248, + 0.46654736467636887, + 0.37001858255133513, + 0.4295138840707998, + 0.5824750726996702, + 0.6255401808284371, + 0.5066243419994632, + 0.4114642213250793, + 0.42740054509997644, + 0.5341717720882942, + 0.5308469185935165, + 0.43196109660627485, + 0.5458629045447625, + 0.5243244339701503, + 0.4041693189572273, + 0.45657115147053307, + 0.42056114826941193, + 0.49785671308478235, + 0.4816397412610536, + 0.37445524700038724, + 0.4433837160256555, + 0.5057891655371334, + 0.43177268330573704, + 0.42148388072171056, + 0.45997510838997413, + 0.41715350608761853, + 0.4074585264242581, + 0.4737278056616237, + 0.5783813777671437, + 0.4561841831043916, + 0.32654094365738057, + 0.6056639952775631, + 0.4289581514942928, + 0.5868976706280238, + 0.5770973561129592, + 0.4129155846038713, + 0.5433871992288744, + 0.6177611303557015, + 0.5317562721588921, + 0.6774226000717227, + 0.5053276526686296, + 0.5625580025375408, + 0.5348099889456787, + 0.5197261664407571, + 0.5688265974156166, + 0.5313713180594631, + 0.39129338299680516, + 0.5117809084499922, + 0.553908729240163, + 0.4606016889549832, + 0.34702348401456434, + 0.49338456139011344, + 0.511706842735052, + 0.4944956359792521, + 0.6673224670835567, + 0.5723593398754807, + 0.5777023031048724, + 0.5913908289100345, + 0.6612748647950621, + 0.5818340618348382, + 0.6371177492276056, + 0.6503509220139992, + 0.637823484617666, + 0.48061538164236917, + 0.5053231176938505, + 0.4569418461330556, + 0.36669265540629475, + 0.3311898474031096, + 0.39926131671308507, + 0.42733205272518665, + 0.374458214827615, + 0.34418456176276396, + 0.5282125759857305, + 0.3370094213085055, + 0.4935239138286013, + 0.3778236562757571, + 0.5620382224767817, + 0.6227881325160574, + 0.5199139262248079, + 0.3701744047687196, + 0.41831653968318305, + 0.5917169130683044, + 0.37580449018758066, + 0.523556870893598, + 0.34738712410250716, + 0.8370475676800893, + 0.5584607205591386, + 0.4528898740735316, + 0.6374106757632498, + 0.5953269144510221, + 0.5222710208313104, + 0.45819760042519614, + 0.5193794714318856, + 0.5978276777457565, + 0.4731096853001956, + 0.6128826161822886, + 0.5712347351042389, + 0.3367353188899693, + 0.5337666969800372, + 0.34155235300615494, + 0.4107518192162701, + 0.35124892200902486, + 0.5497337943520009, + 0.458443819169794, + 0.4054343857127037, + 0.5619499392113696, + 0.5948242669009598, + 0.4176365850897175, + 0.5200082396140733, + 0.4590932593574767, + 0.49520834911341854, + 0.5256097007596607, + 0.5002558875396386, + 0.37903397476584394, + 0.37015397498010216, + 0.6140342316572575, + 0.44884812519775824, + 0.4113592125150419, + 0.549795014557763, + 0.43751159906214027, + 0.4794536264494361, + 0.3606782816167873, + 0.4055932560841409, + 0.457517043388503, + 0.5098589131057746, + 0.5147737852247916, + 0.38845036377798425, + 0.5116227721441441, + 0.34278933983104337, + 0.5380595026514573, + 0.4929363307103818, + 0.5189315928576117, + 0.5304977793558044, + 0.4374248221154195, + 0.49401096426241575, + 0.4121225046322464, + 0.537128988569424, + 0.4475458313002553, + 0.4887746108105329, + 0.5046967945589765, + 0.38715658748651827, + 0.5373397585568875, + 0.30010746053242043, + 0.6747051384093538, + 0.6769084904249414, + 0.5748528836313012, + 0.5654712764818539, + 0.6201156449695308, + 0.6164513805916795, + 0.30585171001993566, + 0.4260477633268339, + 0.3864739208524905, + 0.4339476403171075, + 0.3701563194770208, + 0.37570523440850256, + 0.4324767171997999, + 0.5029899680990534, + 0.4877451236116867, + 0.3896301044804213, + 0.41256428927459815, + 0.34398289251471725, + 0.48035579156281677, + 0.2694697868581285, + 0.4538367715358067, + 0.5088569244350597, + 0.5182816045709805, + 0.5351531469955553, + 0.45555273807780444, + 0.5662528117729493, + 0.3194040681276329, + 0.5500618770567581, + 0.4846754919147008, + 0.4323911603026009, + 0.4128577905076333, + 0.44728235232295116, + 0.45721784443675134, + 0.4674141163674114, + 0.5062751150868501, + 0.5287778366946948, + 0.5675384550825884, + 0.3800964828436867, + 0.42155914226966357, + 0.49485283404273583, + 0.5575123396117165, + 0.5433627155150225, + 0.5495807422101209, + 0.37864700130649354, + 0.6079854140473996, + 0.49959034055350404, + 0.4845527044045093, + 0.620797062334705, + 0.44983267824570167, + 0.3825696878541194, + 0.3441766822747095, + 0.43173213260150506, + 0.4773345087903647, + 0.42733878342537684, + 0.4411981012171495, + 0.4539401695194361, + 0.57051993677676, + 0.6262015010572025, + 0.673073457129509, + 0.5661020296309591, + 0.5814667731298148, + 0.5391009367332887, + 0.6083422474699394, + 0.5476503224446116, + 0.6269603559621899, + 0.39547784961538085, + 0.5152033406215565, + 0.6333387672125228, + 0.6829341804356354, + 0.22996598327880802, + 0.42451578679291335, + 0.4459600820140584, + 0.5484714252013151, + 0.5373984172631083, + 0.5615989494412205, + 0.42260299226718834, + 0.538959165744528, + 0.7558256732549823, + 0.624981469551868, + 0.7920031359022532, + 0.8058994141176054, + 0.6698215397379724, + 0.7676479704434395, + 0.7742167004148578, + 0.4315529787240063, + 0.4427218871243303, + 0.4551170840006893, + 0.5535380184016592, + 0.3763991788621409, + 0.5131944315027879, + 0.40541356151222563, + 0.5112924773205624, + 0.4485402968705762, + 0.4289139633289383, + 0.35164471239524503, + 0.4554177056126135, + 0.4461303374067318, + 0.4790711493918293, + 0.45829493042617775, + 0.35266909834576105, + 0.3747058407001643, + 0.36098283898719696, + 0.3797583428340756, + 0.2936214445581985, + 0.3778546559663948, + 0.33685566774497183, + 0.5587232734855686, + 0.597800468199538, + 0.39943661727939517, + 0.34187015645514596, + 0.2929945075642095, + 0.5446416884373569, + 0.6479126634381118, + 0.6518480758540698, + 0.4025497547862418, + 0.39278085523121026, + 0.2895921319977153, + 0.527118187566571, + 0.3576562705055837, + 0.41834910463266495, + 0.5088024303877619, + 0.5773704932985948, + 0.49482589848636493, + 0.3442389327949961, + 0.4934278721122326, + 0.5519555903893645, + 0.5418595175217243, + 0.4870833893439426, + 0.5380084379258587, + 0.44346489640461606, + 0.5428418554339328, + 0.5167290148701504, + 0.5809505690080113, + 0.38225193276004654, + 0.5090269112465283, + 0.6921731293401314, + 0.47349951066518897, + 0.6878660948604284, + 0.41223890244592476, + 0.5646641314337264, + 0.7225324751538874, + 0.587769921208236, + 0.4272207619189964, + 0.4036636289892361, + 0.37355264761696877, + 0.3479242116599142, + 0.5016139239896082, + 0.6236671609233313, + 0.5019474209836442, + 0.438612333928073, + 0.3647571308161013, + 0.5496589266838074, + 0.5396528335587315, + 0.5779349522918077, + 0.6101609132342789, + 0.5623363280161696, + 0.5254112038570967, + 0.4436470226046206, + 0.5969325210309775, + 0.3315525837569274, + 0.44186820204782756, + 0.5004194562867867, + 0.4434569922446533, + 0.6553371181091249, + 0.659114695108993, + 0.6223672716934351, + 0.6832534721740613, + 0.7583424022305714, + 0.6144958252275963, + 0.7672480583689416, + 0.5456017889849825, + 0.48865715317858816, + 0.4708536629615443, + 0.6130475193725303, + 0.39712114952284844, + 0.5466438975926519, + 0.5593855317130328, + 0.5482289739236702, + 0.5114823577917578, + 0.4742391413424043, + 0.4825731745143173, + 0.5337358594129533, + 0.4479729339649885, + 0.5810494411844718, + 0.5608986677139244, + 0.5742203593979313, + 0.6128459004375366, + 0.6637252589322943, + 0.5945809221809298, + 0.47576021737356466, + 0.6233033946371329, + 0.6025080001418632, + 0.38832099151784016, + 0.4746242598773552, + 0.448396485167088, + 0.4477848637796065, + 0.4309547749058564, + 0.45758678344590153, + 0.5202752342198075, + 0.50897496219968, + 0.51882254846588, + 0.5055688089634197, + 0.33096922618305924, + 0.4837744776258371, + 0.571488215297421, + 0.5182946072722674, + 0.6234505554546366, + 0.6515266978340131, + 0.7494492400687139, + 0.37430813704037713, + 0.4005113726457716, + 0.5452749634148835, + 0.4356287042830577, + 0.3160783807290512, + 0.38414465868202136, + 0.36444807417445674, + 0.6946826363311442, + 0.6450862591379519, + 0.6719334026677128, + 0.3836975128177527, + 0.43222794154317185, + 0.35317783915097484, + 0.6010682138443729, + 0.5102511270687268, + 0.2989746276110164, + 0.35030789506468446, + 0.5059788493960294, + 0.4613739991281406, + 0.530290110891774, + 0.5396913099238634, + 0.4541743701545915, + 0.44376330156685956, + 0.33248815109042373, + 0.700769790965877, + 0.5803139870874471, + 0.7085639256640701, + 0.595524928823045, + 0.3270386239546027, + 0.5512407401335874, + 0.5669277868304423, + 0.5656892713707645, + 0.5178884423198101, + 0.42675112806789284, + 0.5338649773134281, + 0.5823448734817835, + 0.5761674596726407, + 0.5183667751839707, + 0.4000355996169127, + 0.5146097987969234, + 0.42604478546528207, + 0.492610947110533, + 0.6520005160740909, + 0.6002630267561412, + 0.5999226592551462, + 0.4791375794385125, + 0.5076741990861249, + 0.506125180054326, + 0.46242101048172174, + 0.5331005924539132, + 0.4507066739771642, + 0.31776526416523004, + 0.49409999809774496, + 0.5806179309367877, + 0.3029515952713056, + 0.7689497746845161, + 0.4767221465188613, + 0.6174355989273206, + 0.7412843275726878, + 0.6895860615812666, + 0.4147789588920206, + 0.8125915876413794, + 0.5110992991442198, + 0.6824776638665527, + 0.4214234475042711, + 0.606252751234814, + 0.6152914095825027, + 0.6932660226650752, + 0.6471011968675633, + 0.707925060909244, + 0.6312836998648841, + 0.4511410285857698, + 0.6138717905288715, + 0.5687298703177246, + 0.4905286046220725, + 0.7125707856747333, + 0.39192371017877947, + 0.6226354239305493, + 0.6350841750199379, + 0.7026303303839303, + 0.6511548075078026, + 0.7154086044750042, + 0.5918013435093852, + 0.43738343501253196, + 0.5549715673561479, + 0.5653391566393208, + 0.5479706486111202, + 0.4425374649887771, + 0.4075317628338988, + 0.48523108634968093, + 0.5658622840815782, + 0.576868366831497, + 0.4537890421850248, + 0.5920837526388251, + 0.6038466348563525, + 0.4624435394335501, + 0.5878190592410908, + 0.5776118509633655, + 0.5519143439186692, + 0.48994483596912125, + 0.43531289969732334, + 0.46971865227597026, + 0.5341587354779576, + 0.45999708315192417, + 0.38699738769449504, + 0.4859271089872254, + 0.4013715807921409, + 0.5295770542076351, + 0.4563460224614416, + 0.3848607815212198, + 0.3939310914306659, + 0.41547100014440685, + 0.5196313362492782, + 0.49709723673505046, + 0.3998706470694601, + 0.3625211528446466, + 0.4518665727683566, + 0.5996903894819091, + 0.6672919539309837, + 0.35190675866756976, + 0.5690662282659695, + 0.6175661776952609, + 0.6026228083098034, + 0.60460308912309, + 0.45599014592258325, + 0.5747563646856712, + 0.450073265626759, + 0.6189666618832663, + 0.49058407282782646, + 0.6633551042869273, + 0.6005428971118756, + 0.6074414510573056, + 0.34967000059599507, + 0.38253599976883024, + 0.43649883903554454, + 0.5310407115160916, + 0.5064122325843137, + 0.5918215676724193, + 0.37333306853470943, + 0.6143050449903158, + 0.5446651467673329, + 0.584289557565789, + 0.5772272814145313, + 0.5471918176244842, + 0.46137776671706787, + 0.5763340528772644, + 0.6005008596978664, + 0.5845016582899419, + 0.5655884707643717, + 0.47359532174935387, + 0.37879035038404785, + 0.40481814646266817, + 0.40826462710579486, + 0.28276834317411853, + 0.4267538481403876, + 0.5539101117649302, + 0.4685811568942387, + 0.45269541569098365, + 0.48278225153009113, + 0.4404454347465205, + 0.5490251963631454, + 0.5766087866042925, + 0.3929591989951384, + 0.5147571694543327, + 0.5998796585615288, + 0.4034898619537424, + 0.4523284234144721, + 0.5365100155264181, + 0.5115884720732161, + 0.6238265609261741, + 0.3891952355390399, + 0.562524174189266, + 0.4096470573053109, + 0.6671989257613304, + 0.6147313209356058, + 0.6678280162726409, + 0.3487023325699185, + 0.3713413585089505, + 0.3499467746370133, + 0.32882305654602423, + 0.3474800380331936, + 0.5050287763404923, + 0.5445004610663652, + 0.31658273139772763, + 0.4439494965275435, + 0.4650590942176803, + 0.49875530289545283, + 0.5720600136018349, + 0.5387757938264404, + 0.676666346622735, + 0.3455029167352523, + 0.47868078379736867, + 0.49071755865984806, + 0.4556069605650297, + 0.46507402023936273, + 0.5736348020203144, + 0.5310732189729008, + 0.326084861817965, + 0.5716220144724887, + 0.5100868710998487, + 0.4688928967922863, + 0.6175332043711586, + 0.4718091596400774, + 0.3938964713652804, + 0.4298272197159337, + 0.4806283941425301, + 0.4758589004657187, + 0.4397602291458007, + 0.4736720360987965, + 0.6327244036077403, + 0.4899628249921202, + 0.6447403038818112, + 0.5475943140966688, + 0.5448522732834326, + 0.6059560880741112, + 0.5606659847713122, + 0.547836174009582, + 0.5328436515028776, + 0.6651574083876459, + 0.6586101590877438, + 0.5870260956481852, + 0.6073126745616556, + 0.7550402097317834, + 0.491581962588748, + 0.4926970382177293, + 0.3829504192004125, + 0.5615664070453333, + 0.5954519269044833, + 0.5551483187263616, + 0.48710027442028914, + 0.5749124529606431, + 0.4010008367259717, + 0.5147154656652414, + 0.43978828227177486, + 0.3241137230301435, + 0.49888892920331296, + 0.48104735594701686, + 0.4904738373973315, + 0.4942063662134963, + 0.3708658472268396, + 0.6497592306494211, + 0.6794187533945326, + 0.5528869650559285, + 0.5915325024889978, + 0.6624127152543353, + 0.5446041823343922, + 0.34912470556022235, + 0.45716934348418903, + 0.4304648631938759, + 0.40529708436357714, + 0.44448022437058177, + 0.4769101043844391, + 0.5268257275897786, + 0.48488519906588173, + 0.3593748848573296 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.3993223593094076, + "valid_tracks": 4, + "per_track_pearson": [ + 0.4206237321280911, + 0.35895853017995333, + 0.45132903355246046, + 0.3663781413771256 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.598374028316481, + "valid_tracks": 626, + "per_track_pearson": [ + 0.608146176480493, + 0.5964585341302213, + 0.601069715188783, + 0.5974543435242535, + 0.6168956369279598, + 0.6054348461790308, + 0.5273602122934263, + 0.5697563913395591, + 0.6056969981360836, + 0.5999175885941868, + 0.6025296397429195, + 0.6086240891719259, + 0.5786783480383921, + 0.5866634948239704, + 0.6089608492297526, + 0.6008445723959502, + 0.5896105202207524, + 0.5893085206766204, + 0.6161814045103642, + 0.612127706377176, + 0.6267821327038144, + 0.618601202202378, + 0.6147955581762492, + 0.5834949043891693, + 0.5844417052335492, + 0.6045156633588851, + 0.6066384091076867, + 0.587916936732922, + 0.5950809066345926, + 0.598967994590814, + 0.6031444256293096, + 0.6089894246563894, + 0.5911687990144974, + 0.619172811096679, + 0.5891835826476967, + 0.6221196886118587, + 0.5734748753996987, + 0.5996761119648178, + 0.5886474457086358, + 0.6137791985643803, + 0.61481037293609, + 0.5899247236643359, + 0.5705667997465114, + 0.5907869973106034, + 0.610703781915507, + 0.5818509842228051, + 0.6000179443716027, + 0.605931762687253, + 0.60196245621712, + 0.5963007626466053, + 0.578240169075425, + 0.5788366647987729, + 0.5985920505929363, + 0.6046371815348843, + 0.5814738116874074, + 0.5935453816928108, + 0.6053501342156062, + 0.6023382869122504, + 0.6061728789423101, + 0.6025858107557445, + 0.5901811950130517, + 0.6009528845424138, + 0.6056514411077423, + 0.5976236442559092, + 0.5908910294385646, + 0.5891467507225593, + 0.6007893317505458, + 0.6045067146892436, + 0.5871775151839417, + 0.5949563487256606, + 0.6069594242182269, + 0.5992166283954595, + 0.6040130347558303, + 0.5982531479746569, + 0.6139788856764763, + 0.614428780136381, + 0.5964029259629489, + 0.5986460421769261, + 0.5888448154571886, + 0.5895835174069293, + 0.6238954796194439, + 0.6113323337725273, + 0.60380078435539, + 0.5870853591090348, + 0.5948651190568792, + 0.5937882478745832, + 0.6035420142811905, + 0.6038071503381297, + 0.606611512914073, + 0.6015453642290379, + 0.5949185982336673, + 0.594625775221306, + 0.6026873657616535, + 0.5981575397703914, + 0.6099786360290654, + 0.6191229573989323, + 0.5984888086000664, + 0.603797383326283, + 0.6094737537766473, + 0.6044528418942019, + 0.5833205507546074, + 0.5981532272410401, + 0.6116969632373975, + 0.5920502037396703, + 0.6019339288132468, + 0.5895287274391258, + 0.5793817332621202, + 0.6043844943721002, + 0.6039976755834121, + 0.5996538762655853, + 0.5918915398390462, + 0.596939639807433, + 0.6091166865532024, + 0.599753754193039, + 0.6002459537858087, + 0.6045060700151045, + 0.5860146567642415, + 0.5846415855113212, + 0.5823209461177408, + 0.606946413771342, + 0.6152217490125869, + 0.5966584485206029, + 0.6023907621138279, + 0.5960688920966816, + 0.5900970955434711, + null, + 0.6069098134904365, + 0.5911664101964307, + 0.5852111277077364, + 0.6052380198482474, + 0.6043234242976141, + 0.5923918220179305, + 0.5998302153406894, + 0.5920418289361681, + 0.5976118507212103, + 0.5991834134148859, + null, + 0.6061395731927366, + 0.6126614518793305, + 0.5785941656288341, + 0.5877686430437505, + 0.5999193619321616, + 0.5887845509801707, + 0.6098810645937188, + 0.6108756081264802, + 0.5926340757821907, + 0.5906106849881658, + 0.6089121267368048, + 0.5915817002536194, + 0.5976297418883311, + 0.5995052354002289, + 0.6029290923616806, + 0.5919554931502099, + 0.5995989157511576, + 0.6047379683844945, + 0.5982781758158857, + 0.5931857246983728, + 0.6069086701123411, + 0.6041103209339034, + 0.6011367472173758, + 0.5988121708331834, + 0.6040276601965869, + 0.6165261261837848, + 0.6137315513882411, + 0.6027546306105835, + 0.6051573955065771, + 0.61034104611082, + 0.5864417064214393, + 0.5709194592202288, + 0.5875902395514401, + 0.6028276856569225, + 0.600580853801522, + 0.597811464414838, + 0.6022526257489649, + 0.6041941268438604, + 0.5709141918752858, + 0.60391878848329, + 0.6069160492354592, + 0.5859318251649749, + 0.5981818716392284, + 0.6012552611387476, + 0.5989131392739865, + 0.5829038861879611, + 0.6052805861569155, + 0.5906947879668576, + 0.5693207264386159, + 0.6142138812890594, + 0.6003428690670661, + 0.6140493481425744, + 0.5904188897726094, + 0.6096260685862546, + 0.5912995505094849, + 0.6110313851613008, + 0.6051516351652515, + 0.6027642640631127, + 0.6090474086083878, + 0.6047601164738177, + 0.5800691845787107, + 0.6158346817455467, + 0.6156841740220674, + null, + 0.6126390944822739, + 0.6188017914650916, + 0.6064599805643842, + 0.6134664680616864, + 0.6161429165065593, + 0.614139918781838, + 0.6017460867521786, + null, + null, + 0.608965612485588, + 0.6149876250013041, + null, + 0.5877416388313628, + 0.5673865785858307, + 0.5900532976178404, + 0.6105093777039432, + 0.6125253916200257, + 0.6180526007434328, + 0.6066978241125266, + null, + 0.6179198043601696, + 0.5970182671768397, + null, + 0.5474783103778545, + null, + 0.5788415900131683, + 0.5988833749087658, + 0.6108850121442279, + 0.6119802880671226, + null, + 0.5974438422914012, + 0.5973534167642296, + 0.6051230353553827, + null, + 0.6060653374343735, + null, + 0.5987779191251409, + null, + 0.5826009788161483, + 0.6001797929176973, + null, + 0.5942392803194811, + 0.6022305441958836, + null, + null, + 0.612812582825039, + 0.611082100850434, + 0.6068772029683088, + 0.6119408614311568, + null, + 0.6054374772368948, + null, + 0.6086740496169555, + 0.6274701571955903, + null, + 0.5033839932973447, + 0.6048321024139303, + 0.6053664782079964, + null, + 0.599580333919487, + 0.6181497149638336, + null, + 0.6159147304360245, + 0.6153244071922129, + 0.6118420048450195, + null, + null, + null, + null, + 0.6224572630252598, + null, + null, + 0.6046521257702483, + 0.6064448912973203, + 0.6159821478397789, + null, + 0.6160135278807526, + 0.6162463339777694, + null, + null, + 0.6206105990320847, + 0.6058824964734092, + 0.6066253721679203, + 0.6120995980131609, + 0.6182045834028743, + 0.614060330571896, + 0.5939249092408752, + 0.6137717002193211, + 0.605779209451566, + 0.6045995484384694, + 0.6100002875738425, + 0.6167407399087076, + null, + 0.5915082249407894, + 0.6131786844780933, + 0.6077816266037371, + 0.6051323782253725, + 0.5813474250789508, + 0.5646589010279985, + 0.6126958629388348, + 0.6000674778620437, + 0.6040077791113503, + 0.60545874044329, + 0.6089656905055054, + 0.6061494697635866, + null, + 0.5982718228206145, + 0.6253870446375029, + 0.6227952842919698, + null, + 0.6111867639134928, + 0.5228054267562879, + null, + 0.6216285275409265, + null, + 0.5982100499206711, + 0.6084142092812425, + 0.608652553994792, + 0.5894658454169405, + 0.6084490482788253, + 0.6011836675942955, + 0.5905555932855667, + null, + null, + 0.6056980305695641, + 0.5883808778470335, + null, + 0.6084053042398483, + null, + 0.6090756181108049, + 0.6044344586054311, + 0.6023421601060435, + 0.6037483535952064, + null, + null, + null, + null, + 0.615859422827987, + null, + 0.608421785021862, + 0.6157577613514172, + null, + null, + 0.6074152453845166, + 0.6188232950122876, + 0.6099082375502238, + null, + 0.6001373140891992, + 0.6001452140747338, + 0.6050000112712172, + null, + 0.6184623535079897, + null, + null, + 0.6073523912725005, + null, + 0.609004205881856, + null, + null, + 0.600147258345024, + 0.6096802591670355, + 0.6028150926474471, + 0.6060299651578638, + 0.5953761833821383, + null, + 0.5930380921424634, + 0.5989573112232698, + 0.5960516825621209, + 0.5710207855207752, + 0.5687181444900531, + 0.6056369387319176, + 0.6086273099967277, + 0.5298733240790461, + 0.6098811307544633, + 0.5855637521635169, + 0.6115939163511462, + 0.6004683113671825, + 0.5994399090561839, + 0.5983696965416964, + 0.6146675143644975, + 0.5883317265012978, + 0.5857416181860603, + 0.5987427428131074, + 0.5979573137277365, + 0.5953020808319052, + 0.5961323277712761, + 0.6163864203909242, + 0.6338426538695486, + 0.5946076852222342, + 0.5247740434454805, + 0.6115776011611596, + 0.6045627048840712, + 0.5807451979393448, + 0.5996652602837147, + 0.6020071253904613, + 0.5904071452713858, + 0.6017038589259975, + 0.5642212639660082, + 0.5287954465247254, + 0.5968506706151628, + 0.5103527719163049, + 0.6002315423765364, + 0.4953704351822811, + 0.5990974177850493, + 0.595916087802783, + 0.6199858430909977, + 0.630323349788143, + 0.5553165612273496, + 0.5617235716531549, + 0.5874929284421471, + 0.5940042845243685, + 0.5399366970145716, + 0.5671438266844908, + 0.5603463645489831, + 0.5812613095029716, + 0.5551559606117797, + 0.5742715284061852, + 0.529522245271803, + 0.5867664892838933, + 0.5815007970761055, + 0.5442864482116974, + 0.5876717465771387, + 0.612318874725412, + 0.5950249989564947, + 0.5943154195072453, + 0.606707159008578, + 0.5776293872732848, + 0.5980903596896066, + 0.5958288947069217, + 0.6168468514369112, + 0.5979528739695977, + 0.6097939265132671, + 0.609009357991758, + 0.6296716784366841, + 0.6202421647111572, + 0.5912045209696153, + 0.5963775298084646, + 0.5932010704383474, + 0.6031937533624218, + 0.6045576362242338, + 0.5963254499432522, + 0.6058728441617517, + 0.6005728364437491, + 0.5832899893806246, + 0.5819268138676162, + 0.5951927856641323, + 0.5985673104548357, + 0.5815910986273748, + 0.5938403238163408, + 0.6200633826047955, + 0.6057374527195283, + 0.5902326142768564, + 0.5840763202308873, + 0.5948153271255855, + 0.5987199829248507, + 0.589211118183584, + 0.5979161472179564, + 0.5979816813186838, + 0.6068443032465498, + 0.6081776569320797, + 0.6121872130269246, + 0.6047255383883372, + 0.6071827115089636, + 0.6018963554848936, + 0.5942709404909341, + 0.6145749931935622, + 0.6055069761060512, + 0.6189557129316093, + 0.596347064791538, + 0.60393731770929, + 0.6235224147154114, + 0.6063685310365097, + 0.5848720566414228, + 0.5927952775883081, + 0.5993973970927654, + 0.598645802974233, + 0.6019910364943476, + 0.5886200568162497, + 0.5894537429277179, + 0.5934766637345115, + 0.5999698304581719, + 0.6081583379484569, + 0.6033653484840787, + 0.6130432142416639, + 0.5945532729587124, + 0.6069303674796238, + 0.5992817918382917, + 0.6283913102124769, + 0.602237704347952, + 0.5921058852473674, + 0.6189828005646962, + null, + 0.6057037034322587, + 0.5827031477940845, + 0.6024300446230494, + 0.6101687090375968, + 0.6200304411132564, + 0.6091355802713762, + 0.6105870452310955, + 0.6309752839655975, + 0.603281200570369, + 0.6147105102119126, + null, + 0.5815251432343033, + 0.6187559519404108, + 0.6277549894918489, + 0.6137236754425389, + 0.626741125436386, + 0.6200470779151287, + 0.5960783255742856, + 0.6235201138484654, + 0.6377461784232927, + 0.6393489648849978, + 0.599078842461974, + 0.6388086954618127, + 0.6296658958540776, + 0.6321408925528763, + 0.6383469780308096, + 0.6105595222603729, + 0.5848641002025102, + 0.5554032985812553, + 0.5570326810091683, + 0.5569312793435254, + 0.6237808354296889, + 0.6068190608549069, + 0.6144826539123313, + 0.6191379289869821, + 0.6199754048668668, + 0.6073906762853363, + 0.6233310411545838, + 0.6119553813902024, + 0.6125317261400612, + 0.6017583087066196, + 0.5937262458540599, + 0.6075488333485034, + 0.610079339748194, + 0.608323805346969, + 0.6196055166393867, + 0.6187265492534253, + 0.6106371622694252, + 0.6027817424546149, + 0.6083931505144169, + 0.6020314945063239, + 0.6079536294641658, + 0.5885118722443227, + 0.5976139144741964, + 0.6135123041033061, + 0.6185976621369649, + 0.5842212250281111, + 0.6001671809307654, + 0.6108769100674644, + 0.6175858176821646, + 0.6019402707460241, + 0.5924690858535223, + 0.6200436001201365, + 0.6040050927210535, + 0.6122981296877263, + 0.6181349078890834, + 0.5727021333480727, + 0.6119202908660113, + 0.6090340453235383, + 0.6076837117946883, + 0.6144527123003022, + 0.5934087202797551, + 0.584793892922966, + 0.5992849642663954, + null, + 0.5626796511428283, + 0.6336586644546062, + 0.5801768630544527, + 0.6337589408664372, + 0.6010355956744384, + 0.5898407936920722, + 0.5794367103033202, + null, + null, + 0.5985615407155844, + 0.6265340054628618, + null, + 0.5749072269941126, + 0.5525907016716337, + 0.5751995902065624, + 0.5969585484786009, + 0.6076329189189091, + 0.6029847804685236, + 0.577032285379405, + null, + 0.5843168264617313, + 0.5948808933464662, + null, + 0.5555463324196255, + null, + 0.5735440697757253, + 0.578893999655316, + 0.5753932740317661, + 0.5945890382603887, + null, + 0.571381904415326, + 0.5685089921085037, + 0.6100268091108377, + null, + 0.5639136766233769, + null, + 0.5974292959017491, + null, + 0.5842273417309891, + 0.5871903429944134, + null, + 0.5810409577225558, + 0.5718813622194127, + null, + null, + 0.5674292088481974, + 0.630803945596375, + 0.581661685072824, + 0.5852871194688058, + null, + 0.5956319147292023, + null, + 0.5929546409362678, + 0.6188413715451911, + null, + 0.4754242531488534, + 0.6016056003182183, + 0.5741050794300268, + null, + 0.6154026149746853, + 0.6239995563798936, + null, + 0.5805768641577551, + 0.5603368164061793, + 0.582242163583849, + null, + null, + null, + null, + 0.5954480405397139, + null, + null, + 0.6108237239705909, + 0.6067720237174247, + 0.5565120378050903, + null, + 0.5711802882178583, + 0.5747953172024463, + null, + null, + 0.6076611751523254, + 0.6018346183054464, + 0.6056834583748083, + 0.6105980695574609, + 0.6208222603540595, + 0.6272028809106013, + 0.5844532864675407, + 0.6103622094252685, + 0.588600550502004, + 0.6080080420735875, + 0.6100577922704691, + 0.6079690899341419, + null, + 0.5948005936093241, + 0.5943189761562128, + 0.5772608881776438, + 0.5908377263886255, + 0.5939038692712065, + 0.5744228993977091, + 0.5957764867954349, + 0.5687491416517878, + 0.564727272139604, + 0.6085261262191857, + 0.6083136165026138, + 0.6036861809639461, + null, + 0.6045129502746214, + 0.5863575495766402, + 0.5961206870461815, + null, + 0.6075602983608966, + 0.5305484719897923, + null, + 0.6191611111167445, + null, + 0.5610923766024047, + 0.6133152500725133, + 0.6028809539363694, + 0.5805560426199334, + 0.5906786841374937, + 0.5997689570675604, + 0.5968171521132646, + null, + null, + 0.5824079840719594, + 0.5786209699294731, + null, + 0.5935008986535363, + null, + 0.6088391956712784, + 0.591231869824144, + 0.6147538710690826, + 0.5750647003965895, + null, + null, + null, + null, + 0.6277163520548088, + null, + 0.5968334072288175, + 0.6060698711326432, + null, + null, + 0.6213218277236289, + 0.6051751722126111, + 0.6013854046999754, + null, + 0.6193079441071835, + 0.6199618324019188, + 0.5836952896996488, + null, + 0.605873976949835, + null, + null, + 0.6253765647794858, + null, + 0.5915966868991243, + null, + null, + 0.5942500872132964, + 0.5624618582275055, + 0.564427629590653, + 0.5648465104674445, + 0.5704599240753134, + null, + 0.571510125844622 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985, + 0.45277463193657985 + ] + }, + "junctions": { + "mean_pearson": 0.3928286582500866, + "valid_tracks": 626, + "per_track_pearson": [ + 0.4527441908418515, + 0.38845916220935517, + 0.43034461477242697, + 0.40121768041497086, + 0.48131973098822856, + 0.4905981707256096, + 0.1184132377935214, + 0.34287812887716396, + 0.40691867909898305, + 0.3445580865587401, + 0.46814244875547295, + 0.3766961000137205, + 0.31427381251439873, + 0.4068351566805675, + 0.37714117521132334, + 0.3653645010072552, + 0.38485868495362247, + 0.3319536943818701, + 0.37844698595596576, + 0.4762508843885985, + 0.3449277167702112, + 0.38792988817119173, + 0.34373876545906873, + 0.4203091289012534, + 0.37043949444531404, + 0.31038073574012426, + 0.40286178580401116, + 0.34910791661598994, + 0.3736872450537137, + 0.3980563084582112, + 0.4430171764819866, + 0.480779988989803, + 0.448901575995272, + 0.43279466125242455, + 0.44387117065694814, + 0.3957220490647875, + 0.4321054865882123, + 0.37974251537221415, + 0.5128979361494482, + 0.35170414395738586, + 0.4907663088471432, + 0.3862741285193922, + 0.3946977623984719, + 0.4545915681239546, + 0.3657831909383291, + 0.3591775384248519, + 0.3125096187623992, + 0.3959285375006265, + 0.3852945531321564, + 0.3549100141488028, + 0.41340760915615793, + 0.3191001417284229, + 0.37241251701832756, + 0.38899185275436277, + 0.3673078344389809, + 0.40489663283747523, + 0.32166041254754546, + 0.2884441976692649, + 0.3004120859025106, + 0.394153713610745, + 0.2423189238105024, + 0.2720658962702204, + 0.31117510140673255, + 0.2964149347456505, + 0.3931466118620344, + 0.39494623512959165, + 0.4167303719369073, + 0.3838537964183078, + 0.4046135081117089, + 0.3361131509352867, + 0.3038135936254989, + 0.36526715643103047, + 0.3099351678717382, + 0.2898030655190696, + 0.4262192349991798, + 0.34141862847534765, + 0.3406263455331404, + 0.315325804459155, + 0.33995105782597046, + 0.4101141449234315, + 0.4047139490011297, + 0.3510135571986315, + 0.3422598361550024, + 0.4632464793553572, + 0.4545149172831998, + 0.27302187914486453, + 0.3337494317827216, + 0.3486679234959556, + 0.28655278760140485, + 0.3504611803683571, + 0.35217299138007974, + 0.43673526872244955, + 0.30158320558115753, + 0.20001075357891743, + 0.3844778504834223, + 0.29620543018218354, + 0.3954966539752896, + 0.4421509874032246, + 0.4109646842860228, + 0.3637231849264277, + 0.3363126484255959, + 0.49959523890815766, + 0.39589771728046075, + 0.31650450758720056, + 0.3505311741120631, + 0.30761743621059773, + 0.32020868890924814, + 0.41147224337190863, + 0.31817988630415245, + 0.34738830558749484, + 0.3828331119060528, + 0.39155288311087605, + 0.4148337940178971, + 0.37742778137016614, + 0.40752017559863335, + 0.41046652835652603, + 0.32808147183600544, + 0.42093796267279704, + 0.41451566084594227, + 0.34397550628721796, + 0.3168526358475339, + 0.3675029824035938, + 0.400136610561671, + 0.31679632988666884, + 0.40204003047703024, + null, + 0.3536689318073008, + 0.3561201236063614, + 0.2052706709139875, + 0.32584317481015057, + 0.38226916927966637, + 0.2848575362800783, + 0.41756386560812914, + 0.44999864514876786, + 0.3821056118689558, + 0.4530795996517771, + null, + 0.374763938012422, + 0.41598454767585924, + 0.4033529966451197, + 0.4745206403486224, + 0.40510023000826695, + 0.39923069563466607, + 0.3540541681120861, + 0.35709502123703013, + 0.18087338709882514, + 0.39162330027579123, + 0.3994568630529597, + 0.3242046864967938, + 0.30008763993516785, + 0.43946622208751335, + 0.41126468269576166, + 0.30372606189446355, + 0.4293599476056833, + 0.5121415113771126, + 0.49817638496997796, + 0.3846839460886622, + 0.2919716907194573, + 0.3796691238795976, + 0.23762427096703126, + 0.44454661515015487, + 0.3137170156203149, + 0.40470666414286866, + 0.3779709072223098, + 0.3133085998531449, + 0.37826581284971067, + 0.45945538849292333, + 0.29246115009400975, + 0.3500467769171543, + 0.3582235593400881, + 0.37943550627860057, + 0.3730906453628703, + 0.4940867897408681, + 0.28508366062113527, + 0.3186297583888621, + 0.48560935016174583, + 0.24200949373443323, + 0.3014619939694287, + 0.2848129131679566, + 0.43786314703747825, + 0.3178523906419888, + 0.3295382415330988, + 0.4410978041103534, + 0.4624155285430426, + 0.3306146717317934, + 0.38903377361551766, + 0.4417918835028518, + 0.4797162308574257, + 0.4908128655783554, + 0.35650271552962476, + 0.4072492889053302, + 0.3837879467750903, + 0.39159599319954264, + 0.5081748070546617, + 0.5198705054497343, + 0.5122124331067535, + 0.39148038548735964, + 0.31775827107536, + 0.361205020285314, + 0.3550391404575086, + null, + 0.3888996297798193, + 0.5243508421867593, + 0.48110792972894, + 0.43922729244405656, + 0.3246081742312186, + 0.33802496533886306, + 0.33815668112752834, + null, + null, + 0.3091310249098216, + 0.5212756679766238, + null, + 0.3764946047830942, + 0.4668669035520589, + 0.25745984910472175, + 0.337312755492817, + 0.4653838302872985, + 0.4346434760978952, + 0.260045869298966, + null, + 0.45313755833445674, + 0.4609798953000277, + null, + 0.39162701519079124, + null, + 0.38946696733706426, + 0.41961380650647995, + 0.30515373322779477, + 0.3771072465211954, + null, + 0.2591427527390417, + 0.316070101553211, + 0.40367795639687376, + null, + 0.37655780486979235, + null, + 0.4182572343119652, + null, + 0.21091097260703914, + 0.4180067754967069, + null, + 0.3544058802915321, + 0.415366049007666, + null, + null, + 0.3393812809043826, + 0.3954799527377234, + 0.3156687778671718, + 0.3246335576935738, + null, + 0.5126097960083148, + null, + 0.43413330922727705, + 0.40687824769351644, + null, + 0.4056750480705867, + 0.4877306113789765, + 0.47398811488283416, + null, + 0.4294598442372999, + 0.39513415342285846, + null, + 0.5381348259664407, + 0.5336323430192239, + 0.4553204039766447, + null, + null, + null, + null, + 0.3326226270313368, + null, + null, + 0.32410711388410707, + 0.3008122042481019, + 0.49304007075477646, + null, + 0.4475007025976551, + 0.4386087672333695, + null, + null, + 0.4154733577712666, + 0.40877086559407316, + 0.36077524163555247, + 0.44997524637949715, + 0.46352056344132414, + 0.4494671684736642, + 0.4298748004055077, + 0.40325277433344064, + 0.4358178585903498, + 0.33496352928156126, + 0.4256188962522952, + 0.5029036121216444, + null, + 0.3972186034218545, + 0.5124354581684674, + 0.2624420984992222, + 0.3320890197566269, + 0.49340203761082596, + 0.36421752362847665, + 0.35756471428248415, + 0.3302087755823868, + 0.4191519043896539, + 0.29809846072743745, + 0.4767899349910027, + 0.4250995637602032, + null, + 0.43434456452968895, + 0.30117781903337876, + 0.3424447039543272, + null, + 0.3571011790873638, + 0.33972617539317296, + null, + 0.516927099409485, + null, + 0.2934298486252027, + 0.42954433267684317, + 0.3551672188735399, + 0.3363058345939374, + 0.3434179031593132, + 0.42504537885228166, + 0.40275607851270406, + null, + null, + 0.37792844349811155, + 0.38606241575296396, + null, + 0.44461801595343625, + null, + 0.39738779773259275, + 0.40931897818263774, + 0.33125264815240835, + 0.4549044787762007, + null, + null, + null, + null, + 0.31621618397766976, + null, + 0.3556668542618528, + 0.40861991277472537, + null, + null, + 0.48536023643675924, + 0.3075638399980391, + 0.3574783773375803, + null, + 0.5822310270596636, + 0.40434887469235037, + 0.36477793201239067, + null, + 0.428534315484276, + null, + null, + 0.4142026722834832, + null, + 0.2918543870485416, + null, + null, + 0.3378865528687291, + 0.38378431210369723, + 0.3735272973572075, + 0.39982220452408157, + 0.3767828191253859, + null, + 0.3695020792858385, + 0.41066565461002436, + 0.35927262435931906, + 0.6229755545603336, + 0.3761966027889692, + 0.5588559173764732, + 0.40323152471269835, + 0.3207979477087179, + 0.25775373777565214, + 0.4094935405111982, + 0.44438605175447443, + 0.30124624666616523, + 0.42192096223385583, + 0.3825349391929841, + 0.32281538798085835, + 0.33509326707637005, + 0.3335429911585616, + 0.36134055722340114, + 0.5226657325097116, + 0.5167053313070813, + 0.4758259201901521, + 0.49002715316041623, + 0.47995244703599854, + 0.3958615220564984, + 0.5994995373906236, + 0.2989663818031945, + 0.3590526268408857, + 0.4802862477124086, + 0.4641631221076135, + 0.41343380700674537, + 0.27537061542243413, + 0.3607555179314541, + 0.4622632488584313, + 0.40196940133501413, + 0.5784069799067804, + 0.22159587643549122, + 0.428944899922485, + 0.17786433691463582, + 0.3942998750374789, + 0.5729979238702058, + 0.40972712598457306, + 0.41760314884962285, + 0.2920128257929436, + 0.2269505081664603, + 0.23590352571799994, + 0.3645476110328778, + 0.30602738542095254, + 0.294746612532595, + 0.46211844672598984, + 0.23739577739006879, + 0.34438677821370506, + 0.22012640389996008, + 0.30921535318079485, + 0.22857353147291654, + 0.27998576047220664, + 0.23256700504812797, + 0.3867063635587347, + 0.3967732145090445, + 0.3820395980247717, + 0.33933681829229606, + 0.33631112271422736, + 0.43391169657839107, + 0.3119712773241094, + 0.4566904141988454, + 0.36566331683736114, + 0.3658112438249477, + 0.28891636346651794, + 0.43497797525209747, + 0.3884551111950295, + 0.276523487279134, + 0.34382487332305456, + 0.3998639608995414, + 0.3795241641798562, + 0.39726029069666385, + 0.46231427821072063, + 0.3830078093415699, + 0.39836258430402105, + 0.4314168117628092, + 0.519855959736536, + 0.416261520018821, + 0.4437894583815828, + 0.38831233619618044, + 0.43131191245013756, + 0.44672037546745474, + 0.3627920698370947, + 0.3693120944610987, + 0.3842604647827393, + 0.49778896999696176, + 0.46414782848035135, + 0.3765578798138976, + 0.41888253713529205, + 0.48193507151676274, + 0.4053167406704493, + 0.4013995474340343, + 0.3787099820356777, + 0.3346618543541107, + 0.3168009184206479, + 0.4312209092642642, + 0.3663051940390614, + 0.46518999015750817, + 0.509843014680635, + 0.24118253573185397, + 0.3651122762361189, + 0.37548802776326523, + 0.3530539779269945, + 0.44011921838608464, + 0.38917023958763597, + 0.3219259726171402, + 0.4512206625865072, + 0.4402247306089381, + 0.446889968447482, + 0.32014137184879565, + 0.45015610772790027, + 0.3291747821686222, + 0.4121342485684693, + 0.4369845517429804, + 0.31230977236283863, + 0.479050658361934, + 0.2358709336547351, + 0.35956019428034475, + 0.3047978318944363, + 0.527453248730349, + 0.372750468639382, + 0.33748338739172024, + 0.37734804840697694, + 0.5241775354666501, + null, + 0.3991979514216444, + 0.5091266977691157, + 0.28736162953462613, + 0.4783201413922549, + 0.4313040538417444, + 0.30255659298486437, + 0.31337492830539326, + 0.4307354798619232, + 0.40790052986450304, + 0.29047590349417884, + null, + 0.3474305372360699, + 0.29697790092529297, + 0.23425269878318097, + 0.2596046522244266, + 0.3966974606710215, + 0.3286258439362009, + 0.36381043710672495, + 0.4180444142849707, + 0.21415272613194333, + 0.274865882744851, + 0.38146663666174446, + 0.3527265460653647, + 0.3631893805063784, + 0.40963813595237125, + 0.5198370899017483, + 0.40356182202557855, + 0.38469616354751757, + 0.5530703189436209, + 0.47954203439551374, + 0.40174075021501476, + 0.37394638326942886, + 0.37991718459450113, + 0.26736442103373004, + 0.3285998919751843, + 0.28324451494827685, + 0.5094012065420839, + 0.5613750323457843, + 0.37268712173929275, + 0.32651391701985044, + 0.46732973734907984, + 0.45345759834479044, + 0.35685031681664525, + 0.4821188760244813, + 0.3459234921917584, + 0.36399991494864353, + 0.38834297089669, + 0.28037355804701963, + 0.3197278275011466, + 0.3759513630752909, + 0.4206283155117878, + 0.3332138032650315, + 0.30586292146338695, + 0.40332780106285493, + 0.33918901441033444, + 0.30109822148401555, + 0.5408267962455497, + 0.5247612109950806, + 0.4504988510997983, + 0.5507051094350607, + 0.5629047031082, + 0.2926759095534717, + 0.49074973473654226, + 0.4042411356467455, + 0.558752954184127, + 0.41557490960810217, + 0.37037028285449664, + 0.27772380356737136, + 0.3588439517027293, + 0.32607823478298925, + 0.3325571361148629, + 0.4758490312138629, + 0.4734709226272129, + 0.5724690529505309, + null, + 0.40149548376058836, + 0.4622001155045172, + 0.5652718399698639, + 0.5458618086974237, + 0.4573134182242307, + 0.2774681382033885, + 0.4063658936770647, + null, + null, + 0.4527468198451554, + 0.5151682016029275, + null, + 0.21501917006302187, + 0.5007524147248893, + 0.22186536535004958, + 0.35475827595275217, + 0.3560478928759996, + 0.3624914572837944, + 0.3774108486209099, + null, + 0.47318565765018206, + 0.3718663158043465, + null, + 0.44382297257083686, + null, + 0.5862947265174557, + 0.4228631485948293, + 0.3011534359418496, + 0.3495722266270327, + null, + 0.43976543871685386, + 0.3540356851958721, + 0.4955198664355075, + null, + 0.4001091925647389, + null, + 0.37461480470497394, + null, + 0.4237490762208619, + 0.4574070973623907, + null, + 0.43347956657991965, + 0.4703598825681998, + null, + null, + 0.4419048412429747, + 0.40148063909531223, + 0.2214777281418153, + 0.3301032581105701, + null, + 0.49687077000272595, + null, + 0.3164818191789209, + 0.47174814168679385, + null, + 0.22277656500537832, + 0.5320477426651435, + 0.44989079105644825, + null, + 0.43626105650156805, + 0.5276251965996169, + null, + 0.48499641088043444, + 0.5173623678933756, + 0.501516065343592, + null, + null, + null, + null, + 0.481525088110646, + null, + null, + 0.5105373611133895, + 0.47301946997750643, + 0.5429213863128292, + null, + 0.35579396081080317, + 0.5257025671832107, + null, + null, + 0.36190038166292793, + 0.1338746831869994, + 0.4441884451209664, + 0.2612698829936252, + 0.3318879898824561, + 0.4502895929702552, + 0.22440385826452078, + 0.34292741377371, + 0.23479830939299612, + 0.37556941996089266, + 0.46193418809427256, + 0.34421024279314383, + null, + 0.3249458777659795, + 0.5123564676845568, + 0.5159448276138863, + 0.4679410152711867, + 0.24163151714836628, + 0.4768292238937709, + 0.4215167014943634, + 0.42514962708305565, + 0.48535760149059615, + 0.41416430590766057, + 0.4861189943410357, + 0.4071644273881027, + null, + 0.414374023400312, + 0.49306700543376253, + 0.43490520945919886, + null, + 0.548828044970759, + 0.4156149767997362, + null, + 0.43302156935127334, + null, + 0.5357128234894113, + 0.5109872203981902, + 0.32076942237151507, + 0.5042728102060509, + 0.4983046580874034, + 0.5663983996719616, + 0.5643974993732676, + null, + null, + 0.47800079983400606, + 0.5983708109111786, + null, + 0.4207005345137328, + null, + 0.4463162075880367, + 0.46532842385872447, + 0.4850942238558119, + 0.45682501467181486, + null, + null, + null, + null, + 0.36694959126716015, + null, + 0.44430746918814756, + 0.5916714222304109, + null, + null, + 0.49664569264858716, + 0.2709335053072724, + 0.3696610316253299, + null, + 0.397990261442124, + 0.3819516262397864, + 0.4192737097337627, + null, + 0.37601924355028, + null, + null, + 0.42239702588243766, + null, + 0.30808786491370294, + null, + null, + 0.4464514765002175, + 0.5417493476910437, + 0.5770088184399342, + 0.6034881238158256, + 0.48877151209488084, + null, + 0.3933249962912132 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.7284768211920529, + 0.7828947368421053, + 0.7224199288256228, + 0.7151898734177216, + 0.7682119205298014, + 0.7254237288135593, + 0.721830985915493, + 0.7291666666666666, + 0.7435897435897436, + 0.7198581560283688, + 0.7407407407407407, + 0.7050847457627119, + 0.7207792207792207, + 0.746875, + 0.7766666666666666, + 0.756578947368421, + 0.7237762237762237, + 0.7211538461538461, + 0.7515923566878981, + 0.7418300653594772, + 0.7122302158273381, + 0.7387096774193549, + 0.6915584415584416, + 0.7360594795539034, + 0.7188612099644128, + 0.7532894736842105, + 0.782051282051282, + 0.7593220338983051, + 0.75, + 0.7517241379310344, + 0.7054794520547946, + 0.7184466019417476, + 0.7388059701492538, + 0.756838905775076, + 0.717948717948718, + 0.7192429022082019, + 0.7061068702290076, + 0.7366771159874608, + 0.6888111888111889, + 0.7449664429530202, + 0.7372013651877133, + 0.7731958762886598, + 0.7442748091603053, + 0.7555555555555555, + 0.7213114754098361, + 0.722972972972973, + 0.7269372693726938, + 0.7168458781362007, + 0.6985815602836879, + 0.7482993197278912, + 0.7262773722627737, + 0.75, + 0.7121212121212122, + 0.7448979591836735, + 0.7313432835820896, + 0.6836734693877551, + 0.7058823529411765, + 0.7748344370860927, + 0.7226027397260274, + 0.7183098591549296, + 0.7034482758620689, + 0.7540983606557377, + 0.7417218543046358, + 0.7474402730375427, + 0.7114093959731543, + 0.7658227848101266, + 0.7027027027027027, + 0.7348993288590604, + 0.7167235494880546, + 0.7843137254901961, + 0.7437722419928826, + 0.7225806451612903, + 0.7407407407407407, + 0.7549668874172185, + 0.7455197132616488, + 0.7593220338983051, + 0.7652733118971061, + 0.745819397993311, + 0.7448275862068966, + 0.7653061224489796, + 0.7350993377483444, + 0.7563291139240507, + 0.7671232876712328, + 0.7605177993527508, + 0.7578947368421053, + 0.7523219814241486, + 0.75, + 0.7516339869281046, + 0.7571428571428571, + 0.7563291139240507, + 0.7456445993031359, + 0.7594501718213058, + 0.7649122807017544, + 0.7448979591836735, + 0.7331081081081081, + 0.7326388888888888, + 0.7661016949152543, + 0.7197231833910035, + 0.7615894039735099, + 0.7468354430379747, + 0.7166666666666667, + 0.7639344262295082, + 0.7615658362989324, + 0.7699680511182109, + 0.7372013651877133, + 0.77, + 0.6966292134831461, + 0.782608695652174, + 0.760797342192691, + 0.7398648648648649, + 0.7285714285714285, + 0.7550335570469798, + 0.762987012987013, + 0.7288135593220338, + 0.7346938775510204, + 0.7533783783783784, + 0.7206896551724138, + 0.7263513513513513, + 0.7364620938628159, + 0.7610921501706485, + 0.7214285714285714, + 0.7457627118644068, + 0.7279411764705882, + 0.756578947368421, + 0.7125506072874493, + 0.9919354838709677, + 0.7647058823529411, + 0.7247386759581882, + 0.7462121212121212, + 0.753731343283582, + 0.7246376811594203, + 0.7655172413793103, + 0.7386759581881533, + 0.7105263157894737, + 0.7473684210526316, + 0.725752508361204, + 0.9838709677419355, + 0.7389830508474576, + 0.7126865671641791, + 0.7697368421052632, + 0.7700729927007299, + 0.6940298507462687, + 0.7279411764705882, + 0.7275862068965517, + 0.7263157894736842, + 0.7099236641221374, + 0.7047244094488189, + 0.7222222222222222, + 0.6727941176470589, + 0.7420494699646644, + 0.6962962962962963, + 0.7340425531914894, + 0.6996336996336996, + 0.75, + 0.7586206896551724, + 0.7457044673539519, + 0.7234848484848485, + 0.7725631768953068, + 0.7314487632508834, + 0.7153284671532847, + 0.7083333333333334, + 0.7361111111111112, + 0.7586206896551724, + 0.7073170731707317, + 0.7169811320754716, + 0.7410071942446043, + 0.7439446366782007, + 0.7715231788079471, + 0.6981132075471698, + 0.7228070175438597, + 0.7216117216117216, + 0.7098976109215017, + 0.7481203007518797, + 0.7636986301369864, + 0.7461538461538462, + 0.7431906614785992, + 0.7318840579710145, + 0.7212543554006968, + 0.6985294117647058, + 0.7307692307692307, + 0.735191637630662, + 0.7388059701492538, + 0.7084870848708487, + 0.6989247311827957, + 0.7676056338028169, + 0.7822878228782287, + 0.726027397260274, + 0.7306397306397306, + 0.7132616487455197, + 0.7901639344262295, + 0.7256944444444444, + 0.7720588235294118, + 0.7178571428571429, + 0.6866197183098591, + 0.7003484320557491, + 0.6607773851590106, + 0.6764705882352942, + 0.7291666666666666, + 0.738562091503268, + 0.7466666666666667, + 0.9612403100775194, + 0.75, + 0.7220338983050848, + 0.6843853820598007, + 0.7466216216216216, + 0.7588424437299035, + 0.7279151943462897, + 0.6875, + 1.0, + 0.9621212121212122, + 0.7623762376237624, + 0.75, + 0.9769230769230769, + 0.7034482758620689, + 0.6117216117216118, + 0.6824817518248175, + 0.6996587030716723, + 0.7251655629139073, + 0.7172413793103448, + 0.7114093959731543, + 0.967479674796748, + 0.7391304347826086, + 0.738831615120275, + 1.0, + 0.6328125, + 0.9859154929577465, + 0.6763636363636364, + 0.7403508771929824, + 0.7147887323943662, + 0.7171717171717171, + 0.9782608695652174, + 0.6992753623188406, + 0.6936619718309859, + 0.7298245614035088, + 0.968503937007874, + 0.6767676767676768, + 0.9846153846153847, + 0.7272727272727273, + 0.9915254237288136, + 0.7166666666666667, + 0.7181208053691275, + 0.9781021897810219, + 0.6988847583643123, + 0.7278688524590164, + 0.9689922480620154, + 0.96875, + 0.7375886524822695, + 0.7516339869281046, + 0.7254901960784313, + 0.696969696969697, + 0.984, + 0.7137809187279152, + 0.9612403100775194, + 0.7207792207792207, + 0.7044673539518901, + 0.9791666666666666, + 0.6884615384615385, + 0.707395498392283, + 0.6945454545454546, + 0.9710144927536232, + 0.7035714285714286, + 0.7391304347826086, + 0.9834710743801653, + 0.7344262295081967, + 0.7087719298245614, + 0.735593220338983, + 0.9626865671641791, + 0.9844961240310077, + 0.9612403100775194, + 0.9833333333333333, + 0.7275747508305648, + 0.9716312056737588, + 0.9763779527559056, + 0.7312925170068028, + 0.6985815602836879, + 0.7147435897435898, + 0.9916666666666667, + 0.6996587030716723, + 0.7191780821917808, + 1.0, + 0.9846153846153847, + 0.7421602787456446, + 0.6847826086956522, + 0.7190635451505016, + 0.7195945945945946, + 0.7337883959044369, + 0.6904761904761905, + 0.7133105802047781, + 0.698961937716263, + 0.6711864406779661, + 0.7188612099644128, + 0.7707006369426752, + 0.7158671586715867, + 0.9850746268656716, + 0.7263157894736842, + 0.706081081081081, + 0.6964285714285714, + 0.7239057239057239, + 0.7278911564625851, + 0.7222222222222222, + 0.712280701754386, + 0.6827586206896552, + 0.7213114754098361, + 0.7094594594594594, + 0.7028985507246377, + 0.7030716723549488, + 0.983739837398374, + 0.7427652733118971, + 0.7398648648648649, + 0.7628205128205128, + 1.0, + 0.7156549520766773, + 0.6881720430107527, + 0.9669421487603306, + 0.711340206185567, + 0.9925373134328358, + 0.7339743589743589, + 0.7256944444444444, + 0.7326732673267327, + 0.6666666666666666, + 0.7346938775510204, + 0.6986301369863014, + 0.7097902097902098, + 0.9858156028368794, + 1.0, + 0.7124183006535948, + 0.6802721088435374, + 0.9701492537313433, + 0.7111913357400722, + 0.9923076923076923, + 0.7275747508305648, + 0.7046979865771812, + 0.6690909090909091, + 0.7242524916943521, + 0.9714285714285714, + 0.984375, + 0.96, + 0.9761904761904762, + 0.6952054794520548, + 0.9708029197080292, + 0.7093425605536332, + 0.765079365079365, + 0.9826086956521739, + 0.9858156028368794, + 0.7359154929577465, + 0.7009966777408638, + 0.7222222222222222, + 0.992, + 0.7226027397260274, + 0.7003367003367004, + 0.6773049645390071, + 0.9838709677419355, + 0.7770700636942676, + 1.0, + 0.9689922480620154, + 0.726962457337884, + 0.9672131147540983, + 0.7055016181229773, + 0.984251968503937, + 0.9922480620155039, + 0.667953667953668, + 0.7207792207792207, + 0.7319587628865979, + 0.7044673539518901, + 0.6936619718309859, + 0.9922480620155039, + 0.7185430463576159, + 0.7142857142857143, + 0.7228070175438597, + 0.75, + 0.7428571428571429, + 0.8164556962025317, + 0.6868327402135231, + 0.710801393728223, + 0.6415094339622641, + 0.7337883959044369, + 0.7266187050359713, + 0.740484429065744, + 0.7075812274368231, + 0.7363013698630136, + 0.7294520547945206, + 0.7360594795539034, + 0.6996336996336996, + 0.7007042253521126, + 0.6777777777777778, + 0.7178571428571429, + 0.7201365187713311, + 0.7132867132867133, + 0.734982332155477, + 0.782608695652174, + 0.7368421052631579, + 0.6857142857142857, + 0.7418181818181818, + 0.7284768211920529, + 0.7804878048780488, + 0.7455197132616488, + 0.6679841897233202, + 0.743859649122807, + 0.7463235294117647, + 0.717741935483871, + 0.7553191489361702, + 0.7252747252747253, + 0.7836879432624113, + 0.6972111553784861, + 0.6773049645390071, + 0.7062937062937062, + 0.7194244604316546, + 0.6654929577464789, + 0.7773722627737226, + 0.7455830388692579, + 0.7058823529411765, + 0.7319587628865979, + 0.7446043165467626, + 0.7255639097744361, + 0.732824427480916, + 0.7283018867924528, + 0.7076923076923077, + 0.714859437751004, + 0.7472118959107806, + 0.696969696969697, + 0.7651245551601423, + 0.7180451127819549, + 0.7211895910780669, + 0.7769784172661871, + 0.7383512544802867, + 0.701067615658363, + 0.677536231884058, + 0.7560975609756098, + 0.7453183520599251, + 0.764505119453925, + 0.7289377289377289, + 0.7220216606498195, + 0.7389705882352942, + 0.7107142857142857, + 0.6447876447876448, + 0.7034220532319392, + 0.7060931899641577, + 0.7408759124087592, + 0.6847826086956522, + 0.7661016949152543, + 0.7096774193548387, + 0.7132616487455197, + 0.717391304347826, + 0.7103448275862069, + 0.7137809187279152, + 0.7266187050359713, + 0.720754716981132, + 0.726643598615917, + 0.7191780821917808, + 0.6920152091254753, + 0.7, + 0.7269503546099291, + 0.7426470588235294, + 0.7366412213740458, + 0.75, + 0.6788321167883211, + 0.7142857142857143, + 0.7526132404181185, + 0.697508896797153, + 0.7206896551724138, + 0.7323420074349443, + 0.6701754385964912, + 0.6953405017921147, + 0.7058823529411765, + 0.6802973977695167, + 0.7226027397260274, + 0.7060931899641577, + 0.6896551724137931, + 0.708029197080292, + 0.7391304347826086, + 0.7428571428571429, + 0.7210144927536232, + 0.7256944444444444, + 0.7169811320754716, + 0.7444444444444445, + 0.7052631578947368, + 0.708185053380783, + 0.704225352112676, + 0.7265917602996255, + 0.7251655629139073, + 0.7168458781362007, + 0.7275985663082437, + 0.7038461538461539, + 0.6996466431095406, + 0.6653846153846154, + 0.6877470355731226, + 0.6863468634686347, + 0.712280701754386, + 0.7096774193548387, + 0.6959706959706959, + 0.7385159010600707, + 0.6877323420074349, + 0.9916666666666667, + 0.7446808510638298, + 0.7357142857142858, + 0.7624113475177305, + 0.7233201581027668, + 0.7328519855595668, + 0.7412587412587412, + 0.7243816254416962, + 0.6992481203007519, + 0.7526132404181185, + 0.6984732824427481, + 0.9719626168224299, + 0.7421602787456446, + 0.6654411764705882, + 0.7972027972027972, + 0.8118466898954704, + 0.6914498141263941, + 0.6950354609929078, + 0.6853146853146853, + 0.6843971631205674, + 0.6617100371747212, + 0.6977611940298507, + 0.6741573033707865, + 0.6750902527075813, + 0.6959706959706959, + 0.6461538461538462, + 0.6568265682656826, + 0.7132616487455197, + 0.7343173431734318, + 0.7320754716981132, + 0.7689655172413793, + 0.7435897435897436, + 0.7593984962406015, + 0.7263513513513513, + 0.7296296296296296, + 0.6884057971014492, + 0.6923076923076923, + 0.7766990291262136, + 0.7027972027972028, + 0.704119850187266, + 0.6968503937007874, + 0.708904109589041, + 0.7526881720430108, + 0.6448979591836734, + 0.6840148698884758, + 0.7282608695652174, + 0.6777777777777778, + 0.6976744186046512, + 0.7243816254416962, + 0.7158671586715867, + 0.7195121951219512, + 0.7269503546099291, + 0.725, + 0.7097902097902098, + 0.7153284671532847, + 0.7073170731707317, + 0.7286245353159851, + 0.6654411764705882, + 0.6959706959706959, + 0.7570422535211268, + 0.7325102880658436, + 0.7100371747211895, + 0.7194244604316546, + 0.676056338028169, + 0.7428571428571429, + 0.7525773195876289, + 0.7047970479704797, + 0.7385159010600707, + 0.6715867158671587, + 0.6810035842293907, + 0.6762589928057554, + 0.6619718309859155, + 0.6884615384615385, + 0.8, + 0.7542662116040956, + 0.9914529914529915, + 0.7624113475177305, + 0.7575757575757576, + 0.6753731343283582, + 0.7142857142857143, + 0.773972602739726, + 0.7414965986394558, + 0.696113074204947, + 0.9761904761904762, + 0.9666666666666667, + 0.8275862068965517, + 0.7947882736156352, + 0.952, + 0.7192982456140351, + 0.6643598615916955, + 0.6807692307692308, + 0.7303754266211604, + 0.7247386759581882, + 0.7153024911032029, + 0.7385159010600707, + 0.9609375, + 0.7455197132616488, + 0.6962962962962963, + 0.9923076923076923, + 0.6417910447761194, + 1.0, + 0.6813559322033899, + 0.7508896797153025, + 0.7601351351351351, + 0.7052631578947368, + 0.9555555555555556, + 0.6959706959706959, + 0.7192982456140351, + 0.6875, + 0.9922480620155039, + 0.7046263345195729, + 1.0, + 0.696113074204947, + 0.9603174603174603, + 0.7228464419475655, + 0.7090909090909091, + 1.0, + 0.6909090909090909, + 0.7419354838709677, + 0.9925925925925926, + 0.952, + 0.7586206896551724, + 0.7420494699646644, + 0.756578947368421, + 0.7441077441077442, + 0.9915254237288136, + 0.6809338521400778, + 1.0, + 0.7056737588652482, + 0.7276119402985075, + 0.9761904761904762, + 0.6346863468634686, + 0.7237762237762237, + 0.6728624535315985, + 0.9767441860465116, + 0.7137681159420289, + 0.7429577464788732, + 0.9831932773109243, + 0.726962457337884, + 0.7224080267558528, + 0.7419354838709677, + 0.9632352941176471, + 0.9838709677419355, + 0.9920634920634921, + 0.990990990990991, + 0.754325259515571, + 0.9908256880733946, + 0.9724770642201835, + 0.7411347517730497, + 0.7551020408163265, + 0.7601351351351351, + 0.9844961240310077, + 0.7412587412587412, + 0.7417218543046358, + 0.9831932773109243, + 0.96, + 0.7370242214532872, + 0.7560137457044673, + 0.7236363636363636, + 0.7557755775577558, + 0.7368421052631579, + 0.718978102189781, + 0.6764705882352942, + 0.7517006802721088, + 0.6867924528301886, + 0.7094339622641509, + 0.7661016949152543, + 0.6985815602836879, + 1.0, + 0.7534722222222222, + 0.6989247311827957, + 0.7517006802721088, + 0.743859649122807, + 0.6958041958041958, + 0.7097902097902098, + 0.7157534246575342, + 0.7289377289377289, + 0.7483221476510067, + 0.7243816254416962, + 0.7111111111111111, + 0.6909090909090909, + 0.968503937007874, + 0.7392857142857143, + 0.7357142857142858, + 0.7866666666666666, + 0.983739837398374, + 0.7508650519031141, + 0.725752508361204, + 0.9590163934426229, + 0.7311475409836066, + 1.0, + 0.7508305647840532, + 0.7366548042704626, + 0.7668918918918919, + 0.6784313725490196, + 0.7431506849315068, + 0.6902985074626866, + 0.6768060836501901, + 0.9754098360655737, + 0.983739837398374, + 0.7323943661971831, + 0.6917562724014337, + 0.9844961240310077, + 0.7117437722419929, + 1.0, + 0.7337883959044369, + 0.7058823529411765, + 0.7027972027972028, + 0.7279411764705882, + 0.9562043795620438, + 0.9921259842519685, + 1.0, + 0.9906542056074766, + 0.7183098591549296, + 0.9826086956521739, + 0.6884615384615385, + 0.7912457912457912, + 0.959349593495935, + 1.0, + 0.7311827956989247, + 0.7403508771929824, + 0.7071428571428572, + 0.9831932773109243, + 0.7275985663082437, + 0.7212543554006968, + 0.726962457337884, + 1.0, + 0.775974025974026, + 0.983739837398374, + 0.9917355371900827, + 0.7454545454545455, + 0.9833333333333333, + 0.7018867924528301, + 0.963302752293578, + 0.992, + 0.6937269372693727, + 0.754325259515571, + 0.7730263157894737, + 0.7342657342657343, + 0.7457627118644068, + 0.9831932773109243, + 0.7288135593220338 + ] + }, + "contact_maps": { + "mean_pearson": 0.34198331062892884, + "valid_tracks": 28, + "per_track_pearson": [ + 0.3320274873225753, + 0.36239563766113925, + 0.3855406656472823, + 0.315970228099349, + 0.2549116161777331, + 0.28997682019991844, + 0.3500882724326381, + 0.2702547706750472, + 0.4188151407346995, + 0.38413346197839865, + 0.314032613660622, + 0.3691036398660645, + 0.3719472297719302, + 0.2902348464843489, + 0.32574593329752155, + 0.449893253652323, + 0.09719635121648074, + 0.37032469119401784, + 0.3652394648247715, + 0.3656753441736635, + 0.31474885228387284, + 0.318764043557503, + 0.3202943455205856, + 0.38222184077536786, + 0.3862415260989682, + 0.39685918569903844, + 0.36162095351691814, + 0.41127448108722614 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.49692428448153536, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..9454a71b8d7921f7eb062fe72d82172e596d9c15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S031/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.4494588041544943, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5458187779616351, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.045300968039382924, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7366212964318372, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.5682257145305185, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.5624980855302734, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.25235379308670775, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.4341420381315536, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.14577564576390645, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.22321043347781205, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.4797983383164624, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.22751315072494746, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.22563969764497074, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/README.md new file mode 100644 index 0000000000000000000000000000000000000000..ffe7c586aaa55836313ff4635626a3feb7b716f1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/README.md @@ -0,0 +1,115 @@ +# SeqFunc RC108: eight-pass small model with equal forward/reverse-complement inference + +One model with **13,290,504 parameters** serves every family and both entry points. It reads the entire +1,048,576-bp DNA window through a convolutional encoder and 6 transformer +blocks of width 384 over 1,024 tokens. Dense heads predict 128-bp bins over the central +196,608 bp; other heads predict splice sites, sparse usage/junctions and full-window contacts. +A width-32 base-resolution path uses a 21-bp stem and 6 dilated residual depthwise blocks (10,941-bp receptive field). Average/max pooling provides aligned 128-bp residual features to dense, usage and junction heads. Splice logits combine local base-resolution logits with the reference offset decoder. Local average/max features are also averaged into 1-kb tokens and linearly projected into the global trunk before its transformer blocks. The projection is initialized with zero output; any later checkpoint interventions are described in the lineage below. Each transformer attention and MLP update is capped at a per-token RMS of 0.1; smaller updates are unchanged. This parameter-free bound limits residual growth and preserves the incoming sequence features. Junction values use low-rank donor/acceptor factors; contacts use a symmetric +product of global features. Sites and junction endpoints are selected from model scores alone. + +All weights originate from local random initialization (seed 17) and locally trained +checkpoints. Provisional combined recipe: a 10,941-bp local path, local-to-global projection, branch RMS cap 0.1, and fixed training-variance family weights with balanced splice loss. Component ablations are documented in the research ledger. Both scales start from local random initialization; the paired comparison isolates model capacity within this combined recipe. After the first 1,000 examples, both runs use per-local-block activation recomputation; the same mathematical model and objective are preserved. Numerical and memory checks, including an unchanged BF16 repeat, are included in training_history. At40,000examples both scales resume unchanged checkpoints using four bounded I/O threads per data worker. Ordered IDs and all decoded tensor bytes were verified identical; model operations, objective, batch order and LR schedule are unchanged. At the saved small96k/large78k boundaries, an optional training-only byte cache was enabled after exact256example and integrated16example tensor/ID checks. Source compressed bytes are verified on every cache hit, with read-only source fallback. Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. The cache is disposable and is disabled for validation/public inference. 13,290,504 parameters; local width 32, trunk width 384, 6 transformer blocks. Parameters are initialized locally with seed 17. The paired scale uses the same batch-four interval order, objective, learning-rate schedule and full training permutations. Only encoder/trunk/head/local-path width and transformer depth differ between scales. + +This checkpoint records **328,160 training examples in 82,040 optimizer updates**, 8 complete training permutations and +0 intervals into the next permutation. Each permutation includes all **41,020** +eligible FOLD_0 training intervals (folds 2–7), shuffled deterministically without replacement. +Quantized train targets are decoded by the supplied reader. Validation is fold 0; no test labels, +test-derived caches or pretrained model parameters were used. Development labels only evaluate +and select models. Checkpoints before a complete permutation remain intermediate milestones. + +Training uses one H100 80 GB and at most 22 CPUs. The current loader uses 4 workers +and 2 torch CPU threads. Batches contain up to 4 examples; losses are averaged equally over examples, with padded event queries excluded. Every epoch includes its final partial batch. BF16 training uses AdamW (weight decay 1e-4), gradient +clipping at 5, and a 14-epoch cosine learning-rate schedule from +0.0002 to 2e-05, with +4000 warmup training examples. The objective is `variance_balanced`. +Each continuous family uses the reference MSE divided by its mean within-track variance in a fixed, training-only sample of 256 windows (seed 1729). These scales are frozen in the checkpoint; the balanced splice BCE is unchanged. Recorded accumulated training-loop wall time is +1231.6 minutes, including inherited parent time. Runs can share this +same physical GPU with experiments or evaluation. Setup, evaluation and discarded mechanical checks +are additional research compute, documented in the research ledger. Matching training junction queries +uses a verified, bit-identical CPU index construction followed by a bulk GPU transfer. + +The package contains model code, weights, training source/configuration/provenance, SHA-256 checksums, +and the complete CPython 3.11 / PyTorch 2.7.1 CUDA 12.6 Python runtime archive. Both executable entry +points take INPUT_ROOT and OUTPUT_ROOT, extract the vendored runtime temporarily, and run offline. +The host supplies Linux, bash/tar/gzip and the NVIDIA driver. Inference reads only public manifests +and the pinned reference FASTA. No target, splice-coordinate or prediction cache is packaged or read. +Track outputs use float16; perturbation outputs use float32 to preserve small A/B differences. + +The appended validation results cover all 200 intervals and all 2,697 development perturbations, +scored by the unmodified verifier tools. Published anchors use the same label rows. All results, +including regressions, are retained in the package's JSON reports. Limitations include incomplete +training at intermediate checkpoints, a 10,941-bp local receptive field, pooled usage/junction features, +low-rank contact modeling, and uneven variant performance. Functional readouts do not provide the +evolutionary constraint information used by conservation-based variant scorers. +The release has no targets for the 54 GTEx RNA tracks, so those output heads receive no direct +supervision and are excluded from the scored RNA track mean. The verifier's eQTL metrics use +supervised ENCODE RNA tracks selected by public tissue ontology and gene strand. + +Junction selection requires the donor to precede the acceptor on plus-strand transcripts and +follow it on minus-strand transcripts. This fixed biological constraint uses model-selected +coordinates and public strand halves. It changes no model parameters, usage selection or +perturbation computation. The rest of the decoder and candidate budget are unchanged. + +The splice-family output is a probability, using the same fixed link in both entry points: +`sigmoid(raw_logits + offset)`. Four offsets are estimated from 256 training windows (seed 1729) +to correct the marginal prior induced by balanced splice BCE. All network weights are unchanged. +The training prevalence, effective balanced-loss prior, offsets and sampling hash are included in +`splice_calibration.json`. This is an approximate marginal correction because the loss balances +each training window separately. No development labels fit these constants. + +Site candidates remain ranked by the underlying logits, preserving ordering without probability +rounding ties. Usage, junction and all dense/contact computations are unchanged. The verifier +applies its standard sigmoid when scoring splice perturbations; that scoring code is unmodified. +The complete reports below measure the resulting track and variant tradeoffs. The complete reports below determine this checkpoint's track and variant tradeoffs. + +This predeclared late fixed-link follow-up uses the complete 8-epoch small main model. It applies the original four training-only offsets without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. Both entry points use the same link. All200 validation windows and all13 development metrics are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. A publication additionally requires native replay and three sampled runtime estimates below four hours. + +Every output family and both entry points average this same checkpoint on the provided DNA +and its reverse complement, mapped back using vendored public-metadata strand permutations. +There are 13,290,504 unique learned parameters and two deterministic views, +weighted equally without fitting development labels. No additional flanking DNA is read. +Dense and contact values average in the contract's asinh space; splice scores average the +training-prior-corrected probabilities. This approximately doubles inference work. + +Each view selects sparse coordinates independently. Their union averages missing-view values +as zero. Usage retains at most4,096 positions; junctions retain16,384 pairs by averaged own-strand +strength, with deterministic coordinate tie handling. Track arrays are averaged in float32 and +then stored in float16; perturbation arrays remain float32. Sparse recall and score dilution +can both change. The frozen decoder passed the RC17 synthetic, native offline and partial-resume +checks; its exact parent-extraction and source fingerprints are enforced by the builder. +This candidate requires its own complete validation track and development variant reports +before publication. Architecture, learned weights and training history are inherited unchanged. + +## Measured validation results + +Family-mean Pearson: **0.510352** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.595241 | +| dnase | 0.670827 | +| procap | 0.556183 | +| cage | 0.340146 | +| rna_seq | 0.643930 | +| chip_tf | 0.473845 | +| chip_histone | 0.508014 | +| splice_sites | 0.431237 | +| splice_usage | 0.632448 | +| junctions | 0.391177 | +| contact_maps | 0.370824 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5091 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5790 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | -0.0213 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7607 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.6116 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.6469 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.2667 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4521 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1523 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2087 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4617 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.1954 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2232 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..8adab1ac9a55de960ee5252eeb36c4b4b5bdf689 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-small-epoch08/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..39c21d44f620d2c890411295291fda55b041f668 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/package_manifest.json @@ -0,0 +1,68 @@ +{ + "parameters": 13290504, + "source_checkpoint_sha256": "8c38a6101b84fca69ac2f7bd9daa19f1a3a5b4cebe6e26258e12ad2a137ed79c", + "files": { + "README.md": "e00ae1780749fb076f567efe620ddb2e8abf8421d79b9e7a00a9e9db529c1828", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "e97102570c90baf718facde0be67b8dee60e7fc44fb2ff8fbd25e780c909bc4d", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "bdf1099d0812b9a26e4bb108cc81c2dcf40c7ec475712b839e3cc29fad2ca390", + "training_assets_manifest.json": "7e4a33e58edeb9c25fb59ec5b1f3b65ff2b1ca83abb9c3fc0770ec4f57af06c9", + "decoding_policy.json": "57d76f661dadbf958ab071b23062f5f9e6803fa596a18bf746765fc63c271aa9", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/late-fixed-links81-master-plan.json": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "training_history/late-fixed-links81-resolved-parents.json": "719225ce26773880c04343af0e6c61941eb0a0649f8145bf7067748620fe4f5b", + "training_history/late-fixed-links81-parent-check.json": "9d6c82a8f82993fad54c586b29c3bda774e3839e67352c84eb663f819d0b51ea", + "training_history/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "training_history/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60", + "rc_ensemble_policy.json": "504cd05bfd886805df8e8cdc66ceed7534f169d987d44b6179141b1fd2ae9f9f", + "validation_tracks.json": "52fb0735c13904046b56400c4894b672f95d5920f6db01b1b2d101a843c8ee36", + "validation_variants.json": "2358fb31bd4bf7a6780085edab7765fceec958f4a6f1d36f1e802c2627c0c4e3" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..b8ee571ad9dab90aded5077affbb42419869da16 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.py @@ -0,0 +1,92 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import predict_requested + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + group = root.create_group(f"perturbations/p{index:06d}") + for family, values in predict_requested(model, dna, row).items(): + values = np.asarray(values, dtype=np.float32) + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..665fb186f3a62b1ff130abb034ac25a3a924c7c8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/predict.py @@ -0,0 +1,52 @@ +"""Stream one shared-checkpoint forward/reverse-complement mean for every output family.""" +from __future__ import annotations +import argparse +import json +from pathlib import Path +from numcodecs import Blosc +import numpy as np +import torch +import zarr +from prediction_contract import PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import engine_for + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + model = CompactModel(ModelConfig(**payload['config'])) + model.load_state_dict(payload['model'], strict=True) + device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + engine = engine_for(model) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample['example_id'] + values = engine.tracks(sample['dna'][None].to(device)) + group = root.create_group(f"examples/{example_id}") + for name, array in values.items(): + dtype = np.int8 if name == 'junction_half' else np.int32 if name in ( + 'splice_usage_positions', 'junction_donor', 'junction_acceptor') else np.float16 + group.create_dataset(name, data=array.astype(dtype), compressor=COMPRESSOR) + ids.append(example_id) + print(json.dumps({'event': 'predict', 'example': len(ids), 'total': len(dataset)}), flush=True) + root.attrs.update({'schema': PREDICTION_SCHEMA, 'example_ids': ids}) + (output_root/'example_ids.json').write_text(json.dumps(ids)+'\n') + + +if __name__ == '__main__': + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('input_root', type=Path) + p.add_argument('output_root', type=Path) + p.add_argument('--checkpoint', type=Path, default=Path(__file__).resolve().parent/'last.pt') + p.add_argument('--split', default='test', choices=['valid', 'test']) + a = p.parse_args() + main(a.input_root, a.output_root, a.checkpoint, a.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble.py new file mode 100644 index 0000000000000000000000000000000000000000..f076804c66e481eb823e6ca5e721f39fce50bad5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble.py @@ -0,0 +1,111 @@ +"""One shared checkpoint evaluated on DNA and its reverse complement for every output.""" +import json +from pathlib import Path +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from single_prediction import single_prediction +from splice_output import splice_output_scores + + +def contact_reversal(bins): + row, column = np.triu_indices(bins) + lookup = np.empty((bins, bins), dtype=np.int64) + lookup[row, column] = np.arange(len(row)) + lookup[column, row] = np.arange(len(row)) + return lookup[bins-1-row, bins-1-column] + + +def reverse_readout(arrays, permutations, input_length, target_length, contact_bins): + """Map a reverse-complement prediction back into the original genomic coordinates.""" + out = {name: arrays[name][::-1, permutations[name]].copy() for name in DENSE_TRACKS} + out['splice_sites'] = arrays['splice_sites'][::-1, permutations['splice_sites']].copy() + positions = target_length-1-arrays['splice_usage_positions'] + order = np.argsort(positions) + out['splice_usage_positions'] = positions[order].astype(np.int32) + out['splice_usage'] = arrays['splice_usage'][order][:, permutations['splice_usage']].copy() + out['junction_half'] = (1-arrays['junction_half']).astype(np.int8) + for key in ['junction_donor', 'junction_acceptor']: + out[key] = (input_length-1-arrays[key]).astype(np.int32) + out['junctions'] = arrays['junctions'][:, permutations['junctions']].copy() + out['contact_maps'] = arrays['contact_maps'][contact_reversal(contact_bins)][:, permutations['contact_maps']].copy() + return out + + +def combine_readouts(a, b, track_halves): + """Average sparse functions with zero at unlisted coordinates, then enforce output budgets.""" + out = {name: (a[name]+b[name])*np.float32(.5) + for name in [*DENSE_TRACKS, 'splice_sites', 'contact_maps']} + usage_positions, inverse = np.unique(np.concatenate([a['splice_usage_positions'], b['splice_usage_positions']]), return_inverse=True) + assert len(usage_positions) <= MAX_USAGE_SITES + usage = np.zeros((len(usage_positions), a['splice_usage'].shape[-1]), dtype=np.float32) + np.add.at(usage, inverse, np.concatenate([a['splice_usage'], b['splice_usage']])*np.float32(.5)) + out.update(splice_usage_positions=usage_positions.astype(np.int32), splice_usage=usage) + keys = np.concatenate([np.stack([x['junction_half'], x['junction_donor'], x['junction_acceptor']], axis=1) for x in [a, b]]) + keys, inverse = np.unique(keys, axis=0, return_inverse=True) + values = np.zeros((len(keys), len(track_halves)), dtype=np.float32) + np.add.at(values, inverse, np.concatenate([a['junctions'], b['junctions']])*np.float32(.5)) + strength = np.empty(len(keys), dtype=np.float32) + for half in [0, 1]: + keep = keys[:, 0] == half + strength[keep] = values[keep][:, track_halves == half].max(axis=1) + # Stable sorting makes ties deterministic in the canonical coordinate order. + order = np.argsort(-strength, kind='stable')[:MAX_JUNCTION_PAIRS] + keys, values = keys[order], values[order] + out.update(junction_half=keys[:, 0].astype(np.int8), + junction_donor=keys[:, 1].astype(np.int32), + junction_acceptor=keys[:, 2].astype(np.int32), junctions=values) + return out + + +class ReverseComplementMean: + def __init__(self, model): + self.model = model + data = json.loads(Path(__file__).with_name('strand_permutations.json').read_text()) + self.permutations = {name: np.asarray(values, dtype=np.int64) for name, values in data['permutations'].items()} + for name, indices in self.permutations.items(): + assert np.array_equal(indices[indices], np.arange(len(indices))), name + + def tracks(self, dna): + assert len(dna) == 1 + config = self.model.config + original = single_prediction(self.model, dna) + reverse = single_prediction(self.model, dna.flip((1, 2))) + reverse = reverse_readout(reverse, self.permutations, config.input_length, config.target_length, config.contact_bins) + return combine_readouts(original, reverse, self.model.junction_track_half.cpu().numpy()) + + def dense_and_sites(self, dna, families): + """Return central readouts for both alleles; this same probability/asinh averaging defines tracks.""" + config = self.model.config + start = (config.input_length-config.target_length)//2 + predictions = [] + for reverse in [False, True]: + trunk = self.model._trunk(dna.flip((1, 2)) if reverse else dna) + features = None + values = {} + for family in families: + if family == 'splice_sites': + logits = self.model._splice_site_logits(trunk)[:, start:start+config.target_length] + array = splice_output_scores(logits).float().cpu().numpy() + else: + if features is None: + features = self.model._dense_features(trunk) + array = self.model.dense_heads[family](features).float().cpu().numpy() + if reverse: + array = array[:, ::-1][:, :, self.permutations[family]].copy() + values[family] = array + predictions.append(values) + return {family: (predictions[0][family]+predictions[1][family])*np.float32(.5) for family in families} + + +def engine_for(model): + if not hasattr(model, '_rc_prediction_engine'): + # Plain helper, not a child nn.Module; the backbone state dictionary stays unchanged. + model._rc_prediction_engine = ReverseComplementMean(model) + return model._rc_prediction_engine + + +def predict_requested(model, dna, row): + values = engine_for(model).dense_and_sites(dna, row.families) + return {family: array[:, row.splice_start:row.splice_end] if family == 'splice_sites' + else array[:, row.dense_bin_start:row.dense_bin_end] for family, array in values.items()} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..ea28ca2d1886a20b2d9c56a721efe4d1f50b4a2c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/rc_ensemble_policy.json @@ -0,0 +1,35 @@ +{ + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-26T01:07:17.491176+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib81-small-epoch08/package", + "parent_inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 13290504, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/cached_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/cached_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3677c1e8518ea97db0dd6759f3d636f5fff27c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/cached_read_store.py @@ -0,0 +1,71 @@ +"""Optional local byte cache for immutable training Zarr stores. + +Source storage stays read-only. Cached entries contain a key fingerprint, length, +and SHA256 of the original compressed bytes; every hit verifies all three. Writes +replace a private temporary inode atomically. A missing/corrupt/unwritable cache +falls back to the original store, so this layer changes neither values nor keys. +Only training loaders may opt into it; it is not an inference asset. +""" +import hashlib +import os +from pathlib import Path +import struct +import tempfile + +from .direct_read_store import ParallelDirectReadDirectoryStore + +MAGIC = b'SFBC0001' +HEADER = struct.Struct('<8sQ32s32s') + + +class CachedReadDirectoryStore(ParallelDirectReadDirectoryStore): + def __init__(self, path, *, cache_root, namespace, io_workers=4): + super().__init__(path, io_workers=io_workers) + if not namespace: + raise ValueError('An immutable input namespace is required') + self.cache_root = Path(cache_root) + self.namespace = namespace + self.store_key = hashlib.sha256((namespace+'\0'+os.path.abspath(path)).encode()).digest() + self.statistics = {'hit': 0, 'miss': 0, 'corrupt': 0, 'write_failure': 0} + + def cache_location(self, key): + key = self._normalize_key(key) + identity = hashlib.sha256(self.store_key+key.encode()).digest() + name = identity.hex() + return self.cache_root/name[:2]/name[2:], identity + + def __getitem__(self, key): + key = self._normalize_key(key) + destination, identity = self.cache_location(key) + try: + blob = destination.read_bytes() + if len(blob) >= HEADER.size: + magic, size, cached_key, checksum = HEADER.unpack_from(blob) + payload = blob[HEADER.size:] + if (magic == MAGIC and size == len(payload) and cached_key == identity + and hashlib.sha256(payload).digest() == checksum): + self.statistics['hit'] += 1 + return payload + self.statistics['corrupt'] += 1 + except OSError: + pass + self.statistics['miss'] += 1 + payload = super().__getitem__(key) + temporary = None + try: + destination.parent.mkdir(parents=True, exist_ok=True) + with tempfile.NamedTemporaryFile(prefix='.pending-', dir=destination.parent, delete=False) as stream: + temporary = stream.name + stream.write(HEADER.pack(MAGIC, len(payload), identity, hashlib.sha256(payload).digest())) + stream.write(payload) + os.replace(temporary, destination) + temporary = None + except OSError: + self.statistics['write_failure'] += 1 + finally: + if temporary is not None: + try: + os.unlink(temporary) + except OSError: + pass + return payload diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/single_prediction.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/single_prediction.py new file mode 100644 index 0000000000000000000000000000000000000000..a47022029461e5c8097b3a653af8b3407196d67d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/single_prediction.py @@ -0,0 +1,97 @@ +"""The parent single-orientation decoder, preserving float32 until ensemble averaging.""" +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from splice_output import splice_output_scores + +COMPRESSOR = None + +class ArrayCollector: + def __init__(self): + self.arrays = {} + def create_dataset(self, name, *, data, **kwargs): + self.arrays[name] = np.asarray(data) + +def single_prediction(model, dna): + config = model.config + group = ArrayCollector() + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy(), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy(), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy(), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy(), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy(), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + return group.arrays diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/strand_permutations.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/strand_permutations.json new file mode 100644 index 0000000000000000000000000000000000000000..044c090773f20347601dcd4f0983c4dfd4c0ecc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/strand_permutations.json @@ -0,0 +1,5959 @@ +{ + "schema": "seqfunc-strand-mapping/v1", + "source": "public track metadata; no target values", + "source_sha256": "23450e6ff9463c64d271d7014f8a798875bc95186c36aeb34609c7dc1d62cd5f", + "permutations": { + "atac": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166 + ], + "dnase": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304 + ], + "procap": [ + 6, + 7, + 8, + 9, + 10, + 11, + 0, + 1, + 2, + 3, + 4, + 5 + ], + "cage": [ + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272 + ], + "rna_seq": [ + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666 + ], + "chip_tf": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115, + 1116, + 1117, + 1118, + 1119, + 1120, + 1121, + 1122, + 1123, + 1124, + 1125, + 1126, + 1127, + 1128, + 1129, + 1130, + 1131, + 1132, + 1133, + 1134, + 1135, + 1136, + 1137, + 1138, + 1139, + 1140, + 1141, + 1142, + 1143, + 1144, + 1145, + 1146, + 1147, + 1148, + 1149, + 1150, + 1151, + 1152, + 1153, + 1154, + 1155, + 1156, + 1157, + 1158, + 1159, + 1160, + 1161, + 1162, + 1163, + 1164, + 1165, + 1166, + 1167, + 1168, + 1169, + 1170, + 1171, + 1172, + 1173, + 1174, + 1175, + 1176, + 1177, + 1178, + 1179, + 1180, + 1181, + 1182, + 1183, + 1184, + 1185, + 1186, + 1187, + 1188, + 1189, + 1190, + 1191, + 1192, + 1193, + 1194, + 1195, + 1196, + 1197, + 1198, + 1199, + 1200, + 1201, + 1202, + 1203, + 1204, + 1205, + 1206, + 1207, + 1208, + 1209, + 1210, + 1211, + 1212, + 1213, + 1214, + 1215, + 1216, + 1217, + 1218, + 1219, + 1220, + 1221, + 1222, + 1223, + 1224, + 1225, + 1226, + 1227, + 1228, + 1229, + 1230, + 1231, + 1232, + 1233, + 1234, + 1235, + 1236, + 1237, + 1238, + 1239, + 1240, + 1241, + 1242, + 1243, + 1244, + 1245, + 1246, + 1247, + 1248, + 1249, + 1250, + 1251, + 1252, + 1253, + 1254, + 1255, + 1256, + 1257, + 1258, + 1259, + 1260, + 1261, + 1262, + 1263, + 1264, + 1265, + 1266, + 1267, + 1268, + 1269, + 1270, + 1271, + 1272, + 1273, + 1274, + 1275, + 1276, + 1277, + 1278, + 1279, + 1280, + 1281, + 1282, + 1283, + 1284, + 1285, + 1286, + 1287, + 1288, + 1289, + 1290, + 1291, + 1292, + 1293, + 1294, + 1295, + 1296, + 1297, + 1298, + 1299, + 1300, + 1301, + 1302, + 1303, + 1304, + 1305, + 1306, + 1307, + 1308, + 1309, + 1310, + 1311, + 1312, + 1313, + 1314, + 1315, + 1316, + 1317, + 1318, + 1319, + 1320, + 1321, + 1322, + 1323, + 1324, + 1325, + 1326, + 1327, + 1328, + 1329, + 1330, + 1331, + 1332, + 1333, + 1334, + 1335, + 1336, + 1337, + 1338, + 1339, + 1340, + 1341, + 1342, + 1343, + 1344, + 1345, + 1346, + 1347, + 1348, + 1349, + 1350, + 1351, + 1352, + 1353, + 1354, + 1355, + 1356, + 1357, + 1358, + 1359, + 1360, + 1361, + 1362, + 1363, + 1364, + 1365, + 1366, + 1367, + 1368, + 1369, + 1370, + 1371, + 1372, + 1373, + 1374, + 1375, + 1376, + 1377, + 1378, + 1379, + 1380, + 1381, + 1382, + 1383, + 1384, + 1385, + 1386, + 1387, + 1388, + 1389, + 1390, + 1391, + 1392, + 1393, + 1394, + 1395, + 1396, + 1397, + 1398, + 1399, + 1400, + 1401, + 1402, + 1403, + 1404, + 1405, + 1406, + 1407, + 1408, + 1409, + 1410, + 1411, + 1412, + 1413, + 1414, + 1415, + 1416, + 1417, + 1418, + 1419, + 1420, + 1421, + 1422, + 1423, + 1424, + 1425, + 1426, + 1427, + 1428, + 1429, + 1430, + 1431, + 1432, + 1433, + 1434, + 1435, + 1436, + 1437, + 1438, + 1439, + 1440, + 1441, + 1442, + 1443, + 1444, + 1445, + 1446, + 1447, + 1448, + 1449, + 1450, + 1451, + 1452, + 1453, + 1454, + 1455, + 1456, + 1457, + 1458, + 1459, + 1460, + 1461, + 1462, + 1463, + 1464, + 1465, + 1466, + 1467, + 1468, + 1469, + 1470, + 1471, + 1472, + 1473, + 1474, + 1475, + 1476, + 1477, + 1478, + 1479, + 1480, + 1481, + 1482, + 1483, + 1484, + 1485, + 1486, + 1487, + 1488, + 1489, + 1490, + 1491, + 1492, + 1493, + 1494, + 1495, + 1496, + 1497, + 1498, + 1499, + 1500, + 1501, + 1502, + 1503, + 1504, + 1505, + 1506, + 1507, + 1508, + 1509, + 1510, + 1511, + 1512, + 1513, + 1514, + 1515, + 1516, + 1517, + 1518, + 1519, + 1520, + 1521, + 1522, + 1523, + 1524, + 1525, + 1526, + 1527, + 1528, + 1529, + 1530, + 1531, + 1532, + 1533, + 1534, + 1535, + 1536, + 1537, + 1538, + 1539, + 1540, + 1541, + 1542, + 1543, + 1544, + 1545, + 1546, + 1547, + 1548, + 1549, + 1550, + 1551, + 1552, + 1553, + 1554, + 1555, + 1556, + 1557, + 1558, + 1559, + 1560, + 1561, + 1562, + 1563, + 1564, + 1565, + 1566, + 1567, + 1568, + 1569, + 1570, + 1571, + 1572, + 1573, + 1574, + 1575, + 1576, + 1577, + 1578, + 1579, + 1580, + 1581, + 1582, + 1583, + 1584, + 1585, + 1586, + 1587, + 1588, + 1589, + 1590, + 1591, + 1592, + 1593, + 1594, + 1595, + 1596, + 1597, + 1598, + 1599, + 1600, + 1601, + 1602, + 1603, + 1604, + 1605, + 1606, + 1607, + 1608, + 1609, + 1610, + 1611, + 1612, + 1613, + 1614, + 1615, + 1616 + ], + "chip_histone": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115 + ], + "contact_maps": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27 + ], + "splice_sites": [ + 2, + 3, + 0, + 1 + ], + "splice_usage": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ], + "junctions": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ] + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8e314c842cca039da1d2be8da7fff6a146178651 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_assets_manifest.json @@ -0,0 +1,21 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa30b0f9f20f1dd25a2c8426ff18d3afab3ba47f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 32 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 32 +gradient_checkpointing: true +head_hidden_size: 256 +hidden_dim: 384 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 32 +mlp_expansion: 4.0 +num_attention_heads: 6 +num_downsampling_blocks: 7 +num_sequence_blocks: 6 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-integration_check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-integration_check.json new file mode 100644 index 0000000000000000000000000000000000000000..1377348921feb42310e66b133df8ebee9f0d591c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-integration_check.json @@ -0,0 +1,166 @@ +{ + "created_utc": "2026-10-25T03:48:07.725319+00:00", + "preparation_sha256": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "source_sha256": "b0921bfd3d3dc47f3cb82debc0b492d8291693c40e0cef028764c713c1257e85", + "original_full256_report_sha256": "e6620ff13b00a36857dad22fa0c9d1ad5c2b266eca70773650d280998ba20005", + "examples": 16, + "worker_processes": 4, + "ordered_indices": [ + 38509, + 39126, + 38926, + 17432, + 6551, + 22599, + 24749, + 10790, + 7022, + 30789, + 17340, + 40258, + 26222, + 22674, + 23658, + 17167 + ], + "all_returned_training_ids_and_tensor_bytes_exact": true, + "reference_and_cache_both_match_original_report": true, + "public_inference_and_validation_cache_disabled_despite_environment": true, + "default_cache_disabled": true, + "models_and_trainers_unchanged": true, + "rows": { + "reference": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ], + "cached": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ] + }, + "no_active_source_or_training_changes": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-master-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-master-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..60f920df90c63e6614b09745bb36dedab5f40cc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-master-plan.json @@ -0,0 +1,562 @@ +{ + "created_utc": "2026-10-25T16:20:50.709567+00:00", + "members": { + "large-epoch04": { + "capacity": "large", + "epoch": 4, + "previous_epoch": 2, + "parent": "/workspace/research/experiments/scale13-large-epoch04", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch02", + "output": "/workspace/research/experiments/calib81-large-epoch04", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch04" + }, + "small-epoch08": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-small-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch04", + "output": "/workspace/research/experiments/calib81-small-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch08" + }, + "large-epoch08": { + "capacity": "large", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-large-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch04", + "output": "/workspace/research/experiments/calib81-large-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch08" + }, + "small-epoch14": { + "capacity": "small", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-small-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/calib81-small-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch14" + }, + "large-epoch14": { + "capacity": "large", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-large-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch08", + "output": "/workspace/research/experiments/calib81-large-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch14" + } + }, + "capacities": { + "small": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 13290504 + }, + "large": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 96, + "channel_growth": 96, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 1024, + "num_sequence_blocks": 16, + "num_attention_heads": 16, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 768, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 64, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 215626632 + } + }, + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "calibration_reference": "/workspace/research/experiments/calib57/package/splice_calibration.json", + "helper_sha256": { + "/run-state/workspace/research/seqfunc/check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "/run-state/workspace/research/seqfunc/check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "/run-state/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "/run-state/workspace/research/seqfunc/diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "/run-state/workspace/research/seqfunc/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "/run-state/workspace/research/seqfunc/evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "/run-state/workspace/research/seqfunc/finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "/run-state/workspace/research/seqfunc/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "/run-state/workspace/research/seqfunc/make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "/run-state/workspace/research/seqfunc/materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "/run-state/workspace/research/seqfunc/measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "/run-state/workspace/research/seqfunc/promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "/run-state/workspace/research/seqfunc/publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "/run-state/workspace/research/seqfunc/resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "/run-state/workspace/research/seqfunc/resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "/run-state/workspace/research/seqfunc/run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "/run-state/workspace/research/seqfunc/run_fixed_link_followup69.py": "3234a647219aae3df72afee9e5d9ed671d9db8b08adafff58bbffcbebcafb40d", + "/run-state/workspace/research/seqfunc/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "/run-state/workspace/research/seqfunc/verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "evidence_sha256": { + "/workspace/research/studies/fixed_link69/plan.json": "45541635f549af8ce6c3db3fc7c857720970a5ee35e9de3c6a57f41dbe1f56bc", + "/workspace/research/studies/fixed_link61/plan.json": "100e4da9bf4faa715e901a0aa928dab65fc19f60970312e3e0eb548fb5b70860", + "/workspace/research/studies/fixed_link57/complete-summary.json": "5bba2a23162a77a523777f19de0be819d44cb44246c8c5e2f4bef913f5ee7a87", + "/workspace/research/studies/scale13/study-cached-io.json": "88aa6d95ef93776ed3551d68d98b1a31d726aa4b0270edbfc16b3f16ceeb937f", + "/workspace/research/studies/scale13/loss_scales.json": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42", + "/workspace/data/public/train.parquet": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "/workspace/research/experiments/calib57/package/splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5", + "/workspace/research/experiments/scale13-large-epoch04/package/training_provenance.json": "6589e14349cbd5419c814aa3d292c81f5328f1441bc75bc3ea7cae2b114dcdac" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "native_latency_template": { + "created_utc": "2026-10-25T09:54:57.944795+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.", + "scope": "Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "The rawlarge2 parent timing is contextual only. Require fresh candidate replay and allthree estimates <=14400seconds; no estimate is a guaranteed upper bound." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "bounded_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "native_timeout_seconds": 1800, + "materialization_timeout_seconds": 1800, + "parent_minimum_reward_gain": 1e-06, + "publication_minimum_reward_gain": 1e-06, + "hypothesis": "Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.", + "admission": "Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.", + "validation": "Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.", + "diagnostics": "After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.", + "nonimprovement_policy": "A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.", + "limitations": [ + "The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.", + "The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.", + "Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.", + "No test labels, variant-label training, external parameters or inference caches are used." + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-parent-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-parent-check.json new file mode 100644 index 0000000000000000000000000000000000000000..4e85ff1231bb0b51a26679743837f33f5732f829 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-parent-check.json @@ -0,0 +1,53 @@ +{ + "created_utc": "2026-10-25T16:21:07.054587+00:00", + "plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "actual_completed_parent_checks": [ + { + "experiment": "/workspace/research/experiments/scale13-small-epoch04", + "capacity": "small", + "epoch": 4, + "inference_signature": { + "last.pt": "ae6a15c21e8d9b5900f986f28f91b00962c0030ffc952340b1b3264eef1a6659", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.44610804541871724, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch04/evaluation.json": "3193c974189a7ad77a06d0abf64a2c3e002953470660ca20f30d97271e3ffd0a", + "/workspace/research/experiments/scale13-small-epoch04/evaluation-package.json": "e06f0c731873e9ef1a94722d29651c180eefc6c1b54650b52622bad525dd5da7", + "/workspace/research/experiments/scale13-small-epoch04/valid-score/report.json": "3a45ae11ac8ad3897b926283073b5e755b3013a7796aa9837954dec842d7a942", + "/workspace/research/experiments/scale13-small-epoch04/valid-variant-score/report.json": "19224eae63273b62b0421fc48c32704b86418a2bfb21adda2a90449cfa0988d8", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5" + } + }, + { + "experiment": "/workspace/research/experiments/scale13-large-epoch02", + "capacity": "large", + "epoch": 2, + "inference_signature": { + "last.pt": "6e7c605f2683881b9f59571f042cb7df93a3a0f224d8f2cb4e37be0a123cdb27", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4231233468773853, + "sha256": { + "/workspace/research/experiments/scale13-large-epoch02/evaluation.json": "d2336bfba9e6788cb2ab7dcf01834d825846920c21db43b763f3f5a4de056afb", + "/workspace/research/experiments/scale13-large-epoch02/evaluation-package.json": "cad26f6c8167a550d9a11057e69f2feca299cb091bfce64a95b403a627271610", + "/workspace/research/experiments/scale13-large-epoch02/valid-score/report.json": "eeec21adce86c14356ed29a5bc0e76e84f847bceba41c0faefa8e3a8d6793d4d", + "/workspace/research/experiments/scale13-large-epoch02/valid-variant-score/report.json": "f806730a18850e289cb95020db0ac6f69d4203c8ab53c2ef31845fd798a36f28", + "/workspace/research/experiments/scale13-large-epoch02/package/training_provenance.json": "1eb6eafaaec3f950708cabd7dc43a3ed22e67be602d7c036fe0fbbe0fbbff802" + } + } + ], + "gpu_used": false, + "packages_changed": false, + "no_new_predictions_or_parameter_fitting": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-resolved-parents.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-resolved-parents.json new file mode 100644 index 0000000000000000000000000000000000000000..241cd648df020f91b007aefc0cf0e8d9f188765a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late-fixed-links81-resolved-parents.json @@ -0,0 +1,58 @@ +{ + "master_plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "member": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-small-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch04", + "output": "/workspace/research/experiments/calib81-small-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch08" + }, + "receipts": { + "parent": { + "experiment": "/workspace/research/experiments/scale13-small-epoch08", + "capacity": "small", + "epoch": 8, + "inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4622727224599102, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch08/evaluation.json": "048a4f8e805c8476b4240977d2e6c6e800dd8a509ec1183f099e9726cc51cd11", + "/workspace/research/experiments/scale13-small-epoch08/evaluation-package.json": "84a37d41701a015643c86a120d0c8b1794af718fce1478824f8aaf0798411cf5", + "/workspace/research/experiments/scale13-small-epoch08/valid-score/report.json": "2baaccea669daefc1fde4adec4477eeb813e1e17fc5d64ed3dc1f8b54dff676f", + "/workspace/research/experiments/scale13-small-epoch08/valid-variant-score/report.json": "e204ff744b560b4df16ad6c114205480864faffac1a5e22c519f4a1960638d76", + "/workspace/research/experiments/scale13-small-epoch08/package/training_provenance.json": "ca8cc8c1d364643632183fc676d1abbd8ca285730fe678465c634fe7dae17d01" + } + }, + "previous": { + "experiment": "/workspace/research/experiments/scale13-small-epoch04", + "capacity": "small", + "epoch": 4, + "inference_signature": { + "last.pt": "ae6a15c21e8d9b5900f986f28f91b00962c0030ffc952340b1b3264eef1a6659", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.44610804541871724, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch04/evaluation.json": "3193c974189a7ad77a06d0abf64a2c3e002953470660ca20f30d97271e3ffd0a", + "/workspace/research/experiments/scale13-small-epoch04/evaluation-package.json": "e06f0c731873e9ef1a94722d29651c180eefc6c1b54650b52622bad525dd5da7", + "/workspace/research/experiments/scale13-small-epoch04/valid-score/report.json": "3a45ae11ac8ad3897b926283073b5e755b3013a7796aa9837954dec842d7a942", + "/workspace/research/experiments/scale13-small-epoch04/valid-variant-score/report.json": "19224eae63273b62b0421fc48c32704b86418a2bfb21adda2a90449cfa0988d8", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5" + } + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..28c50d9e3a33f6accbd9ce3cdeca8bdbb9b481a6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/late_fixed_links81.py @@ -0,0 +1,145 @@ +"""Predeclare fixed-prior output links for the remaining main training checkpoints.""" +import argparse +from datetime import datetime, timezone +import hashlib +import json +import math +from pathlib import Path + +from verify_package import inference_signature, verify_package + +ROOT = Path('/workspace/research') +STUDY = ROOT / 'studies/late_fixed_links81' +HERE = Path(__file__).resolve().parent + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def write_json(path, value): + path = Path(path) + temporary = path.with_suffix(path.suffix + '.tmp') + temporary.write_text(json.dumps(value, indent=2, allow_nan=False) + '\n') + temporary.replace(path) + + +def parent_ready(path): + path = Path(path) + try: + return (json.loads((path / 'pipeline-state.json').read_text())['phase'] == 'complete' + and (path / 'evaluation.json').exists()) + except (FileNotFoundError, json.JSONDecodeError, KeyError): + return False + + +def parent_receipt(path, capacity, epoch, plan): + path = Path(path) + assert parent_ready(path), str(path) + package = path / 'package' + verify_package(package) + signature = inference_signature(package) + assert json.loads((path / 'evaluation.json').read_text())['package_inference_signature'] == signature + provenance = json.loads((package / 'training_provenance.json').read_text()) + assert provenance['config'] == plan['capacities'][capacity]['config'] + assert provenance['global_step'] == epoch * 41020 and provenance['optimizer_steps'] == epoch * 10255 + assert provenance['epoch'] == epoch and provenance['offset'] == 0 + assert provenance['seed'] == 17 and provenance['schedule_epochs'] == 14 + assert provenance['objective'] == 'variance_balanced' and provenance['examples'] == 41020 + assert provenance['training_arguments']['batch_size'] == 4 + assert provenance['family_loss_scales'] == plan['family_loss_scales'] + source = provenance['provenance'] + assert source['split'] == 'train' and source['pretrained'] is False and source['world_size'] == 1 + assert source['train_manifest_sha256'] == plan['train_manifest_sha256'] + manifest = json.loads((package / 'package_manifest.json').read_text()) + assert manifest['parameters'] == plan['capacities'][capacity]['parameters'] + assert not any((package / name).exists() for name in [ + 'splice_calibration.json', 'checkpoint_ensemble_policy.json', 'rc_ensemble_policy.json']) + tracks = json.loads((path / 'valid-score/report.json').read_text()) + variants = json.loads((path / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + assert all(math.isfinite(value['model']) for value in variants.values()) + return dict(experiment=str(path), capacity=capacity, epoch=epoch, inference_signature=signature, + reward=tracks['overall_mean_family_pearson'], + sha256={str(p): digest(p) for p in [path / 'evaluation.json', path / 'evaluation-package.json', + path / 'valid-score/report.json', path / 'valid-variant-score/report.json', + package / 'training_provenance.json']}) + + +def prepare(): + STUDY.mkdir(exist_ok=True) + assert not (STUDY / 'plan.json').exists() + link69_path = ROOT / 'studies/fixed_link69/plan.json' + link69 = json.loads(link69_path.read_text()) + link61_path = ROOT / 'studies/fixed_link61/plan.json' + link61 = json.loads(link61_path.read_text()) + members = {} + for capacity, epoch, previous in [('large', 4, 2), ('small', 8, 4), ('large', 8, 4), + ('small', 14, 8), ('large', 14, 8)]: + key = f'{capacity}-epoch{epoch:02d}' + members[key] = dict(capacity=capacity, epoch=epoch, previous_epoch=previous, + parent=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{epoch:02d}'), + previous_raw=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{previous:02d}'), + output=str(ROOT / 'experiments' / f'calib81-{key}'), study=str(STUDY / key)) + (STUDY / key).mkdir(exist_ok=True) + capacities, evidence = {}, [link69_path, link61_path, ROOT / 'studies/fixed_link57/complete-summary.json', + ROOT / 'studies/scale13/study-cached-io.json', ROOT / 'studies/scale13/loss_scales.json', + Path('/workspace/data/public/train.parquet'), ROOT / 'experiments/calib57/package/splice_calibration.json'] + for capacity in ['small', 'large']: + package = ROOT / 'experiments' / f'scale13-{capacity}-epoch04/package' + provenance_path = package / 'training_provenance.json' + provenance = json.loads(provenance_path.read_text()) + capacities[capacity] = dict(config=provenance['config'], parameters=13290504 if capacity == 'small' else 215626632) + evidence.append(provenance_path) + helpers = set(link69['helper_sha256']) | { + 'late_fixed_links81.py', 'run_late_fixed_links81.py', 'diagnose_track_uncertainty.py', + 'diagnose_remaining_variant_uncertainty.py'} + materialization = dict(link69['materialization']) + materialization['installation'] = 'Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.' + latency = dict(link61['native_latency_template']) + latency['scope'] = 'Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.' + latency['memory_evidence'] = 'One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.' + latency['minimum_free_GiB'] = 16 + plan = dict(created_utc=datetime.now(timezone.utc).isoformat(), members=members, capacities=capacities, + train_manifest_sha256=digest('/workspace/data/public/train.parquet'), + family_loss_scales=json.loads((ROOT / 'studies/scale13/loss_scales.json').read_text())['family_loss_scales'], + training_prior=link69['training_prior'], training_prior_sha256=link69['training_prior_sha256'], + calibration_reference=str(ROOT / 'experiments/calib57/package/splice_calibration.json'), + helper_sha256={str(HERE / name): digest(HERE / name) for name in sorted(helpers)}, + evidence_sha256={str(path): digest(path) for path in evidence}, materialization=materialization, + native_latency_template=latency, evaluation_lock=str(ROOT / 'studies/scale13/evaluation.lock'), + bounded_lock=str(ROOT / 'studies/junction_discovery39/bounded-diagnostic.lock'), + minimum_free_GiB=16, native_timeout_seconds=1800, materialization_timeout_seconds=1800, + parent_minimum_reward_gain=1e-6, publication_minimum_reward_gain=1e-6, + hypothesis='Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.', + admission='Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.', + validation='Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.', + diagnostics='After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.', + nonimprovement_policy='A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.', + limitations=['The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.', + 'The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.', + 'Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.', + 'No test labels, variant-label training, external parameters or inference caches are used.']) + write_json(STUDY / 'plan.json', plan) + print(json.dumps(dict(plan=str(STUDY / 'plan.json'), sha256=digest(STUDY / 'plan.json'), members=list(members))), flush=True) + + +def audit(): + assert not (STUDY / 'parent-check.json').exists() + plan = json.loads((STUDY / 'plan.json').read_text()) + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + receipts = [parent_receipt(ROOT / 'experiments' / name, capacity, epoch, plan) + for name, capacity, epoch in [('scale13-small-epoch04', 'small', 4), ('scale13-large-epoch02', 'large', 2)]] + write_json(STUDY / 'parent-check.json', dict(created_utc=datetime.now(timezone.utc).isoformat(), + plan_sha256=digest(STUDY / 'plan.json'), actual_completed_parent_checks=receipts, + gpu_used=False, packages_changed=False, no_new_predictions_or_parameter_fitting=True)) + print(json.dumps(dict(phase='parent_audit_complete', report_sha256=digest(STUDY / 'parent-check.json'))), flush=True) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('action', choices=['prepare', 'audit']) + globals()[parser.parse_args().action]() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..3e3973883128f4aea458210d43790edce69d380a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_late_fixed_links81.py @@ -0,0 +1,268 @@ +"""Execute one predeclared late fixed-link follow-up after its complete raw parent.""" +import argparse +from datetime import datetime, timezone +import fcntl +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from check_splice_link_reuse import check_link_only +from late_fixed_links81 import ROOT, STUDY, HERE, digest, write_json, parent_ready, parent_receipt +from verify_package import inference_signature, verify_package + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('member') + args = parser.parse_args() + plan_path = STUDY / 'plan.json' + plan = json.loads(plan_path.read_text()) + member = plan['members'][args.member] + study, parent, previous, output = [Path(member[k]) for k in ['study','parent','previous_raw','output']] + lock = (study / 'controller.lock').open('a') + fcntl.flock(lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + complete = study / 'complete.json' + if complete.exists(): + assert json.loads(complete.read_text())['master_plan_sha256'] == digest(plan_path) + print(json.dumps(dict(phase='already_complete_no_rerun', member=args.member)), flush=True) + return + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + assert digest(plan['training_prior']) == plan['training_prior_sha256'] + audit = json.loads((STUDY / 'parent-check.json').read_text()) + assert audit['plan_sha256'] == digest(plan_path) and len(audit['actual_completed_parent_checks']) == 2 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2', + PYTHONNOUSERSITE='1', PYTHONDONTWRITEBYTECODE='1') + for key in ['PYTHONPATH','PYTHONHOME','SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def status(phase, **extra): + record = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), member=args.member, **extra) + write_json(study / 'state.json', record) + print(json.dumps(record), flush=True) + + def finish(phase, **extra): + write_json(complete, dict(phase=phase, master_plan_sha256=digest(plan_path), + created_utc=datetime.now(timezone.utc).isoformat(), **extra)) + status(phase, **extra) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study / f'{phase}.log').open('a') as stream: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=stream, stderr=subprocess.STDOUT, start_new_session=True) + try: + code = process.wait(timeout=timeout) + assert code == 0, (phase, code) + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try:process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL); process.wait() + + def wait_memory(): + while int(subprocess.check_output(['nvidia-smi','--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB'] * 1024: + time.sleep(5) + + status('waiting_for_complete_declared_raw_parents') + while not (parent_ready(parent) and parent_ready(previous)): + time.sleep(10) + receipts = dict(parent=parent_receipt(parent, member['capacity'], member['epoch'], plan), + previous=parent_receipt(previous, member['capacity'], member['previous_epoch'], plan)) + resolved_path = study / 'resolved-parents.json' + resolved = dict(master_plan_sha256=digest(plan_path), member=member, receipts=receipts) + if resolved_path.exists():assert json.loads(resolved_path.read_text()) == resolved + else:write_json(resolved_path, resolved) + admitted = receipts['parent']['reward'] > receipts['previous']['reward'] + plan['parent_minimum_reward_gain'] + gate = dict(admitted=admitted, parent_reward=receipts['parent']['reward'], + previous_raw_reward=receipts['previous']['reward'], minimum_gain=plan['parent_minimum_reward_gain']) + if not (study / 'parent-gate.json').exists():write_json(study / 'parent-gate.json', gate) + else:assert json.loads((study / 'parent-gate.json').read_text()) == gate + if not admitted: + finish('rejected_at_parent_gain_gate', **gate) + return + output.mkdir(exist_ok=True) + package = output / 'package' + construction_path = study / 'construction-check.json' + if not construction_path.exists(): + if package.exists(): + try:verify_package(package);check_link_only(parent / 'package', package) + except (OSError,ValueError,KeyError,AssertionError): + package.rename(package.with_name('package-incomplete-' + str(time.time_ns()))) + if not package.exists(): + execute('build', [HERE / 'make_splice_probability_candidate.py', parent / 'package', package, + '--training-prior', plan['training_prior']]) + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] + assert json.loads((package / 'package_manifest.json').read_text())['parameters'] == plan['capacities'][member['capacity']]['parameters'] + before = (parent / 'package/predict.py').read_text() + marker = 'from reference.alphagenome_50m.model import CompactModel, ModelConfig\n' + old = 'data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16),' + new = 'data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16),' + assert before.count(marker) == before.count(old) == 1 + expected = before.replace(marker, marker + 'from splice_output import splice_output_scores\n').replace(old, new) + assert (package / 'predict.py').read_text() == expected + for name in ['infer.sh','perturb.sh']: + assert digest(package / name) == digest(parent / 'package' / name) + assert json.loads((package / 'splice_calibration.json').read_text()) == json.loads(Path(plan['calibration_reference']).read_text()) + history = package / 'training_history'; history.mkdir(exist_ok=True) + additions = {'late-fixed-links81-master-plan.json': plan_path, + 'late-fixed-links81-resolved-parents.json': resolved_path, + 'late-fixed-links81-parent-check.json': STUDY / 'parent-check.json', + 'late_fixed_links81.py': HERE / 'late_fixed_links81.py', + 'run_late_fixed_links81.py': Path(__file__)} + for name, source in additions.items(): + (history / name).write_bytes(source.read_bytes()) + readme = package / 'README.md' + body = readme.read_text().replace( + 'In preliminary\ndiagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed.', + "The complete reports below determine this checkpoint's track and variant tradeoffs.") + note = (f'\nThis predeclared late fixed-link follow-up uses the complete {member["epoch"]}-epoch ' + f'{member["capacity"]} main model. It applies the original four training-only offsets ' + 'without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. ' + 'Both entry points use the same link. All200 validation windows and all13 development metrics ' + 'are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. ' + 'A publication additionally requires native replay and three sampled runtime estimates below four hours.\n') + if note.strip() not in body:body += note + readme.write_text(body) + manifest = json.loads((package / 'package_manifest.json').read_text()) + for name in ['README.md', *['training_history/' + name for name in additions]]: + manifest['files'][name] = digest(package / name) + write_json(package / 'package_manifest.json', manifest) + verification = verify_package(package) + write_json(construction_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), resolved_parents_sha256=digest(resolved_path), + package_inference_signature=inference_signature(package), exact_link_only_checks=checks, + parameters=manifest['parameters'], network_updates=0, prior_refitted=False, package_verified=True)) + construction = json.loads(construction_path.read_text()) + assert construction['master_plan_sha256'] == digest(plan_path) + assert construction['resolved_parents_sha256'] == digest(resolved_path) + verify_package(package) + signature = inference_signature(package) + assert construction['package_inference_signature'] == signature + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] and checks['candidate'] == signature + native = study / 'native-recovery' + if not (native / 'report.json').exists() or not (output / 'valid-variants').exists(): + status('waiting_for_bounded_native_and_derivation_slot') + with Path(plan['bounded_lock']).open('a') as probe_lock: + fcntl.flock(probe_lock, fcntl.LOCK_EX);wait_memory() + if not (native / 'report.json').exists(): + execute('native', [HERE / 'check_native_recovery.py', package, native], plan['native_timeout_seconds']) + if not (output / 'valid-variants').exists(): + materialization = dict(plan['materialization'], candidates=[dict(parent=str(parent), candidate=str(output), + parent_inference_signature=checks['parent'], candidate_inference_signature=signature)]) + path = study / 'materialization-plan.json' + if path.exists():assert json.loads(path.read_text()) == materialization + else:write_json(path, materialization) + execute('materialize', [HERE / 'materialize_splice_link_outputs_v2.py', path], plan['materialization_timeout_seconds']) + proof = json.loads((native / 'report.json').read_text()) + assert proof['package_inference_signature'] == signature and proof['all_arrays_bit_identical'] + assert proof['native_vendored_entrypoints_passed'] and proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks':30,'perturbations':22} + derivation = json.loads((output / 'valid-variants/perturb-link-derivation.json').read_text()) + assert derivation['checks']['candidate'] == signature and derivation['checks']['parent'] == checks['parent'] + assert derivation['native_arrays_bit_exact'] == 36 and derivation['queries'] == 2697 + if not (output / 'evaluation.json').exists(): + status('waiting_for_full_evaluation_slot') + with Path(plan['evaluation_lock']).open('a') as evaluation_lock: + fcntl.flock(evaluation_lock, fcntl.LOCK_EX);wait_memory() + execute('evaluate', [HERE / 'evaluate_package.py', package, output, + '--resume-partial-tracks','--resume-partial-perturbations']) + assert json.loads((output / 'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output / 'valid-score/report.json').read_text()) + variants = json.loads((output / 'valid-variant-score/report.json').read_text()) + raw_tracks = json.loads((parent / 'valid-score/report.json').read_text()) + raw_variants = json.loads((parent / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None and len(variants) == 13 + affected = {'sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'} + for family, report in tracks['families'].items(): + if family != 'splice_sites':assert report == raw_tracks['families'][family],family + for name, report in variants.items(): + assert math.isfinite(report['model']) + for field in ['rows','perturbations','metric','anchors']:assert report[field] == raw_variants[name][field],(name,field) + if name not in affected:assert report == raw_variants[name],name + comparison_path = study / 'matched-comparison.json' + if not comparison_path.exists(): + write_json(comparison_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), reward=tracks['overall_mean_family_pearson'], + reward_delta=tracks['overall_mean_family_pearson']-raw_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-raw_tracks['families'][k]['mean_pearson'] for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-raw_variants[k]['model'] for k,v in variants.items()}, + unaffected_family_reports_exact=10, unaffected_variant_reports_exact=8, package_inference_signature=signature)) + decision_path = study / 'publication-decision.json' + if not decision_path.exists(): + current = json.loads(Path('/workspace/submission/validation_tracks.json').read_text())['overall_mean_family_pearson'] + eligible = tracks['overall_mean_family_pearson'] > current + plan['publication_minimum_reward_gain'] + runtime_passed = None + if eligible: + latency = study / 'native-latency';latency.mkdir(exist_ok=True) + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency / 'plan.json').exists():assert json.loads((latency / 'plan.json').read_text()) == timing_plan + else:write_json(latency / 'plan.json', timing_plan) + if not (latency / 'report.json').exists(): + execute('native-latency', [HERE / 'measure_native_latency.py', latency / 'plan.json']) + timing = json.loads((latency / 'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + names = ['development_mixture_seconds','slowest_observed_stratum_mean_seconds','pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in names) + runtime_passed = max(estimates[k] for k in names) <= 14400 + write_json(study / 'runtime-gate.json', dict(passed=runtime_passed, estimates=estimates, + report_sha256=digest(latency / 'report.json'), scope='Shared-load extrapolations, not guaranteed upper bounds.')) + if runtime_passed:execute('publish', [HERE / 'promote_if_better.py', package, output]) + write_json(decision_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward=tracks['overall_mean_family_pearson'], current_reward_at_review=current, + eligible_at_review=eligible, runtime_passed=runtime_passed, + publication='Atomic promoter rechecks the current strongest package; see publish.log if run.')) + for phase, helper in [('tracks','diagnose_track_uncertainty.py'), + ('remaining-variants','diagnose_remaining_variant_uncertainty.py')]: + report_path = study / f'paired-{phase}.json' + marker_path = study / f'paired-{phase}-complete.json' + if marker_path.exists(): + assert json.loads(marker_path.read_text())['output_sha256'] == digest(report_path) + continue + def diagnostic_complete(): + try: + report = json.loads(report_path.read_text()) + assert Path(report['candidate']).resolve() == output.resolve() + assert Path(report['control']).resolve() == parent.resolve() + assert report['replicates'] == 2000 and report['seed'] == 1729 + assert report['diagnostic_only_no_parameter_fitting'] + leaf = 'valid-score/report.json' if phase == 'tracks' else 'valid-variant-score/report.json' + assert report['report_sha256'] == {'candidate':digest(output / leaf),'control':digest(parent / leaf)} + if phase == 'tracks': + assert len(report['point_deltas']) == 12 and 'reward' in report['point_deltas'] + assert set(report['bootstrap']) == {'chromosome','chromosome_10Mb_blocks'} + for result in report['bootstrap'].values(): + assert set(result['deltas']) == set(report['point_deltas']) + else: + assert set(report['evaluations']) == affected | {'caqtl','bqtl_spi1'} + for item in report['evaluations'].values(): + assert set(item['bootstrap']) == {'variant_id','genomic_block'} + for result in item['bootstrap'].values(): + assert result['reference_replicates_checked'] >= 32 + assert result['maximum_absolute_metric_error'] < 1e-10 + return True + except (FileNotFoundError,json.JSONDecodeError,AssertionError,KeyError): + return False + if not diagnostic_complete(): + execute('paired-' + phase, [HERE / helper, output, parent, report_path, + '--replicates','2000','--seed','1729']) + assert diagnostic_complete(), ('incomplete diagnostic', phase) + write_json(marker_path, dict(output_sha256=digest(report_path), master_plan_sha256=digest(plan_path))) + finish('complete', reward=tracks['overall_mean_family_pearson'], + matched_comparison_sha256=digest(comparison_path), publication_decision=json.loads(decision_path.read_text()), + diagnostics='All11families/sevenvariants, with eight unaffected official variant reports exact; not anall13bootstrap.') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..8a387fd3ba1fb0eddf3c22ca10c91600e21aef6b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/training_provenance.json @@ -0,0 +1,154 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 8, + "offset": 0, + "global_step": 328160, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-small", + "config": "/workspace/research/studies/scale13/scale13-small.yaml", + "resume": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 4, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 73897.99778079404, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 82040, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + }, + "inference_ensemble": { + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-26T01:07:17.491176+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib81-small-epoch08/package", + "parent_inference_signature": { + "last.pt": "4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 13290504, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..1fda528c8fec80dd01846baae286d46ac958e7ef --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5952409681697739, + "valid_tracks": 167, + "per_track_pearson": [ + 0.7162916999380372, + 0.7378131755710019, + 0.46699415787558957, + 0.6626763805848139, + 0.47957350257270126, + 0.479734970407924, + 0.4970940262195392, + 0.5806761387439995, + 0.5945559355563002, + 0.6612908227175676, + 0.5966167277374154, + 0.5733124705230662, + 0.40675584760839256, + 0.5393093982109629, + 0.6972448939720401, + 0.5206384267448704, + 0.571427589034842, + 0.5411875262060158, + 0.5735208209497836, + 0.5582663077892955, + 0.5268427160926525, + 0.4881797414935095, + 0.5675492364446837, + 0.542266094075655, + 0.5411826026333192, + 0.5200700483376121, + 0.4979223942506745, + 0.5188049841637596, + 0.5292335565958949, + 0.5907420963366347, + 0.5061780776365333, + 0.54566843877601, + 0.5002178229711586, + 0.4189104743577026, + 0.3889577261261861, + 0.6036152391869629, + 0.5403740943228217, + 0.6504639220820564, + 0.53133849091403, + 0.5230685397228485, + 0.6018918450069887, + 0.6033332320590037, + 0.5200906883922227, + 0.4960469327596779, + 0.5655896055250774, + 0.5834668219683927, + 0.45171367042805194, + 0.5049124310779565, + 0.49924874303004113, + 0.5436910491525456, + 0.5137663407255305, + 0.47975884767519156, + 0.5787304391010564, + 0.6144656911205332, + 0.47363664926808624, + 0.6995247122798669, + 0.8148938847977888, + 0.5690791707911121, + 0.6082186617599318, + 0.6564837960682501, + 0.7841626636666975, + 0.5185378695182153, + 0.7059709297201038, + 0.6948720310498536, + 0.6495835728937592, + 0.5253599152984737, + 0.588731099201362, + 0.6528456337618102, + 0.8042670293921365, + 0.723895397413535, + 0.5414320595662097, + 0.49411686552664247, + 0.46875660027410404, + 0.5568405560665932, + 0.5179132461091569, + 0.507800654182973, + 0.5130976234103813, + 0.5541379987387496, + 0.6002391543735935, + 0.5087339645850157, + 0.5919280723722237, + 0.6142512770392202, + 0.5300508921314372, + 0.5034292095606672, + 0.5389522245394467, + 0.5305387751988354, + 0.507282883387854, + 0.4981489434996194, + 0.46806681761506297, + 0.52316346287167, + 0.5263133952224991, + 0.5221247138595094, + 0.5018335109956202, + 0.520670656308489, + 0.5825749870608544, + 0.4092541230028647, + 0.4618243592000733, + 0.4505531123381647, + 0.510117701170264, + 0.5263238883242346, + 0.47488682339670657, + 0.47868899427620376, + 0.5229174210101531, + 0.4821024397066225, + 0.5630076650717932, + 0.569938281494487, + 0.5557322825378548, + 0.5380281375462908, + 0.5623675659444302, + 0.5716059042720775, + 0.6248342817480647, + 0.5455725737736211, + 0.5396750774878486, + 0.45799673541935665, + 0.48360739405646347, + 0.45727464014276564, + 0.4968241423382643, + 0.4995821953035191, + 0.5154506989807865, + 0.49201775926195657, + 0.6715319972962915, + 0.6875149095461972, + 0.7530742575663724, + 0.7435988081742322, + 0.733881602968143, + 0.591628821904697, + 0.8711441593742649, + 0.7425951935015277, + 0.7797602608468281, + 0.694855239313783, + 0.7386695732961909, + 0.5770165427982271, + 0.6062197773920772, + 0.5793290692503432, + 0.7249703931577279, + 0.7594908083475518, + 0.7117573044292618, + 0.554571305916985, + 0.43884590394519296, + 0.82904952017837, + 0.8000152790456309, + 0.820510085346221, + 0.813988547843336, + 0.5902924167875748, + 0.8814629681982501, + 0.7915931249653146, + 0.758562348612762, + 0.7475919730806571, + 0.8501340371492583, + 0.8296446446512509, + 0.6351075812265518, + 0.6383163340424183, + 0.739137985667799, + 0.767264478744637, + 0.7840442465604132, + 0.6225356869487015, + 0.7568311546206139, + 0.6114874284477966, + 0.613371649947908, + 0.7028035028944712, + 0.617362506125388, + 0.790612869670294, + 0.7922909151359582, + 0.6615220593868638, + 0.6098597120450119, + 0.6073739151483419, + 0.7560995701651257 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.670826748766955, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5824667519383658, + 0.9332862050760857, + 0.8001780882065873, + 0.5432750140956559, + 0.5572677682888533, + 0.5666389251019613, + 0.9467600684064935, + 0.7403824678826315, + 0.48445624100265156, + 0.3714184864098835, + 0.8976867336939629, + 0.7277344881543165, + 0.4525709287940511, + 0.5430615177457364, + 0.5255623614010002, + 0.8238204703541911, + 0.8073194868224742, + 0.8584431616618042, + 0.8779839076782129, + 0.9052363457964263, + 0.719780750270716, + 0.6427492449816675, + 0.9143462335579083, + 0.5565913330779558, + 0.5721370129849686, + 0.7923190484135452, + 0.47653560397598266, + 0.6938056461695301, + 0.5505119287805298, + 0.5780493458403257, + 0.6566001248402331, + 0.49318612488376196, + 0.8520823995960877, + 0.6383315900141471, + 0.8903059441030028, + 0.8983080427069753, + 0.46612114110393843, + 0.8391657244704934, + 0.884072804992339, + 0.8716414033178751, + 0.840989871389998, + 0.6811564200542722, + 0.7166070380605885, + 0.6587977600621455, + 0.5573563112009491, + 0.6511197140501481, + 0.746067258864535, + 0.5191079480417531, + 0.620891368701055, + 0.42748272910458973, + 0.40866438832703367, + 0.5918371582360831, + 0.5306464318591985, + 0.8366990770588079, + 0.466559028007227, + 0.43191083511402056, + 0.4933912174928751, + 0.5447282353495627, + 0.5107703866531816, + 0.5448786986569913, + 0.48470097595324824, + 0.6365974030272086, + 0.5284612773691266, + 0.5187370514258022, + 0.5483516592851078, + 0.5552641043968543, + 0.583391432272753, + 0.5132177305472924, + 0.41198475159446085, + 0.4562563223873448, + 0.5295115940016831, + 0.6779815373434065, + 0.5408602183213013, + 0.663455887880042, + 0.3935410335653646, + 0.8232091389425696, + 0.9037956355266964, + 0.6913924967921895, + 0.47292873670754615, + 0.4887216450989171, + 0.5674317014006338, + 0.5418335609187044, + 0.48154863338697795, + 0.5209647717705627, + 0.6071574464161671, + 0.5235803619052978, + 0.6434748929351894, + 0.5333454817238383, + 0.6630510385028762, + 0.5333813208427338, + 0.659665570036695, + 0.43356634954339773, + 0.7447939908361002, + 0.503934487916297, + 0.734364718214861, + 0.5324655215527793, + 0.7774735366271533, + 0.7466491988467144, + 0.40313874302981495, + 0.41190978370612263, + 0.5061715396007448, + 0.4007828689013087, + 0.3839316760596631, + 0.7549264521380602, + 0.7287835991444441, + 0.4328291272949076, + 0.5195467976237489, + 0.6235801598402857, + 0.8581219446496978, + 0.5813746259265256, + 0.5974259156315641, + 0.6134253527693644, + 0.5440442995635103, + 0.47075712763634275, + 0.7973051978717113, + 0.6061106308686245, + 0.5243419077269591, + 0.580237213268375, + 0.6727447375078085, + 0.49984188491551573, + 0.5614839933566723, + 0.5249556194956881, + 0.5914481585209005, + 0.5780706135448355, + 0.6470545488673821, + 0.5826425132705844, + 0.6751917610435396, + 0.696800593287926, + 0.719443461086955, + 0.6622432637110794, + 0.49337774093564113, + 0.6351599112153216, + 0.6715339592254975, + 0.8103669590687673, + 0.47353644212379636, + 0.36250629840195553, + 0.6254050506281729, + 0.5079439281621954, + 0.6712757214479562, + 0.6343860151482099, + 0.6440012691979903, + 0.5881293666335096, + 0.5153611469993181, + 0.797204337355543, + 0.6548318163775092, + 0.7745009151031941, + 0.4256280547124505, + 0.4034047772554786, + 0.4306035671404724, + 0.38576507139960586, + 0.906277409397454, + 0.3532301252616735, + 0.9116447150042035, + 0.7861912090194451, + 0.4934191193580781, + 0.3869093832432726, + 0.38831750479188437, + 0.3947179503561932, + 0.6813315322568991, + 0.7859597859493263, + 0.7760711860585147, + 0.6951931283659918, + 0.5929581957649741, + 0.5147579852068106, + 0.7069897426688841, + 0.6473356640225174, + 0.5543628563189079, + 0.6735618470187656, + 0.4876180262393008, + 0.5254375872253078, + 0.4453282233580886, + 0.8425773035488456, + 0.49829254182475874, + 0.5110548132638932, + 0.6859197795996214, + 0.7228380623550654, + 0.5564276125228771, + 0.37644013166577533, + 0.8868788937069464, + 0.4079372012358814, + 0.7995983963247714, + 0.5373996312239001, + 0.6382877867835853, + 0.7620860980157632, + 0.48075303587879464, + 0.6028263720134345, + 0.6413658926653443, + 0.8415043451972446, + 0.4455319407473947, + 0.6443784981690422, + 0.6092850011830547, + 0.7402016431031928, + 0.9057226660594492, + 0.47804017742706195, + 0.6261388938014555, + 0.5265931811554839, + 0.6832197244352894, + 0.7520728510287547, + 0.5918126313761135, + 0.49458012146651953, + 0.8822385790519297, + 0.8945506310963736, + 0.7669060690463697, + 0.8508796062040602, + 0.8602843979630396, + 0.8515022707588459, + 0.8634631731741025, + 0.39927507482507996, + 0.424972091864971, + 0.5479723580830785, + 0.5005156469654147, + 0.4945830901148757, + 0.6485466583481941, + 0.8770318898019306, + 0.8599644581421348, + 0.7170408330388153, + 0.8434546444867099, + 0.8413473611871792, + 0.8716785122663792, + 0.9173589041002427, + 0.5848166993212536, + 0.680574816675155, + 0.729434774583992, + 0.7025271405419559, + 0.8413342014462121, + 0.6333379292077493, + 0.922904487673823, + 0.895605692838129, + 0.8834802549658353, + 0.9115694673879363, + 0.9011000282938454, + 0.8902852211595299, + 0.8462098221124233, + 0.8712108651090682, + 0.9205035211886355, + 0.49923603610056205, + 0.9355486587178291, + 0.890649093078824, + 0.8349715429626435, + 0.6526134921282833, + 0.8816959115514925, + 0.5669373114103688, + 0.7623451861174072, + 0.8840063136947602, + 0.6496405935895815, + 0.6403125761200732, + 0.7327803379401018, + 0.5613629684140392, + 0.5965195970657386, + 0.7576728010490189, + 0.8977696897861605, + 0.7856343814799945, + 0.8336267370642684, + 0.674041429364152, + 0.8803172132825764, + 0.8213382315103144, + 0.8513542899912828, + 0.9157303932797292, + 0.8394300558047425, + 0.9105050389349986, + 0.8811262923874229, + 0.5758224123551081, + 0.8879657658223847, + 0.619855963269435, + 0.5908799305676532, + 0.8942263748676001, + 0.9143209717646846, + 0.8940970129018061, + 0.6701740115673687, + 0.6848323657859168, + 0.697562939018539, + 0.8107372287618676, + 0.9076909446351265, + 0.7554505730748008, + 0.9252933924034147, + 0.9300227529719179, + 0.647673594771781, + 0.37163331401572575, + 0.5631254455487876, + 0.908619919814417, + 0.5053697491363519, + 0.6642523980649295, + 0.8964828247353532, + 0.8432571697163289, + 0.6908537842785079, + 0.8765397768569035, + 0.904680971832611, + 0.8998892166165057, + 0.8911823196194996, + 0.8741934183551183, + 0.9022849936216465, + 0.8707484469665733, + 0.940906248683442, + 0.7951577591160633, + 0.846225019705623, + 0.917720892187818, + 0.6116932947610603, + 0.6196649275165738, + 0.6206533858220158, + 0.5904888859993278, + 0.9173593537716039, + 0.8160972790431693, + 0.8582436479313579, + 0.8145263649957645, + 0.7822111610638648 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5561831780553281, + "valid_tracks": 12, + "per_track_pearson": [ + 0.523536379779026, + 0.5705213466993627, + 0.5361541869235945, + 0.5665946471959361, + 0.567029863849686, + 0.5959088217200769, + 0.5114726212131325, + 0.5630665630248907, + 0.5229219672665067, + 0.5613836900667785, + 0.540153855959525, + 0.6154541929654223 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.3401463358374285, + "valid_tracks": 546, + "per_track_pearson": [ + 0.33922519936924705, + 0.38061794973390983, + 0.3128784842557797, + 0.2065332623377263, + 0.11844066942702197, + 0.3251380351746678, + 0.2235118964826846, + 0.21878855366853908, + 0.3383838259468754, + 0.3219921881234121, + 0.3629830943284621, + 0.2152854493944153, + 0.2280195173915163, + 0.18337403434081312, + 0.19064502967790203, + 0.2357573941746563, + 0.2869749585548755, + 0.3978219214323346, + 0.3766199903187956, + 0.3790941976670184, + 0.30203718387083384, + 0.30512639740851016, + 0.44762658771296976, + 0.25406027382276003, + 0.3066198130191791, + 0.3642626554149573, + 0.32438596981276874, + 0.36157280986040474, + 0.37558247350474644, + 0.3937861507985989, + 0.24499156738945058, + 0.4124354695250736, + 0.2566029242193943, + 0.22540841856003282, + 0.417018276159388, + 0.34868149963680184, + 0.3905810328433971, + 0.43889876329518107, + 0.38877950789393206, + 0.40805296631933186, + 0.4570362134732815, + 0.37966617099453503, + 0.3623109341235046, + 0.3137859141777968, + 0.22203712824657, + 0.30097445738860734, + 0.31884762812104034, + 0.3646190343396537, + 0.19893658558685445, + 0.2505107288253443, + 0.3931457517563688, + 0.38720458933318763, + 0.3919605397971376, + 0.2812545975342322, + 0.4521204743449956, + 0.4413411942287034, + 0.2659217727799998, + 0.45043211067777433, + 0.3124822113893087, + 0.4311864597210517, + 0.41258069820370197, + 0.2773758817931145, + 0.3576902224599577, + 0.30087017744054056, + 0.3756953166329538, + 0.43233226498664223, + 0.42923073670132345, + 0.2602235437969275, + 0.2048687536626288, + 0.17798844568792255, + 0.338683086770354, + 0.3862460184009496, + 0.3556422136275623, + 0.33274407454616844, + 0.38328495475570057, + 0.30899861291339875, + 0.36288986080506525, + 0.3516041382602295, + 0.4158028871897335, + 0.3990691779824238, + 0.35062145032324216, + 0.2049585755144693, + 0.35940662696794334, + 0.45838546843713907, + 0.20488479319149477, + 0.36147959511712074, + 0.32715542376688134, + 0.4467972390651169, + 0.2961894401860228, + 0.37942509819718867, + 0.3095077651687737, + 0.35749530047355643, + 0.37050638575327743, + 0.18663766030821582, + 0.39224568989137854, + 0.3251246886815894, + 0.37970575229433917, + 0.3722288203467799, + 0.34281994434641155, + 0.22239199370910076, + 0.35357993494844014, + 0.36295088462340697, + 0.39704236664136144, + 0.4477300637596801, + 0.4151417726631952, + 0.39530694048760506, + 0.2971741979292083, + 0.4022195876054256, + 0.3067341715046226, + 0.31368385622147377, + 0.32714841290931235, + 0.3298560527034187, + 0.3543129414797267, + 0.35351438195425516, + 0.29400953624012954, + 0.37909373000379987, + 0.35134989275219414, + 0.35954771792658746, + 0.3511202815523055, + 0.38590770933399865, + 0.39951927479014304, + 0.3730862473645405, + 0.27499751128031835, + 0.3156794343005407, + 0.33659907411670376, + 0.3837860808531958, + 0.36118493767862847, + 0.39219692421262675, + 0.4584722967404524, + 0.39906869010225826, + 0.40910104596638003, + 0.31162273123186596, + 0.41186768024575454, + 0.43424348809778607, + 0.4607765366546753, + 0.34256183985840816, + 0.4414556626259667, + 0.3498946697905253, + 0.19314908346133702, + 0.4027763710004049, + 0.4113299180676122, + 0.35434862544968543, + 0.4057805517399682, + 0.3116478001641571, + 0.3339631317237731, + 0.35623776252107875, + 0.34940711315756845, + 0.4680872496497471, + 0.333413757719176, + 0.3301461618839739, + 0.4117128110242245, + 0.31946863753647087, + 0.37692183029385706, + 0.37640902027205236, + 0.3483998701962421, + 0.3362496980113014, + 0.4378747471388163, + 0.4666699970245334, + 0.3223466592419855, + 0.3006106946104221, + 0.25820786004214435, + 0.3293192782281211, + 0.33383442635416166, + 0.3855410382637437, + 0.3332225865977922, + 0.31169676099699356, + 0.3546996747391966, + 0.2769261834255426, + 0.21639083841642812, + 0.3985059660099338, + 0.35122759726815056, + 0.3003501094821228, + 0.3591529684969189, + 0.4555813796937788, + 0.3649519065226908, + 0.21875239307582842, + 0.24594741768855435, + 0.33148885718671517, + 0.40867639243622095, + 0.415249004058515, + 0.4020281870201112, + 0.41541252668395623, + 0.2690384076333056, + 0.4032514360468392, + 0.35719063981857985, + 0.28484085519566094, + 0.34561356131535226, + 0.2484049305950529, + 0.455367022866902, + 0.38384391613529545, + 0.3126542560182467, + 0.24610094986182154, + 0.2391998201790172, + 0.34909333684851784, + 0.3651906371930594, + 0.31117730135003013, + 0.3555212548234165, + 0.378520208303794, + 0.26839452502716793, + 0.3568669724403861, + 0.18370985052327976, + 0.18747858124275582, + 0.17046422238009604, + 0.1592093914805894, + 0.22927787671754551, + 0.34963487817013256, + 0.34916786366896685, + 0.2987472736380806, + 0.1626275668298449, + 0.3145663890032205, + 0.40862605691060916, + 0.41997484660073686, + 0.42796165870396924, + 0.3822765821133199, + 0.3757667352418622, + 0.29926089861135463, + 0.4121989648542746, + 0.41749495475395126, + 0.375362850339652, + 0.4155947345175214, + 0.3783047029691091, + 0.29849114804949434, + 0.39034810825767013, + 0.3809978343187143, + 0.4639195052474912, + 0.42599886427596906, + 0.36465457322936606, + 0.35929608425183074, + 0.3975389131229863, + 0.44409688828495764, + 0.28647662354924697, + 0.3030676182131297, + 0.4359439266547981, + 0.34066294871032976, + 0.3173922281274717, + 0.33662732004886864, + 0.44763806947120294, + 0.2957920571438482, + 0.3902421945861769, + 0.378702141868775, + 0.17866634952291102, + 0.2974203343667321, + 0.23596534675440858, + 0.3867845161094579, + 0.3351950349743249, + 0.3740432228793286, + 0.167612009619754, + 0.4120644094787427, + 0.3698486173696145, + 0.4020932990438707, + 0.39576492278660746, + 0.26748819267451307, + 0.4484499404244617, + 0.2030009635513551, + 0.39345287952882535, + 0.3050165758448437, + 0.4475407686226269, + 0.40899429366080087, + 0.42885257805536026, + 0.4679627306310408, + 0.34719487090162077, + 0.3786432793808298, + 0.1750275364228902, + 0.38487577513299237, + 0.41367348048530134, + 0.29382571567626115, + 0.18313702074525917, + 0.18834187804780675, + 0.2188035749415489, + 0.35041287411041816, + 0.42490849015914406, + 0.2177925803460138, + 0.16727550614100936, + 0.3304298474640794, + 0.38364175819081336, + 0.33807198525256493, + 0.2361328238142295, + 0.11100121798101754, + 0.35483786953142027, + 0.23847762911909479, + 0.25949007728676854, + 0.34889381601195096, + 0.31285344631564677, + 0.379980236100797, + 0.21755456571215642, + 0.24223432124872965, + 0.2153625965823042, + 0.2096050034039562, + 0.22909779248008258, + 0.28911534597335703, + 0.3864691065817536, + 0.374241780226105, + 0.35918616729406655, + 0.2950051538104177, + 0.3009229077986128, + 0.4380055855536379, + 0.2419542717427726, + 0.281004367362813, + 0.387770059308297, + 0.32484556984923296, + 0.36129822747664764, + 0.3716506954040799, + 0.38150387354653614, + 0.23581700325784488, + 0.39341001754056937, + 0.2498303459160427, + 0.22626227777521601, + 0.3946476618245519, + 0.36801056013515976, + 0.42092937048596357, + 0.43177968666813005, + 0.4579144391807912, + 0.4013909934188057, + 0.44955775286161254, + 0.3715844376840719, + 0.3604581242617318, + 0.34644551738623647, + 0.2542315994172277, + 0.3207606779960412, + 0.3589164103273535, + 0.4119965453760567, + 0.20378761342740842, + 0.26391509995753876, + 0.3557012756436401, + 0.38776070749767755, + 0.3701335084767562, + 0.27914753509161455, + 0.44588514984444183, + 0.42285147656384137, + 0.2765849326421481, + 0.4451581651062534, + 0.30349922365192467, + 0.4332079686457453, + 0.4007469937593188, + 0.2770798196169379, + 0.36454318513954864, + 0.3104920594870745, + 0.3879148847323319, + 0.4259766606590602, + 0.4133253771878114, + 0.25904575213282516, + 0.21220916159581918, + 0.17719126278195022, + 0.3344649961652956, + 0.38596478675686136, + 0.3476184730753689, + 0.32171439778436034, + 0.3872081956034402, + 0.3022422470532035, + 0.34242337915678894, + 0.330440990143425, + 0.39931966447464884, + 0.3927935699520181, + 0.3470649377460116, + 0.21321010722073425, + 0.36462230752508656, + 0.44972174432954115, + 0.21556167835183335, + 0.36532192749874176, + 0.36223325661435674, + 0.4473708198780622, + 0.2784280293125777, + 0.39081056430589434, + 0.30249732055429307, + 0.35142685101217286, + 0.37657251628010113, + 0.1949127304508734, + 0.3843292703718169, + 0.31201265885441487, + 0.35115373647090137, + 0.382267609270122, + 0.3349677687046966, + 0.2192852642146742, + 0.35827383071487817, + 0.36990738495571956, + 0.3828190812337879, + 0.4322531026353025, + 0.4029748624524925, + 0.402758354526942, + 0.28384357536594534, + 0.37025821983235957, + 0.29617108147043586, + 0.308388325033374, + 0.31552879541156054, + 0.322335214721725, + 0.35497898565667235, + 0.3607659393145743, + 0.2829381469820122, + 0.365203060043917, + 0.33010227685526355, + 0.35159352973956604, + 0.337799022935566, + 0.37949842288828634, + 0.3776252496048217, + 0.36569339806016243, + 0.27293925074596953, + 0.32315472299957204, + 0.343078554769425, + 0.39086775101250654, + 0.3462012786576161, + 0.40102236090426513, + 0.45243985838156076, + 0.4252147146865567, + 0.38962651467254483, + 0.2937518794859986, + 0.37135950728761175, + 0.40616870098725494, + 0.4411847401636594, + 0.34803232573666354, + 0.4337522583037407, + 0.3453369047814434, + 0.21751215750111522, + 0.39641654162871415, + 0.3868405064885193, + 0.34020060355139425, + 0.4113562000454203, + 0.30063553256591957, + 0.33440231283009253, + 0.34577685848997397, + 0.34407846097523753, + 0.4554272914418634, + 0.37283751681469135, + 0.32733448492014305, + 0.41948489273804007, + 0.31996095525384854, + 0.36726323830403157, + 0.3814081134481363, + 0.34832894204304576, + 0.35090689107472406, + 0.42097557163780286, + 0.46293116005159995, + 0.3241927433046679, + 0.33000703356302574, + 0.2714534350510163, + 0.31060409879922535, + 0.34406154214992357, + 0.3937638260351429, + 0.36583119803265135, + 0.30373201420675766, + 0.3540946549008876, + 0.3046256121012867, + 0.21073459153674884, + 0.43229527377669297, + 0.3567866600641036, + 0.304782816609809, + 0.353143310252913, + 0.4522891017215409, + 0.3527378283633797, + 0.2062868469906408, + 0.2508612356287503, + 0.3344792006128085, + 0.3685165374852033, + 0.41608529895210494, + 0.39916171526275424, + 0.4148025304592785, + 0.2573492152691245, + 0.39956876691885823, + 0.3618865386562254, + 0.31833914651923473, + 0.34611909895618415, + 0.2626862961047996, + 0.44460825505102614, + 0.37983334895439014, + 0.3181219242328092, + 0.2692539555189274, + 0.23121691632287605, + 0.2921364395604423, + 0.38418113029929063, + 0.32713206394276667, + 0.36977999697350894, + 0.37686877504054944, + 0.28286313793790174, + 0.34032562049138515, + 0.1741233075746553, + 0.151419638021492, + 0.14791585917968544, + 0.14322898905244835, + 0.19228886995684932, + 0.36046047095814626, + 0.3196442581570812, + 0.29326920542484564, + 0.1600201628361177, + 0.3310598230454542, + 0.41838792796159413, + 0.4473893812668846, + 0.45412992220749376, + 0.40110425782828607, + 0.3969278611732812, + 0.3128795048694821, + 0.4259747897766163, + 0.4213332069388025, + 0.3733341653336091, + 0.4192034930823133, + 0.38953297470965387, + 0.3266895310223949, + 0.41060743775326075, + 0.37047141684641743, + 0.49231001860686235, + 0.4261044713436465, + 0.39185995532526124, + 0.3652186971897284, + 0.3759138012904136, + 0.4059163890643882, + 0.2352987413627185, + 0.23567544845893285, + 0.4483449472370212, + 0.3490732726548134, + 0.33144837983416786, + 0.3352804554848704, + 0.4495785463494323, + 0.3198409003281558, + 0.37333337964655683, + 0.36473250731098816, + 0.1705432691664648, + 0.2746573792790693, + 0.2261955792755677, + 0.400557953723511, + 0.3080087612025879, + 0.380315761404772, + 0.1704761336318688, + 0.4075526784906513, + 0.3768575138029348, + 0.4017313524826458, + 0.40582145913615525, + 0.29133779813854477, + 0.4893818218963004, + 0.19890006953405195, + 0.3810938214823534, + 0.29409067986086895, + 0.43509830739933153, + 0.4207670293118084, + 0.45698250176331884, + 0.45662621964611183, + 0.35962683903676945, + 0.37757423162110243, + 0.1658872579907592, + 0.3777401482598134, + 0.40494652665458125, + 0.3203540770278961, + 0.17608631079196735, + 0.17646496491470842, + 0.20999294529890192, + 0.3543403466116621, + 0.4175153334064538, + 0.20390272692543618, + 0.15111832194420913 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.6439298864530274, + "valid_tracks": 613, + "per_track_pearson": [ + 0.6311652303732685, + 0.6996361421110132, + 0.6230323625675874, + 0.6693159337565515, + 0.684653252852836, + 0.48278551628734356, + 0.6264039199347694, + 0.6914547468120326, + 0.6768613752064179, + 0.703889648932899, + 0.6477142456979146, + 0.6246134080424677, + 0.5355051228453901, + 0.6929059330059539, + 0.6609365252210786, + 0.6750047074728998, + 0.7131306394557738, + 0.6426512327636713, + 0.6472491189023688, + 0.6185408774129926, + 0.6127040340254797, + 0.686613262637094, + 0.6338177166829132, + 0.6168988402843284, + 0.5102950829574582, + 0.6993345181133618, + 0.5960859765425134, + 0.6505625691144289, + 0.6292618271338358, + 0.6457880820660542, + 0.6516303523841522, + 0.6665673838213392, + 0.6648160499646066, + 0.5748457371191175, + 0.6662701541744898, + 0.54808991752518, + 0.6577502940013419, + 0.6602106887557999, + 0.6084450541614155, + 0.6495337609252196, + 0.7052875069197182, + 0.6113031224264319, + 0.6980476878144687, + 0.703711865069493, + 0.6365255872364336, + 0.552850338156772, + 0.6105514678968342, + 0.6784960826094633, + 0.6485618000737103, + 0.6819690650610947, + 0.6669943156243056, + 0.6766939324661563, + 0.5827410011323038, + 0.6357408005366126, + 0.5097599458438228, + 0.6464226995267726, + 0.6255401186886784, + 0.700766510777159, + 0.7163082533881228, + 0.745959825205146, + 0.7411842102858058, + 0.6722134163872882, + 0.7026097764198065, + 0.6098206707837559, + 0.6839882979029225, + 0.6752715131393642, + 0.7016637032349378, + 0.7034724683210403, + 0.6807723239528364, + 0.72869977841581, + 0.6944123290559995, + 0.7060271094803138, + 0.6649854008227133, + 0.6774592617410584, + 0.6750514860263648, + 0.6705007240056514, + 0.6947380297400642, + 0.7044367046737624, + 0.6732720127368016, + 0.7288680509317066, + 0.6790705206250285, + 0.6058105728808307, + 0.7397458886093141, + 0.716083440794688, + 0.617999162238066, + 0.6609964479539491, + 0.7273096859169647, + 0.6434883293547388, + 0.6961692524983493, + 0.6617368165373875, + 0.5650064124431478, + 0.709484767210907, + 0.7259948185531816, + 0.7059258581895853, + 0.7086406698055334, + 0.7173628658993517, + 0.7368871171756882, + 0.7153261586264624, + 0.6616585020902132, + 0.5838608076317832, + 0.5674071622158637, + 0.586868382915664, + 0.6282950116317908, + 0.682399683531573, + 0.604877706832279, + 0.6471370290769675, + 0.699135736685107, + 0.6881266113977981, + 0.6848127372114288, + 0.6394101228212667, + 0.6710267406473899, + 0.6716516643026312, + 0.6802291674817222, + 0.6495346743079389, + 0.6687732275356127, + 0.6835370918731659, + 0.68815139836671, + 0.6801662187457114, + 0.6329965148496642, + 0.6760693249315639, + 0.6825167666752161, + 0.6153900105531879, + 0.6149359057068335, + 0.6455082729463056, + 0.7100111980019451, + 0.5541098897260618, + 0.5690120080513004, + 0.5636205618942761, + 0.5749420250755888, + 0.6083488910117324, + 0.7104556726178415, + 0.6374648726993707, + 0.6560002059114355, + 0.6128786061008542, + 0.6509897602387975, + 0.690082349486682, + 0.6603029639182927, + 0.6875041452283229, + 0.6402834812271327, + 0.681581262478773, + 0.6134267025486662, + 0.6238685798648251, + 0.7000456326234279, + 0.6624608557040896, + 0.5823478073300735, + 0.6000642972176501, + 0.6030575043599065, + 0.6233617668186695, + 0.569077295312179, + 0.6455792592794681, + 0.5979570544270114, + 0.629709304129978, + 0.6536415259011134, + 0.5695739144098485, + 0.6224452719852309, + 0.6169951399711576, + 0.6359965395064945, + 0.63530386669552, + 0.6276801689946653, + 0.6920999835433574, + 0.6896151844450543, + 0.6835084991436869, + 0.6758007617223297, + 0.5942184988409526, + 0.5895166612636781, + 0.5931240930254703, + 0.5977591464107297, + 0.6129980264950619, + 0.638346221768261, + 0.6394516631146249, + 0.7357526461714453, + 0.6153522212067277, + 0.7325183498750305, + 0.6407222379217786, + 0.5885309899563176, + 0.6162499535123582, + 0.5968969644027762, + 0.6768539296426636, + 0.5585632412188046, + 0.5278843460412769, + 0.5390059735780844, + 0.6861657251092601, + 0.59050605973921, + 0.6880979052599805, + 0.5559110282873582, + 0.6388833481669433, + 0.654607678432059, + 0.47568843487487983, + 0.5894756311196546, + 0.6178468542813212, + 0.6984846688714946, + 0.656773572231199, + 0.6267453858428695, + 0.580197217705264, + 0.6694140217237052, + 0.6177796398664418, + 0.639304513270512, + 0.42306353787399115, + 0.6496947675378477, + 0.6318828013006876, + 0.6398009959997037, + 0.7168115286111473, + 0.6931466993624797, + 0.6764844631762842, + 0.6794966738588082, + 0.6839466329761069, + 0.6768659028724014, + 0.4065167924986336, + 0.6632289769619713, + 0.6469739067439503, + 0.6488620043717025, + 0.6276949162376848, + 0.6936392351387327, + 0.6966990618808312, + 0.584665238095939, + 0.6699539197117723, + 0.6680312499222693, + 0.6100467772229069, + 0.7309445558834744, + 0.7148043649008612, + 0.6923244183488004, + 0.7598924918393174, + 0.6624672948026182, + 0.6682599779108911, + 0.6609095355774228, + 0.6405401657674724, + 0.6633797748136365, + 0.6437700491964155, + 0.5660302499777273, + 0.7291395274868394, + 0.6220397914021409, + 0.6763546787329574, + 0.6917028182512635, + 0.5575297467351756, + 0.7248295656563697, + 0.5201640894423085, + 0.5077766002170037, + 0.6854715102620416, + 0.7042615666793752, + 0.675265846177047, + 0.6820002776596781, + 0.7197435048551943, + 0.6895773761904632, + 0.7468173063317536, + 0.36008263540763447, + 0.6272396060718989, + 0.6592642762996866, + 0.682777682373249, + 0.6176039733186491, + 0.6036007918298524, + 0.5891077135672104, + 0.6165102904524812, + 0.6187942005399862, + 0.625510980766917, + 0.6730368287286993, + 0.6474048208262129, + 0.5871206080460094, + 0.6923363069093756, + 0.6396609188151275, + 0.6836828388726678, + 0.6080198308759206, + 0.6100568707779853, + 0.7242054686926073, + 0.7280190383808716, + 0.6609304540847665, + 0.7224690938819798, + 0.7358396961186751, + 0.57607374289673, + 0.6705994112114636, + 0.6493168037352, + 0.5167157063905735, + 0.6350478309616614, + 0.7223052033721454, + 0.7186448298322756, + 0.7210137670526688, + 0.715222415451588, + 0.46774209907570957, + 0.5936862034284359, + 0.7024199569960915, + 0.6664898208595893, + 0.7180857237289932, + 0.6403714398988293, + 0.5896589643715768, + 0.5561377492264853, + 0.6723743701303297, + 0.6413840397891314, + 0.6532831672068791, + 0.708068957021758, + 0.6913840054668193, + 0.7227285319919989, + 0.6330180952295411, + 0.6391771086124965, + 0.7010665484014897, + 0.6411093846801325, + 0.580126739597622, + 0.5026631941748197, + 0.6916222794720411, + 0.7278262494437514, + 0.7381356206488195, + 0.6567606371159119, + 0.6708678903500128, + 0.6237174429588994, + 0.6960173086225377, + 0.6013694722041969, + 0.6383454791461071, + 0.7032079283232944, + 0.6531386589723002, + 0.7143781784328355, + 0.7215047229297085, + 0.6759502140657322, + 0.700232109440537, + 0.7018646374418042, + 0.6635212523120488, + 0.6813143318552612, + 0.6497243954078149, + 0.5982270881457953, + 0.6027250909872722, + 0.6411926693858208, + 0.7044668309055513, + 0.6465506857581066, + 0.6853425161307591, + 0.7275676251898412, + 0.7094441020044076, + 0.5463140056790117, + 0.6184248418962416, + 0.4744201664112501, + 0.6476150159225526, + 0.6373608249048452, + 0.6764049800398816, + 0.7182676595813035, + 0.6951916007072991, + 0.7180920607698394, + 0.7135984475905987, + 0.7306366121122522, + 0.598991937513838, + 0.7141758390823904, + 0.7079183713894165, + 0.744328527615456, + 0.750706999933269, + 0.721505525136978, + 0.7512974589157679, + 0.7077868305683287, + 0.7174009122882871, + 0.6553607008400736, + 0.7347820228585845, + 0.7073689781990878, + 0.691558501765781, + 0.7176294233737512, + 0.7508831570906187, + 0.6067991066420001, + 0.6957426184017719, + 0.6598712235284864, + 0.6138852065108634, + 0.747675320215999, + 0.7287284666067975, + 0.6645688496000961, + 0.664584190390158, + 0.6767786735754436, + 0.6427178654980575, + 0.7153750492853053, + 0.620863876682915, + 0.6062610013025866, + 0.6985959684915511, + 0.6787387871840128, + 0.679724377203451, + 0.6487867543171908, + 0.6766726549637974, + 0.6889876810344809, + 0.6757650931106525, + 0.6234907813431251, + 0.5875531585651034, + 0.5226765946258016, + 0.536132222725463, + 0.5636969549950875, + 0.7321101426132157, + 0.5930945325839323, + 0.6089930284716802, + 0.7474670657664865, + 0.6575806265813608, + 0.686377093306164, + 0.6164109881881573, + 0.6699717443369502, + 0.6354735965739425, + 0.6378009002416889, + 0.6682075993833968, + 0.7595556588448624, + 0.6961974827430056, + 0.7020894701247858, + 0.6484891907307889, + 0.6070338073753708, + 0.6515047413715637, + 0.7270871494929931, + 0.5904202201292434, + 0.6084030423365385, + 0.6186708073581546, + 0.7279512365280603, + 0.5341119965957218, + 0.54751457790799, + 0.5360617009423793, + 0.5376805102399907, + 0.626678217366642, + 0.7154004976906267, + 0.5979788219225772, + 0.6403257162637847, + 0.7175866962353918, + 0.6788398266832277, + 0.7081443857426878, + 0.6516283145600354, + 0.7081608413377806, + 0.6385677226798622, + 0.6830224415670008, + 0.576386276189456, + 0.6423245264206084, + 0.7038133655203115, + 0.6433284439537568, + 0.5703940191986665, + 0.5793807940616962, + 0.6422265726587361, + 0.6682694838088455, + 0.677391155241979, + 0.5970883133878105, + 0.5797783972521136, + 0.6464012367012054, + 0.7065263912388532, + 0.521682146559956, + 0.5945604683572735, + 0.6272897737217568, + 0.707286669272236, + 0.611963559226466, + 0.6850882902984438, + 0.6518922461745192, + 0.6895867349750088, + 0.6817500829986542, + 0.686658965605654, + 0.4904575109607943, + 0.4916668570262101, + 0.49453466338983154, + 0.49348128475729175, + 0.6051140350914117, + 0.6430897628048567, + 0.6382433370018128, + 0.7815475805692015, + 0.6364867487315039, + 0.7554209968214576, + 0.636763568358016, + 0.6116339099227361, + 0.6008134955369661, + 0.5893181004797813, + 0.7019192965507144, + 0.5178138242897448, + 0.5401842988821398, + 0.5078036368807437, + 0.6942999504937676, + 0.5485266146381085, + 0.6143866246648813, + 0.5415646024840368, + 0.6357925598599631, + 0.6799612887745744, + 0.4853350056087486, + 0.5906300540442545, + 0.5661791725931586, + 0.6758939227120054, + 0.6789237478035355, + 0.5783321590361208, + 0.5541936740146078, + 0.6758387218992151, + 0.6446227044497316, + 0.6506793789881239, + 0.40594347150752425, + 0.6898131120234746, + 0.6797944577079308, + 0.6075755442086119, + 0.7581340962356524, + 0.6735277767453852, + 0.6787992679029775, + 0.6935677942162112, + 0.6777303314435908, + 0.661508600050517, + 0.33421467215110906, + 0.6923272020001513, + 0.615835088443443, + 0.72832668707877, + 0.702879464489104, + 0.7787707691692695, + 0.7748746173744053, + 0.6230459431824816, + 0.7091213529675447, + 0.7534670282362268, + 0.6408850541598453, + 0.7977182629430397, + 0.6836877942387685, + 0.6335595317370925, + 0.7295116235787419, + 0.6099661496691091, + 0.5924338106530737, + 0.591023779601265, + 0.5630237653760812, + 0.5464548638163912, + 0.5778401418577628, + 0.47773856643679746, + 0.7469885693465597, + 0.5282945949208807, + 0.6521510576596125, + 0.7418759708237209, + 0.549087366892827, + 0.7350968388426264, + 0.49201653717878674, + 0.506941661268339, + 0.6924436106838676, + 0.6729876912064358, + 0.6223143421987263, + 0.6326269645018138, + 0.7321741041707579, + 0.7352915381959462, + 0.8046372390110326, + 0.3625478628611472, + 0.6339771003598587, + 0.6506775077190697, + 0.6830074558690192, + 0.6667032943474864, + 0.6074943009682713, + 0.6084637954786308, + 0.6141532400345093, + 0.6260932084135691, + 0.6208942763866703, + 0.6776744927768312, + 0.6092359730038119, + 0.5152465323414446, + 0.6988563901185233, + 0.6579200329930595, + 0.7506243404278552, + 0.5484868560877993, + 0.495846317244993, + 0.6679006411631577, + 0.679032176857248, + 0.6585936260424434, + 0.7239222044566251, + 0.758450477319606, + 0.5070167420813925, + 0.6822276366441015, + 0.6628553875201578, + 0.49603226327357675, + 0.681249713910141, + 0.4020023187884614, + 0.5744398442075725, + 0.6331949464353368, + 0.5776140149375509, + 0.6446543890721471, + 0.6119345644057751, + 0.6644690748148135, + 0.6794734855685182, + 0.5772369425529547, + 0.5539750797763395, + 0.555704418974647, + 0.5512870699738949, + 0.5433284031366713, + 0.5434638203871118, + 0.5534083871387947, + 0.5486282377090235, + 0.5225633653487102, + 0.4988192175828015, + 0.6565774770952948, + 0.6282122883407897, + 0.6285061296112683, + 0.6284637168052007, + 0.6417865821682873, + 0.6240281378025956, + null, + 0.6412251716650419, + 0.685769758764334, + null, + 0.5338454479394473, + 0.6841006806804789, + 0.5364715554598604, + 0.6719166353490038, + 0.6908884113247101, + 0.6823313661652726, + 0.6752935052119857, + 0.6769498234012113, + 0.6748799625923213, + 0.657783121144161, + 0.6267783553996111, + 0.6225442177404369, + 0.5693862248330648, + 0.6485363575275899, + 0.5311718284080329, + 0.6558928856559502, + 0.6368891349310642, + null, + 0.6458868114062062, + null, + null, + 0.6381935552701257, + null, + 0.6556950517952422, + 0.6284805706658781, + null, + 0.6724881863548532, + null, + null, + null, + null, + null, + null, + null, + 0.6561122053545686, + null, + null, + null, + null, + 0.5681803545545172, + null, + 0.6487470987243968, + null, + 0.633908011614238, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6742480271777149, + 0.5005187807401799, + 0.6274702682350173, + 0.6595605473543372, + 0.5772424558538098, + 0.6835761412096995, + null, + 0.6707555336119831, + null, + 0.6572262952693935, + null, + null, + 0.6743638304600614, + 0.6879877622068917, + 0.6120614942438052, + 0.5974388785021405, + null, + null, + 0.6241024809878639, + null, + null, + 0.655695760229505, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6639458931872652, + 0.6503128050253917, + 0.6663188421231435, + 0.634349891961928, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.4738454010394988, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.25538926840159076, + 0.3921201626276131, + 0.3279809460818927, + 0.49719904555500183, + 0.44260045534323933, + 0.32877025757496575, + 0.43281694121883274, + 0.506465056944795, + 0.42166763925127004, + 0.43950296951936296, + 0.41548154077375554, + 0.4343509605521734, + 0.30803393017577313, + 0.5113928003500388, + 0.42733807277911984, + 0.43432654111570074, + 0.505583351440439, + 0.39980632278076905, + 0.5342372143515803, + 0.4295565142821136, + 0.42609280895088975, + 0.36894900132464664, + 0.3414346013871383, + 0.5544796958577486, + 0.398145061636397, + 0.4514780242116092, + 0.3721432491126965, + 0.3991179441212512, + 0.4364019725272235, + 0.45832668231123724, + 0.4628991206346552, + 0.44214623272590137, + 0.5114750598523516, + 0.30917532455761776, + 0.6834893621302248, + 0.2994623311114575, + 0.26832056299597173, + 0.19844706365123746, + 0.3550820647986884, + 0.24362178319486352, + 0.2999863735332245, + 0.48366693312370757, + 0.3299159469168723, + 0.4318697863939187, + 0.3179899330584598, + 0.41861462978648023, + 0.3222540070177899, + 0.242550948046035, + 0.4904467230375315, + 0.4630854369494051, + 0.4089664867999297, + 0.3815639381310185, + 0.40257959183352776, + 0.4469108253143683, + 0.4411682922814872, + 0.45517193075330387, + 0.3779980892126048, + 0.5392497141308182, + 0.5548494227664953, + 0.5555379191505609, + 0.34491201386042186, + 0.224751700179868, + 0.41841395484642085, + 0.44669704597489873, + 0.292897220791319, + 0.3764089404003292, + 0.3703319850101118, + 0.49810477597588954, + 0.5189854726770403, + 0.4664934889762381, + 0.38856230137861825, + 0.7287077814417748, + 0.37937155793503763, + 0.41117686698717837, + 0.41466431143899257, + 0.4535721777694698, + 0.3389520645180934, + 0.6775886093466026, + 0.43920548577317575, + 0.46508192119385544, + 0.4203027452283173, + 0.4394519182913112, + 0.5049532538196518, + 0.470838842294271, + 0.3974134858436613, + 0.34534358623585537, + 0.4722821064606324, + 0.49275966810696714, + 0.8115415032775714, + 0.5034892527865305, + 0.41826804902901366, + 0.36783166528102, + 0.4885338268745111, + 0.4139739842354182, + 0.45400660773239276, + 0.5741825959731478, + 0.3615581888903258, + 0.4169300544475417, + 0.42469215380849895, + 0.35904078584303706, + 0.26435192474080155, + 0.3947334704456107, + 0.3533717557045557, + 0.3366137316380426, + 0.38473640628931094, + 0.27680411288786955, + 0.3468677280897522, + 0.37424221406647684, + 0.39323302833615437, + 0.4537398521825768, + 0.27872236587972316, + 0.39123349582052536, + 0.364659239764113, + 0.41901422205493577, + 0.466968787261504, + 0.45898285579392645, + 0.43166979740937783, + 0.44325321061873413, + 0.4172932200340206, + 0.4745063733961004, + 0.4249923743045773, + 0.4571092808321518, + 0.4437455805865342, + 0.3638440632118701, + 0.42247455558083896, + 0.3808277704731779, + 0.41171671316746705, + 0.38432553681311954, + 0.4160192935940944, + 0.5498903342623018, + 0.47078825630211246, + 0.3915367027622456, + 0.4534986451326951, + 0.6544612183173312, + 0.390370724284756, + 0.45238828337135006, + 0.4560563113061968, + 0.34280605457507685, + 0.4063833000275305, + 0.38458816919323796, + 0.4169836645450335, + 0.4598793731982041, + 0.584494995684198, + 0.39845860489616647, + 0.4639715059959903, + 0.5291694334980847, + 0.4795086369322358, + 0.34350757098838414, + 0.4609117567734264, + 0.4863747595752448, + 0.4689921404441009, + 0.4004108255451794, + 0.5040790920001553, + 0.5162333294529252, + 0.4552136779564974, + 0.4539344616556386, + 0.3669717898382378, + 0.42216062406063465, + 0.4210464067902036, + 0.47737658769513547, + 0.4834074429405067, + 0.41019765954008336, + 0.4404735010675083, + 0.45267207914467816, + 0.4768720253203202, + 0.41941722509227314, + 0.4134168915036088, + 0.5024551716670217, + 0.3465956988302997, + 0.46980174176449335, + 0.44742071614415746, + 0.39449679435438395, + 0.5119263594913331, + 0.4066686858088748, + 0.3688946741600991, + 0.46268336256250064, + 0.3323387539986321, + 0.4897147157840247, + 0.503312660719372, + 0.3765086604503784, + 0.4809393923682401, + 0.46606582537751373, + 0.38232354442765704, + 0.4606892468445108, + 0.39112924406338173, + 0.4847906497497024, + 0.4795833454871428, + 0.3494760133166724, + 0.3810900274879333, + 0.4533878952208789, + 0.38777557140758107, + 0.41460749949700315, + 0.47078149925314366, + 0.5156133892231265, + 0.40201606695066266, + 0.39306191622700554, + 0.4020268739164775, + 0.4005728967396741, + 0.312899093661483, + 0.5114914323798215, + 0.41815746649206914, + 0.3913783913507741, + 0.4321401042976921, + 0.38095935930961805, + 0.4878145017165497, + 0.4264584571498459, + 0.5014326414361108, + 0.4175991250729862, + 0.37245662458804396, + 0.5404962819740863, + 0.41251178064262983, + 0.4821113931670492, + 0.4500236005565377, + 0.46418509634818333, + 0.8678052244095881, + 0.5058686947140061, + 0.4239730917005596, + 0.469267673178445, + 0.4805242424504233, + 0.5597567706119579, + 0.6245015720204307, + 0.6394235115412646, + 0.4580086952314006, + 0.41358159799836397, + 0.506103884246044, + 0.7419390961949175, + 0.3832799098008994, + 0.410737395015233, + 0.5601730207878278, + 0.535803690029266, + 0.6000874865749632, + 0.47696938116505244, + 0.3595669471356012, + 0.429172878182593, + 0.6124467299840075, + 0.47464046700220014, + 0.3898724181235045, + 0.6240846685025383, + 0.5661179422133211, + 0.470393553721893, + 0.5102934509935433, + 0.7203858783534659, + 0.5045387681345401, + 0.5335809274332439, + 0.4635503492132533, + 0.7698582579129863, + 0.3567092737202845, + 0.6502136898740012, + 0.42849886059423475, + 0.5609219313500785, + 0.45888186044199014, + 0.39989106509024136, + 0.7095305761331835, + 0.5303164054039017, + 0.36656694738427653, + 0.3952124579468412, + 0.6570912871483172, + 0.4117455114958413, + 0.46000804712922166, + 0.5151822087492214, + 0.41606580708955526, + 0.5106847485406991, + 0.5126772368072894, + 0.4488010646351969, + 0.4082658508224382, + 0.48735710629078055, + 0.3871953937741841, + 0.3687587061206911, + 0.6394863141066002, + 0.4036401661457953, + 0.4694402483455347, + 0.5861829680567325, + 0.4192932275215914, + 0.6681076960106518, + 0.3876804213313681, + 0.38951034036985555, + 0.5706146292255886, + 0.5121456716092768, + 0.5918016215082625, + 0.7367868181038342, + 0.6807638060577742, + 0.43610651709191695, + 0.4856338775579918, + 0.5803393249951528, + 0.3831954760505808, + 0.3712879326371089, + 0.5778123207338196, + 0.5502868533147314, + 0.7470352982511488, + 0.5064988572685645, + 0.3917709263421258, + 0.48305380503803536, + 0.3941868740105093, + 0.4891427874952529, + 0.714940277037793, + 0.6389567276376927, + 0.5605708819049982, + 0.5497672948165436, + 0.5146988389540184, + 0.5285830332463017, + 0.7484196147321652, + 0.3301437933695372, + 0.46514352082949306, + 0.35214813904159836, + 0.6223877491617026, + 0.4090379259796956, + 0.744385243642573, + 0.45136554835152903, + 0.45341788327000443, + 0.705707432133441, + 0.532406192344929, + 0.36271372907080485, + 0.5798029126973806, + 0.4294992999563837, + 0.524863793909835, + 0.5097245844266188, + 0.3668564382243502, + 0.5171146891204869, + 0.42488128295162964, + 0.47359868746066247, + 0.4351583289704761, + 0.383097663536643, + 0.4402693150467559, + 0.545129484247152, + 0.6034947726973316, + 0.6741969258600174, + 0.4504044777160895, + 0.33582025094419193, + 0.4810079354411958, + 0.48839531051860746, + 0.4470216909528132, + 0.49856271104190303, + 0.5085609922284597, + 0.40583254088014376, + 0.35683357028961465, + 0.5322454757863901, + 0.7706391810113541, + 0.5424268851938866, + 0.6177701646952733, + 0.6021084543185213, + 0.776576172899508, + 0.5834234263291974, + 0.38454281888526154, + 0.602113258124661, + 0.4135019968808623, + 0.5560705705035697, + 0.5019359406356844, + 0.4140885219270271, + 0.5160645909487434, + 0.7910911949136128, + 0.2602287502445447, + 0.4871548728521156, + 0.4913266862836096, + 0.635512890531214, + 0.3990034802168482, + 0.34424576816393165, + 0.45461438617263206, + 0.7296877083077643, + 0.4155677191563353, + 0.46478685826194754, + 0.5212188872021545, + 0.4750817257617859, + 0.4496439680776316, + 0.519052212377307, + 0.6266967837853672, + 0.479041222135833, + 0.4337629617991163, + 0.5684233309300424, + 0.473156850182586, + 0.5851450493894786, + 0.5723704019599891, + 0.5477979455335148, + 0.5692078271759192, + 0.364874936203573, + 0.3885675248743056, + 0.5840444770126391, + 0.34680971997240434, + 0.7051491285567403, + 0.3476390097314651, + 0.5775570451660299, + 0.6074591900661088, + 0.6954167906901996, + 0.3324217684628142, + 0.39929394842749383, + 0.37133712858422163, + 0.4213449210209628, + 0.350934317935135, + 0.3344602048954689, + 0.6335938439223937, + 0.41464114429218407, + 0.6893750191674455, + 0.5962527950074312, + 0.4787016842298767, + 0.605511191744122, + 0.39965848131984, + 0.5676997129382384, + 0.5514956156551325, + 0.49092180162099514, + 0.40723164330159417, + 0.4825382481348562, + 0.5250818279994875, + 0.5477080838393362, + 0.4140282086699012, + 0.3816087020889987, + 0.643323768472536, + 0.4949035787322079, + 0.6664304153237268, + 0.6603327272327423, + 0.5385418968027332, + 0.3590867567597524, + 0.3942744095650705, + 0.3913117748617899, + 0.5267124848959585, + 0.7522824979264693, + 0.4882396180304985, + 0.40249255718073507, + 0.7502823183006909, + 0.46759475130291844, + 0.6342423986247696, + 0.5458564465050849, + 0.455941833647483, + 0.5137167439468586, + 0.47733267495601306, + 0.4915774483886509, + 0.587442887546598, + 0.5589330958394354, + 0.5476924481517115, + 0.46923910537485136, + 0.4890188455668142, + 0.5430838182067945, + 0.510427773510251, + 0.7736820428157761, + 0.5062406195314776, + 0.5064885755965116, + 0.4582835096364069, + 0.33096740415022935, + 0.6707902881028849, + 0.4333231986071464, + 0.37272283165638215, + 0.5323985396754558, + 0.48068921366126083, + 0.37250936049799394, + 0.49694577654117905, + 0.45841468556424114, + 0.7844591532789967, + 0.33202795190644374, + 0.42096650688632986, + 0.3856408034447364, + 0.4473430162566241, + 0.4197290569878234, + 0.33970361226875134, + 0.44329071270807774, + 0.298982970975841, + 0.44199797169949173, + 0.74987033407927, + 0.3524269968446602, + 0.42370859235842945, + 0.5172945125938418, + 0.43608751205748814, + 0.4227251392015458, + 0.49488339469615783, + 0.3708681174845073, + 0.5362903323964683, + 0.5481598594860627, + 0.28153334660786905, + 0.6477599265786309, + 0.6996424638010424, + 0.5223143400235216, + 0.5018134138444408, + 0.5541897833164753, + 0.44918837396456035, + 0.47372596105461773, + 0.5830718384812243, + 0.7863507031401352, + 0.42432368925064856, + 0.6184208906948254, + 0.3720919580473917, + 0.47439622929153347, + 0.502797615251089, + 0.5063381969332494, + 0.4175083506406224, + 0.3670990722666949, + 0.36445182682245114, + 0.41261545525916704, + 0.5923255700365013, + 0.482235259783354, + 0.35843382313957695, + 0.4224788192601194, + 0.47722121889621405, + 0.32890538791107776, + 0.5629797937990799, + 0.5179435977568351, + 0.4451146599238932, + 0.4677861435766331, + 0.3509965261340087, + 0.44409286367518397, + 0.5241249509447494, + 0.5982166739358488, + 0.4421438041970041, + 0.7865408830588481, + 0.3899762757491495, + 0.5662581468961418, + 0.745990864611613, + 0.642895763443838, + 0.3346237771184627, + 0.5223520973343299, + 0.4327297624010676, + 0.37764080097976976, + 0.3975983186963222, + 0.4094692570382853, + 0.4848909424463719, + 0.5091328216825386, + 0.3935843886397508, + 0.6262987893427645, + 0.7737256682588181, + 0.4514211885848965, + 0.46846723445657357, + 0.4545857103647784, + 0.5206275803616371, + 0.5355251858972819, + 0.5016059597276221, + 0.5166320593643104, + 0.3840880834469107, + 0.6723613258693903, + 0.4998753334152432, + 0.5662840488027564, + 0.7219993042219907, + 0.347935490836353, + 0.6021359468499707, + 0.4027313333659011, + 0.4772545780633968, + 0.3603041522810812, + 0.5316108189360499, + 0.5132765120544633, + 0.35768069636885363, + 0.45794930266734696, + 0.42314373333703437, + 0.5467988506431015, + 0.6054649503981757, + 0.48071386744997946, + 0.4814890636642889, + 0.4307911848256574, + 0.7482488424883504, + 0.5213588839711278, + 0.5517982978236631, + 0.4814857229421367, + 0.4777367009397341, + 0.6249040803175304, + 0.3743255899213512, + 0.5803933949366622, + 0.3984735905786713, + 0.5619498828281253, + 0.4431422458762131, + 0.37342004229732895, + 0.3885063920228889, + 0.4898244738063877, + 0.3919325457283379, + 0.5205420956881492, + 0.7418151901463982, + 0.7326029944631791, + 0.42580025225048496, + 0.3843701107249271, + 0.7094910866716037, + 0.462570626002134, + 0.40869141041194906, + 0.5401037360535443, + 0.3756365776339895, + 0.6084234543722947, + 0.4446499449813761, + 0.5789817752160795, + 0.5984616386069309, + 0.47256907104832274, + 0.5264191535276836, + 0.6643021558489844, + 0.5360650246384799, + 0.38709089860832807, + 0.4446819197161582, + 0.33239579714586315, + 0.5525348504156636, + 0.3676136868725336, + 0.5336377441147698, + 0.5154533276017087, + 0.41473558455830917, + 0.7284426348006748, + 0.3522848774869921, + 0.7339659994971298, + 0.4670283624351947, + 0.5530583957241498, + 0.7503836607949784, + 0.496656365827433, + 0.5642502146861567, + 0.43622871302138533, + 0.7115513826129989, + 0.43751660251082675, + 0.3592564878072522, + 0.5782500027731234, + 0.536655382394432, + 0.6505699767551177, + 0.5608793818700429, + 0.4222550341397331, + 0.4990774492052882, + 0.3267088803258071, + 0.48097315167997395, + 0.5769092166841001, + 0.5285156210516795, + 0.3793767787193233, + 0.5573880954623268, + 0.5969768678018001, + 0.581312698116121, + 0.45573774016289437, + 0.4231427103424669, + 0.5208735163862784, + 0.4776047463642484, + 0.5682113822953259, + 0.5081748906243974, + 0.3862086427281742, + 0.42846504838699856, + 0.5254129570912668, + 0.41911007207902334, + 0.7440672753203522, + 0.5351953395485345, + 0.4699443905811269, + 0.5127718921675516, + 0.4210059842960961, + 0.4619324703383334, + 0.551342456839092, + 0.5633184591945284, + 0.7173245956011696, + 0.5546106070191488, + 0.5854741110486305, + 0.5435345338683852, + 0.5122287801263057, + 0.5879423641424278, + 0.4248905547265978, + 0.6489225559701819, + 0.6060563948728255, + 0.7307720428900432, + 0.5359410763192188, + 0.5576117698715324, + 0.5512449557257337, + 0.4642601963503375, + 0.5472262989862529, + 0.37266244317346264, + 0.6910418917221063, + 0.5021535694811278, + 0.7431079649850277, + 0.7681673025377177, + 0.5082907665397466, + 0.47010241632860356, + 0.5154911611427732, + 0.3848556071143853, + 0.5193603211433787, + 0.7650111992488497, + 0.5012991693613668, + 0.5327787704591161, + 0.5964684488046305, + 0.5380703021996947, + 0.3750771517039048, + 0.6236483829790476, + 0.4503552394625301, + 0.41314851872053393, + 0.740551074435461, + 0.4350761349352856, + 0.457520890631992, + 0.4218434297155709, + 0.49811792476764877, + 0.5730846883910647, + 0.6424150115837847, + 0.6877353923425524, + 0.6325225539741595, + 0.5990598507080022, + 0.4784864042250334, + 0.617945181814672, + 0.5754788532379997, + 0.785841795804321, + 0.7508292879976409, + 0.39161202500713793, + 0.40575491196392016, + 0.38417474701932586, + 0.5359140454348621, + 0.4339594928164325, + 0.7943042417696284, + 0.39921406807184884, + 0.43858534401178395, + 0.5886812641521125, + 0.5018130611646977, + 0.581225786021317, + 0.6363060145713698, + 0.48798951995907575, + 0.6265980276575455, + 0.401314447370994, + 0.696725863317981, + 0.7871682313943853, + 0.7708320802510743, + 0.5978623811650238, + 0.4534282532695071, + 0.5740695123685018, + 0.7505767250963352, + 0.7007193006269115, + 0.6372930869679286, + 0.5086499034321968, + 0.48420114842080936, + 0.5081164192451627, + 0.743367841377248, + 0.44418900813870915, + 0.4568077433619656, + 0.3460654173099244, + 0.6130642531102909, + 0.3437088327170478, + 0.44043288857107665, + 0.6818468002298835, + 0.5526926918326653, + 0.6939572146194444, + 0.5628006589998561, + 0.5334851942462436, + 0.6772791607104569, + 0.6446667299452653, + 0.5782629538844257, + 0.4026834086499928, + 0.6575053696600305, + 0.5376980365779276, + 0.42542131300555597, + 0.3779012421339678, + 0.49950303032878574, + 0.6484424379501375, + 0.5352028908033977, + 0.6260591383341295, + 0.5894218972685734, + 0.4443399662045021, + 0.43093014770634014, + 0.6985268070849687, + 0.52082493596942, + 0.6258426998706469, + 0.7316042429781074, + 0.4337664913200796, + 0.4074222214964677, + 0.404442245376489, + 0.40233114519846913, + 0.7218992232401656, + 0.7431720361601102, + 0.5760902767276574, + 0.4398874934462356, + 0.49147219283426546, + 0.41223827109410494, + 0.7889534445898608, + 0.3745787618751776, + 0.5032246099041813, + 0.3630670441299954, + 0.5283389481484552, + 0.40012254711139156, + 0.5800864901917175, + 0.5079723557341146, + 0.37782966830943393, + 0.7082014797437278, + 0.5271273889768728, + 0.4487536008020466, + 0.3655595798593098, + 0.38001672714731666, + 0.4314553774486092, + 0.44384052563261567, + 0.39711975834663304, + 0.39086303069393, + 0.42580171404168754, + 0.41949373567388265, + 0.43912381457927757, + 0.3622923219534416, + 0.4689085354958491, + 0.2279838183890776, + 0.3154596340655683, + 0.39124287600871305, + 0.40211853426497424, + 0.4737293511843228, + 0.4719095560155636, + 0.402928351326551, + 0.4258410512264139, + 0.3902385774098196, + 0.6931330544702605, + 0.3871307151173445, + 0.3580684777399544, + 0.7937641967816155, + 0.38066006000250385, + 0.466230483136527, + 0.3874922741707393, + 0.4598021811395152, + 0.32746325866221043, + 0.44776238614485053, + 0.3707429925137666, + 0.5128590306943269, + 0.29646285284217266, + 0.4263497368567924, + 0.41207098893023714, + 0.5282102521542167, + 0.4395591760753334, + 0.48268035249929875, + 0.32136331106730515, + 0.4991428908980327, + 0.4396573779884011, + 0.397029385544716, + 0.4832258245315489, + 0.42634682890648073, + 0.48861499565059757, + 0.4363179913804237, + 0.41330391020568225, + 0.8131597265735397, + 0.3618983205652883, + 0.5033624886056185, + 0.3783387681433026, + 0.4467152111725743, + 0.3624725608643719, + 0.4308227890886404, + 0.47163306303277863, + 0.5758612579934353, + 0.35794881728481515, + 0.40308096815060496, + 0.3317178154939855, + 0.42254062306260615, + 0.4914002221994632, + 0.4745712236095475, + 0.499789547818958, + 0.39319698934258196, + 0.3876218006070534, + 0.4809048103137342, + 0.4547662921702662, + 0.46342569622088187, + 0.3517750088723678, + 0.33773452013003014, + 0.5067403035216447, + 0.468181671364727, + 0.4407744792480297, + 0.4160078114079723, + 0.43918531275614214, + 0.841668141771693, + 0.34686327977819836, + 0.3923583629607154, + 0.3898236448668048, + 0.3693204741220879, + 0.4133278186915009, + 0.4115682085212245, + 0.3641159536344045, + 0.37896839316899705, + 0.36660006691385966, + 0.4922689501495019, + 0.4704079972029902, + 0.46529082457777743, + 0.4906221104463937, + 0.49627782359024647, + 0.4101261810391929, + 0.5784324252556164, + 0.4607336407786526, + 0.4945024930600591, + 0.5175146622850906, + 0.47376524459600255, + 0.44161291315126994, + 0.34488604055457545, + 0.4407774123892036, + 0.5523095441398314, + 0.5070555460438123, + 0.3789591726404385, + 0.4072913287685582, + 0.44252839321765697, + 0.4943634636963285, + 0.38077846257666564, + 0.42838292063937583, + 0.5132933882254526, + 0.5672598272817209, + 0.3787367855053105, + 0.4465361431050851, + 0.385383684750104, + 0.3886884912282649, + 0.3877384233569937, + 0.7980125111509719, + 0.4319263610760323, + 0.46681971395589494, + 0.38481977857360455, + 0.4011645464044659, + 0.4320862352117046, + 0.7138605634919566, + 0.478698736299654, + 0.45653534570645043, + 0.4684701522103684, + 0.4411002077723422, + 0.47955116735488995, + 0.48260646869626517, + 0.5150902922318366, + 0.31861283405978613, + 0.3935132594380333, + 0.46875860236664785, + 0.47643495902285565, + 0.4380496768960412, + 0.43838827310614026, + 0.44052790769063266, + 0.3768316042451849, + 0.4167026806396325, + 0.42792961673411906, + 0.435597905537116, + 0.6253941720383611, + 0.4118523347436358, + 0.44056292907435035, + 0.36793485423013084, + 0.6054541454224927, + 0.4255899219080924, + 0.3988759653402887, + 0.4630876854355987, + 0.48561841912803594, + 0.46973312677487244, + 0.391362804689827, + 0.5717601072351686, + 0.5009999578543084, + 0.3758893927650816, + 0.467805253041772, + 0.47202244459259246, + 0.5125873442982927, + 0.4216986774538935, + 0.48148235848368076, + 0.3760232677335837, + 0.6708976694356198, + 0.4135247388967973, + 0.47104709054702076, + 0.49597917454076523, + 0.4460518457789878, + 0.4410784638407053, + 0.4697012245549779, + 0.36200930115823426, + 0.42050899903574557, + 0.6140228052049415, + 0.4161765633863671, + 0.4040424840066599, + 0.534316303129203, + 0.44510765299960564, + 0.43182965795577644, + 0.6418961014021964, + 0.44164705644627944, + 0.5088437934487695, + 0.4607638948625003, + 0.43166306949870703, + 0.5244289776980418, + 0.42701274284880364, + 0.5134486127438186, + 0.4382032674142375, + 0.5040116430556013, + 0.39553958645292564, + 0.5029370406969906, + 0.4517111224510073, + 0.3782860881762755, + 0.3540710749932628, + 0.41946491741420755, + 0.45562181232324844, + 0.5706315884985635, + 0.48211300267910306, + 0.4704640760051945, + 0.5154055053661492, + 0.4214440159611449, + 0.4186143624730604, + 0.5039556169310504, + 0.38996919184999873, + 0.42867229510797217, + 0.5116650694934782, + 0.4248742495274583, + 0.5326576660289154, + 0.5171175990880909, + 0.42704494811389665, + 0.43151499598108956, + 0.536434154966171, + 0.44747879803574336, + 0.4601680903488498, + 0.3689935275539948, + 0.5204446083694569, + 0.6153649250876736, + 0.7995988569247896, + 0.45051839078534683, + 0.44066731596281633, + 0.447899539628508, + 0.41246267528248937, + 0.38471839840686434, + 0.3994989132405055, + 0.49028953142471576, + 0.4891415330283885, + 0.4471509889121063, + 0.4672654006297148, + 0.5153953763756953, + 0.5375694694892321, + 0.3982981220182013, + 0.3874225522594546, + 0.4849483866704681, + 0.43522455657159526, + 0.41223855146107413, + 0.4504306811010101, + 0.42022854435312446, + 0.4339777823069599, + 0.3384946942355329, + 0.5287243837258213, + 0.4605786618947161, + 0.4210466057252403, + 0.4240592689049878, + 0.3376606640775978, + 0.48687767307554725, + 0.32306737315282524, + 0.5553201868747015, + 0.43665233857513147, + 0.36949846581171425, + 0.48476049211429434, + 0.45173647284989854, + 0.3795535761136708, + 0.6208634549016497, + 0.546145341176547, + 0.3683477613250462, + 0.5063193731051799, + 0.4428104561405973, + 0.46581235269012067, + 0.481604738080613, + 0.3315896860359651, + 0.421856329163265, + 0.5291435809707166, + 0.40477122941887345, + 0.46492394968800504, + 0.44674962570486554, + 0.48130747665639023, + 0.49751896645687493, + 0.4220568163022201, + 0.5233993942756422, + 0.4908906307574255, + 0.4275585966372007, + 0.36394706791042286, + 0.48775711852150716, + 0.38823285642445504, + 0.42621998665752386, + 0.43401810712014555, + 0.4959433790064563, + 0.5554454698495033, + 0.4874220844724737, + 0.44236992903423145, + 0.5047532312149088, + 0.4643923608923504, + 0.45741981396380293, + 0.5394096130333352, + 0.4221132146977757, + 0.35170711360462426, + 0.5113734642050379, + 0.45390465381840656, + 0.47215259588390657, + 0.4617744651770299, + 0.4686044084825612, + 0.5177885625791188, + 0.39390338258804325, + 0.46287433914918197, + 0.3787579474032885, + 0.3858600788203685, + 0.3704142963914873, + 0.4525789288624914, + 0.3904453892367649, + 0.4614750100024176, + 0.5397335720487777, + 0.46592008461302287, + 0.501189601789328, + 0.7829411033946885, + 0.5763808932299425, + 0.46972157506007073, + 0.4313829310330704, + 0.48200974086463017, + 0.4923132640645533, + 0.37046649408079735, + 0.5886649345239808, + 0.4960573923328752, + 0.4838957475923229, + 0.48955529611351223, + 0.4283407216496079, + 0.5197592229863426, + 0.5276371998913856, + 0.4363988026630108, + 0.5547767475735843, + 0.5270946341415698, + 0.41299747563606265, + 0.45982860850449814, + 0.5381441568906365, + 0.44764778153300966, + 0.4794473891312921, + 0.3804563463759052, + 0.5728009572182892, + 0.5361975542158247, + 0.8275712792640806, + 0.5497423788731816, + 0.47209810347944325, + 0.297351091812743, + 0.48826727353559124, + 0.44357487170510135, + 0.6196699725903654, + 0.36940500167868284, + 0.5727376522895108, + 0.36951055608503625, + 0.41420140184285437, + 0.40480585070527825, + 0.3822754234691584, + 0.4701907635909022, + 0.39203566455195915, + 0.4527669195681369, + 0.4640373090733523, + 0.44253087989596285, + 0.3975320737058808, + 0.39251492825078216, + 0.43631880360886843, + 0.385893084790118, + 0.5704353313759635, + 0.44519286168309596, + 0.5314013071532329, + 0.4433475448924991, + 0.4040439011521977, + 0.5737360966955919, + 0.4980671588821529, + 0.403850227238359, + 0.6064320979524179, + 0.3366257405616503, + 0.47532339317641714, + 0.38710300420346244, + 0.41583025995861883, + 0.3531695696610845, + 0.42460584479801694, + 0.5168586993769851, + 0.3773074988598525, + 0.7020093090896204, + 0.472714017874302, + 0.40905905924746505, + 0.4404880261780644, + 0.40820268445240926, + 0.6149017933591945, + 0.3464434588482877, + 0.4167460130866358, + 0.408985440044662, + 0.46318712419607166, + 0.3943717275715448, + 0.43835001638597293, + 0.34258688977892393, + 0.4899581279731097, + 0.771924461416548, + 0.5216013076622383, + 0.4052818957923767, + 0.42679495841233817, + 0.45902173623210984, + 0.3681889302690476, + 0.3993014462559055, + 0.47424380118958764, + 0.757662824945957, + 0.5699862245877435, + 0.47872542969884185, + 0.4165829216856514, + 0.40063996586948286, + 0.4357573805031639, + 0.36268115463431744, + 0.4929601738687282, + 0.5457035524777782, + 0.4188126661323441, + 0.41280703345296343, + 0.4042565695318422, + 0.523800737022167, + 0.3764500318038119, + 0.5652620559375116, + 0.4930962798942766, + 0.32369811320504904, + 0.4281093852079686, + 0.4482734482849524, + 0.5397561300329807, + 0.5488528935776315, + 0.2569496332434837, + 0.7322117828859177, + 0.7175430038269263, + 0.22031433892939276, + 0.34955794662191253, + 0.4500707998353079, + 0.4806059971878298, + 0.32872997381671437, + 0.429834029499102, + 0.47714043007216145, + 0.48645064262737286, + 0.4996167767857965, + 0.395380374634432, + 0.44832883825971753, + 0.37930535137168747, + 0.3390490870987263, + 0.4549582113509466, + 0.40721424220297797, + 0.4030677828444223, + 0.5266976872730901, + 0.45458764416005415, + 0.5479864784183549, + 0.4591881577993812, + 0.617209190273705, + 0.4451166346734969, + 0.4920774198818558, + 0.5145348995330573, + 0.47171028471530907, + 0.3536563731861596, + 0.6390939016705858, + 0.6414541117684078, + 0.46751788276083733, + 0.3922583425798389, + 0.41693132554106305, + 0.5482160948816973, + 0.47676298948859114, + 0.5245888417568866, + 0.5254790765314123, + 0.49260112951193796, + 0.4593718659963346, + 0.39515105173434234, + 0.4283064528194998, + 0.5842682800240897, + 0.508084131649815, + 0.4854625107052417, + 0.5084796778987607, + 0.42256929619606487, + 0.2813594506438819, + 0.36750364158786114, + 0.40976585238321506, + 0.5465840260564737, + 0.4845337943631025, + 0.38265970764452995, + 0.4919403551565449, + 0.3895668897952402, + 0.3504084170925642, + 0.41883392441898676, + 0.43528527818261004, + 0.4079186187184786, + 0.4584170536024022, + 0.63844034489199, + 0.4089678682546266, + 0.4675170464097557, + 0.5836790612349003, + 0.4186281928178577, + 0.5291615260874005, + 0.6115256698134609, + 0.4738986170877109, + 0.5100145189191716, + 0.5019314013009807, + 0.4616715223559741, + 0.3524113212107678, + 0.38041426595165173, + 0.5092064163692851, + 0.5250091310251185, + 0.4991248506833951, + 0.3948506298788111, + 0.48766273560786355, + 0.5060206306826077, + 0.5313016389194052, + 0.4193844245334035, + 0.48028082236661435, + 0.3979026334135996, + 0.49332566322477633, + 0.459238291973399, + 0.4354111515827019, + 0.4001476384797818, + 0.4791585540506041, + 0.4876627558392861, + 0.3220692439217204, + 0.4094253217851557, + 0.45135728655873447, + 0.4276342804058128, + 0.4928714421144133, + 0.5096278301338482, + 0.5145304677239053, + 0.47574628607326563, + 0.5170053817037358, + 0.4191105641533643, + 0.48455417715077787, + 0.4915739413117749, + 0.4525713470631448, + 0.4609221936450938, + 0.47081129653424836, + 0.46793568676553204, + 0.4850329205550465, + 0.4956592827831252, + 0.7305915191151047, + 0.5566247567879367, + 0.517593305004625, + 0.5479463223447467, + 0.5325464257853795, + 0.5128728087720931, + 0.5499922681200071, + 0.37953853376137986, + 0.42215254563627536, + 0.48475730588896015, + 0.3629458610462018, + 0.46540287669733826, + 0.5135548259671704, + 0.45426287618034805, + 0.4580829362526784, + 0.4846892838309741, + 0.34787300517467884, + 0.5625377140270676, + 0.503105009031151, + 0.40503924272629505, + 0.4558254515398339, + 0.4040514040895765, + 0.3754452611713226, + 0.47823036730520696, + 0.468773274832641, + 0.44580253528115826, + 0.48015502447606623, + 0.4145419149889127, + 0.501692046965636, + 0.3861500442083169, + 0.40794082189292663, + 0.4477480613655979, + 0.5131513419405603, + 0.5054838166560949, + 0.4853050254106726, + 0.42576847330149065, + 0.3688006091574859, + 0.3705884106003286, + 0.46274573102816896, + 0.4281960874058689, + 0.4680363477998709, + 0.500149634422308, + 0.4475967789835776, + 0.42375015940227284, + 0.37557447302289737, + 0.4283895723710986, + 0.5476993197350021, + 0.4262783349243901, + 0.2189947230825509, + 0.38084075120231653, + 0.46114452762032454, + 0.3256595609403827, + 0.31641264180182277, + 0.47944050423570433, + 0.47740600529102956, + 0.34842331923761716, + 0.3274617995319502, + 0.41532220205032183, + 0.4080312493675272, + 0.5533272723398822, + 0.33489965239706715, + 0.342124760072201, + 0.36677584622547543, + 0.3846080546536604, + 0.29240986676266856, + 0.35751395566955424, + 0.33476795874106974, + 0.3716600067540278, + 0.3412101294408066, + 0.4593695867436564, + 0.36818620276898767, + 0.4107089772264469, + 0.41265812863002327, + 0.5027402668643168, + 0.3574514269111899, + 0.4201327517645565, + 0.4124008716019579, + 0.42153469352948614, + 0.4168378578475842, + 0.42375191534827966, + 0.4127591455698563, + 0.6507111550425976, + 0.362251160594062, + 0.46398908397046706, + 0.5942295213337939, + 0.5082001891823514, + 0.37873094131489377, + 0.48376208859687325, + 0.4344368355431122, + 0.4775175260138621, + 0.48831307850833267, + 0.373138335131899, + 0.5910444994885623, + 0.4459257196318419, + 0.44338651901821413, + 0.37802801213898507, + 0.45005069633017686, + 0.4119221249448491, + 0.4761998334426833, + 0.5104900035928708, + 0.4659229980706639, + 0.4831505973264057, + 0.5806911106449015, + 0.459351568677257, + 0.4259353415212689, + 0.5222379303071375, + 0.5907002724631887, + 0.5444454067683879, + 0.4582104333737668, + 0.5923821984487457, + 0.5782550104212657, + 0.39824846942962144, + 0.4113516870218821, + 0.45955130040599385, + 0.3620849267835652, + 0.6868821691692076, + 0.4990043084558347, + 0.4451199636442686, + 0.3689133227541178, + 0.47951118096762163, + 0.41520815253735477, + 0.5344039557461031, + 0.4527752706018967, + 0.48359845707698723, + 0.547240021447599, + 0.35869596351949834, + 0.43519240895439476, + 0.5672798256374627, + 0.3650879312962449, + 0.44669714773985936, + 0.3439125672347152, + 0.3903767237517224, + 0.41899729299204175, + 0.4192807404260913, + 0.491589459444821, + 0.36436377262543873, + 0.3450198222251375, + 0.4536248752035587, + 0.4171606140074862, + 0.391173885772363, + 0.46898585583063873, + 0.48462050158746395, + 0.44433115600312767, + 0.5428234968541288, + 0.5737914102062042, + 0.480330339542722, + 0.3762304194665403, + 0.49409641559907125, + 0.457205960013108, + 0.41218985010994447, + 0.45237211360594337, + 0.34511747665726306, + 0.34581018463330415, + 0.3267935866703807, + 0.3840712369805719, + 0.4499436910156588, + 0.6594140022494454, + 0.4497998753896334, + 0.4409589060260232, + 0.3101361909915304, + 0.3728311897711998, + 0.33146478898498355, + 0.42507617098295886, + 0.3848671225226373, + 0.40746952509007195, + 0.36970589811538357, + 0.43419241925539065, + 0.4114531356649939, + 0.43290127699094116, + 0.427129516145276, + 0.3904998704969913, + 0.5132104147274585, + 0.5086861600181468, + 0.4007627559041198, + 0.4456948290894919, + 0.4595489152823072, + 0.4042719444934848, + 0.37932757598579925, + 0.5233533368653325, + 0.4626915442882737, + 0.5919312433713427, + 0.46616861453906383, + 0.4591981144537696, + 0.45815892763583105, + 0.4656628902865604, + 0.36419637455499027, + 0.6309019947979818, + 0.4213780773795039, + 0.24119387223569885, + 0.4938876936003874, + 0.6112184869374256, + 0.38763079738399675, + 0.5028096782070648, + 0.7692741551619767, + 0.46432189959957804, + 0.5412654095001562, + 0.32569679652172495, + 0.5073218040304613, + 0.37019559551360554, + 0.3901264286091799, + 0.37066112432429227, + 0.3230852324258993, + 0.4610577305066026, + 0.27048484267085854, + 0.4422940777972294, + 0.4701408265273227, + 0.44994860477972765, + 0.4167064226974009, + 0.4003634465099385, + 0.3762080370851736, + 0.4781121264274003, + 0.5345877065391672, + 0.5569066058183186, + 0.37456178839414256, + 0.4668634699508898, + 0.43963439058877374, + 0.7417099496390339, + 0.4354966989014325, + 0.5692417387062598, + 0.7424938658913709, + 0.7218664508388603, + 0.48548746310523316, + 0.41093882130714876, + 0.3647558992265856, + 0.4712101027749145, + 0.5158915490774334, + 0.5037272305340112, + 0.5917530866514882, + 0.43778919338427275, + 0.5624029188364726, + 0.4852817534516938, + 0.4119176304862642, + 0.7096797766645327, + 0.6886792661429898, + 0.5112602756476619, + 0.43272330961630884, + 0.4868162294183191, + 0.4358559456080667, + 0.47077548436573596, + 0.6525370849027693, + 0.6779851810521923, + 0.44268167816282605, + 0.4522875354877165, + 0.45246382768193133, + 0.4055853077708553, + 0.5072692183010187, + 0.3812381808466232, + 0.5500184912211991, + 0.5452642329950911, + 0.381666054947333, + 0.47020392258886173, + 0.4841471056749226, + 0.40422868957441604, + 0.5508576143120008, + 0.5360203466437876, + 0.44439207912150014, + 0.3979056408885032, + 0.5061740217084781, + 0.47023510699938886, + 0.38331696641715357, + 0.363839248196877, + 0.3098220515190491, + 0.26877365998476, + 0.32199181047209086, + 0.5076702876791032, + 0.5607726182509826, + 0.42240549283452566, + 0.3325014131833631, + 0.4268966999383317, + 0.5084359721957296, + 0.5077873815121668, + 0.5338339869799057, + 0.428365165713974, + 0.5180070825687962, + 0.5632107427462731, + 0.43418732565594464, + 0.5960025446454206, + 0.5508706206891646, + 0.4710795096613773, + 0.449878050234594, + 0.39948748560847586, + 0.269559050172743, + 0.315087630367402, + 0.3037696544955132, + 0.5697262082752579, + 0.48182882693165896, + 0.4430644896352711, + 0.5736654317028738, + 0.5950692485867984, + 0.34377208587954694, + 0.5002152482837473, + 0.40282888010228374, + 0.7305044081983816, + 0.45837014055214276, + 0.47004581375891097, + 0.3822965551404915, + 0.4698405873764922, + 0.5119814822877177, + 0.4582925117970679, + 0.4295028610543086, + 0.5125811323725087, + 0.47286950064123406, + 0.5868266307682021, + 0.4409778235337301, + 0.5161903058443624, + 0.46618789830009405, + 0.4514736707656242, + 0.43403186248828335, + 0.4950792950540884, + 0.4592525925432495, + 0.5123947474759227, + 0.39216198722965784, + 0.37036467465327955, + 0.3659195641944773, + 0.44060514212682605, + 0.4855642972855972, + 0.4236400107150137, + 0.49437806773221954, + 0.5175895815169407, + 0.47360720534821077, + 0.4432165310202626, + 0.45981177521860417, + 0.2829205413753541, + 0.45431419362462583, + 0.3343294063287899, + 0.47462439212926344, + 0.2487339172685465, + 0.6796922974430679, + 0.3440842909850687, + 0.524048700571701, + 0.6043709110400495, + 0.5252933626627071, + 0.29009785185790443, + 0.48678716951240336, + 0.38884541329946304, + 0.7856532515305669, + 0.30495985702461864, + 0.5692843459876376, + 0.5341318140700317, + 0.5631556016250095, + 0.6446053091689257, + 0.5064202941171835, + 0.6637258540383785, + 0.4873698249301387, + 0.44444285306911124, + 0.3885740140935141, + 0.4856990277039211, + 0.34136848737048253, + 0.729533510534357 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.5080138511022613, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.5894123703224724, + 0.7680751326929405, + 0.7417081829101788, + 0.8243945529833355, + 0.7516336078664528, + 0.6980293171134263, + 0.7496864513393088, + 0.6876639751450326, + 0.44042640353139917, + 0.727766408228179, + 0.39715860741256503, + 0.527204589569849, + 0.4371820516743961, + 0.4057417247029655, + 0.44633975980979085, + 0.3653630538801399, + 0.4148237305767207, + 0.6023553789086589, + 0.6009605313859059, + 0.4883555173479847, + 0.37119912935652655, + 0.6986396066460054, + 0.4884987004568605, + 0.6162003701006447, + 0.3993032797325312, + 0.3591599667858095, + 0.46966980522455176, + 0.5769105481460239, + 0.6678850117939042, + 0.5233451914186424, + 0.34844059059751126, + 0.4459988367574106, + 0.44985825051828104, + 0.39828544399285476, + 0.5058744987740539, + 0.38772115758209896, + 0.5094720389743043, + 0.5702253604759775, + 0.6345111803804628, + 0.6365057581882823, + 0.7982606594993786, + 0.7888494295407448, + 0.7834371262574216, + 0.6739846659313542, + 0.5056982433553153, + 0.5697058545679491, + 0.7891893616707754, + 0.4506749338140741, + 0.43178209867984685, + 0.5724078251061375, + 0.4765846329375246, + 0.7840499475769429, + 0.8000532093715029, + 0.5052151191547936, + 0.5095564434148796, + 0.4309065433434666, + 0.42068353729476543, + 0.46905197783664304, + 0.4658586688277218, + 0.5922433377593004, + 0.5564620925642896, + 0.5199885887699296, + 0.52629464182745, + 0.5641298640385773, + 0.6756916752152344, + 0.4651864113987617, + 0.4223116887092415, + 0.6402291950096591, + 0.4676633435455379, + 0.5498775176682363, + 0.5661990896246295, + 0.5461923209933802, + 0.5130252762180022, + 0.5788661435306176, + 0.48566732767048953, + 0.5909893967268899, + 0.3728966279860746, + 0.40995584106848737, + 0.4591390397329659, + 0.3687373778730283, + 0.4877352483361368, + 0.6137537478031413, + 0.442740122666356, + 0.4270391627267165, + 0.4301534235782809, + 0.5210939596336182, + 0.5104443696961954, + 0.48775183431210895, + 0.4000149849585672, + 0.4351715850500309, + 0.39193582566681184, + 0.38964044609983006, + 0.5632496728027913, + 0.5846529951007905, + 0.4655276958088889, + 0.6238283429815699, + 0.41420749120209294, + 0.5960305923039745, + 0.4548551286841599, + 0.5087860274941737, + 0.4954661201648947, + 0.5684386233165216, + 0.35009921318974446, + 0.5109482582352807, + 0.3819705700649906, + 0.39562312235319386, + 0.5260933089907601, + 0.5658066775955373, + 0.3824434188851883, + 0.817997188370617, + 0.8116926160443433, + 0.816376322738495, + 0.8349090484002949, + 0.5940356259836104, + 0.713414339448727, + 0.7289372190323111, + 0.7659707129536845, + 0.7191439934168404, + 0.594661589368632, + 0.6703597603519807, + 0.5814801036331655, + 0.7811604116255708, + 0.673140528139856, + 0.4602427571361908, + 0.36586413456450084, + 0.34481926783610234, + 0.3767403919347382, + 0.5915669775056707, + 0.4649222889708378, + 0.3268260122031279, + 0.5102459633866424, + 0.4775643505143354, + 0.5310316176062635, + 0.42442313273118876, + 0.44922012240971876, + 0.517236029112669, + 0.5083148328851449, + 0.3999226519241192, + 0.43036737114798707, + 0.43431776141558376, + 0.40162388794373544, + 0.3480429648948281, + 0.47418995607016545, + 0.6215702089079267, + 0.33932828853605235, + 0.585492223753991, + 0.39161720548842993, + 0.5186614833705572, + 0.4745559614711091, + 0.5879792722470594, + 0.3535915343320998, + 0.5264030818651395, + 0.36796726724353485, + 0.3948366839172049, + 0.45587053513311593, + 0.39005405467737997, + 0.3061458355230482, + 0.2279585940780106, + 0.39782235196137555, + 0.3955352563390811, + 0.5684006111360085, + 0.5611903967457961, + 0.634296624123968, + 0.5658985814567318, + 0.4088129958375349, + 0.5327156785381173, + 0.5807620679028992, + 0.5707237120382408, + 0.47519734871630065, + 0.5164802048943596, + 0.4618980529425321, + 0.33972806685383056, + 0.5214302506323066, + 0.49499107927311764, + 0.31367458681142824, + 0.33548860543271897, + 0.539467677072544, + 0.6101025716621897, + 0.5958803963895338, + 0.5206590687731404, + 0.36295114305156934, + 0.596370213233873, + 0.541575905818712, + 0.5547956468121236, + 0.3439833506401796, + 0.5951509577303011, + 0.5055585933463626, + 0.5555976437749266, + 0.28986099141947624, + 0.4793517775103618, + 0.3754643853004301, + 0.3088386751385059, + 0.5719953518917378, + 0.4842832821816547, + 0.5956173265161917, + 0.4238281852182136, + 0.6042100426571275, + 0.3995243281983729, + 0.5013966627625148, + 0.4324950775887962, + 0.3684570319602946, + 0.3049111425772501, + 0.4173782237249425, + 0.4115008470600789, + 0.6530405014154064, + 0.39722009967447097, + 0.3288063841428355, + 0.5062995651208754, + 0.6041359322350568, + 0.4465807285353169, + 0.4946563562291815, + 0.474139703738099, + 0.3894046423246138, + 0.42034330226018873, + 0.40768052916886566, + 0.5627142525321419, + 0.6434668795227536, + 0.6452558252047563, + 0.6055427573314448, + 0.38129546031072964, + 0.5531847076569533, + 0.49264019679482973, + 0.3129055349051735, + 0.37557120934206584, + 0.40575554169786066, + 0.2600830018286943, + 0.5759237933766774, + 0.36442214522345223, + 0.3605149420981834, + 0.5626922508088462, + 0.32389765632301437, + 0.30917541575290414, + 0.5663237310063667, + 0.44273058513479324, + 0.35628566411847074, + 0.6528277852816768, + 0.5821265981436846, + 0.33652735778260784, + 0.3448541039853627, + 0.5094795249378156, + 0.35288440570200946, + 0.48320279083806544, + 0.35230493589284434, + 0.5374500339758318, + 0.4859106482854587, + 0.42603114944982823, + 0.3907546748819868, + 0.363044239830724, + 0.48999848870887436, + 0.3822552114935078, + 0.4716205049997844, + 0.36096478041767704, + 0.43113507470170814, + 0.3896902036110998, + 0.4032687838976266, + 0.4851043825200128, + 0.44775998933361477, + 0.4147138456086605, + 0.46854401584799366, + 0.4799824802977234, + 0.3635275455030491, + 0.5745657830753365, + 0.36490194405850546, + 0.39745828656783116, + 0.4302874678227335, + 0.358034032056197, + 0.39276588633073517, + 0.5536619838415873, + 0.30696701009055033, + 0.3836394147312074, + 0.3729813387747543, + 0.38496857313762284, + 0.3995307746731445, + 0.43570949426691463, + 0.40980554210931974, + 0.5206313055463012, + 0.3293317306486617, + 0.4962405796137819, + 0.535845305592695, + 0.34613118682558935, + 0.3407819322366794, + 0.6460767758576523, + 0.5230153417275272, + 0.3151226856561061, + 0.5898889639404857, + 0.5440727478445176, + 0.5156529805740817, + 0.5256848669326595, + 0.6018233406235223, + 0.5665571995580392, + 0.5850684070346748, + 0.5736646684050345, + 0.5766382071779033, + 0.46124660737659345, + 0.5222567898183985, + 0.585405797097173, + 0.5230343771077328, + 0.4926826618185762, + 0.6667074464853694, + 0.3638856285092496, + 0.6241910521102645, + 0.681804910744344, + 0.7407726752994647, + 0.714618678820989, + 0.550933678106, + 0.5885583129952344, + 0.4492521877515217, + 0.7078361406239582, + 0.5162442251155284, + 0.5281953555300033, + 0.36582505139339755, + 0.550712750454999, + 0.4639509244672636, + 0.47847492471982633, + 0.5978029170275023, + 0.3799011327291813, + 0.45764820987486976, + 0.36186850973468715, + 0.4867731033345604, + 0.41806693540955536, + 0.3716031967817945, + 0.4179551718261845, + 0.3091669998904241, + 0.3767542105260163, + 0.41220412206482676, + 0.4194986546459361, + 0.4070655620349635, + 0.41749574717564863, + 0.6336725269919864, + 0.5318469920374266, + 0.4892705236461799, + 0.5549207743279321, + 0.5211787695008046, + 0.3596367742655367, + 0.33737748928553757, + 0.6246123839785584, + 0.4384711687560624, + 0.3711680037151338, + 0.4887332693096717, + 0.6105653075598453, + 0.6373759879668142, + 0.48229111883890136, + 0.3899258889004977, + 0.5178317356434265, + 0.4126355562836974, + 0.37978667194141746, + 0.5406301320603898, + 0.3819547811562322, + 0.8299738396437404, + 0.5555728599714864, + 0.4583175830083443, + 0.45954148047496596, + 0.48796955100193284, + 0.48327511953387475, + 0.44944280545496845, + 0.5627218546295039, + 0.4319314640543975, + 0.5860035617875534, + 0.4900794737867277, + 0.5567321596119855, + 0.6036287360472481, + 0.5061307489431731, + 0.5227728347409162, + 0.43614910640710425, + 0.4531363468747479, + 0.6432235312285748, + 0.6131313495657958, + 0.46516031271497754, + 0.6291796003621736, + 0.5060801816397689, + 0.596241460080734, + 0.4840004017052337, + 0.5228153058875247, + 0.46994689257888717, + 0.598758946877123, + 0.613490557572255, + 0.656820029316039, + 0.5799208224060692, + 0.5747212885175028, + 0.5629753760594342, + 0.5271817842101203, + 0.4398353509560321, + 0.4379649389717979, + 0.36939287577688895, + 0.4293369355796594, + 0.35764821227618065, + 0.39009195136009694, + 0.4131071368109833, + 0.5679191652253114, + 0.3967826978361276, + 0.3703334616133506, + 0.43380912988949377, + 0.6014907680197107, + 0.6692022713704553, + 0.6245529752437398, + 0.6132087314653019, + 0.44746183847889964, + 0.6274431744923312, + 0.4912154744088129, + 0.6575104093571993, + 0.5072798400446236, + 0.6207367786232099, + 0.48478428361363324, + 0.603985608924173, + 0.6275321869123189, + 0.5342066291377304, + 0.5264882375933715, + 0.5656064833523511, + 0.6328662248282055, + 0.5793828136177791, + 0.4564342752478502, + 0.43461763250136354, + 0.3824629817620969, + 0.4125697559293593, + 0.5670365148348296, + 0.6075720607976575, + 0.3580133598167808, + 0.6625388127542586, + 0.6389755958279071, + 0.520764862803018, + 0.4416668565846553, + 0.5121042346586586, + 0.612130895888318, + 0.5763711962331297, + 0.5232638662530656, + 0.867042772412465, + 0.5107424174265595, + 0.6158235291245193, + 0.5665757023948768, + 0.6006218748483331, + 0.5053658370481813, + 0.43704653161615536, + 0.6384252780759323, + 0.49827331238812966, + 0.46317672420962314, + 0.4911507150984972, + 0.48163372191700504, + 0.4949170808223236, + 0.4574169749831987, + 0.4893812115654351, + 0.47673851341773643, + 0.5034120738051722, + 0.3464454979176856, + 0.5170581887878125, + 0.355693223786896, + 0.4282037753210482, + 0.3189497819399535, + 0.5205908930459122, + 0.4130129419081543, + 0.35904823420888593, + 0.5902330230690496, + 0.43370240327028864, + 0.31814492820236534, + 0.5785566434770867, + 0.5009465957307825, + 0.5050321716131819, + 0.5599155478465692, + 0.4437979849126085, + 0.4519504288503658, + 0.6712277720477284, + 0.4556520989676147, + 0.5684543372738635, + 0.4717034059890399, + 0.33288625142566947, + 0.31432899203218484, + 0.4323735171842055, + 0.5354332387787769, + 0.6334805975078418, + 0.6209839070521334, + 0.5183165908013969, + 0.5808949003047087, + 0.5119908993684045, + 0.542323859900519, + 0.48012970727940346, + 0.4604022588355171, + 0.4735435421456267, + 0.4574848060073297, + 0.5145278353567557, + 0.44652796117936716, + 0.4851930312349406, + 0.571172539442148, + 0.506596034539973, + 0.48400819292653846, + 0.4100603197920836, + 0.44776333405937646, + 0.5396192890031875, + 0.5956233526939315, + 0.5792970926388196, + 0.5932783632879873, + 0.6120364317678076, + 0.6432112195098524, + 0.6801113846507428, + 0.6627630567076453, + 0.5009370583668493, + 0.5464846363227359, + 0.5281982555559702, + 0.5658711101978333, + 0.4372961933049859, + 0.5406670864248784, + 0.5058346594606752, + 0.5962419248665034, + 0.43113800051410406, + 0.6061459616725714, + 0.666406878805297, + 0.6356866316466235, + 0.5859343457870565, + 0.6593956677768503, + 0.5601647134874005, + 0.48657181182278947, + 0.4126703487989875, + 0.47216955239776676, + 0.3753419959884144, + 0.4396996095582384, + 0.5911610456516714, + 0.6309381514692429, + 0.5120023659777433, + 0.4162398941519814, + 0.4349643006470848, + 0.5416777904620507, + 0.5384985963277659, + 0.43411456810834353, + 0.5534806110615687, + 0.5246322597853532, + 0.4055650722279046, + 0.4571867520168195, + 0.4272408783093784, + 0.502915530449846, + 0.4835869462996916, + 0.37852868351965213, + 0.4530426586834751, + 0.514855301401544, + 0.44048862207238104, + 0.4291060632054143, + 0.47087202817317314, + 0.42523813518837383, + 0.41247384849276064, + 0.47735298437533225, + 0.5891674521063786, + 0.4601988341442441, + 0.3288485386595437, + 0.6120944625355408, + 0.4330583664018283, + 0.5927493056297944, + 0.5809744143124798, + 0.4171091136384715, + 0.5506573171300452, + 0.6279884489257336, + 0.5372438937302652, + 0.6850763951379728, + 0.5152009351200947, + 0.5708304376123033, + 0.542129933900715, + 0.5282824467156417, + 0.5774098062460654, + 0.5417437880347429, + 0.4005441078128907, + 0.5255780598366869, + 0.5643695562779758, + 0.4680240831808521, + 0.35279030087028523, + 0.49857875621843817, + 0.5254063686169578, + 0.5036429234671777, + 0.6737650829261809, + 0.5773031470317288, + 0.583943875067783, + 0.6057098695955397, + 0.6660528299184364, + 0.586659665909058, + 0.6428259734310949, + 0.657336036327393, + 0.6441050851638861, + 0.4927321188747843, + 0.5099842495807054, + 0.46472467000165585, + 0.37140337210364555, + 0.3345731163353805, + 0.40466948225178956, + 0.43125108065311396, + 0.37645799183651485, + 0.34519556358453124, + 0.5314591990295835, + 0.33884705328689074, + 0.4971285623160484, + 0.3810566241859527, + 0.567279043844881, + 0.6272027134657545, + 0.5223316369055907, + 0.3801504044609583, + 0.42268996300695594, + 0.5926401891847954, + 0.38073731468818606, + 0.533907083994536, + 0.35350673694506934, + 0.8406084484205747, + 0.5642357260346942, + 0.45816735178740686, + 0.6459773150696352, + 0.6047410824854852, + 0.5273334991074818, + 0.46297974130941316, + 0.5254384571835169, + 0.6081503143782847, + 0.47781073499590176, + 0.6195599917145185, + 0.5764026803269884, + 0.3372527719966718, + 0.5389063697333819, + 0.34400374665712513, + 0.41431458860702775, + 0.3553109010871457, + 0.5548748217569424, + 0.4630871491364715, + 0.41418455536264553, + 0.5763195276270127, + 0.6052653247988942, + 0.42425048971064167, + 0.5242859991599016, + 0.4704027731350888, + 0.5000238690134505, + 0.5280107328050506, + 0.5086250327706704, + 0.38310053094379826, + 0.3779125601381163, + 0.6198789349771395, + 0.4601525859947791, + 0.4146321409476432, + 0.5524218869255879, + 0.44097897150445114, + 0.48264676257019057, + 0.36254860056776184, + 0.4094536900714801, + 0.4621722264296072, + 0.5167259698354636, + 0.5184179437245486, + 0.39217958641454354, + 0.5154760908752912, + 0.3472756307963731, + 0.5435829587697081, + 0.49727478069654885, + 0.5242727510508631, + 0.5340231973445866, + 0.44054941492177674, + 0.4952103212412098, + 0.4167373232733561, + 0.5409593089337876, + 0.45448362674132, + 0.4967671521445003, + 0.5087461505900461, + 0.39300897176571625, + 0.5411759505246085, + 0.30879316083820235, + 0.6755044450609892, + 0.6777054783265233, + 0.5760821255191191, + 0.5673284706191534, + 0.6218986949656119, + 0.617741910658001, + 0.30778861950727854, + 0.4331349708212048, + 0.3908025998832029, + 0.43839912307537193, + 0.37356306413406953, + 0.3782162659925782, + 0.4374155439718679, + 0.508586433406161, + 0.49129316305030857, + 0.3930970779264924, + 0.4164561948522161, + 0.3462948536688352, + 0.4818746375928835, + 0.2711292888808239, + 0.4564405490784158, + 0.5148621560060921, + 0.524788192395808, + 0.5380516814188838, + 0.46042548362836, + 0.5699338616543672, + 0.32515179749683815, + 0.5553210816067168, + 0.49155434762812944, + 0.4365585533534503, + 0.41868734004011027, + 0.4529531080940684, + 0.46609824921870213, + 0.47328465574315975, + 0.5072158513132606, + 0.532276684350413, + 0.5689696831979846, + 0.38218713594540127, + 0.42094435053119156, + 0.5002190731083904, + 0.5582878908631291, + 0.5459714325837051, + 0.5560078303532723, + 0.38552689463518497, + 0.6165706425786573, + 0.5040833140249332, + 0.4847200935512793, + 0.6300390204953846, + 0.45841786862244155, + 0.3913498439716891, + 0.35255385552215485, + 0.43775901176819426, + 0.48195405180518447, + 0.4340676989210015, + 0.446313330125983, + 0.45929118541517167, + 0.5800497369637548, + 0.6274778576517595, + 0.6807482433327282, + 0.5726848978954406, + 0.5871727874645127, + 0.5449801110532408, + 0.6157368711256727, + 0.5540306496528713, + 0.6351304177742665, + 0.39643899295839785, + 0.5235280399445765, + 0.6406143024851599, + 0.6853238611409189, + 0.23176634298667886, + 0.4302828361516269, + 0.45809635912298013, + 0.5697173209393932, + 0.5527868604337993, + 0.57142799480702, + 0.42733186188476857, + 0.5421851911109112, + 0.7584140349720585, + 0.6268728848362288, + 0.7945747729921383, + 0.8088924059627045, + 0.6714608401704627, + 0.7707353241499066, + 0.7768463820651154, + 0.4349344289663494, + 0.4467474469276649, + 0.4581915520269597, + 0.5563392914654304, + 0.3782785678459021, + 0.5150831087211126, + 0.4094658053142089, + 0.5178749779985761, + 0.45315861543011715, + 0.43086317938274366, + 0.35506717758296386, + 0.4582136921269877, + 0.4530335967488371, + 0.4859699983397078, + 0.4643051508915309, + 0.3556565041566969, + 0.3766216442215466, + 0.3646953833176389, + 0.3849622256341348, + 0.29486415770652014, + 0.37911414899094453, + 0.34189670116213944, + 0.5635422063679816, + 0.6061978733135192, + 0.4029926733579349, + 0.3462435238036658, + 0.29651803163727697, + 0.549259757109353, + 0.6562193574974631, + 0.654086824787313, + 0.40756513012467127, + 0.39445084357192023, + 0.2910209349650054, + 0.5339215838802046, + 0.35885780116346216, + 0.4249720705007928, + 0.5118912242596669, + 0.582937199373766, + 0.4975784783845059, + 0.34666931551424623, + 0.49904825799294344, + 0.5585014784476284, + 0.5452179038257732, + 0.4918162834940736, + 0.5416315108949971, + 0.44724785168395104, + 0.5501234241256259, + 0.521343919631204, + 0.587129134889581, + 0.3921995504125928, + 0.5210454715524759, + 0.7084223147702408, + 0.4786663727607817, + 0.6935813285358365, + 0.4158739669773593, + 0.5680074087540272, + 0.7258067634760502, + 0.5932285378584264, + 0.43249793279864274, + 0.41018994627543776, + 0.3761652205512484, + 0.35252205629712, + 0.5106119373605774, + 0.6320351949202033, + 0.5071094294281115, + 0.4411194595611289, + 0.3670705975480676, + 0.5587125955138279, + 0.5440555812899004, + 0.5818043958926307, + 0.6126735422719222, + 0.5687573596235681, + 0.526892778366989, + 0.4563432239345059, + 0.6053415155347163, + 0.34023846177241984, + 0.45398791849086884, + 0.5105623230083259, + 0.44848053286630674, + 0.6693018035071073, + 0.6692618055107404, + 0.6258881487992124, + 0.6860353708865355, + 0.7599517887043398, + 0.6169602018813689, + 0.769327633022932, + 0.5546255052296345, + 0.4953320966361486, + 0.4746584422918453, + 0.6202236527785316, + 0.4077677469215119, + 0.5552041128199987, + 0.5669558271458941, + 0.5538723506188916, + 0.5161461172359793, + 0.4784270116600123, + 0.4859857901613686, + 0.5363944330236441, + 0.4503612094350895, + 0.5959912057352136, + 0.5732946319130808, + 0.5841179534797436, + 0.6229275396662748, + 0.6718802402552384, + 0.612110758292962, + 0.4889457667669399, + 0.6389277688994595, + 0.6159589447354025, + 0.392594205506028, + 0.4785441162102573, + 0.4532650964451074, + 0.45131241414876466, + 0.4347999434874022, + 0.46313904140038364, + 0.5288184224863024, + 0.5153293797579583, + 0.5252229822811436, + 0.510834705889777, + 0.333609032831099, + 0.4860948177884061, + 0.5802722181834733, + 0.52622005414949, + 0.6275952931654531, + 0.6571793806846719, + 0.7517424663823195, + 0.3822322073534541, + 0.4091198914425073, + 0.5571723809453396, + 0.44161986476711584, + 0.31903108199794145, + 0.3877286573163239, + 0.3713779910092758, + 0.696879803484888, + 0.6499542057931481, + 0.6747368959855413, + 0.38760278211185645, + 0.4354499303237207, + 0.3646389995548863, + 0.6181348415895562, + 0.5237488371819735, + 0.3040699145144434, + 0.3526842679296661, + 0.5165026444249314, + 0.4695087857758301, + 0.5401298427697686, + 0.5468292153551099, + 0.46425285585499104, + 0.4499730219091615, + 0.33648240346664243, + 0.7030781434499743, + 0.5857639802744269, + 0.7116730969109755, + 0.5976998517896267, + 0.32989571009792323, + 0.5548607335965616, + 0.5738365292659687, + 0.5699669655581439, + 0.5217474924084411, + 0.42978697223559226, + 0.5373426345143397, + 0.5904059083426597, + 0.5807332380214013, + 0.5221521683468814, + 0.40280167022932795, + 0.5254128567533598, + 0.4355361226709282, + 0.5025358811172396, + 0.6542048573574307, + 0.6083571598111794, + 0.6026204061535492, + 0.48178576671472667, + 0.5114984925093109, + 0.5132801351692425, + 0.46664815354500894, + 0.5372762803672042, + 0.4539276045300009, + 0.3251428430266344, + 0.505684575834985, + 0.5911712390753402, + 0.3065711339386387, + 0.7711058539147149, + 0.4775676316117374, + 0.6227362174530082, + 0.7439951727349874, + 0.6911104552417928, + 0.4182015870475997, + 0.8148278431169128, + 0.5148742236173707, + 0.6859488872340126, + 0.42594762248241214, + 0.6204347354378653, + 0.6372440830576898, + 0.7104307371299073, + 0.6595142829706369, + 0.7142880524448376, + 0.63483478769404, + 0.4633747667391017, + 0.6351132456269438, + 0.5813106396237782, + 0.498854912054586, + 0.7192678669792261, + 0.3958523018273691, + 0.6351787672146914, + 0.6503572589939127, + 0.7201347121071864, + 0.6614932970204354, + 0.7229444835071435, + 0.5992415956017239, + 0.4417420502650074, + 0.5653960978937286, + 0.5670616966934344, + 0.5523807843315705, + 0.4459799031176895, + 0.4088354319957824, + 0.4870691982199181, + 0.5692652488872746, + 0.5803139104374133, + 0.4583311458780564, + 0.594690547911929, + 0.6059473762572698, + 0.46612696893420047, + 0.5954013257901352, + 0.5810542086298904, + 0.5566652618089712, + 0.4932611777599108, + 0.44383453857263144, + 0.4837758280351004, + 0.5480119758318768, + 0.46684817725346217, + 0.39004753292269007, + 0.4911574506467848, + 0.4073056966882412, + 0.5404212242654584, + 0.4628474278763118, + 0.3884991035239702, + 0.40125999104081156, + 0.42452934546758725, + 0.5307797804625102, + 0.5040726873430162, + 0.40394119383138305, + 0.3666415969911816, + 0.4558360870492865, + 0.6042865482000124, + 0.6704522233764899, + 0.3541496737693549, + 0.5729444510968926, + 0.6258469318958433, + 0.6076504111113817, + 0.6092931773966729, + 0.4587562531837773, + 0.5831766984366489, + 0.4562227909684163, + 0.6253479232580292, + 0.49550259150841064, + 0.6653380084522479, + 0.6117749261625066, + 0.6162125849222156, + 0.35371530119290245, + 0.38803322629627585, + 0.44090642572541416, + 0.5342223921434719, + 0.5145520942678491, + 0.5948267066504659, + 0.3770646124869722, + 0.6160392636028876, + 0.5482270533773752, + 0.5913147143094326, + 0.5819125044316998, + 0.5513579695067433, + 0.46454551380244485, + 0.5801567825275327, + 0.607901732218832, + 0.5890887147599904, + 0.5691138700140148, + 0.4761498777431086, + 0.38349550360869256, + 0.413484311813722, + 0.4123687407246798, + 0.2837170440624349, + 0.43242418216157036, + 0.5611150817344456, + 0.47310114319948154, + 0.4579570913461949, + 0.48671504070870864, + 0.4459719648514652, + 0.5573590565608928, + 0.5857578811776646, + 0.40080366058627237, + 0.5200221327112726, + 0.6049305555483904, + 0.41637829639802726, + 0.4734532257870093, + 0.5487810057132871, + 0.5235591050976213, + 0.6322624375328605, + 0.3956918609376224, + 0.5839605681271062, + 0.4282561027177997, + 0.683689782640079, + 0.6266070634956993, + 0.6761686139353383, + 0.35452949799526273, + 0.37446648027077206, + 0.3515645061422708, + 0.3307753031406501, + 0.3631188820270901, + 0.5163927095991266, + 0.5561843776305808, + 0.3205423545937168, + 0.44763445171028116, + 0.47246714199934253, + 0.5136023604147525, + 0.583841179307132, + 0.550019446867078, + 0.6860869055775181, + 0.35125910075756717, + 0.48181388775198125, + 0.5065843033266156, + 0.47711970047613755, + 0.47966502034203146, + 0.5873904366385765, + 0.5377596864390047, + 0.3302023292672014, + 0.582740589876202, + 0.5213286250598583, + 0.4772087978424331, + 0.6232998693558848, + 0.47523668889358256, + 0.40506389059865844, + 0.4376356593154528, + 0.4907183072462159, + 0.48380079917979185, + 0.4455729916256905, + 0.48325192953370716, + 0.648319773693314, + 0.49721293942335815, + 0.6517488364069872, + 0.559557712917537, + 0.5611124788030545, + 0.6216765462758794, + 0.5699721559242914, + 0.5539179425388828, + 0.5369106382204499, + 0.67211007016156, + 0.6738364788956418, + 0.5932206573584738, + 0.6115783147174274, + 0.7609990349332139, + 0.49433287002501763, + 0.5110469579818836, + 0.3940658129382347, + 0.5733300506255685, + 0.605207412840217, + 0.5634290915200013, + 0.4928340223823078, + 0.5850573792937258, + 0.4105195504076307, + 0.521282467448056, + 0.4447962320279245, + 0.3328806795113488, + 0.5067819302872354, + 0.4882408696227352, + 0.49521140675907577, + 0.49853420177112184, + 0.3804927292776563, + 0.6607134425348401, + 0.6935547377370571, + 0.5620133856682801, + 0.5984304460332481, + 0.6677934422533197, + 0.5494097348810824, + 0.3566576060296595, + 0.4628681926644577, + 0.4358004307190133, + 0.4095476691487938, + 0.4522299238953462, + 0.4930455566646224, + 0.5372430572766593, + 0.4914152241282144, + 0.3622967919109001 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.4312373396122038, + "valid_tracks": 4, + "per_track_pearson": [ + 0.469610031425627, + 0.3956720978533213, + 0.47034813870847747, + 0.3893190904613894 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.6324484478571953, + "valid_tracks": 626, + "per_track_pearson": [ + 0.6506915067626478, + 0.6361102969742879, + 0.6436584709833866, + 0.6441615293414488, + 0.6638684370567619, + 0.6464699264620332, + 0.5578957413765223, + 0.6068095967912513, + 0.6473724067064136, + 0.6426171473408151, + 0.6430638075559633, + 0.6497448886438635, + 0.6158516169322898, + 0.629120189917183, + 0.6477310438155289, + 0.6398472570360622, + 0.6327695878108475, + 0.6273133775136017, + 0.6583045007103215, + 0.6521847940416949, + 0.6680598984959822, + 0.6579235625512694, + 0.6542095062890427, + 0.6244643466147747, + 0.6298065345731824, + 0.647901942412604, + 0.6458829668707665, + 0.6339674542063293, + 0.6379541926913062, + 0.6379607253696858, + 0.6393884405379037, + 0.652085190101088, + 0.623270194207214, + 0.6635988905122363, + 0.6296131246663684, + 0.6643739262695131, + 0.6130293769572996, + 0.6393560948787319, + 0.6306603350116586, + 0.654180942484429, + 0.6580154553848266, + 0.6263680622107297, + 0.6114970255199723, + 0.6297789356009589, + 0.6490669582042488, + 0.616832540961528, + 0.6408522408835884, + 0.6491592335629431, + 0.6431044743181307, + 0.6443478904995162, + 0.6190151434604767, + 0.621219927252441, + 0.6353410436160017, + 0.64628912241011, + 0.6193606130224133, + 0.635563048104736, + 0.6416445505746177, + 0.64447757343365, + 0.6489328130464487, + 0.6411390567318894, + 0.6395799138776375, + 0.6421155956464715, + 0.6486984356576625, + 0.6408568356828271, + 0.6402991423353261, + 0.6340382718017614, + 0.6419593413520656, + 0.6480604227741464, + 0.6290497680526881, + 0.6336260267431953, + 0.6496530793610514, + 0.6430170751513317, + 0.6437154140189014, + 0.6357829644851737, + 0.6532328481302859, + 0.6561949947049773, + 0.6369660619842393, + 0.6455345247594543, + 0.6321317446636813, + 0.635466798497668, + 0.6661874632466125, + 0.6551916222733202, + 0.6449905021898248, + 0.6339965023578279, + 0.6308865771091177, + 0.6376595123223295, + 0.645816669502184, + 0.6485441534220243, + 0.6477899363231991, + 0.642762318700232, + 0.6317426821177196, + 0.6360868424795053, + 0.6453040394106726, + 0.6412323434799752, + 0.6469420287686424, + 0.6611352782616534, + 0.6428356813009056, + 0.6442207707481505, + 0.6536621604291628, + 0.6454940182595084, + 0.6244161494558411, + 0.6376272021241116, + 0.6526953928481902, + 0.6349348969080342, + 0.6466526147915932, + 0.6303905600130063, + 0.6209812786561276, + 0.6496316415049319, + 0.6451731086371736, + 0.6377710964378124, + 0.6326771991297113, + 0.6408374703152171, + 0.6569681667391393, + 0.6446118068971255, + 0.6468031793521543, + 0.6444241418220337, + 0.629154017024647, + 0.6315683342844981, + 0.6268400481203793, + 0.6466597975885231, + 0.652735315407763, + 0.6425733083342781, + 0.6471590059264897, + 0.6395793368398883, + 0.62512798802988, + null, + 0.6484392281881511, + 0.628994029629333, + 0.6287799496869414, + 0.6475175418463556, + 0.6395416073327791, + 0.6337837850739225, + 0.643222292898593, + 0.6386431018968428, + 0.6323926585804811, + 0.6322277266333043, + null, + 0.6452598641239778, + 0.6500930426238359, + 0.6159327501399897, + 0.6247784907679585, + 0.6411132081483358, + 0.6314980737403963, + 0.6489489265484508, + 0.6498184088380494, + 0.6260113844236899, + 0.6293568680119298, + 0.6447243798179895, + 0.6321266937374067, + 0.63618154579057, + 0.6414735430617571, + 0.6430595774332734, + 0.6327444301832882, + 0.6386565910680204, + 0.6440444765900797, + 0.6368971801819489, + 0.6362272855154857, + 0.6403663647467956, + 0.6456594176030106, + 0.643325127322863, + 0.6373071534104167, + 0.647254296002267, + 0.6588317856357301, + 0.6520152294792274, + 0.6327138518029866, + 0.6407862369546949, + 0.6464380699880965, + 0.6206361641317457, + 0.6115062311645298, + 0.630022500956309, + 0.6449971409815158, + 0.642364091023281, + 0.6391499778337371, + 0.6405930908764387, + 0.6414939647798263, + 0.6086964415222419, + 0.6352134964837676, + 0.6407128493974845, + 0.6259650614390906, + 0.6400215626323348, + 0.6419033875057603, + 0.6392501725386426, + 0.6148628097185225, + 0.6444288104942781, + 0.6328542749053335, + 0.6093739868677958, + 0.6555387231285073, + 0.6406783160483217, + 0.6535966169108919, + 0.6338431036069906, + 0.6510278752149986, + 0.6379259748413341, + 0.6543389768334708, + 0.6513114045536115, + 0.6461954360369605, + 0.6534235836877115, + 0.6500350360442062, + 0.6194738838344758, + 0.6538473086871875, + 0.6588290979488239, + null, + 0.6550221010683337, + 0.6651836804953154, + 0.6506704614214185, + 0.6577384713242973, + 0.6603981023996841, + 0.6607312945986263, + 0.642140054164437, + null, + null, + 0.6536840687696629, + 0.6605530519590798, + null, + 0.6312990543266889, + 0.611548133091866, + 0.6296885865091467, + 0.6575139896908122, + 0.6581046872205166, + 0.6624433595024228, + 0.652654565504419, + null, + 0.6637953107849444, + 0.6460897835833928, + null, + 0.5925635609682013, + null, + 0.619364336426388, + 0.6467228721491951, + 0.6568513821602701, + 0.6605609370828385, + null, + 0.6400898023861609, + 0.6374474882061584, + 0.6490295974334692, + null, + 0.645927999527949, + null, + 0.6417700912435679, + null, + 0.6204031685817875, + 0.6421009730066809, + null, + 0.63581116678224, + 0.645873169583204, + null, + null, + 0.6557202878956649, + 0.6572257612907099, + 0.6447008716622652, + 0.651716765665791, + null, + 0.6493352301464829, + null, + 0.6525871434704913, + 0.6733721898098496, + null, + 0.5482594682762607, + 0.6506713793972753, + 0.6527581099389806, + null, + 0.6472843649314882, + 0.6661991476849227, + null, + 0.6647985525107304, + 0.6637443601478038, + 0.662985978355305, + null, + null, + null, + null, + 0.6674974689715457, + null, + null, + 0.650965382483269, + 0.6527066170931334, + 0.6614048711783405, + null, + 0.664253255797193, + 0.6617455391366613, + null, + null, + 0.6632584006811011, + 0.6515905434252807, + 0.6538592212827574, + 0.6575595464345282, + 0.6672026369929669, + 0.6599224185645646, + 0.6358661868061308, + 0.6585980739061845, + 0.6478482601347368, + 0.6466232807634796, + 0.6571350310778155, + 0.6614449004024379, + null, + 0.6327219059399402, + 0.6545580016416482, + 0.6504251768526012, + 0.6481209274695631, + 0.6203683397279096, + 0.606063844789649, + 0.6583051464003847, + 0.6462766479159994, + 0.6490665693472422, + 0.652848422169634, + 0.6550942965432206, + 0.6522988601162314, + null, + 0.6448859622282509, + 0.6708644449479578, + 0.6702839384252607, + null, + 0.6551461175574833, + 0.5577872309421724, + null, + 0.6627451430073, + null, + 0.64138736480591, + 0.653743064321591, + 0.6565384720445253, + 0.6328034120404762, + 0.6540272799279528, + 0.6464112794724393, + 0.6361612825861902, + null, + null, + 0.6478629393401057, + 0.628014407138213, + null, + 0.655172216841194, + null, + 0.6562933389251967, + 0.6499791724559447, + 0.6427830410959982, + 0.6457580065540776, + null, + null, + null, + null, + 0.6626867159503617, + null, + 0.6513360493613862, + 0.6581482872916373, + null, + null, + 0.6537554219317643, + 0.66051993609676, + 0.6509891622919032, + null, + 0.6444611363152554, + 0.6469368049833064, + 0.6447333721160438, + null, + 0.6639322657679412, + null, + null, + 0.6541188760834589, + null, + 0.6522189915974065, + null, + null, + 0.6413986296587562, + 0.6513514978628203, + 0.6455589889933765, + 0.6473682758124627, + 0.6382917121718157, + null, + 0.6351720587509616, + 0.6421036456343817, + 0.627879053426429, + 0.5870249564582416, + 0.5847572164195026, + 0.6378358601585478, + 0.6340217823097022, + 0.5512425708386735, + 0.6263404646677887, + 0.620381874408989, + 0.6365191087663246, + 0.6299862491983285, + 0.6287443064924317, + 0.6255616637382992, + 0.6439095703721085, + 0.5960156274627636, + 0.5940212568443094, + 0.6232100569492295, + 0.6322339944668697, + 0.635399636122906, + 0.6332261241465637, + 0.6440585469218848, + 0.6533207347330547, + 0.6067737046477659, + 0.5368989410909222, + 0.6414255619908711, + 0.6348403389690046, + 0.6178151772420452, + 0.6479302682836722, + 0.6119732568635466, + 0.5961317357808065, + 0.6137556153081941, + 0.5813937308019393, + 0.5479346674402639, + 0.6173505734323458, + 0.5284130469111534, + 0.6224735748421695, + 0.5151516387129906, + 0.642879654617352, + 0.6274001503567929, + 0.6319798251119272, + 0.6504520953474009, + 0.5737629889526331, + 0.5833139424808674, + 0.6113033109900138, + 0.6234783983873747, + 0.5557124441080906, + 0.5954694380591127, + 0.572800812467681, + 0.6062368744944029, + 0.5754484720042642, + 0.595791137998882, + 0.5518308671534786, + 0.6150238140127684, + 0.6053451081045736, + 0.5624620981932891, + 0.6089227327071184, + 0.6355909680431194, + 0.6266543015142757, + 0.6211069437705936, + 0.6404605619644146, + 0.6085289471544255, + 0.6487786341504067, + 0.6400461671901225, + 0.6422602823137364, + 0.6259420013648278, + 0.6350893944332945, + 0.6331725980979138, + 0.6565029721235744, + 0.645978867936049, + 0.6014890103263346, + 0.6422759244857548, + 0.629121270968258, + 0.626028393680655, + 0.6310777112767879, + 0.6185879435775763, + 0.6270457643127227, + 0.6285700661751314, + 0.6215506472518639, + 0.6159643240456643, + 0.6313067795195564, + 0.6312186861502217, + 0.6154484637437171, + 0.6198801128855883, + 0.6442364392794191, + 0.6307535385888668, + 0.6287697454498159, + 0.6123068804653735, + 0.620121601396555, + 0.6241437122458925, + 0.6259671096565275, + 0.626164284470416, + 0.6248964909260014, + 0.6352764072241425, + 0.628707915180335, + 0.6325617936631011, + 0.631783100668009, + 0.633003133580014, + 0.6379134135992744, + 0.6282665759140275, + 0.6528894322460312, + 0.6256652744316568, + 0.6494521686449305, + 0.628508723158339, + 0.6303416052138839, + 0.6501553483451155, + 0.6361815772667229, + 0.608683842579277, + 0.624094804328465, + 0.6183618188519728, + 0.6167363065505872, + 0.6281604948260212, + 0.6062128015927818, + 0.6121233074994926, + 0.615709230745756, + 0.622072911163408, + 0.6383271881525543, + 0.6277287838582822, + 0.633730208826638, + 0.6087629054333508, + 0.6286648984587854, + 0.6300485377121875, + 0.6529125568609222, + 0.633426532470006, + 0.6207207210856149, + 0.6499417162629947, + null, + 0.6394492711490287, + 0.5982889543178594, + 0.6302275419868961, + 0.635570641299477, + 0.6416519763610365, + 0.6360027578731564, + 0.6376603204445525, + 0.6561646144478618, + 0.6424427635213876, + 0.647288195861996, + null, + 0.606498598300182, + 0.6460213524133558, + 0.6544419345279726, + 0.6481978788033097, + 0.6469725716719692, + 0.6435012942853434, + 0.6223596664132428, + 0.6476468274826123, + 0.6635660385505906, + 0.6571076927803315, + 0.628389960871777, + 0.6673549835821146, + 0.6556175284123823, + 0.6479784518312722, + 0.6633369672707221, + 0.6355097813582028, + 0.6076517911806204, + 0.5749554482587721, + 0.5791050877469188, + 0.5734420328938861, + 0.6463973279202383, + 0.632931434904136, + 0.6387191532517633, + 0.639532326490651, + 0.641397111886552, + 0.6422529178621743, + 0.6431055764267244, + 0.6359970795203147, + 0.6347391048350529, + 0.6405289519071409, + 0.617706618899482, + 0.6303190846536398, + 0.6303000696390497, + 0.6334859180380892, + 0.656206134092761, + 0.6451787961328329, + 0.6365779251838002, + 0.6162109292432193, + 0.6345481743332125, + 0.6393801702546587, + 0.6466676123267033, + 0.616247631215732, + 0.6030270438686368, + 0.6335833129424678, + 0.6387619734896561, + 0.6036065146875856, + 0.6341262936604515, + 0.6370347042418462, + 0.6454983628607955, + 0.6362992058735174, + 0.6067946308535734, + 0.6466499496572617, + 0.6261719739301546, + 0.6412901719829022, + 0.6453687699138979, + 0.6092415597224139, + 0.6326099024942761, + 0.6254932131206495, + 0.6241497144607295, + 0.6391634830526071, + 0.6204918456350661, + 0.6260514157808026, + 0.630364454438091, + null, + 0.599712897705185, + 0.6547971395869131, + 0.6008948945388529, + 0.656840389273931, + 0.6341918123627974, + 0.628108038027685, + 0.6036406711117693, + null, + null, + 0.6284166795070153, + 0.6495839031376638, + null, + 0.6097363342869585, + 0.5767943184089032, + 0.6027748952981536, + 0.6144230299607988, + 0.6260803782227433, + 0.6256236479845262, + 0.6200110267079861, + null, + 0.6221412807477803, + 0.6304963543400296, + null, + 0.5784596165889963, + null, + 0.6012214262212502, + 0.6196512799474682, + 0.5993464245948142, + 0.6188681617114994, + null, + 0.5976962438951351, + 0.6016530671592522, + 0.6340446373417945, + null, + 0.5742911051661005, + null, + 0.6153445748340867, + null, + 0.6076494710177002, + 0.6135107782090108, + null, + 0.5967365786265443, + 0.614009883839155, + null, + null, + 0.6035850853457789, + 0.6493667359537014, + 0.6005557964320316, + 0.6039057592433209, + null, + 0.6143708287669922, + null, + 0.6132693443105613, + 0.6334753989036728, + null, + 0.49297540703222537, + 0.6178730177646519, + 0.5939353401176963, + null, + 0.6291275012446653, + 0.6402447614784937, + null, + 0.6194211594695958, + 0.5991382113957412, + 0.6204225158914246, + null, + null, + null, + null, + 0.6324477527722054, + null, + null, + 0.6420140895061333, + 0.6341394522318372, + 0.5971404968796962, + null, + 0.6148445522578013, + 0.6154523477326242, + null, + null, + 0.6357401724354705, + 0.6248266000413817, + 0.6350015762253775, + 0.6279212809906826, + 0.6476640070091333, + 0.6551559834307112, + 0.5955990373257389, + 0.6372852605346154, + 0.600065429272146, + 0.6332896721636732, + 0.6301324713337105, + 0.6277710261026951, + null, + 0.6134293123287295, + 0.6078776563024472, + 0.6195062601660181, + 0.6226558339430818, + 0.6165443533220173, + 0.617552491098009, + 0.6265081294324146, + 0.6003068059598348, + 0.6026277456104563, + 0.6335235744505707, + 0.6297150437170774, + 0.6235521586606042, + null, + 0.6215995613732207, + 0.6207560246416957, + 0.6322433139785846, + null, + 0.6265917548588074, + 0.5570433543389289, + null, + 0.6438888923125459, + null, + 0.5949806928424651, + 0.6370121501318932, + 0.6277193277044938, + 0.5997164927876492, + 0.6255569189626659, + 0.6114836485393027, + 0.6081674690478835, + null, + null, + 0.6127193125671908, + 0.620105706260431, + null, + 0.6153375885112867, + null, + 0.6337867695383513, + 0.6107635419347485, + 0.6355679864373665, + 0.617380488424883, + null, + null, + null, + null, + 0.6574091373322206, + null, + 0.6244435750011632, + 0.6346523974674346, + null, + null, + 0.647357700515133, + 0.6171869769998196, + 0.6148928692741403, + null, + 0.6350832732763435, + 0.6412700875264558, + 0.601800535196748, + null, + 0.6403497313305749, + null, + null, + 0.6515509666552718, + null, + 0.6038405816006077, + null, + null, + 0.6091771334498337, + 0.6038290576701001, + 0.6077939395267746, + 0.6068343606962787, + 0.6125035045978043, + null, + 0.6017564002123306 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183, + 0.3629294073235183 + ] + }, + "junctions": { + "mean_pearson": 0.39117726448912443, + "valid_tracks": 626, + "per_track_pearson": [ + 0.4437559496412534, + 0.4015339814256498, + 0.39911127204297425, + 0.3748546869580531, + 0.48915112750362133, + 0.5060965720948611, + 0.09908315622045462, + 0.3412707623611853, + 0.3801993682763689, + 0.36839919320968906, + 0.47554910250198135, + 0.4030210926780169, + 0.3304769352952609, + 0.4149042031978289, + 0.3859460549384045, + 0.3918576023962365, + 0.3632712353580988, + 0.3225620938918891, + 0.4096219751248398, + 0.456912600632001, + 0.36213104135978735, + 0.40896496577718144, + 0.292137677827478, + 0.40487443321422456, + 0.42531969835104166, + 0.3385512600931679, + 0.4221720688347053, + 0.3526914956167478, + 0.3660846146949838, + 0.4265380392769948, + 0.4943321873393342, + 0.4577031193732875, + 0.3915353200513305, + 0.42777294149522355, + 0.44067672648629713, + 0.37837210075546474, + 0.394913062432971, + 0.39324297094649463, + 0.4811678475303704, + 0.3635348334403937, + 0.5093486119576236, + 0.3840977986104453, + 0.38413429724871845, + 0.4131792370404873, + 0.38014138619960974, + 0.34001590257609166, + 0.30420248111362536, + 0.36666625563259936, + 0.36978381664413046, + 0.3317713143675528, + 0.3676453402120181, + 0.30788808007805285, + 0.35411021975782253, + 0.3948052809048716, + 0.33519484337485905, + 0.4057018868417423, + 0.3052239922653218, + 0.31142990984708846, + 0.3281092191738205, + 0.43712576002385034, + 0.24428363427768116, + 0.2812918027738905, + 0.3206224030087173, + 0.31050468615576443, + 0.41573117138990534, + 0.46517687972325444, + 0.40248863241067195, + 0.397996932302236, + 0.4195997132531739, + 0.35691869234755835, + 0.28584711114665107, + 0.3888069365758893, + 0.364554786664238, + 0.31336054379057865, + 0.41138958462202424, + 0.3937244092564332, + 0.40416342995148946, + 0.3473350865578514, + 0.3163571511368961, + 0.4679190298529381, + 0.4398730426103313, + 0.34494674956429533, + 0.38178195814076826, + 0.48559130654848665, + 0.48941624786161225, + 0.28603397923312046, + 0.3877341170630064, + 0.3726636645895429, + 0.35335625116555597, + 0.3679727548341711, + 0.3965516377352916, + 0.45318291933128274, + 0.33667004484930346, + 0.16486805693467385, + 0.4087166815047189, + 0.30287090190669425, + 0.4137869248247093, + 0.4423026719811673, + 0.41631553257822956, + 0.40629968538801925, + 0.3694650345723613, + 0.5169153031520065, + 0.40612443938965165, + 0.339241831176954, + 0.37633289645320517, + 0.325887717450551, + 0.30244946692557423, + 0.4522566485816034, + 0.3345023815532202, + 0.37009551741336794, + 0.37098610087668177, + 0.43017842543620877, + 0.4452550676582491, + 0.3953390069000345, + 0.4014363270218895, + 0.4310667153788374, + 0.3602138577240989, + 0.4271358031986407, + 0.42559597904182256, + 0.3383474129847989, + 0.31702826488829733, + 0.4100354614661541, + 0.4605090663664345, + 0.3555170390975957, + 0.36254530354790826, + null, + 0.361890096771059, + 0.38518060687407163, + 0.20591124633893548, + 0.2940042581248664, + 0.4170959272062311, + 0.2639072838731923, + 0.36877985167799465, + 0.45403420849991255, + 0.37952830223943973, + 0.4742332451266013, + null, + 0.34439217425511376, + 0.4085212691838545, + 0.3743919682500813, + 0.4342067029484234, + 0.4697357614061822, + 0.41012636034894406, + 0.34266428620079903, + 0.38902733738986267, + 0.16471179622183765, + 0.39727593143446416, + 0.3788165366972781, + 0.2774054670793739, + 0.3207299595658127, + 0.4477496170818458, + 0.43767303388785683, + 0.3030900889923224, + 0.43282244940034875, + 0.5166676548588778, + 0.47195301976356735, + 0.3686615907935693, + 0.274770654333142, + 0.3820749998328969, + 0.21685437738003682, + 0.466923194277449, + 0.3470998043620534, + 0.4302525675791163, + 0.37727386731347096, + 0.37365586989254757, + 0.41084247575891913, + 0.48640679040710255, + 0.2752675083648222, + 0.3048209757086531, + 0.33541968110337816, + 0.41050642659934306, + 0.3659096875918258, + 0.4951343465591845, + 0.2943056378761408, + 0.3535379227649697, + 0.4302026794447786, + 0.23647476979268517, + 0.2899322415278244, + 0.269684760291122, + 0.40673549872228726, + 0.337554954575174, + 0.32596033859534124, + 0.41163009661691624, + 0.4721399519704558, + 0.3347733943427186, + 0.37383265274583044, + 0.4465658309101197, + 0.5130920588267198, + 0.49790991141074786, + 0.4309107588185718, + 0.42639243618615025, + 0.34157882289925556, + 0.3749073200399563, + 0.4455861359327394, + 0.5127041671064794, + 0.4510993176157823, + 0.3688364280769933, + 0.3295244523523375, + 0.35476310714360815, + 0.34369197144193486, + null, + 0.3598088351302803, + 0.5094880516058897, + 0.48064080612300597, + 0.4555489879877688, + 0.36046954605653947, + 0.39809533016740134, + 0.3917158046204845, + null, + null, + 0.34061513145060324, + 0.510255757267679, + null, + 0.34201699470265895, + 0.465684973194765, + 0.32252033645580375, + 0.3174834637411525, + 0.4031834963266631, + 0.4307405841607514, + 0.25221406946174857, + null, + 0.4290352671991202, + 0.5079845694849621, + null, + 0.369840960002537, + null, + 0.4157186995159219, + 0.3567173208749635, + 0.26493323453278345, + 0.3720839368837386, + null, + 0.23988092144281, + 0.31036193564821346, + 0.4099686519193185, + null, + 0.37693461890355595, + null, + 0.44586010081397504, + null, + 0.2197976134243885, + 0.42102123076954556, + null, + 0.3645563899407145, + 0.4061074029659764, + null, + null, + 0.33356939751774456, + 0.4253898809757637, + 0.26271181643101293, + 0.3321275338842759, + null, + 0.5026647811855669, + null, + 0.4299800173592416, + 0.36579056369439866, + null, + 0.4295631860837408, + 0.4409270023239098, + 0.49474410380164224, + null, + 0.4332104388667933, + 0.35525221500857884, + null, + 0.5260524765978482, + 0.5274929775117617, + 0.4761031157074411, + null, + null, + null, + null, + 0.3614686117971314, + null, + null, + 0.30865240407700434, + 0.2875307662933858, + 0.5083003024944471, + null, + 0.41312616954463555, + 0.4206687597918552, + null, + null, + 0.4659776873410861, + 0.3680135194914366, + 0.38002346534968606, + 0.3681637778391016, + 0.4657462944962455, + 0.45560751105688363, + 0.35951537473206846, + 0.4000787320371034, + 0.3889495005752987, + 0.2901297481372651, + 0.4129218301929332, + 0.5621907632213464, + null, + 0.3402594007975599, + 0.5359813437770816, + 0.2634567782925063, + 0.34380861020897996, + 0.47110385188607395, + 0.37804090583402644, + 0.3462842587979104, + 0.31757007782506946, + 0.420275174652117, + 0.345548155579318, + 0.46292665668169725, + 0.43387393262243384, + null, + 0.4074287518347942, + 0.33079553273125506, + 0.33322768964387156, + null, + 0.29457019853262584, + 0.3365000405464247, + null, + 0.5333883963861756, + null, + 0.28902788712970956, + 0.44446677208971386, + 0.37830925979387964, + 0.33159955222788984, + 0.36443373280145064, + 0.3741413320287871, + 0.3463022200305168, + null, + null, + 0.38929468235770737, + 0.3673177366126545, + null, + 0.44622244331817795, + null, + 0.4346778928079972, + 0.38346688902613296, + 0.32675850639920123, + 0.4381167196978658, + null, + null, + null, + null, + 0.307655685545505, + null, + 0.37923372649544407, + 0.40387872383865975, + null, + null, + 0.4890883751770534, + 0.2841055260839946, + 0.29724161613354405, + null, + 0.5743017803352057, + 0.4080710831925524, + 0.4093293870046907, + null, + 0.45066027032748535, + null, + null, + 0.42773210722892907, + null, + 0.2907896484976109, + null, + null, + 0.3457220735533087, + 0.39092580700465107, + 0.3772575303369059, + 0.39743412029499103, + 0.37362765502250617, + null, + 0.3622003543396794, + 0.40411255989549405, + 0.34178828225405994, + 0.6255665506128398, + 0.4291539797005459, + 0.5166879161135356, + 0.38766942058692977, + 0.39077947515678446, + 0.28154022497698405, + 0.38120908237114376, + 0.4307980467628928, + 0.29100178304107643, + 0.4189270228001471, + 0.39696657032913935, + 0.30531460867342874, + 0.3132195283467738, + 0.3266432767372959, + 0.3468018379718477, + 0.5148922801615953, + 0.4992841342251494, + 0.45607495209179305, + 0.4547363249381875, + 0.4676328199234261, + 0.35315378153389193, + 0.5647980771067197, + 0.2679408492961876, + 0.34440507223243183, + 0.476700876109693, + 0.44980787901476427, + 0.39116242225375686, + 0.269108339257314, + 0.35300962888863535, + 0.44242104598626375, + 0.4505566196448623, + 0.5344150700758616, + 0.2805456108201789, + 0.40759425473181143, + 0.21960072589190338, + 0.3955244962360293, + 0.5532642872829681, + 0.41862161512163776, + 0.40235919359643624, + 0.28106469952699076, + 0.2737921017145485, + 0.2862832421869472, + 0.3758588025716551, + 0.2872821405124225, + 0.27410318703304093, + 0.45671866342142853, + 0.23320445248702518, + 0.3949787004651145, + 0.21777276455215172, + 0.35446298407236354, + 0.29500945462365064, + 0.3209898910680427, + 0.2690542950258337, + 0.3584619589308829, + 0.4079584611062344, + 0.34352482426475295, + 0.3265594802115594, + 0.31255233253233594, + 0.3935125863947894, + 0.2944158021669289, + 0.4114332657052813, + 0.3476119711851563, + 0.343616602923998, + 0.2864615678092904, + 0.40983632173516704, + 0.4027639614013271, + 0.29234138499614154, + 0.3109909959142144, + 0.37489246340186, + 0.34118534918150223, + 0.3814121812097932, + 0.4534050686928232, + 0.39910603420246343, + 0.45757766336845496, + 0.4059297229106111, + 0.5054767750726433, + 0.4046139927899586, + 0.4025405920092609, + 0.36725032088954335, + 0.4155857168575704, + 0.42397679713666153, + 0.33241269739945933, + 0.3916954416153057, + 0.36918255541794953, + 0.47287373312460584, + 0.44113537570552647, + 0.3596733668630001, + 0.38947398747569834, + 0.44150781875887724, + 0.39504362182748687, + 0.37992073198818876, + 0.3978194797048984, + 0.3685617812675933, + 0.3325383614759932, + 0.41462778023499924, + 0.41570050301446526, + 0.4625692512951213, + 0.47786952187502063, + 0.2282764284306363, + 0.384449764597322, + 0.36594290944713903, + 0.3640709088309385, + 0.4276648184567552, + 0.3810132421770382, + 0.26748518774179736, + 0.4369849118192715, + 0.43173361346245875, + 0.43591014833751496, + 0.3095806289720787, + 0.4606414847755498, + 0.3417225436558158, + 0.3876516574044314, + 0.4278339362423836, + 0.36167451970210807, + 0.4741717649495822, + 0.22848614868735398, + 0.32293036712548406, + 0.31713068989716, + 0.5193443900605218, + 0.345435804829175, + 0.38452626674955254, + 0.35022748350065747, + 0.48875921169277564, + null, + 0.4007196453101875, + 0.4939514508721663, + 0.33295315256120295, + 0.48894943046119976, + 0.4036878032496111, + 0.30134602588131876, + 0.3403111344288697, + 0.4227843387777727, + 0.3792952539380554, + 0.34274855561335515, + null, + 0.40433772374184096, + 0.3316434868469045, + 0.2931902760440231, + 0.3152802217688068, + 0.38569225833278753, + 0.3044978018629046, + 0.31439396938919495, + 0.43740352097809904, + 0.2575121181496892, + 0.2859167633397801, + 0.3617088217166998, + 0.3554226337059609, + 0.34592452535894436, + 0.406996037288884, + 0.5144879979546051, + 0.3704796720776341, + 0.37089944324981966, + 0.5496505074887422, + 0.49607567176112644, + 0.3644620041339581, + 0.3799219312160398, + 0.38197577834772267, + 0.3790634729711227, + 0.38734666936251994, + 0.35298667452415877, + 0.5203483188049787, + 0.5371228940109355, + 0.3985645525586167, + 0.36650144275100144, + 0.46857656237046624, + 0.41499100098215835, + 0.3746577698227425, + 0.45165087082640326, + 0.32060023109668806, + 0.348510113664279, + 0.3503578784836523, + 0.26877261865501884, + 0.3490679757602009, + 0.40956682770860636, + 0.40618149926517266, + 0.3633563082962763, + 0.35265341149608687, + 0.39697838551933257, + 0.3131794744957534, + 0.31047994694427855, + 0.5133383938642524, + 0.5108912193001135, + 0.4211610415072188, + 0.5589047211472611, + 0.5446061512994534, + 0.2683886870631891, + 0.4768966735576897, + 0.36604720039767635, + 0.5659816152341377, + 0.4016162716074907, + 0.3358288469992484, + 0.30585807867287895, + 0.36380018052186436, + 0.3238317145295804, + 0.34305410198032726, + 0.44948571625424777, + 0.42748307118118706, + 0.5611333186668549, + null, + 0.39614815709454965, + 0.40004275052940613, + 0.5499960828321369, + 0.5260467867829665, + 0.4531483755884356, + 0.25738588575047133, + 0.3928560621803439, + null, + null, + 0.4590253682012496, + 0.5263791056319871, + null, + 0.22574743630396374, + 0.4589509154180955, + 0.1862775006287814, + 0.3514140043243893, + 0.35756244217602196, + 0.3629418127486805, + 0.366621760314832, + null, + 0.47055503859194614, + 0.36033157037917796, + null, + 0.43210525359856744, + null, + 0.5503388679282339, + 0.41305536407695964, + 0.2815058255339444, + 0.375116546059763, + null, + 0.4285051742418369, + 0.36528474063337535, + 0.47173819759048136, + null, + 0.4361498631385562, + null, + 0.36010940704553596, + null, + 0.39161661066339715, + 0.4449375333998955, + null, + 0.40511563451256327, + 0.45650288885331963, + null, + null, + 0.4442499031904852, + 0.414850301147699, + 0.22645005351606698, + 0.37521017212802976, + null, + 0.4822834849617776, + null, + 0.31583335424910564, + 0.4746011528764561, + null, + 0.21016337598406987, + 0.510759628770388, + 0.4399903838789547, + null, + 0.3886187111506182, + 0.5256333224882792, + null, + 0.48267046586797463, + 0.5090001356981245, + 0.4929264585091801, + null, + null, + null, + null, + 0.4706061767110396, + null, + null, + 0.4787876932743838, + 0.46747819870835117, + 0.5055836053868511, + null, + 0.34909049097547035, + 0.531990307405772, + null, + null, + 0.3545479418314884, + 0.14410365602422537, + 0.4139953930351641, + 0.2697318214533862, + 0.32991211368702855, + 0.45874149887696375, + 0.18372807884268236, + 0.35833263731230874, + 0.226951875773368, + 0.3816971001228711, + 0.43333756992000927, + 0.3652057721171139, + null, + 0.3268861021387281, + 0.5202383563602815, + 0.5069917175620603, + 0.48543879712691945, + 0.22770593092585775, + 0.4614659501368044, + 0.43307213890045293, + 0.4150184615383423, + 0.4554858551776062, + 0.4145924902808705, + 0.4726995347786753, + 0.4111466000795599, + null, + 0.4109993278526277, + 0.47699177738861287, + 0.42505637400420343, + null, + 0.5268079912434742, + 0.390119203745289, + null, + 0.4217083125088822, + null, + 0.5271119206409144, + 0.49675928640344147, + 0.33987183388933767, + 0.47793921204919115, + 0.47764746851116746, + 0.5137827303352475, + 0.5535038858403653, + null, + null, + 0.47004909409409723, + 0.5680809770622063, + null, + 0.4175752915127892, + null, + 0.4201032950589581, + 0.45894185144519223, + 0.48493571161093413, + 0.4503475992266625, + null, + null, + null, + null, + 0.3819717780893279, + null, + 0.3950502183417029, + 0.574779865463776, + null, + null, + 0.486418838885885, + 0.25334441485254894, + 0.3767034622711646, + null, + 0.3885417177504702, + 0.40450991452098334, + 0.38530537159399325, + null, + 0.3531787817251646, + null, + null, + 0.40903557450179656, + null, + 0.30216953991178913, + null, + null, + 0.42114483542487613, + 0.5195116921264826, + 0.5709089475051822, + 0.5891180151967241, + 0.4713410583740648, + null, + 0.37279657023068313 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.7549668874172185, + 0.7927631578947368, + 0.7437722419928826, + 0.740506329113924, + 0.8013245033112583, + 0.752542372881356, + 0.75, + 0.75, + 0.7884615384615384, + 0.7411347517730497, + 0.7474747474747475, + 0.7288135593220338, + 0.7532467532467533, + 0.778125, + 0.7833333333333333, + 0.7598684210526315, + 0.7412587412587412, + 0.7435897435897436, + 0.7707006369426752, + 0.7581699346405228, + 0.7302158273381295, + 0.7548387096774194, + 0.724025974025974, + 0.758364312267658, + 0.7330960854092526, + 0.7927631578947368, + 0.8173076923076923, + 0.7966101694915254, + 0.777027027027027, + 0.7655172413793103, + 0.7534246575342466, + 0.7605177993527508, + 0.75, + 0.78419452887538, + 0.7509157509157509, + 0.7287066246056783, + 0.7175572519083969, + 0.7523510971786834, + 0.6888111888111889, + 0.7583892617449665, + 0.764505119453925, + 0.7800687285223368, + 0.7442748091603053, + 0.762962962962963, + 0.7540983606557377, + 0.75, + 0.7638376383763837, + 0.7455197132616488, + 0.7375886524822695, + 0.7585034013605442, + 0.7481751824817519, + 0.7683823529411765, + 0.7196969696969697, + 0.7687074829931972, + 0.7313432835820896, + 0.7278911564625851, + 0.7335640138408305, + 0.7814569536423841, + 0.7431506849315068, + 0.7394366197183099, + 0.7275862068965517, + 0.7868852459016393, + 0.7649006622516556, + 0.757679180887372, + 0.7516778523489933, + 0.7879746835443038, + 0.7331081081081081, + 0.7550335570469798, + 0.7440273037542662, + 0.7777777777777778, + 0.7508896797153025, + 0.7516129032258064, + 0.7744107744107744, + 0.7913907284768212, + 0.7562724014336918, + 0.8033898305084746, + 0.7845659163987139, + 0.7725752508361204, + 0.7655172413793103, + 0.7993197278911565, + 0.7781456953642384, + 0.7848101265822784, + 0.7773972602739726, + 0.7669902912621359, + 0.7649122807017544, + 0.7616099071207431, + 0.773972602739726, + 0.7745098039215687, + 0.7821428571428571, + 0.7848101265822784, + 0.7700348432055749, + 0.7835051546391752, + 0.7684210526315789, + 0.7585034013605442, + 0.7601351351351351, + 0.7361111111111112, + 0.7830508474576271, + 0.7231833910034602, + 0.7814569536423841, + 0.7848101265822784, + 0.7366666666666667, + 0.7770491803278688, + 0.7580071174377224, + 0.792332268370607, + 0.7849829351535836, + 0.79, + 0.7303370786516854, + 0.7792642140468228, + 0.7674418604651163, + 0.7331081081081081, + 0.7464285714285714, + 0.7684563758389261, + 0.7564935064935064, + 0.7457627118644068, + 0.7653061224489796, + 0.7702702702702703, + 0.7379310344827587, + 0.7466216216216216, + 0.7436823104693141, + 0.78839590443686, + 0.75, + 0.7830508474576271, + 0.7279411764705882, + 0.7828947368421053, + 0.7327935222672065, + 0.9758064516129032, + 0.7958477508650519, + 0.7491289198606271, + 0.7651515151515151, + 0.7686567164179104, + 0.7427536231884058, + 0.7931034482758621, + 0.7491289198606271, + 0.7443609022556391, + 0.7649122807017544, + 0.725752508361204, + 0.967741935483871, + 0.7694915254237288, + 0.7350746268656716, + 0.7763157894736842, + 0.791970802919708, + 0.7350746268656716, + 0.7426470588235294, + 0.7413793103448276, + 0.743859649122807, + 0.7290076335877863, + 0.7125984251968503, + 0.7256944444444444, + 0.7058823529411765, + 0.7455830388692579, + 0.7148148148148148, + 0.75177304964539, + 0.7326007326007326, + 0.7694805194805194, + 0.7620689655172413, + 0.738831615120275, + 0.7121212121212122, + 0.779783393501805, + 0.7491166077738516, + 0.7554744525547445, + 0.7121212121212122, + 0.7465277777777778, + 0.780564263322884, + 0.7282229965156795, + 0.7358490566037735, + 0.762589928057554, + 0.7577854671280276, + 0.804635761589404, + 0.720754716981132, + 0.7403508771929824, + 0.7472527472527473, + 0.7064846416382252, + 0.7819548872180451, + 0.7773972602739726, + 0.75, + 0.7665369649805448, + 0.7391304347826086, + 0.7526132404181185, + 0.7132352941176471, + 0.7552447552447552, + 0.7317073170731707, + 0.7388059701492538, + 0.7269372693726938, + 0.7204301075268817, + 0.7887323943661971, + 0.8081180811808119, + 0.7363013698630136, + 0.7474747474747475, + 0.7383512544802867, + 0.8032786885245902, + 0.7256944444444444, + 0.7977941176470589, + 0.7535714285714286, + 0.6936619718309859, + 0.7247386759581882, + 0.6784452296819788, + 0.6838235294117647, + 0.7465277777777778, + 0.7745098039215687, + 0.77, + 0.9689922480620154, + 0.7763157894736842, + 0.7457627118644068, + 0.6943521594684385, + 0.75, + 0.7813504823151125, + 0.7455830388692579, + 0.7138157894736842, + 0.9915966386554622, + 0.9772727272727273, + 0.7920792079207921, + 0.775974025974026, + 0.9769230769230769, + 0.7413793103448276, + 0.6446886446886447, + 0.7153284671532847, + 0.7337883959044369, + 0.7384105960264901, + 0.7344827586206897, + 0.7483221476510067, + 0.983739837398374, + 0.7453416149068323, + 0.7525773195876289, + 0.9774436090225563, + 0.67578125, + 0.9929577464788732, + 0.6909090909090909, + 0.743859649122807, + 0.7464788732394366, + 0.7306397306397306, + 0.9710144927536232, + 0.717391304347826, + 0.704225352112676, + 0.7298245614035088, + 0.9763779527559056, + 0.7037037037037037, + 0.9692307692307692, + 0.7435064935064936, + 1.0, + 0.7433333333333333, + 0.7449664429530202, + 0.9708029197080292, + 0.7211895910780669, + 0.7508196721311475, + 0.9689922480620154, + 0.9765625, + 0.7624113475177305, + 0.7745098039215687, + 0.7549019607843137, + 0.7205387205387206, + 0.992, + 0.7385159010600707, + 0.9767441860465116, + 0.737012987012987, + 0.7285223367697594, + 0.9861111111111112, + 0.7038461538461539, + 0.7266881028938906, + 0.7090909090909091, + 0.9855072463768116, + 0.7142857142857143, + 0.7558528428093646, + 0.9917355371900827, + 0.760655737704918, + 0.7052631578947368, + 0.7627118644067796, + 0.9626865671641791, + 0.9922480620155039, + 0.9689922480620154, + 0.9833333333333333, + 0.7574750830564784, + 0.950354609929078, + 0.968503937007874, + 0.7551020408163265, + 0.7163120567375887, + 0.7339743589743589, + 0.9833333333333333, + 0.7201365187713311, + 0.7534246575342466, + 0.992, + 0.9846153846153847, + 0.7630662020905923, + 0.717391304347826, + 0.7391304347826086, + 0.7297297297297297, + 0.7474402730375427, + 0.7142857142857143, + 0.7303754266211604, + 0.7301038062283737, + 0.6949152542372882, + 0.7259786476868327, + 0.7547770700636943, + 0.7380073800738007, + 0.9925373134328358, + 0.7614035087719299, + 0.7331081081081081, + 0.7142857142857143, + 0.7575757575757576, + 0.7687074829931972, + 0.7361111111111112, + 0.7192982456140351, + 0.7275862068965517, + 0.7475409836065574, + 0.7398648648648649, + 0.717391304347826, + 0.7167235494880546, + 0.991869918699187, + 0.7491961414790996, + 0.7668918918918919, + 0.7788461538461539, + 0.9906542056074766, + 0.7412140575079872, + 0.7204301075268817, + 0.9752066115702479, + 0.7457044673539519, + 0.9925373134328358, + 0.7724358974358975, + 0.7430555555555556, + 0.7623762376237624, + 0.7137681159420289, + 0.7619047619047619, + 0.7157534246575342, + 0.7167832167832168, + 0.9858156028368794, + 1.0, + 0.7222222222222222, + 0.7210884353741497, + 0.9850746268656716, + 0.7364620938628159, + 0.9692307692307692, + 0.7375415282392026, + 0.7281879194630873, + 0.6909090909090909, + 0.7441860465116279, + 0.9714285714285714, + 0.984375, + 0.976, + 0.9761904761904762, + 0.7191780821917808, + 0.9781021897810219, + 0.726643598615917, + 0.7904761904761904, + 0.9739130434782609, + 0.9929078014184397, + 0.7288732394366197, + 0.7308970099667774, + 0.7185185185185186, + 0.976, + 0.75, + 0.7407407407407407, + 0.7092198581560284, + 0.9919354838709677, + 0.8121019108280255, + 0.9850746268656716, + 0.9767441860465116, + 0.7406143344709898, + 0.9836065573770492, + 0.7119741100323624, + 0.968503937007874, + 0.9767441860465116, + 0.694980694980695, + 0.7532467532467533, + 0.7525773195876289, + 0.7457044673539519, + 0.7147887323943662, + 1.0, + 0.7417218543046358, + 0.7180451127819549, + 0.7473684210526316, + 0.7644927536231884, + 0.7642857142857142, + 0.8132911392405063, + 0.7153024911032029, + 0.710801393728223, + 0.6566037735849056, + 0.7542662116040956, + 0.737410071942446, + 0.7439446366782007, + 0.7184115523465704, + 0.7431506849315068, + 0.75, + 0.724907063197026, + 0.6923076923076923, + 0.7007042253521126, + 0.6888888888888889, + 0.7392857142857143, + 0.7406143344709898, + 0.7097902097902098, + 0.7314487632508834, + 0.7759197324414716, + 0.7593984962406015, + 0.6964285714285714, + 0.7563636363636363, + 0.7450331125827815, + 0.7735191637630662, + 0.7526881720430108, + 0.6837944664031621, + 0.7649122807017544, + 0.7573529411764706, + 0.7217741935483871, + 0.74822695035461, + 0.7326007326007326, + 0.7872340425531915, + 0.7211155378486056, + 0.7021276595744681, + 0.7202797202797203, + 0.7302158273381295, + 0.6830985915492958, + 0.791970802919708, + 0.7561837455830389, + 0.7019607843137254, + 0.7457044673539519, + 0.7482014388489209, + 0.7368421052631579, + 0.7366412213740458, + 0.7320754716981132, + 0.7230769230769231, + 0.7349397590361446, + 0.7509293680297398, + 0.7045454545454546, + 0.7615658362989324, + 0.7255639097744361, + 0.7472118959107806, + 0.7769784172661871, + 0.7526881720430108, + 0.7153024911032029, + 0.7065217391304348, + 0.759581881533101, + 0.7677902621722846, + 0.7747440273037542, + 0.7362637362637363, + 0.7472924187725631, + 0.75, + 0.7321428571428571, + 0.6602316602316602, + 0.7224334600760456, + 0.7060931899641577, + 0.7481751824817519, + 0.6992753623188406, + 0.7796610169491526, + 0.7204301075268817, + 0.7311827956989247, + 0.7318840579710145, + 0.7241379310344828, + 0.7385159010600707, + 0.7338129496402878, + 0.7471698113207547, + 0.7577854671280276, + 0.7363013698630136, + 0.7186311787072244, + 0.7148148148148148, + 0.7588652482269503, + 0.7389705882352942, + 0.7480916030534351, + 0.7638888888888888, + 0.6970802919708029, + 0.7326007326007326, + 0.7456445993031359, + 0.7330960854092526, + 0.7344827586206897, + 0.7323420074349443, + 0.6947368421052632, + 0.7204301075268817, + 0.7335640138408305, + 0.7100371747211895, + 0.7294520547945206, + 0.7204301075268817, + 0.6934865900383141, + 0.718978102189781, + 0.7536231884057971, + 0.7464285714285714, + 0.75, + 0.7395833333333334, + 0.720754716981132, + 0.762962962962963, + 0.7228070175438597, + 0.7330960854092526, + 0.7147887323943662, + 0.7191011235955056, + 0.7549668874172185, + 0.7383512544802867, + 0.7455197132616488, + 0.7115384615384616, + 0.7243816254416962, + 0.6884615384615385, + 0.7035573122529645, + 0.6937269372693727, + 0.7298245614035088, + 0.7096774193548387, + 0.7032967032967034, + 0.7526501766784452, + 0.6728624535315985, + 0.9916666666666667, + 0.75177304964539, + 0.7392857142857143, + 0.7624113475177305, + 0.7312252964426877, + 0.7545126353790613, + 0.7622377622377622, + 0.7385159010600707, + 0.706766917293233, + 0.7630662020905923, + 0.6908396946564885, + 0.9813084112149533, + 0.7665505226480837, + 0.6727941176470589, + 0.8006993006993007, + 0.818815331010453, + 0.7137546468401487, + 0.7163120567375887, + 0.7027972027972028, + 0.7021276595744681, + 0.6802973977695167, + 0.7014925373134329, + 0.6891385767790262, + 0.6787003610108303, + 0.7032967032967034, + 0.6461538461538462, + 0.6531365313653137, + 0.7347670250896058, + 0.7343173431734318, + 0.7433962264150943, + 0.7620689655172413, + 0.7472527472527473, + 0.7781954887218046, + 0.7297297297297297, + 0.737037037037037, + 0.6992753623188406, + 0.6996336996336996, + 0.7993527508090615, + 0.7167832167832168, + 0.7078651685393258, + 0.7086614173228346, + 0.7226027397260274, + 0.7670250896057348, + 0.6448979591836734, + 0.6951672862453532, + 0.7427536231884058, + 0.6888888888888889, + 0.7286821705426356, + 0.7491166077738516, + 0.7306273062730627, + 0.7357723577235772, + 0.7304964539007093, + 0.7392857142857143, + 0.7377622377622378, + 0.7445255474452555, + 0.7247386759581882, + 0.7286245353159851, + 0.6948529411764706, + 0.6959706959706959, + 0.7605633802816901, + 0.7407407407407407, + 0.7063197026022305, + 0.7158273381294964, + 0.6901408450704225, + 0.7571428571428571, + 0.7560137457044673, + 0.7343173431734318, + 0.7526501766784452, + 0.6937269372693727, + 0.6917562724014337, + 0.7158273381294964, + 0.676056338028169, + 0.6961538461538461, + 0.8, + 0.7610921501706485, + 1.0, + 0.7553191489361702, + 0.7575757575757576, + 0.6791044776119403, + 0.7216117216117216, + 0.7842465753424658, + 0.7414965986394558, + 0.7067137809187279, + 0.9841269841269841, + 0.9833333333333333, + 0.8213166144200627, + 0.7882736156351792, + 0.976, + 0.7333333333333333, + 0.6747404844290658, + 0.7038461538461539, + 0.7406143344709898, + 0.735191637630662, + 0.7402135231316725, + 0.7385159010600707, + 0.9765625, + 0.7455197132616488, + 0.7037037037037037, + 0.9923076923076923, + 0.664179104477612, + 1.0, + 0.6779661016949152, + 0.7651245551601423, + 0.7668918918918919, + 0.7157894736842105, + 0.9851851851851852, + 0.7252747252747253, + 0.7192982456140351, + 0.7022058823529411, + 1.0, + 0.7224199288256228, + 0.981651376146789, + 0.6996466431095406, + 0.9761904761904762, + 0.7228464419475655, + 0.7163636363636363, + 1.0, + 0.7054545454545454, + 0.7491039426523297, + 0.9777777777777777, + 0.976, + 0.7551724137931034, + 0.734982332155477, + 0.7828947368421053, + 0.7441077441077442, + 0.9915254237288136, + 0.6926070038910506, + 1.0, + 0.6985815602836879, + 0.7238805970149254, + 0.9603174603174603, + 0.6678966789667896, + 0.7237762237762237, + 0.6951672862453532, + 0.9844961240310077, + 0.7137681159420289, + 0.75, + 0.9747899159663865, + 0.7406143344709898, + 0.7324414715719063, + 0.7455197132616488, + 0.9705882352941176, + 0.9758064516129032, + 1.0, + 0.963963963963964, + 0.7647058823529411, + 0.9724770642201835, + 0.9908256880733946, + 0.7446808510638298, + 0.7585034013605442, + 0.7567567567567568, + 0.9922480620155039, + 0.7517482517482518, + 0.7350993377483444, + 0.9831932773109243, + 0.976, + 0.7577854671280276, + 0.7525773195876289, + 0.7345454545454545, + 0.7623762376237624, + 0.7473684210526316, + 0.7372262773722628, + 0.6764705882352942, + 0.7585034013605442, + 0.690566037735849, + 0.7245283018867924, + 0.7661016949152543, + 0.7163120567375887, + 1.0, + 0.7673611111111112, + 0.7311827956989247, + 0.7551020408163265, + 0.7508771929824561, + 0.7237762237762237, + 0.7202797202797203, + 0.726027397260274, + 0.73992673992674, + 0.7516778523489933, + 0.7385159010600707, + 0.7444444444444445, + 0.7090909090909091, + 0.9763779527559056, + 0.7428571428571429, + 0.7357142857142858, + 0.79, + 0.991869918699187, + 0.754325259515571, + 0.7224080267558528, + 0.9754098360655737, + 0.7377049180327869, + 0.9918032786885246, + 0.7674418604651163, + 0.7508896797153025, + 0.7702702702702703, + 0.7176470588235294, + 0.75, + 0.7014925373134329, + 0.6730038022813688, + 0.9672131147540983, + 0.983739837398374, + 0.7359154929577465, + 0.6917562724014337, + 1.0, + 0.7153024911032029, + 1.0, + 0.7337883959044369, + 0.7231833910034602, + 0.7132867132867133, + 0.7316176470588235, + 0.9708029197080292, + 0.9763779527559056, + 1.0, + 0.9719626168224299, + 0.7183098591549296, + 0.9826086956521739, + 0.7038461538461539, + 0.7878787878787878, + 0.967479674796748, + 0.9846153846153847, + 0.7419354838709677, + 0.7473684210526316, + 0.7178571428571429, + 0.9831932773109243, + 0.7240143369175627, + 0.7421602787456446, + 0.726962457337884, + 1.0, + 0.7922077922077922, + 0.983739837398374, + 1.0, + 0.7418181818181818, + 0.9833333333333333, + 0.7132075471698113, + 0.9724770642201835, + 0.992, + 0.7011070110701108, + 0.7474048442906575, + 0.7730263157894737, + 0.7517482517482518, + 0.7423728813559322, + 0.9915966386554622, + 0.7491525423728813 + ] + }, + "contact_maps": { + "mean_pearson": 0.370823809575012, + "valid_tracks": 28, + "per_track_pearson": [ + 0.36444504387438365, + 0.3943547228133399, + 0.4144006751027216, + 0.3537238639446819, + 0.2810657075695973, + 0.3245713195496299, + 0.3790729065539176, + 0.28491302088305226, + 0.4457316346131339, + 0.4166898696553033, + 0.3366161135267191, + 0.40389570390595614, + 0.4016770435247096, + 0.3232589238095367, + 0.3638756867696226, + 0.4629574788030322, + 0.10398242664144128, + 0.3982823081846042, + 0.39549435643210035, + 0.3982678940543707, + 0.3458295567555356, + 0.3498790323625699, + 0.3476583238226044, + 0.41291210203730805, + 0.41927909937644003, + 0.4255553336907286, + 0.39444800590446855, + 0.44022851393882356 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.5103521119052553, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..b6057fc80f8a570287623eb6410f328035d52638 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S033/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.509114534999072, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.579043329043329, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": -0.021271425225878243, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7607243206750722, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.6115744563161862, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.6469271677911642, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.2667347914432191, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.45205403927784293, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.15230859964725899, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2086600517867549, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.46168311385204225, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.1954016677296886, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.22317998495733932, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-CPU-construction-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-CPU-construction-check.json new file mode 100644 index 0000000000000000000000000000000000000000..a9d30ed4d23cd2ca9bb0d4e0953bcc1531f4553b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-CPU-construction-check.json @@ -0,0 +1,35 @@ +{ + "utc": "2026-10-26T07:37:39.350375+00:00", + "parameters": 215626632, + "actual_large8_weights_sha256": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "candidate_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "synthetic_input_bp": 8192, + "synthetic_target_bp": 4096, + "base_model_batch": 4, + "actual_allele_and_N_edits": true, + "forward_and_reverse_inputs": true, + "all11_base_outputs_finite": true, + "actual_native_track_arrays_exact_manual_fixed_mean": 15, + "actual_native_coordinate_and_strand_roundtrip_exact": true, + "paired_dense_and_splice_arrays_exact_manual_fixed_mean": 8, + "all_learned_tensors_and_CPU_RNG_unchanged": true, + "public_junction_metadata_sha256": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4", + "exact_RC108_wrapper_and_permutation_assets": true, + "CUDA_used": false, + "training_updates": 0, + "labels_read": false, + "caveat": "CPU short-input compatibility of the actual large model and unchanged RC operation. It does not establish full1Mb GPU memory/runtime or accuracy; all candidate native/full/runtime gates remain required." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-paired-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-paired-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..1b97d4216f63aec62c1362157337659bfb497c11 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-paired-plan.json @@ -0,0 +1,142 @@ +{ + "created_utc": "2026-10-26T07:40:05.079170+00:00", + "dependencies": [ + "/workspace/research/experiments/rc_large114/evaluation.json", + "/workspace/research/studies/rc_large114/parent-comparison.json", + "/workspace/research/studies/rc_large114/runtime-gate.json" + ], + "source_sha256": { + "run_planned_diagnostics.py": "9c0fc164753231538d83d502a505958f05874c916ca6060c00e1c471a316013c", + "diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "diagnose_variant_uncertainty_weighted_io.py": "575a3b22b61b23965032dca0e07e56c21b50b41dcac2322efea49c4f7fcd1271", + "diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "weighted_variant_metrics.py": "28fe99e06ea545dcb6e28173e35b59600d9af8a7a11f1c48a11e7477f930e7f6", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "summarize_paired_diagnostics.py": "8575ce9df387d7c12b6037b35a2bf9ee133e4d99a8e8fff4f50a1c181c09d12d" + }, + "phases": [ + { + "name": "parent-tracks", + "arguments": [ + "/workspace/research/seqfunc/diagnose_track_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json" + }, + { + "name": "parent-dense-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_variant_uncertainty_weighted_io.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json", + "--replicates", + "2000", + "--seed", + "1729", + "--evaluations", + "eqtl_sign", + "eqtl_causality", + "eqtl_coefficient", + "enhancer_gene", + "traitgym_mendelian", + "traitgym_complex" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json" + }, + { + "name": "parent-remaining-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json" + }, + { + "name": "parent-summary", + "arguments": [ + "/workspace/research/seqfunc/summarize_paired_diagnostics.py", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json", + "/workspace/research/studies/rc_large114/diagnostics/parent/summary.json" + ], + "output": "/workspace/research/studies/rc_large114/diagnostics/parent/summary.json" + }, + { + "name": "smallRC108-tracks", + "arguments": [ + "/workspace/research/seqfunc/diagnose_track_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json" + }, + { + "name": "smallRC108-dense-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_variant_uncertainty_weighted_io.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json", + "--replicates", + "2000", + "--seed", + "1729", + "--evaluations", + "eqtl_sign", + "eqtl_causality", + "eqtl_coefficient", + "enhancer_gene", + "traitgym_mendelian", + "traitgym_complex" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json" + }, + { + "name": "smallRC108-remaining-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json" + }, + { + "name": "smallRC108-summary", + "arguments": [ + "/workspace/research/seqfunc/summarize_paired_diagnostics.py", + "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json", + "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json", + "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json", + "/workspace/research/studies/rc_large114/diagnostics/smallRC108/summary.json" + ], + "output": "/workspace/research/studies/rc_large114/diagnostics/smallRC108/summary.json" + } + ], + "method": "2000 paired clustered draws,seed1729; original-point and first32-replicate verifier parity required by existing helpers. No fitting, label-derived inference assets or GPU work.", + "scope": "Serial original four-phase all11/all13 analyses against two fixed controls,2000/1729. Shared candidate moments collected once; no second concurrent observer.", + "caveat": "Exploratory postselection on repeatedly inspected development data. Resampling schemes are not independent replications; zero-inclusive intervals are not equivalence. No variant-label fitting, pretrained external parameters, test data, new sequence flank or inference cache. One GPU and unchanged existing trainers. All11/all13 versus RC108. Both use eightpasses/seed17/originalschedule/alltrainingintervals, four identical prior offsets and equalF/RC inference. Architecture includes globalcapacity andlocalwidth32-to64 differences; compute and numerical histories differ. One trajectory per size, not independent replication or equalcompute." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-proposed-recipe.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-proposed-recipe.json new file mode 100644 index 0000000000000000000000000000000000000000..f607d8fd875571d0970ed2260059a37ce6ac1497 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S038/training_history/rc-large114-proposed-recipe.json @@ -0,0 +1,64 @@ +{ + "schema": "seqfunc-large-eight-pass-fixed-RC114-proposal/v1", + "created_utc": "2026-10-26T07:33:19.178799+00:00", + "name": "Fixed equal forward/reverse-complement inference on the eight-pass large model", + "parent": "/workspace/research/experiments/calib81-large-epoch08", + "primary_control": "/workspace/research/experiments/calib81-large-epoch08", + "secondary_control": "/workspace/research/experiments/rc108", + "parameters": 215626632, + "additional_training_updates": 0, + "inherited_complete_passes": 8, + "inherited_examples": 328160, + "inherited_optimizer_updates": 82040, + "seed": 17, + "orientation_weights": [ + 0.5, + 0.5 + ], + "change": "Apply the exact unchanged RC17/85/108 inference assets to sole calibrated81-large8. Every family and both entrypoints use the same two deterministic orientations; retain learned weights, four training-derived prior offsets, nearest global-feature decoder and original per-view sparse selection. No interpolation, SWA, joint-position or motif-mask composition.", + "architecture": "Global width1024/16transformerblocks, localwidth64/sixlocalblocks, local-to-trunk bridge, residualRMS cap0.1; same1,048,576bp input and all11families.", + "motivation": "RC108 improves reward and10families against its forward eight-pass small parent, with all13 variant intervals includingzero. At equal eight-pass exposure, rawlarge8 improves missense and worsens enhancer linking versus rawsmall8. Test whether the same fixed orientation averaging helps the large model and report every variant outcome. Neither additivity nor variant rescue is assumed.", + "primary_comparison": "All11/all13 versus the exact calibrated large eight-pass forward parent; isolates the fixed orientation operation on identical weights.", + "secondary_comparison": "All11/all13 versus RC108. Both use eightpasses/seed17/originalschedule/alltrainingintervals, four identical prior offsets and equalF/RC inference. Architecture includes globalcapacity andlocalwidth32-to64 differences; compute and numerical histories differ. One trajectory per size, not independent replication or equalcompute.", + "required_before_execution": [ + "Complete parent200track/all2697variant reports and parent own52/36derivation checks", + "Complete original parent fixed-link paired checks and exact unaffected family/variant reports", + "Accept either parent measured runtime pass or its explicit nonwinner runtime skip; this cannot stand in for the new candidate runtime check", + "CPU actual-large-weight finite/compatibility checks, exact decoder asset and parent-weight fingerprints, replayable package and correct README", + "Fresh one-H100 memory/time/CPU resource review including large14 andseed110, current hashed operator recovery deadline and evaluation reserve" + ], + "required_candidate_gates": [ + "Own52native/offline/partial-recovery arrays", + "Full200trackwindows/2697developmentqueries/all11/all13 and same-row publishedanchors", + "Own624native arrays on the existing fixed16window/96query fixture", + "Allthree sampled combined1000window/6587query runtime estimates<=14400s", + "One serial eight-phase all11/all13 paired observer against both fixed controls,2000replicates/seed1729", + "Atomic monotonic publication only after complete score/native/runtime gates; retain strongest existing package otherwise" + ], + "training_admitted": false, + "GPU_evaluation_admitted": false, + "controller_launched": false, + "parent_report_status_at_proposal": { + "evaluation.json": { + "exists": true, + "sha256": "bd0f0a864278e1086930bed08cf19b1ae19b61b68188d1cf2bc3d289d44a9618" + }, + "valid-score/report.json": { + "exists": true, + "sha256": "8cdb4e7ad48dc7b2c4c06622b1aa170557164d5777cf81b7481ce06f7bc3cf0c" + }, + "valid-variant-score/report.json": { + "exists": true, + "sha256": "fbcc4d8eca03106bb29703effdf2d5278f3a2f513341be2b2aab3ff41cff0965" + } + }, + "evidence_sha256": { + "/workspace/research/studies/rc108/completion-observation.json": "457ae0d8c35be775f898a3f19ef2d97df407290d423f57fb2a625bbe5f086490", + "/workspace/research/studies/rc108/diagnostics/summary.json": "0c754eac92f646930be8c29c6c4dc170713fbc1f17fcab10bf57c74b048a2486", + "/workspace/research/studies/scale13/epoch08-learning-report/report.json": "c13f184b573712c2dad47bca6366c353a46455e93ada6e0c6806c34949baf580", + "/workspace/research/studies/linear_training94/completion-observation.json": "6e47294a7710fad979349792d4c144a44c0eb75386e3e5a78b1ef9bd67e8924b", + "/workspace/research/seqfunc/make_rc_candidate.py": "522b03f803136ca230cea9675b902d57fbc88718dad1bfdf9ec0b88568b4e637", + "/workspace/research/seqfunc/run_fixed_rc_with_runtime.py": "5c123ec89d4c0bd354170b17077533cd87e42e09ee703d812053b1645001c7e9" + }, + "limitations": "Exploratory postselection on repeatedly inspected development data. Resampling schemes are not independent replications; zero-inclusive intervals are not equivalence. No variant-label fitting, pretrained external parameters, test data, new sequence flank or inference cache. One GPU and unchanged existing trainers." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/README.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/README.md new file mode 100644 index 0000000000000000000000000000000000000000..81a68811e45a4295280e0c95303588f6b5dadc49 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/README.md @@ -0,0 +1,84 @@ +# SeqFunc RC model with selective linear feature decoding + +One model with **215,626,632 unique learned parameters** reads 1,048,576 bp and serves +all eleven families, 5,930 tracks and both offline entrypoints. It uses the same fixed +equal forward/reverse-complement views as RC114. The only new operation is the exact +decoder from experiments111/96/113: global features are interpolated at bin centers from +1 kb to128 bp for dense, splice-usage and junction outputs. The original nearest +splice classifier and contact path remain unchanged. No weights or constants are fitted. + +The architecture has a strided convolutional encoder (initial width96, growth96), sixteen width1024 transformer blocks +over1,024 global tokens, a width64 base-resolution path with a21-bp stem and six dilated +residual blocks (10,941-bp receptive field), and a local-to-global bridge. Attention and +MLP updates use RMS cap0.1. Dense outputs cover the central196,608bp; contacts and junctions +use low-rank factors. Each orientation applies the four inherited training-prior splice +offsets before averaging probabilities. Dense and contact values average in asinh space. +Sparse coordinates are predicted independently in each view; the union treats an unlisted +view as zero, keeping at most4,096 usage positions and16,384 junction pairs. + +All weights originate from local random initialization, seed17, with **328,160 examples / +82,040 AdamW updates**, eight complete passes through all41,020 FOLD_0 training intervals +in folds2–7. The supplied reader decodes quantized targets per example. Batch4 BF16 training +uses fixed family-variance scales estimated from256 training windows and balanced splice +BCE. AdamW uses weight decay1e-4, clipping5, and the original14-pass cosine schedule from +2e-4 to2e-5 with4,000 warmup examples. The source checkpoint records108,193.2591 accumulated +training-loop seconds (30.05h) on one shared H10080GB within22CPUs. Setup/evaluation cost is +additional. Training recovery preserves checked saved state/order but is not bit-identical. +This candidate adds zero parameters and zero training updates. Restore the recorded +pre-decoder model source to replay the inherited training recipe. + +Run `infer.sh INPUT_ROOT OUTPUT_ROOT` or `perturb.sh INPUT_ROOT OUTPUT_ROOT` offline. +The runtime archive vendors CPython3.11, PyTorch2.7.1/CUDA12.6 and dependencies; the host +supplies Linux, bash/tar/gzip and the NVIDIA driver. Both entrypoints use public manifests +and the pinned reference only, with no targets, supplied splice coordinates, prediction +cache, extra flanking sequence or network access. Track arrays use float16 after float32 +orientation averaging; perturbation arrays remain float32. + +Candidate validation is pending. Its own52-array native/recovery check, six unchanged +native site/contact arrays versus RC114, full200 validation windows, all2,697 development +queries/all13 metrics with same-row anchors,624 native timing arrays, and all three4h +runtime estimates are required. Splice-site/contact family reports and five splice-based +variant reports must exactly match RC114. An all11/all13 paired comparison to RC114 is +required. Parent scores are not candidate scores. The smaller113 comparison reduced MendelianAP despite +track gains; all newvariant gains andregressions must be measured. GPU evaluation waits for +RC115 full/native/runtime/allpaired completion and a fresh single-H100 resource review. + +Limitations include one-seed exploratory development selection, pooled local features, +finite local context, sparse-union dilution, low-rank contacts and uneven variant effects. +Fifty-four GTEx RNA heads lack direct target supervision and are excluded from scored RNA +means; eQTL scoring uses supervised ENCODE RNA tracks. The model has no conservation input. +Interval overlap with zero does not show equivalence; sampled runtimes are not upper bounds. + +## Measured validation results + +Family-mean Pearson: **0.522771** on 200 validation intervals. + +| Family | Pearson | +|---|---:| +| atac | 0.596442 | +| dnase | 0.673208 | +| procap | 0.564250 | +| cage | 0.359384 | +| rna_seq | 0.660058 | +| chip_tf | 0.475414 | +| chip_histone | 0.512257 | +| splice_sites | 0.431157 | +| splice_usage | 0.641912 | +| junctions | 0.455959 | +| contact_maps | 0.380437 | + +| Development evaluation | Metric | Model | Published anchors on the same rows | +|---|---|---:|---| +| eqtl_sign | auroc | 0.5707 | AlphaGenome: 0.8264 | +| eqtl_causality | auroc | 0.5433 | AlphaGenome: 0.8325 | +| eqtl_coefficient | spearman | 0.2471 | AlphaGenome: 0.4134 | +| sqtl_causality | auprc | 0.7536 | AlphaGenome: 0.8447 | +| clinvar_noncoding | auprc | 0.6457 | AlphaGenome: 0.9410 | +| clinvar_splice_region | auprc | 0.7048 | AlphaGenome: 0.7225 | +| clinvar_missense | auprc | 0.3220 | AlphaGenome: 0.3738 | +| mfass | auprc | 0.4950 | AlphaGenome: 0.7343 | +| caqtl | auprc | 0.1378 | AlphaGenome: 0.8076 | +| bqtl_spi1 | auprc | 0.2215 | AlphaGenome: 0.8167 | +| enhancer_gene | auprc | 0.4080 | AlphaGenome: 0.9594 | +| traitgym_mendelian | auprc | 0.2198 | Borzoi_L2_L2.plus.all: 0.5026; Enformer_L2_L2.plus.all: 0.5453; CADD.plus.RawScore: 0.7189; GPN-MSA_absLLR.plus.score: 0.6729 | +| traitgym_complex | auprc | 0.2710 | Borzoi_L2_L2.plus.all: 0.2945; Enformer_L2_L2.plus.all: 0.2692; CADD.plus.RawScore: 0.4134; GPN-MSA_absLLR.plus.score: 0.4214 | diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_bridge.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_bridge.py new file mode 100644 index 0000000000000000000000000000000000000000..2c2bea67a58e19676fa73ba2a0607002ea2d5419 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_bridge.py @@ -0,0 +1,48 @@ +"""Attach a zero-output local-to-global connection to our own training checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +assert s['schema'] == 'seqfunc-single/v1' +parent_config = ModelConfig(**s['config']) +assert parent_config.local_width > 0 and not parent_config.local_to_trunk +torch.manual_seed(s['seed']) +parent = CompactModel(parent_config) +config = replace(parent_config, local_to_trunk=True) +model = CompactModel(config) +missing = model.load_state_dict(s['model'], strict=False) +assert not missing.unexpected_keys +assert set(missing.missing_keys) == {'local_trunk_projection.weight', 'local_trunk_projection.bias'} +old_names = list(dict(parent.named_parameters())) +new_names = list(dict(model.named_parameters())) +assert new_names[:len(old_names)] == old_names +assert len(s['optimizer']['param_groups']) == 1 +group = s['optimizer']['param_groups'][0] +assert group['params'] == list(range(len(old_names))) +group['params'] = list(range(len(new_names))) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +intervention = {'kind': 'local_to_global_trunk', 'parent_sha256': digest, + 'step': s['global_step'], 'initial_outputs_unchanged': True, + 'computation': 'average aligned local features into 1-kb tokens, zero linear projection, add before transformer'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(intervention) +s.update(config=config.to_dict(), model=model.state_dict(), intervention=intervention, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(intervention, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps({'parameters': count_parameters(model), **intervention}), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_local.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_local.py new file mode 100644 index 0000000000000000000000000000000000000000..aa082537004dbc398c4a78c1bd0203970ce5e42f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/attach_local.py @@ -0,0 +1,52 @@ +"""Attach a zero-initialized local residual path to a checkpoint trained in this workspace.""" +import argparse +from dataclasses import replace +import hashlib +from pathlib import Path + +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--width', type=int, default=32) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1', 'only locally trained single-GPU checkpoints' + parent_config = ModelConfig(**state['config']) + assert parent_config.local_width == 0 + torch.manual_seed(state['seed']) + parent = CompactModel(parent_config) + config = replace(parent_config, local_width=args.width) + model = CompactModel(config) + result = model.load_state_dict(state['model'], strict=False) + assert not result.unexpected_keys + assert all(k.startswith(('local_path.', 'local_projection.')) for k in result.missing_keys) + parent_names = list(dict(parent.named_parameters())) + names = list(dict(model.named_parameters())) + assert names[:len(parent_names)] == parent_names, 'optimizer parameter order changed' + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(parent_names))) + group['params'] = list(range(len(names))) + state['model'] = model.state_dict() + state['config'] = config.to_dict() + state['intervention'] = { + 'kind': 'base_resolution_local_path', 'width': args.width, + 'parent_sha256': hashlib.sha256(args.checkpoint.read_bytes()).hexdigest(), + 'attachment_step': state['global_step'], 'initial_outputs_unchanged': True, + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output / 'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + _atomic_save(state, args.output / 'attached.pt') + print({'parameters': count_parameters(model), **state['intervention']}) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/batched_training.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/batched_training.py new file mode 100644 index 0000000000000000000000000000000000000000..e5c609c3680baa61f04739e9f22c03a801d7b294 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/batched_training.py @@ -0,0 +1,38 @@ +"""Batch variable event queries while retaining equal-per-example reference losses.""" +import torch +from torch.nn import functional as F +from reference.alphagenome_50m.train_eval import _model_inputs, _losses +from train_single import balanced_splice_loss + +def list_collate(samples): + return samples + +def prepare_batch(samples): + """Prepare on CPU before a bulk transfer; padding never contributes to a loss.""" + prepared = [_model_inputs(sample) for sample in samples] + max_usage = max(item[0]['usage_positions'].shape[1] for item in prepared) + max_queries = max(item[0]['junction_queries'].shape[1] for item in prepared) + inputs = { + 'dna': torch.cat([x[0]['dna'] for x in prepared]), + 'usage_positions': torch.cat([F.pad(x[0]['usage_positions'], (0, max_usage-x[0]['usage_positions'].shape[1])) for x in prepared]), + 'junction_site_positions': torch.cat([x[0]['junction_site_positions'] for x in prepared]), + 'junction_queries': torch.cat([F.pad(x[0]['junction_queries'], (0, 0, 0, max_queries-x[0]['junction_queries'].shape[1])) for x in prepared]), + } + return inputs, [x[1] for x in prepared], [x[2] for x in prepared] + +def equal_example_losses(outputs, samples, junction_targets, junction_valid, *, balanced=True, scales=None): + per_example = [] + for i, (sample, target, valid) in enumerate(zip(samples, junction_targets, junction_valid)): + sliced = {name:values[i:i+1] for name,values in outputs.items()} + sliced['splice_usage'] = sliced['splice_usage'][:, :sample['usage_positions'].numel()] + sliced['junctions'] = sliced['junctions'][:, :target.shape[1]] + losses = _losses(sliced, sample, target, valid) + if balanced: + losses['splice_sites'] = balanced_splice_loss(sliced['splice_sites'], sample['splice_sites'][None]) + per_example.append(losses) + raw = {name:torch.stack([loss[name] for loss in per_example]).mean() + for name in outputs} + raw['total'] = torch.stack(list(raw.values())).mean() + scaled = {name:value * (scales or {}).get(name, 1.) for name,value in raw.items() if name != 'total'} + scaled['total'] = torch.stack(list(scaled.values())).mean() + return scaled, raw diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_residual_cap.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_residual_cap.py new file mode 100644 index 0000000000000000000000000000000000000000..cd282275702aaf5d5eede4b3904add879aad846a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_residual_cap.py @@ -0,0 +1,40 @@ +"""Change only a parameter-free branch bound on our own trained checkpoint.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('checkpoint', type=Path) +p.add_argument('output', type=Path) +p.add_argument('--cap', type=float, default=0.1) +a = p.parse_args() +s = torch.load(a.checkpoint, map_location='cpu', weights_only=True) +config = ModelConfig(**s['config']) +assert config.residual_rms_cap == 0 and a.cap > 0 +config = replace(config, residual_rms_cap=a.cap) +model = CompactModel(config) +model.load_state_dict(s['model'], strict=True) +assert len(s['optimizer']['param_groups'][0]['params']) == len(list(model.parameters())) +assert all(torch.equal(v, s['model'][k]) for k, v in model.state_dict().items()) +with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() +change = {'kind': 'bounded_transformer_residuals', 'parent_sha256': digest, + 'step': s['global_step'], 'cap': a.cap, + 'parameters_optimizer_rng_unchanged': True, + 'computation': 'each attention and MLP update multiplied by min(1, cap/token_RMS)'} +history = list(s.get('interventions', [])) +if s.get('intervention') and s['intervention'] not in history: + history.append(s['intervention']) +history.append(change) +s.update(config=config.to_dict(), intervention=change, interventions=history) +a.output.mkdir(parents=True, exist_ok=False) +(a.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) +(a.output/'intervention.json').write_text(json.dumps(change, indent=2) + '\n') +_atomic_save(s, a.output/'start.pt') +print(json.dumps(change), flush=True) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_splice_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_splice_context.py new file mode 100644 index 0000000000000000000000000000000000000000..1a6927a721cd0f49f08768666b710dcdab774bd7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/change_splice_context.py @@ -0,0 +1,39 @@ +"""Create a controlled fork that changes only the splice context computation.""" +import argparse +import hashlib +import json +from pathlib import Path +import torch +import yaml + +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from reference.alphagenome_50m.train_eval import _atomic_save + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + args = p.parse_args() + state = torch.load(args.checkpoint, map_location='cpu', weights_only=True) + assert state['schema'] == 'seqfunc-single/v1' and state['config']['local_width'] > 0 + state['config']['splice_context_mode'] = 'uniform' + config = ModelConfig(**state['config']) + model = CompactModel(config) + model.load_state_dict(state['model'], strict=True) + with args.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + state['intervention'] = { + 'kind': 'uniform_splice_context', 'parent_sha256': digest, + 'step': state['global_step'], 'parameters_and_optimizer_unchanged': True, + 'computation': 'mean legacy offset weights into a per-class global bias; remove pooled local features from site context', + } + args.output.mkdir(parents=True, exist_ok=False) + (args.output/'config.yaml').write_text(yaml.safe_dump(config.to_dict())) + (args.output/'intervention.json').write_text(json.dumps(state['intervention'], indent=2)) + _atomic_save(state, args.output/'start.pt') + print(json.dumps(state['intervention']), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoder117_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoder117_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..efe62f3c8649b09610f12da79253f2fcf9270978 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoder117_policy.json @@ -0,0 +1,26 @@ +{ + "utc": "2026-10-26T10:57:56.711123+00:00", + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "candidate_model_sha256": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "parameters": 215626632, + "inherited_examples": 328160, + "inherited_optimizer_steps": 82040, + "inherited_complete_passes": 8, + "additional_training_updates": 0, + "only_change": "Replace RC114 nearest global-feature decoding for dense, usage and junction heads with the exact selective linear-bin-center operation from completed111/96. Keep its original nearest splice classifier, contacts, learned tensors, four training-derived offsets and equal forward/RC views.", + "scope": "One fixed network and the same two orientation views serve every family and both entrypoints. No learned or fitted mixture, routing, new splice prior, motif mask, SWA, extra flanking DNA, variant-label training or additional SGD.", + "limitations": "Exploratory one-seed postselection. Interaction can worsen any changed readout. Full native replay, runtime and accuracy must be measured independently; unchanged paths require exact complete reports." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoding_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoding_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..0cd25054c8aa9bbac84fcfa8063f85abe9790b7f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/decoding_policy.json @@ -0,0 +1,9 @@ +{ + "kind": "junction_transcript_direction", + "parent_package": "/workspace/research/experiments/scale13-large-epoch08/raw-package", + "weights_unchanged": true, + "plus": "donor < acceptor", + "minus": "donor > acceptor", + "inference_inputs": "model-selected coordinates and public track strand halves only", + "training_evidence": "64 training windows: 30,616 positive junction queries, all in transcript direction" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/extend_local_context.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/extend_local_context.py new file mode 100644 index 0000000000000000000000000000000000000000..3651698e8a120b3308566d05b4523733fa621d48 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/extend_local_context.py @@ -0,0 +1,70 @@ +"""Extend our trained local path with identity-initialized dilated residual blocks.""" +import argparse +from dataclasses import replace +import hashlib +import json +from pathlib import Path +import torch +import yaml +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save + +def extend(state, blocks): + old_config = ModelConfig(**state['config']) + assert old_config.local_width and blocks > old_config.local_context_blocks + config = replace(old_config, local_context_blocks=blocks) + torch.manual_seed(state['seed']) + parent = CompactModel(old_config) + model = CompactModel(config) + missing = model.load_state_dict(state['model'], strict=False) + new_prefixes = tuple(f'local_path.blocks.{i}.' for i in range(old_config.local_context_blocks, blocks)) + assert not missing.unexpected_keys + assert all(k.startswith(new_prefixes) for k in missing.missing_keys) + for i in range(old_config.local_context_blocks, blocks): + torch.nn.init.zeros_(model.local_path.blocks[i].contract.weight) + torch.nn.init.zeros_(model.local_path.blocks[i].contract.bias) + old_names = list(dict(parent.named_parameters())) + new_names = list(dict(model.named_parameters())) + assert len(state['optimizer']['param_groups']) == 1 + group = state['optimizer']['param_groups'][0] + assert group['params'] == list(range(len(old_names))) + # New blocks shift following norm/head IDs; preserve moments by parameter name. + remap = {old_id: new_names.index(name) for old_id, name in enumerate(old_names)} + state['optimizer']['state'] = {remap[old_id]: values + for old_id, values in state['optimizer']['state'].items()} + group['params'] = list(range(len(new_names))) + for name, value in state['model'].items(): + assert torch.equal(model.state_dict()[name], value) + state.update(config=config.to_dict(), model=model.state_dict()) + return state, model, remap + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('checkpoint', type=Path) + p.add_argument('output', type=Path) + p.add_argument('--blocks', type=int, default=6) + a = p.parse_args() + state = torch.load(a.checkpoint, map_location='cpu', weights_only=True) + old_blocks = ModelConfig(**state['config']).local_context_blocks + state, model, remap = extend(state, a.blocks) + with a.checkpoint.open('rb') as f: + digest = hashlib.file_digest(f, 'sha256').hexdigest() + change = {'kind': 'extended_local_context', 'parent_sha256': digest, + 'step': state['global_step'], 'blocks_before': old_blocks, 'blocks_after': a.blocks, + 'receptive_field_bp': 21+8*sum(4**i for i in range(a.blocks)), + 'parent_parameters_optimizer_moments_rng_preserved': True, + 'new_blocks_initially_identity': True, 'optimizer_ids_remapped_by_parameter_name': True} + history = list(state.get('interventions', [])) + if state.get('intervention') and state['intervention'] not in history: + history.append(state['intervention']) + history.append(change) + state.update(intervention=change, interventions=history) + a.output.mkdir(parents=True, exist_ok=False) + (a.output/'config.yaml').write_text(yaml.safe_dump(state['config'])) + (a.output/'intervention.json').write_text(json.dumps(change, indent=2)+'\n') + (a.output/'optimizer_id_remap.json').write_text(json.dumps(remap, indent=2)+'\n') + _atomic_save(state, a.output/'start.pt') + print(json.dumps({'parameters': count_parameters(model), **change}), flush=True) + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/infer.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..2a412a9ad8667d6dcb94bb57d9959d61d43ddd34 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/infer.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: infer.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/predict.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/package_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/package_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..e391a4eb9a1909f72b15d83ac70c3972e10cf540 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/package_manifest.json @@ -0,0 +1,96 @@ +{ + "parameters": 215626632, + "source_checkpoint_sha256": "181f13f7ab8b82826950afbafe1edc2a8bad78f3a02a9d4efaafff5c9c7ca537", + "files": { + "README.md": "b1480e9b37b710226f4a4307903db175e6fea6d2d13328036470ff2e42d0f7be", + "attach_bridge.py": "0740f17d8041d4886ad4437e83f1b781e941cc76a15db7cd5280a5983c36fc7c", + "attach_local.py": "d0774e2711840783267914789ed303a3b61b64381b72d61f0886edbc04974f15", + "batched_training.py": "217255d8f1d4fbaacc9a9bcf7ce252432ffa834b6eb47b7eed1b7471bbf3d5f7", + "change_residual_cap.py": "7c277370f54f352fd9df81ef241bfb100e123d57a56c87d5b85eac420c55be2d", + "change_splice_context.py": "4a7a1bf746b17abb75e0c611b4b19740541f825fe63a750fc44684351efadf32", + "extend_local_context.py": "aa720fdbd8625a900f24bc2c007ec0acf8c2164591971d007d2c01a58792d701", + "infer.sh": "ef810532b1e72b85fbe283e4f52c177bb70025977dbb2345d75837c9bbd51e37", + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "perturb.sh": "113ec2383276a6e83fe1e4e8b0fa914f8d535d5dd0f5b2c34240140312dd5fb4", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/__init__.py": "45f3cf93f48859104e70bb58d717130dfd6521a45d2aba06b1a726afc0dc51dd", + "reference/alphagenome_50m/__init__.py": "a5a7e45d5940c3b201017c6322128e9f47efd37e1d29f717da2c1f787688eff9", + "reference/alphagenome_50m/alphagenome_50m.yaml": "3cac32e9f55e352dff78bb85ac892fdae1f9e412d9bba3e945750210c5252eee", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/model.py": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "reference/alphagenome_50m/reference_large.yaml": "68084d299b39418034c6bbf71db174c3ffa68bbeea8779c1694e44ec71f3529c", + "reference/alphagenome_50m/run_ddp.sh": "96d35a24a4e6789311501fb98203807ac0c4429375c44d5c6671908270c93bb0", + "reference/alphagenome_50m/tiny_overfit.yaml": "f7a104c3929de0530a9a8f73f66d794cca595aefad8ef262e09ecbeb9cfce6de", + "reference/alphagenome_50m/train_eval.py": "95d2dad279b23ed9c0c8a4f070035cf53c27eb8fe27da64c441343717174f784", + "runtime.tar.gz": "718da4c374e0c552b6e429b226c90b2fded5c900d95577d190d4df520aaf3901", + "train_batched.py": "04a872cd66a16cd7c1f672392ac50e75f60536a763526f569cea56786a32c372", + "train_single.py": "59702a64b7f395c8adc50d6b907d3186b270cfc94454dbf9502f8e391e796497", + "training_config.yaml": "4616441df1c88ea6a9f911a4226fea581c2c610e90ba583a34705e961b06d735", + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b", + "training_loss_scales.json": "4a0e3480d69aa61d1d1f17c72c2f02ef830b651a0051c3a49aca8e862a192e3d", + "training_provenance.json": "5dd8009455266cc21bacd5a4c2ffa71085366047f5a461e1899473f0e1386d78", + "training_assets_manifest.json": "7e4a33e58edeb9c25fb59ec5b1f3b65ff2b1ca83abb9c3fc0770ec4f57af06c9", + "decoding_policy.json": "d7c2a1d2e81ad12279412c2076dc0ec9b6065dd456cf881c3af5eea5a3bf7c68", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "training_history/late-fixed-links81-master-plan.json": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "training_history/late-fixed-links81-resolved-parents.json": "1abeb4dda4b52232869aa1405c8a0747091505a9abda48add9c4130016271dbe", + "training_history/late-fixed-links81-parent-check.json": "9d6c82a8f82993fad54c586b29c3bda774e3839e67352c84eb663f819d0b51ea", + "training_history/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "training_history/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60", + "rc_ensemble_policy.json": "473b330cec287e6db11a77593160f16927b5562cf2b03d016cc05cd92372f44c", + "training_history/rc-large114-proposed-recipe.json": "6d068b2dfd9f8cafbd42cb23ec05505f06cfb9c08aed677038b9d7e6853f82ca", + "training_history/rc-large114-training-replay.md": "669c742992dcbc90e79b4e1ddb095c7e63fff3af3d5c637ac39623230c99f004", + "training_history/make_rc_candidate.py": "522b03f803136ca230cea9675b902d57fbc88718dad1bfdf9ec0b88568b4e637", + "training_history/prepare_rc_large114.py": "b942dde822add66bdac0686411cc54c2cbdbd31315a1bd62b8e7c6bb1127f608", + "training_history/rc-large114-study.json": "b49304893c91038388dfa797be368f347c6ddc1793e9b377572630c03db1aa2f", + "training_history/rc-large114-admission.json": "d48ea57465a7578e1f16b380ca3d2221024a957eb61368fbd089bb1b281cc986", + "training_history/rc-large114-CPU-construction-check.json": "5d0ddbd0993914889b1b2ea5bd83576b5ca0c4459a82e2c0b694ac812e8c999e", + "training_history/rc-large114-fixture-check.json": "12b06b35e555e3b78f9bee40acf9ae887714832dc52428ea29cddf7871784691", + "training_history/rc-large114-paired-plan.json": "81c750c4e1302d17a897d3474ee6b62c2274a65f89cf1fa5376510ea387c6332", + "training_history/freeze_rc_large114_execution.py": "1da2453683f36685296f4653af5e6bb1042f051a27c02b33b200d664577b2566", + "training_history/run_fixed_rc_with_runtime.py": "5c123ec89d4c0bd354170b17077533cd87e42e09ee703d812053b1645001c7e9", + "decoder117_policy.json": "94e562b194076f8fe2e41cfd0dbbc3b3268c5bf8f923ba3769ce9f8403c32d7e", + "training_history/prepare_decoder111.py": "afa19189b21fa7780f5f801df3ef0216b1d004bca9e17a034b8606c28420989e", + "training_history/make_linear_upsample83.py": "2048f81143e187051b20471e3fed1b2068e36ef9f4bbc92b66b51c040fb5da0f", + "training_history/make_splice_grid96.py": "2ae38b29e6c055b724b156240ba38ed22cd93ebdb1bf6da490d427f82c6577ec", + "training_history/prepare_rc_large_decoder117.py": "74b74d9513ae1a6a021926b7cffbc440875e986cd71f4065dbf5609c5f4feb06", + "training_history/decoder117-training-replay.md": "296d4eb47fcb56227086b4d277b5e1efd1e22cb76afd22180b0250bfebedaff6", + "training_history/decoder117-proposed-recipe.json": "b95ffe4953d0c11d4510002bda739e3e7a0f20893faa11d987d8c5548cb0e71b", + "training_history/pre-decoder117-model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "training_history/run_decoder111.py": "0c88342f7a15deb2d112139b99200fea34cdca69d9bcdce8239f18e8ef5e64b7", + "training_history/decoder117-controller-reuse.json": "c9a1733c9f84b24a3cada2312f8389753ca051aac3a11e10b9bc7ad5315becbc", + "training_history/decoder117-admission.json": "d1aaa1773c77106fbbbeef8d268c3774ab74a5a5f9569c6526af836fcfa2d227", + "training_history/decoder117-resource-review.json": "0946f74de81094208f216e9a8172e5e46ab87be027d4d11299089232bba9ce00", + "training_history/check_decoder111_parent.py": "053a3f6a6e84bd81d8103833b748b325833de0090887be82e1cb6f7c943a73be", + "training_history/freeze_rc_large_decoder117_execution.py": "051c37636a70cdae24afb89c5b09481b4fec463f8886fda2aec88d38e8a824c7", + "training_history/decoder117-fixture-check.json": "65641f8679d5329706867c2180e1b8d3889770fb1d9d137e779bbf5d99422e0f", + "training_history/decoder117-CPU-construction-check.json": "174ec3758eae1f0ca89d9cf24eb2c8ec9f9915beb937266030412eeff5f13603", + "training_history/decoder117-preparation.json": "7bf5507dc289020b30ba55a8df028535e03cc1ef20c3596d2e77a70716d9f5ce", + "validation_tracks.json": "b7f93bc4b280119a08400c294d1a8a57f797cb479b30cc5be4d208e62a6b4eb2", + "validation_variants.json": "054ed2e5fdb0a1feef5d309eb64c3a39c5049044eebbc589b9b10ee577263ab0" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.py new file mode 100644 index 0000000000000000000000000000000000000000..b8ee571ad9dab90aded5077affbb42419869da16 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.py @@ -0,0 +1,92 @@ +"""Predict both sequences of every perturbation (contract: perturbations/v1). + + python perturb.py INPUT_ROOT OUTPUT_ROOT --checkpoint last.pt + +INPUT_ROOT holds `public/perturbations.parquet` and `reference/GRCh38.p13.genome.fa` (in the +workspace, `/workspace/data` works too: the queries are then read from `variant_evals/public/`). Each row is a +1,048,576-bp window starting at `input_start` and an edit spanning [edit_start, edit_end) in genomic +0-based coordinates: + +- `snv`: sequence A carries `allele_a` at edit_start, sequence B carries `allele_b` (both are written + explicitly; either may differ from the reference genome); +- `mask`: sequence A is the reference, sequence B has the interval replaced by N. + +For each perturbation the output group `perturbations/p{index:06d}` holds `a/` and +`b/` for every family in `families`: dense families sliced to bins +[dense_bin_start, dense_bin_end) of the central target window, and `splice_sites` sliced to bp +[splice_start, splice_end). Values are the same quantities as predictions.zarr (asinh space for +dense tracks, splice-site scores for splice_sites). +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from numcodecs import Blosc +import numpy as np +import pandas as pd +import pysam +import torch +import zarr + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import predict_requested + +PERTURBATION_SCHEMA = "genome-sequence-to-function-perturbations/v1" +COMPRESSOR = Blosc(cname="zstd", clevel=3, shuffle=Blosc.BITSHUFFLE) + + +def sequences(fasta, row) -> tuple[str, str]: + length = fasta.get_reference_length(row.chrom) + start, end = int(row.input_start), int(row.input_start) + 1_048_576 + sequence = fasta.fetch(row.chrom, max(0, start), min(length, end)).upper() + sequence = "N" * max(0, -start) + sequence + "N" * max(0, end - length) + lo, hi = int(row.edit_start) - start, int(row.edit_end) - start + if row.edit == "snv": + return sequence[:lo] + row.allele_a + sequence[hi:], sequence[:lo] + row.allele_b + sequence[hi:] + if row.edit == "mask": + return sequence, sequence[:lo] + "N" * (hi - lo) + sequence[hi:] + raise ValueError(f"unknown edit {row.edit!r}") + + +def main(input_root: Path, output_root: Path, checkpoint: Path, queries: Path | None = None) -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config) + model.load_state_dict(payload["model"], strict=True) + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + model.to(device).eval() + if queries is None: + queries = input_root / "public/perturbations.parquet" + if not queries.exists(): + queries = input_root / "variant_evals/public/perturbations.parquet" + rows = pd.read_parquet(queries) + fasta = pysam.FastaFile(str(input_root / "reference/GRCh38.p13.genome.fa")) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "perturbations.zarr"), mode="w") + target_start = (config.input_length - config.target_length) // 2 + with torch.inference_mode(): + for index, row in enumerate(rows.itertuples(index=False)): + pair = sequences(fasta, row) + dna = torch.from_numpy(np.stack([AlphaGenomeDataset._one_hot(s) for s in pair])).to(device) + group = root.create_group(f"perturbations/p{index:06d}") + for family, values in predict_requested(model, dna, row).items(): + values = np.asarray(values, dtype=np.float32) + for side, array in zip("ab", values): + group.create_dataset(f"{side}/{family}", data=array, compressor=COMPRESSOR) + print(json.dumps({"event": "perturb", "done": index + 1, "total": len(rows)}), flush=True) + root.attrs.update({"schema": PERTURBATION_SCHEMA, + "perturbation_ids": rows.perturbation_id.astype(str).tolist()}) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0]) + parser.add_argument("input_root", type=Path) + parser.add_argument("output_root", type=Path) + parser.add_argument("--checkpoint", type=Path, default=Path(__file__).resolve().parent / "last.pt") + parser.add_argument("--queries", type=Path, help="another query table, e.g. the development split") + args = parser.parse_args() + main(args.input_root, args.output_root, args.checkpoint, args.queries) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.sh new file mode 100644 index 0000000000000000000000000000000000000000..f07c4714532c2d33e7a419de3877e5daacee89fd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/perturb.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail +[[ $# == 2 ]] || { echo "usage: perturb.sh INPUT_ROOT OUTPUT_ROOT" >&2; exit 2; } +package_dir=$(cd "$(dirname "$0")" && pwd) +runtime_dir=$(mktemp -d "${TMPDIR:-/tmp}/seqfunc-runtime.XXXXXXXX") +trap 'rm -rf "$runtime_dir"' EXIT +tar -xzf "$package_dir/runtime.tar.gz" -C "$runtime_dir" +unset PYTHONPATH +export PYTHONHOME="$runtime_dir" PYTHONNOUSERSITE=1 PYTHONDONTWRITEBYTECODE=1 +export LD_LIBRARY_PATH="$runtime_dir/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" +export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4 OPENBLAS_NUM_THREADS=1 NUMEXPR_NUM_THREADS=4 +"$runtime_dir/bin/python" -s "$package_dir/perturb.py" "$1" "$2" --checkpoint "$package_dir/last.pt" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/predict.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..665fb186f3a62b1ff130abb034ac25a3a924c7c8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/predict.py @@ -0,0 +1,52 @@ +"""Stream one shared-checkpoint forward/reverse-complement mean for every output family.""" +from __future__ import annotations +import argparse +import json +from pathlib import Path +from numcodecs import Blosc +import numpy as np +import torch +import zarr +from prediction_contract import PREDICTION_SCHEMA +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig +from rc_ensemble import engine_for + +COMPRESSOR = Blosc(cname="zstd", clevel=7, shuffle=Blosc.BITSHUFFLE) + + +def main(input_root: Path, output_root: Path, checkpoint: Path, split: str = "test") -> None: + payload = torch.load(checkpoint, map_location="cpu", weights_only=True) + model = CompactModel(ModelConfig(**payload['config'])) + model.load_state_dict(payload['model'], strict=True) + device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + model.to(device).eval() + dataset = AlphaGenomeDataset(input_root, split, include_targets=False) + model.set_junction_track_halves(dataset.junction_track_halves) + engine = engine_for(model) + output_root.mkdir(parents=True, exist_ok=True) + root = zarr.open_group(str(output_root / "predictions.zarr"), mode="w") + ids = [] + with torch.inference_mode(): + for sample in dataset: + example_id = sample['example_id'] + values = engine.tracks(sample['dna'][None].to(device)) + group = root.create_group(f"examples/{example_id}") + for name, array in values.items(): + dtype = np.int8 if name == 'junction_half' else np.int32 if name in ( + 'splice_usage_positions', 'junction_donor', 'junction_acceptor') else np.float16 + group.create_dataset(name, data=array.astype(dtype), compressor=COMPRESSOR) + ids.append(example_id) + print(json.dumps({'event': 'predict', 'example': len(ids), 'total': len(dataset)}), flush=True) + root.attrs.update({'schema': PREDICTION_SCHEMA, 'example_ids': ids}) + (output_root/'example_ids.json').write_text(json.dumps(ids)+'\n') + + +if __name__ == '__main__': + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('input_root', type=Path) + p.add_argument('output_root', type=Path) + p.add_argument('--checkpoint', type=Path, default=Path(__file__).resolve().parent/'last.pt') + p.add_argument('--split', default='test', choices=['valid', 'test']) + a = p.parse_args() + main(a.input_root, a.output_root, a.checkpoint, a.split) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/prediction_contract.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/prediction_contract.py new file mode 100644 index 0000000000000000000000000000000000000000..2455424db002b840cf0d69456930840a7bc5f3cb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/prediction_contract.py @@ -0,0 +1,23 @@ +"""Public prediction shapes; no trusted splice coordinates are exposed.""" + +from __future__ import annotations + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} + +PREDICTION_SCHEMA = "genome-sequence-to-function-predictions/v4" +MAX_USAGE_SITES = 4_096 +MAX_JUNCTION_PAIRS = 16_384 +# Fixed-shape arrays per example; splice usage and junctions are sparse (see contract.md). +FIXED_SHAPES = { + **{name: (1_536, tracks) for name, tracks in DENSE_TRACKS.items()}, + "splice_sites": (196_608, 4), + "contact_maps": (131_328, 28), +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble.py new file mode 100644 index 0000000000000000000000000000000000000000..f076804c66e481eb823e6ca5e721f39fce50bad5 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble.py @@ -0,0 +1,111 @@ +"""One shared checkpoint evaluated on DNA and its reverse complement for every output.""" +import json +from pathlib import Path +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from single_prediction import single_prediction +from splice_output import splice_output_scores + + +def contact_reversal(bins): + row, column = np.triu_indices(bins) + lookup = np.empty((bins, bins), dtype=np.int64) + lookup[row, column] = np.arange(len(row)) + lookup[column, row] = np.arange(len(row)) + return lookup[bins-1-row, bins-1-column] + + +def reverse_readout(arrays, permutations, input_length, target_length, contact_bins): + """Map a reverse-complement prediction back into the original genomic coordinates.""" + out = {name: arrays[name][::-1, permutations[name]].copy() for name in DENSE_TRACKS} + out['splice_sites'] = arrays['splice_sites'][::-1, permutations['splice_sites']].copy() + positions = target_length-1-arrays['splice_usage_positions'] + order = np.argsort(positions) + out['splice_usage_positions'] = positions[order].astype(np.int32) + out['splice_usage'] = arrays['splice_usage'][order][:, permutations['splice_usage']].copy() + out['junction_half'] = (1-arrays['junction_half']).astype(np.int8) + for key in ['junction_donor', 'junction_acceptor']: + out[key] = (input_length-1-arrays[key]).astype(np.int32) + out['junctions'] = arrays['junctions'][:, permutations['junctions']].copy() + out['contact_maps'] = arrays['contact_maps'][contact_reversal(contact_bins)][:, permutations['contact_maps']].copy() + return out + + +def combine_readouts(a, b, track_halves): + """Average sparse functions with zero at unlisted coordinates, then enforce output budgets.""" + out = {name: (a[name]+b[name])*np.float32(.5) + for name in [*DENSE_TRACKS, 'splice_sites', 'contact_maps']} + usage_positions, inverse = np.unique(np.concatenate([a['splice_usage_positions'], b['splice_usage_positions']]), return_inverse=True) + assert len(usage_positions) <= MAX_USAGE_SITES + usage = np.zeros((len(usage_positions), a['splice_usage'].shape[-1]), dtype=np.float32) + np.add.at(usage, inverse, np.concatenate([a['splice_usage'], b['splice_usage']])*np.float32(.5)) + out.update(splice_usage_positions=usage_positions.astype(np.int32), splice_usage=usage) + keys = np.concatenate([np.stack([x['junction_half'], x['junction_donor'], x['junction_acceptor']], axis=1) for x in [a, b]]) + keys, inverse = np.unique(keys, axis=0, return_inverse=True) + values = np.zeros((len(keys), len(track_halves)), dtype=np.float32) + np.add.at(values, inverse, np.concatenate([a['junctions'], b['junctions']])*np.float32(.5)) + strength = np.empty(len(keys), dtype=np.float32) + for half in [0, 1]: + keep = keys[:, 0] == half + strength[keep] = values[keep][:, track_halves == half].max(axis=1) + # Stable sorting makes ties deterministic in the canonical coordinate order. + order = np.argsort(-strength, kind='stable')[:MAX_JUNCTION_PAIRS] + keys, values = keys[order], values[order] + out.update(junction_half=keys[:, 0].astype(np.int8), + junction_donor=keys[:, 1].astype(np.int32), + junction_acceptor=keys[:, 2].astype(np.int32), junctions=values) + return out + + +class ReverseComplementMean: + def __init__(self, model): + self.model = model + data = json.loads(Path(__file__).with_name('strand_permutations.json').read_text()) + self.permutations = {name: np.asarray(values, dtype=np.int64) for name, values in data['permutations'].items()} + for name, indices in self.permutations.items(): + assert np.array_equal(indices[indices], np.arange(len(indices))), name + + def tracks(self, dna): + assert len(dna) == 1 + config = self.model.config + original = single_prediction(self.model, dna) + reverse = single_prediction(self.model, dna.flip((1, 2))) + reverse = reverse_readout(reverse, self.permutations, config.input_length, config.target_length, config.contact_bins) + return combine_readouts(original, reverse, self.model.junction_track_half.cpu().numpy()) + + def dense_and_sites(self, dna, families): + """Return central readouts for both alleles; this same probability/asinh averaging defines tracks.""" + config = self.model.config + start = (config.input_length-config.target_length)//2 + predictions = [] + for reverse in [False, True]: + trunk = self.model._trunk(dna.flip((1, 2)) if reverse else dna) + features = None + values = {} + for family in families: + if family == 'splice_sites': + logits = self.model._splice_site_logits(trunk)[:, start:start+config.target_length] + array = splice_output_scores(logits).float().cpu().numpy() + else: + if features is None: + features = self.model._dense_features(trunk) + array = self.model.dense_heads[family](features).float().cpu().numpy() + if reverse: + array = array[:, ::-1][:, :, self.permutations[family]].copy() + values[family] = array + predictions.append(values) + return {family: (predictions[0][family]+predictions[1][family])*np.float32(.5) for family in families} + + +def engine_for(model): + if not hasattr(model, '_rc_prediction_engine'): + # Plain helper, not a child nn.Module; the backbone state dictionary stays unchanged. + model._rc_prediction_engine = ReverseComplementMean(model) + return model._rc_prediction_engine + + +def predict_requested(model, dna, row): + values = engine_for(model).dense_and_sites(dna, row.families) + return {family: array[:, row.splice_start:row.splice_end] if family == 'splice_sites' + else array[:, row.dense_bin_start:row.dense_bin_end] for family, array in values.items()} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble_policy.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble_policy.json new file mode 100644 index 0000000000000000000000000000000000000000..e884027651302fbdddc33de84f5f2bdaca17ef27 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/rc_ensemble_policy.json @@ -0,0 +1,35 @@ +{ + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-26T07:36:00.650275+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib81-large-epoch08/package", + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 215626632, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..4879a0e1a3a8a8050bb48360febee152cd141be8 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/__init__.py @@ -0,0 +1 @@ +"""The reference model of the alphagenome-reference baseline.""" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/__init__.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a86a139babb6e899bd5102bbf3c628e1232576b2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/__init__.py @@ -0,0 +1,5 @@ +"""Compact 1-Mb, eleven-family AlphaGenome reference model.""" + +from .model import CompactModel, ModelConfig, count_parameters + +__all__ = ["CompactModel", "ModelConfig", "count_parameters"] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/alphagenome_50m.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/alphagenome_50m.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a59a4d0c65752d45eb464d129c6ec9d559665e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/alphagenome_50m.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 32 +channel_growth: 32 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 384 +num_sequence_blocks: 6 +num_attention_heads: 6 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 256 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/cached_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/cached_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..9c3677c1e8518ea97db0dd6759f3d636f5fff27c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/cached_read_store.py @@ -0,0 +1,71 @@ +"""Optional local byte cache for immutable training Zarr stores. + +Source storage stays read-only. Cached entries contain a key fingerprint, length, +and SHA256 of the original compressed bytes; every hit verifies all three. Writes +replace a private temporary inode atomically. A missing/corrupt/unwritable cache +falls back to the original store, so this layer changes neither values nor keys. +Only training loaders may opt into it; it is not an inference asset. +""" +import hashlib +import os +from pathlib import Path +import struct +import tempfile + +from .direct_read_store import ParallelDirectReadDirectoryStore + +MAGIC = b'SFBC0001' +HEADER = struct.Struct('<8sQ32s32s') + + +class CachedReadDirectoryStore(ParallelDirectReadDirectoryStore): + def __init__(self, path, *, cache_root, namespace, io_workers=4): + super().__init__(path, io_workers=io_workers) + if not namespace: + raise ValueError('An immutable input namespace is required') + self.cache_root = Path(cache_root) + self.namespace = namespace + self.store_key = hashlib.sha256((namespace+'\0'+os.path.abspath(path)).encode()).digest() + self.statistics = {'hit': 0, 'miss': 0, 'corrupt': 0, 'write_failure': 0} + + def cache_location(self, key): + key = self._normalize_key(key) + identity = hashlib.sha256(self.store_key+key.encode()).digest() + name = identity.hex() + return self.cache_root/name[:2]/name[2:], identity + + def __getitem__(self, key): + key = self._normalize_key(key) + destination, identity = self.cache_location(key) + try: + blob = destination.read_bytes() + if len(blob) >= HEADER.size: + magic, size, cached_key, checksum = HEADER.unpack_from(blob) + payload = blob[HEADER.size:] + if (magic == MAGIC and size == len(payload) and cached_key == identity + and hashlib.sha256(payload).digest() == checksum): + self.statistics['hit'] += 1 + return payload + self.statistics['corrupt'] += 1 + except OSError: + pass + self.statistics['miss'] += 1 + payload = super().__getitem__(key) + temporary = None + try: + destination.parent.mkdir(parents=True, exist_ok=True) + with tempfile.NamedTemporaryFile(prefix='.pending-', dir=destination.parent, delete=False) as stream: + temporary = stream.name + stream.write(HEADER.pack(MAGIC, len(payload), identity, hashlib.sha256(payload).digest())) + stream.write(payload) + os.replace(temporary, destination) + temporary = None + except OSError: + self.statistics['write_failure'] += 1 + finally: + if temporary is not None: + try: + os.unlink(temporary) + except OSError: + pass + return payload diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/dataset.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/dataset.py new file mode 100644 index 0000000000000000000000000000000000000000..8a3f7d4644e5d9f02162bac9e859443ba6c33658 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/dataset.py @@ -0,0 +1,312 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +import os +from concurrent.futures import ThreadPoolExecutor + + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(zarr.DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(store=ParallelDirectReadDirectoryStore(str(path), io_workers=4), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + # Optional training-only compressed-byte cache; off for all public inference. + self.training_bytecache = None + requested_cache = os.environ.get('SEQFUNC_TRAIN_BYTECACHE') + if requested_cache and split == 'train' and include_targets: + import hashlib + import json + from .cached_read_store import CachedReadDirectoryStore + manifests = {str(path): hashlib.sha256(path.read_bytes()).hexdigest() for path in + [self.root/'public/train.parquet', self.root/'labels/train.parquet']} + namespace = hashlib.sha256(json.dumps(manifests, sort_keys=True).encode()).hexdigest() + self.training_bytecache = (requested_cache, namespace) + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + path = self.root / relative + if self.training_bytecache is not None and not str(path).endswith('.zarr.zip'): + from .cached_read_store import CachedReadDirectoryStore + cache_root, namespace = self.training_bytecache + cached_store = CachedReadDirectoryStore(str(path), cache_root=cache_root, + namespace=namespace, io_workers=4) + group, store = zarr.open_group(store=cached_store, mode='r'), None + else: + group, store = _open_part(path) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/direct_read_store.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/direct_read_store.py new file mode 100644 index 0000000000000000000000000000000000000000..c560febc6a15909a523c8e2f3d1f861aeec309a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/direct_read_store.py @@ -0,0 +1,64 @@ +"""Read-only Zarr directory store avoiding redundant per-chunk stat calls.""" +import os +from concurrent.futures import ThreadPoolExecutor + +from zarr.storage import DirectoryStore + +_pools = {} + + +def process_pool(workers): + key = (os.getpid(), workers) + if key not in _pools: + _pools[key] = ThreadPoolExecutor(max_workers=workers, thread_name_prefix='zarr-read') + return _pools[key] + + +class DirectReadDirectoryStore(DirectoryStore): + def __getitem__(self, key): + key = self._normalize_key(key) + try: + return self._fromfile(os.path.join(self.path, key)) + except (FileNotFoundError, IsADirectoryError, NotADirectoryError): + raise KeyError(key) from None + + def getitems(self, keys, *, contexts): + result = {} + for key in keys: + try: + result[key] = self[key] + except KeyError: + pass + return result + + def __setitem__(self, key, value): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def __delitem__(self, key): + raise PermissionError('DirectReadDirectoryStore is read-only') + + def is_writeable(self): + return False + + def is_erasable(self): + return False + + +class ParallelDirectReadDirectoryStore(DirectReadDirectoryStore): + """Bounded I/O threads shared by all read stores within one loader process.""" + def __init__(self, path, *, io_workers=4): + super().__init__(path) + if io_workers < 1: + raise ValueError('io_workers must be positive') + self.io_workers = io_workers + + def getitems(self, keys, *, contexts): + def fetch(key): + try: + return key, self[key] + except KeyError: + return key, None + if self.io_workers == 1: + return super().getitems(keys, contexts=contexts) + return {key: value for key, value in process_pool(self.io_workers).map(fetch, keys) + if value is not None} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/layers.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ef7b4e45a81eb774e4e92ac36e30c5abdebbe46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/layers.py @@ -0,0 +1,134 @@ +"""Shared neural-network layers for the single compact AlphaGenome model.""" + +from __future__ import annotations + +import math + +import torch +from torch import Tensor, nn +from torch.nn import functional as F + + +class ChannelRMSNorm(nn.Module): + def __init__(self, channels: int, eps: float = 1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(channels)) + self.eps = eps + + def forward(self, x: Tensor) -> Tensor: + variance = x.float().square().mean(dim=1, keepdim=True) + normalized = x * torch.rsqrt(variance + self.eps).to(x.dtype) + return normalized * self.weight[None, :, None].to(x.dtype) + + +class StandardizedConv1d(nn.Conv1d): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__(in_channels, out_channels, kernel_size, padding="same", bias=True) + self.scale = nn.Parameter(torch.ones(out_channels)) + nn.init.trunc_normal_(self.weight, std=1.0 / math.sqrt(in_channels * kernel_size)) + nn.init.trunc_normal_(self.bias, std=1e-4) + + def forward(self, x: Tensor) -> Tensor: + centered = self.weight - self.weight.mean(dim=(1, 2), keepdim=True) + variance = centered.float().square().mean(dim=(1, 2), keepdim=True) + fan_in = self.weight.shape[1] * self.weight.shape[2] + multiplier = self.scale[:, None, None] * torch.rsqrt( + (fan_in * variance).clamp_min(1e-4) + ).to(self.weight.dtype) + return F.conv1d(x, centered * multiplier, self.bias, padding="same") + + +class ConvBlock(nn.Module): + def __init__(self, in_channels: int, out_channels: int, kernel_size: int): + super().__init__() + self.norm = ChannelRMSNorm(in_channels) + self.conv = StandardizedConv1d(in_channels, out_channels, kernel_size) + + def forward(self, x: Tensor) -> Tensor: + return self.conv(F.gelu(self.norm(x), approximate="tanh")) + + +def _apply_rope(x: Tensor) -> Tensor: + rotary_dim = x.shape[-1] - x.shape[-1] % 2 + if rotary_dim == 0: + return x + positions = torch.arange(x.shape[-2], device=x.device, dtype=torch.float32) + inverse_frequency = 1.0 / ( + 10_000 ** ( + torch.arange(0, rotary_dim, 2, device=x.device, dtype=torch.float32) + / rotary_dim + ) + ) + angles = torch.outer(positions, inverse_frequency) + cos = angles.cos()[None, None, :, :].to(x.dtype) + sin = angles.sin()[None, None, :, :].to(x.dtype) + even, odd = x[..., :rotary_dim:2], x[..., 1:rotary_dim:2] + rotated = torch.stack((even * cos - odd * sin, even * sin + odd * cos), dim=-1) + return torch.cat((rotated.flatten(-2), x[..., rotary_dim:]), dim=-1) + + +class MultiHeadAttention(nn.Module): + def __init__(self, config): + super().__init__() + self.num_heads = config.num_attention_heads + self.head_dim = config.attention_head_dim + inner_dim = self.num_heads * self.head_dim + self.norm = nn.RMSNorm(config.hidden_dim) + self.qkv = nn.Linear(config.hidden_dim, 3 * inner_dim, bias=False) + self.output = nn.Linear(inner_dim, config.hidden_dim) + self.dropout = config.dropout + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + batch, length, _ = x.shape + qkv = self.qkv(self.norm(x)).view( + batch, length, 3, self.num_heads, self.head_dim + ) + q, k, value = qkv.unbind(dim=2) + attended = F.scaled_dot_product_attention( + _apply_rope(q.transpose(1, 2)), + _apply_rope(k.transpose(1, 2)), + value.transpose(1, 2), + attn_mask=attention_bias, + dropout_p=self.dropout if self.training else 0.0, + ) + return self.output(attended.transpose(1, 2).reshape(batch, length, -1)) + + +class MLPBlock(nn.Module): + def __init__(self, config): + super().__init__() + expanded = int(config.hidden_dim * config.mlp_expansion) + self.norm = nn.RMSNorm(config.hidden_dim) + self.input = nn.Linear(config.hidden_dim, expanded) + self.output = nn.Linear(expanded, config.hidden_dim) + self.dropout = nn.Dropout(config.dropout) + nn.init.trunc_normal_(self.output.weight, std=1e-3) + nn.init.zeros_(self.output.bias) + + def forward(self, x: Tensor) -> Tensor: + return self.dropout( + self.output(F.gelu(self.input(self.norm(x)), approximate="tanh")) + ) + + +def bound_residual_rms(update: Tensor, cap: float) -> Tensor: + """Cap each token's branch RMS, preserving small updates and their directions.""" + if cap == 0: + return update + rms = update.float().square().mean(dim=-1, keepdim=True).clamp_min(1e-12).sqrt() + scale = (cap / rms).clamp(max=1.0) + return update * scale.to(update.dtype) + + +class SequenceBlock(nn.Module): + def __init__(self, config): + super().__init__() + self.attention = MultiHeadAttention(config) + self.mlp = MLPBlock(config) + self.residual_rms_cap = config.residual_rms_cap + + def forward(self, x: Tensor, attention_bias: Tensor | None = None) -> Tensor: + x = x + bound_residual_rms(self.attention(x, attention_bias), self.residual_rms_cap) + return x + bound_residual_rms(self.mlp(x), self.residual_rms_cap) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/model.py new file mode 100644 index 0000000000000000000000000000000000000000..d71fb760a1db4042cc85d73831b98e3e17a9f8bd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/model.py @@ -0,0 +1,467 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _global_dense_features(self, trunk: Tensor) -> Tensor: + # Coarse features are centered in 1-kb bins; align_corners=False maps their + # centers onto the aligned 128-bp grid and extends boundary values constantly. + decoded = self.dense_decoder(trunk) + return F.interpolate(decoded.transpose(1, 2), scale_factor=self.config.dense_upsample, + mode='linear', align_corners=False).transpose(1, 2) + + def _dense_features(self, trunk: Tensor) -> Tensor: + first = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = self._global_dense_features(trunk)[:, first:first + self.config.dense_bins] + if self.config.local_width: + features = features + self._local_features[:, first:first + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self._global_dense_features(trunk) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + # The splice classifier retains the feature grid used during training. + # Dense, usage and junction readouts keep the separate interpolated grid. + features = self.dense_decoder(trunk.repeat_interleave(self.config.dense_upsample, dim=1)) + if self.config.local_width: + features = features + self._local_features + logits = self.splice_site_head(features) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self._gather_bins(self._global_dense_features(trunk), positions, self.config.dense_bin_size) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/reference_large.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/reference_large.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e403757c49eb6e6419927210c7053710964213d1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/reference_large.yaml @@ -0,0 +1,23 @@ +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 96 +channel_growth: 96 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 1024 +num_sequence_blocks: 16 +num_attention_heads: 16 +attention_head_dim: 64 +mlp_expansion: 4.0 +dropout: 0.0 +head_hidden_size: 768 +junction_rank: 32 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: true diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/run_ddp.sh b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/run_ddp.sh new file mode 100644 index 0000000000000000000000000000000000000000..c4702e9366e907b7791063e77dcfe281411ac8cd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/run_ddp.sh @@ -0,0 +1,16 @@ +#!/usr/bin/env bash +# run_ddp.sh DATA_DIR OUTPUT_DIR CONFIG EPOCHS [GPUS=8] [WORKERS_PER_GPU=12] +# Train the reference on every train interval with one example per GPU per step, then score +# validation with the verifier's own scorer. Runs inside the task image (torch, torchrun). +set -Eeuo pipefail +DATA=$1; OUT=$2; CONFIG=$3; EPOCHS=$4; GPUS=${5:-8}; WORKERS=${6:-12} +BASELINE=$(cd "$(dirname "$0")/../.." && pwd) +mkdir -p "$OUT" +cd "$BASELINE" +# Loopback rendezvous: works offline, where --standalone cannot resolve the hostname. +torchrun --nnodes 1 --nproc-per-node "$GPUS" --master-addr 127.0.0.1 --master-port "${MASTER_PORT:-29511}" -m reference.alphagenome_50m.train_eval train \ + --input "$DATA" --config "$CONFIG" --output "$OUT" --epochs "$EPOCHS" --workers "$WORKERS" \ + --log-every 200 --save-every 2000 2>&1 | tee -a "$OUT/train.log" +python predict.py "$DATA" "$OUT/valid-pred" --checkpoint "$OUT/last.pt" --split valid +python /workspace/evaluation/score.py --predictions "$OUT/valid-pred/predictions.zarr" \ + --labels "$DATA" --split valid --output "$OUT/valid-score" diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/tiny_overfit.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/tiny_overfit.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c25f9ffbb76102c970e5767922400f7ec2b6c53e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/tiny_overfit.yaml @@ -0,0 +1,24 @@ +# Keep the frozen 1-Mb input and every output track; shrink only learned widths. +input_length: 1048576 +target_length: 196608 +dense_bin_size: 128 +trunk_bin_size: 1024 +contact_bin_size: 2048 +stem_stride: 8 +encoder_width: 8 +channel_growth: 4 +num_downsampling_blocks: 7 +conv_kernel_size: 5 +dna_kernel_size: 15 +hidden_dim: 32 +num_sequence_blocks: 1 +num_attention_heads: 2 +attention_head_dim: 16 +mlp_expansion: 2.0 +dropout: 0.0 +head_hidden_size: 32 +junction_rank: 8 +splice_usage_tracks: 734 +junction_tracks: 734 +contact_tracks: 28 +gradient_checkpointing: false diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/train_eval.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/train_eval.py new file mode 100644 index 0000000000000000000000000000000000000000..92f9a3f315013630fdd2e03fb6840a6c77840edd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/reference/alphagenome_50m/train_eval.py @@ -0,0 +1,537 @@ +"""Train and stream-evaluate the <=50M compact AlphaGenome baseline.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import numpy as np +import torch +from torch import Tensor +from torch.nn import functional as F +from torch.utils.data import DataLoader, Subset + +from .dataset import ( + AlphaGenomeDataset, + DENSE_FAMILIES, + single_example_collate, +) +from .model import ( + ALL_FAMILIES, + CompactModel, + ModelConfig, + count_parameters, +) + + +DEFAULT_CONFIG = Path(__file__).with_name("alphagenome_50m.yaml") + + +def _device(name: str) -> torch.device: + if name == "auto": + name = "cuda" if torch.cuda.is_available() else "cpu" + if name == "cuda" and not torch.cuda.is_available(): + raise RuntimeError("--device cuda requested but CUDA is unavailable") + return torch.device(name) + + +def _autocast(device: torch.device): + enabled = device.type == "cuda" + dtype = torch.bfloat16 if enabled and torch.cuda.is_bf16_supported() else torch.float16 + return torch.autocast(device_type=device.type, dtype=dtype, enabled=enabled) + + +def _distributed() -> tuple[int, int, int]: + """(rank, local_rank, world_size) under torchrun; (0, 0, 1) otherwise.""" + world = int(os.environ.get("WORLD_SIZE", "1")) + if world > 1 and not torch.distributed.is_initialized(): + torch.distributed.init_process_group("nccl") + return int(os.environ.get("RANK", "0")), int(os.environ.get("LOCAL_RANK", "0")), world + + +def _loader(dataset, *, workers: int, shuffle: bool, sampler=None) -> DataLoader: + return DataLoader( + dataset, + batch_size=1, + shuffle=shuffle and sampler is None, + sampler=sampler, + num_workers=workers, + persistent_workers=workers > 0, + pin_memory=torch.cuda.is_available(), + collate_fn=single_example_collate, + ) + + +def _to_device(value, device: torch.device): + if isinstance(value, Tensor): + return value.to(device, non_blocking=True) + if isinstance(value, dict): + return {key: _to_device(item, device) for key, item in value.items()} + return value + + +def _match_coordinate_queries( + scoring: dict[str, Tensor], positions: Tensor, track_halves: Tensor +) -> tuple[Tensor, Tensor, Tensor]: + """Match trusted coordinate queries to typed position axes like the grader.""" + device = positions.device + tracks = scoring["track"].detach().cpu().numpy() + halves_cpu = track_halves.detach().cpu().numpy()[tracks] + # Build indices on CPU, then copy once. Writing three CUDA scalars per valid query + # otherwise launches thousands of tiny kernels per example, especially costly when + # training and evaluation share one GPU. The matching semantics are unchanged. + queries = np.zeros((len(tracks), 3), dtype=np.int64) + queries[:, 2] = tracks + valid = scoring["valid"].clone() + missing = np.ones(len(tracks), dtype=bool) + position_rows = positions.detach().cpu().numpy() + donors_cpu = scoring["donor_position"].detach().cpu().numpy() + acceptors_cpu = scoring["acceptor_position"].detach().cpu().numpy() + valid_indices = np.flatnonzero(valid.detach().cpu().numpy()) + lookups = [ + {int(coordinate): index for index, coordinate in enumerate(axis) if coordinate >= 0} + for axis in position_rows + ] + for index in valid_indices: + half = int(halves_cpu[index]) + row = lookups[2 * half].get(int(donors_cpu[index])) + column = lookups[2 * half + 1].get(int(acceptors_cpu[index])) + if row is not None and column is not None: + queries[index, 0] = row + queries[index, 1] = column + missing[index] = False + return (torch.from_numpy(queries).to(device).unsqueeze(0), valid, + torch.from_numpy(missing).to(device)) + + +def _model_inputs(sample: dict) -> tuple[dict, Tensor, Tensor]: + queries, query_valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + sample["site_positions"], + sample["junction_track_halves"], + ) + if (query_valid & missing).any(): + raise ValueError("trusted scoring coordinates do not match target positions") + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + return { + "dna": sample["dna"].unsqueeze(0), + "usage_positions": sample["usage_positions"].unsqueeze(0), + "junction_site_positions": sample["site_positions"].unsqueeze(0), + "junction_queries": queries, + }, junction_target, query_valid + + +def _losses(outputs: dict[str, Tensor], sample: dict, junction_target: Tensor, + junction_valid: Tensor) -> dict[str, Tensor]: + losses = {} + for family in DENSE_FAMILIES: + target = torch.asinh(sample["dense"][family]["values"]).unsqueeze(0) + mask = sample["dense"][family]["mask"].unsqueeze(0) + residual = torch.where(mask, outputs[family] - target, 0.0) + losses[family] = residual.square().sum() / mask.sum().clamp_min(1) + losses["splice_sites"] = F.binary_cross_entropy_with_logits( + outputs["splice_sites"], sample["splice_sites"].unsqueeze(0) + ) + usage_target = torch.asinh(sample["splice_usage"]["values"]).unsqueeze(0) + usage_mask = sample["splice_usage"]["mask"].unsqueeze(0) + usage_residual = torch.where(usage_mask, outputs["splice_usage"] - usage_target, 0.0) + losses["splice_usage"] = usage_residual.square().sum() / usage_mask.sum().clamp_min(1) + junction_residual = outputs["junctions"][0] - junction_target[0] + losses["junctions"] = ( + junction_residual[junction_valid].square().mean() + if junction_valid.any() else junction_residual.sum() * 0 + ) + contact_target = torch.asinh(sample["contact_maps"]).unsqueeze(0) + contact_mask = sample["contact_map_mask"].unsqueeze(0) + contact_residual = torch.where( + contact_mask, outputs["contact_maps"] - contact_target, 0.0 + ) + losses["contact_maps"] = contact_residual.square().sum() / contact_mask.sum().clamp_min(1) + losses["total"] = torch.stack([losses[name] for name in ALL_FAMILIES]).mean() + return losses + + +def _validation_inference( + model: CompactModel, sample: dict +) -> tuple[dict[str, Tensor], Tensor, Tensor, Tensor]: + """Use the same predicted-position path as a formal v3 submission.""" + trunk = model._trunk(sample["dna"].unsqueeze(0)) + dense_features = model._dense_features(trunk) + outputs = { + family: model.dense_heads[family](dense_features).float() + for family in DENSE_FAMILIES + } + all_sites = model._splice_site_logits(trunk) + target_start = (model.config.input_length - model.config.target_length) // 2 + outputs["splice_sites"] = all_sites[:, target_start:target_start + model.config.target_length] + usage_bins = model.splice_usage_head(dense_features).float() + usage_indices = torch.div(sample["usage_positions"], model.config.dense_bin_size, + rounding_mode="floor") + outputs["splice_usage"] = usage_bins[:, usage_indices] + predicted_positions = model.select_junction_positions(all_sites) + + matched, valid, missing = _match_coordinate_queries( + sample["junction_scoring"], + predicted_positions[0], + sample["junction_track_halves"], + ) + junction_target = torch.asinh( + sample["junction_scoring"]["target_value"] + ).unsqueeze(0) + junction_prediction = torch.zeros(len(valid), device=valid.device) + if (~missing & valid).any(): + selected = ~missing & valid + junction_prediction[selected] = model._junction_predictions( + trunk, predicted_positions, matched[:, selected] + )[0] + outputs["junctions"] = junction_prediction.unsqueeze(0) + + contact = trunk.reshape( + 1, model.config.contact_bins, -1, model.config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + outputs["contact_maps"] = (0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row])).float() + return outputs, junction_target, valid, missing + + +def _atomic_save(payload: dict, path: Path) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_name(f".{path.name}.tmp") + torch.save(payload, temporary) + os.replace(temporary, path) + + +def _sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _resume_position(payload: dict, *, examples: int, epochs: int) -> tuple[int, int]: + epoch = int(payload["epoch"]) + step = int(payload["global_step"]) + kind = payload.get("checkpoint_kind") + if kind not in (None, "epoch") or step != epoch * examples: + raise ValueError("only a completed-epoch checkpoint can resume training") + if payload.get("examples", examples) != examples or payload.get("schedule_epochs", epochs) != epochs: + raise ValueError("checkpoint data size or training schedule differs from this run") + return epoch, step + + +def train(args: argparse.Namespace) -> None: + # Multi-GPU with `torchrun --nproc-per-node N train_eval.py train ...`: one example per GPU per + # step (data parallel), rank 0 writes checkpoints and logs. Model size is bounded by the + # training budget, not by a parameter cap. + rank, local_rank, world = _distributed() + config = ModelConfig.from_yaml(args.config) + device = torch.device("cuda", local_rank) if world > 1 else _device(args.device) + if world > 1: + torch.cuda.set_device(device) + torch.manual_seed(args.seed) + model = CompactModel(config).to(device) + parameters = count_parameters(model) + optimizer = torch.optim.AdamW( + model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay + ) + scaler = torch.amp.GradScaler("cuda", enabled=device.type == "cuda") + full_dataset = AlphaGenomeDataset(args.input, "train") + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + start_epoch, global_step = 0, 0 + if args.resume: + payload = torch.load(args.resume, map_location="cpu", weights_only=True) + if payload.get("config") != config.to_dict(): + raise ValueError("checkpoint configuration does not match --config") + model.load_state_dict(payload["model"], strict=True) + optimizer.load_state_dict(payload["optimizer"]) + scaler.load_state_dict(payload.get("scaler", {})) + start_epoch, global_step = _resume_position(payload, examples=size, epochs=args.epochs) + per_rank = math.ceil(size / world) + steps = max(1, per_rank * args.epochs) + module = model + if world > 1: + model = torch.nn.parallel.DistributedDataParallel( + model, device_ids=[local_rank], find_unused_parameters=True) + sampler = (torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=world, rank=rank, + shuffle=True, seed=args.seed) + if world > 1 else None) + stop_after_epoch = args.stop_after_epoch or args.epochs + output = args.output.resolve() + output.mkdir(parents=True, exist_ok=True) + started = time.monotonic() + if rank == 0: + print(json.dumps({"event": "train_start", "parameters": parameters, "examples": size, + "device": str(device), "world_size": world}), flush=True) + + for epoch in range(start_epoch, stop_after_epoch): + model.train() + if sampler is not None: + sampler.set_epoch(epoch) + loader = _loader(dataset, workers=args.workers, shuffle=True, sampler=sampler) + for sample in loader: + sample = _to_device(sample, device) + inputs, junction_target, junction_valid = _model_inputs(sample) + progress = min((global_step + 1) / steps, 1.0) + learning_rate = args.min_learning_rate + 0.5 * ( + args.learning_rate - args.min_learning_rate + ) * (1.0 + math.cos(math.pi * progress)) + for group in optimizer.param_groups: + group["lr"] = learning_rate + optimizer.zero_grad(set_to_none=True) + with _autocast(device): + predictions = model(**inputs) + losses = _losses(predictions, sample, junction_target, junction_valid) + if not torch.isfinite(losses["total"]): + raise FloatingPointError(f"non-finite loss at step {global_step}") + scaler.scale(losses["total"]).backward() + scaler.unscale_(optimizer) + gradient_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_gradient_norm) + scaler.step(optimizer) + scaler.update() + global_step += 1 + if rank == 0 and (global_step == 1 or global_step % args.log_every == 0): + print( + json.dumps( + { + "event": "train_step", + "epoch": epoch + 1, + "step": global_step, + "loss": float(losses["total"].detach()), + "gradient_norm": float(gradient_norm), + "learning_rate": learning_rate, + "elapsed_seconds": time.monotonic() - started, + } + ), + flush=True, + ) + if rank == 0 and global_step % args.save_every == 0: + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch, + "global_step": global_step, + "checkpoint_kind": "progress", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "progress.pt", + ) + if world > 1: + torch.distributed.barrier() + if rank != 0: + continue + _atomic_save( + { + "schema": "alphagenome-50m-real/v3", + "model": module.state_dict(), + "optimizer": optimizer.state_dict(), + "scaler": scaler.state_dict(), + "config": config.to_dict(), + "epoch": epoch + 1, + "global_step": global_step, + "checkpoint_kind": "epoch", + "examples": size, + "schedule_epochs": args.epochs, + }, + output / "last.pt", + ) + if world > 1: + torch.distributed.barrier() + torch.distributed.destroy_process_group() + if rank != 0: + return + checkpoint = output / "last.pt" + manifest = { + "schema": "alphagenome-50m-real/v3", + "checkpoint": {"path": checkpoint.name, "sha256": _sha256(checkpoint)}, + "parameters": parameters, + "config": config.to_dict(), + "output_families": list(ALL_FAMILIES), + "epochs": args.epochs, + "completed_epochs": stop_after_epoch, + "global_step": global_step, + } + (output / "model_manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "train_complete", **manifest}), flush=True) + + +class PerTrackMoments: + def __init__(self, tracks: int): + self.n = np.zeros(tracks, dtype=np.int64) + self.sum_x = np.zeros(tracks, dtype=np.float64) + self.sum_y = np.zeros(tracks, dtype=np.float64) + self.sum_x2 = np.zeros(tracks, dtype=np.float64) + self.sum_y2 = np.zeros(tracks, dtype=np.float64) + self.sum_xy = np.zeros(tracks, dtype=np.float64) + + def update(self, prediction: Tensor, target: Tensor, mask: Tensor | None = None) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1, prediction.shape[-1]) + y = target.detach().double().cpu().numpy().reshape(-1, target.shape[-1]) + valid = np.ones_like(y, dtype=bool) if mask is None else mask.detach().cpu().numpy().reshape(y.shape).astype(bool) + self.n += valid.sum(axis=0) + self.sum_x += np.where(valid, x, 0).sum(axis=0, dtype=np.float64) + self.sum_y += np.where(valid, y, 0).sum(axis=0, dtype=np.float64) + self.sum_x2 += np.where(valid, x * x, 0).sum(axis=0, dtype=np.float64) + self.sum_y2 += np.where(valid, y * y, 0).sum(axis=0, dtype=np.float64) + self.sum_xy += np.where(valid, x * y, 0).sum(axis=0, dtype=np.float64) + + def update_indexed(self, prediction: Tensor, target: Tensor, track: Tensor) -> None: + x = prediction.detach().double().cpu().numpy().reshape(-1) + y = target.detach().double().cpu().numpy().reshape(-1) + indices = track.detach().cpu().numpy().reshape(-1).astype(np.int64) + self.n += np.bincount(indices, minlength=len(self.n)) + self.sum_x += np.bincount(indices, weights=x, minlength=len(self.n)) + self.sum_y += np.bincount(indices, weights=y, minlength=len(self.n)) + self.sum_x2 += np.bincount(indices, weights=x * x, minlength=len(self.n)) + self.sum_y2 += np.bincount(indices, weights=y * y, minlength=len(self.n)) + self.sum_xy += np.bincount(indices, weights=x * y, minlength=len(self.n)) + + def report(self) -> dict: + numerator = self.n * self.sum_xy - self.sum_x * self.sum_y + denominator = np.sqrt( + np.maximum(self.n * self.sum_x2 - self.sum_x**2, 0) + * np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + ) + target_ss = np.maximum(self.n * self.sum_y2 - self.sum_y**2, 0) + scorable = (self.n >= 2) & (target_ss > 0) + pearson = np.full_like(numerator, np.nan) + nonconstant = scorable & (denominator > 0) + pearson[nonconstant] = numerator[nonconstant] / denominator[nonconstant] + pearson[scorable & ~nonconstant] = 0.0 + return { + "mean_pearson": float(pearson[scorable].mean()) if scorable.any() else 0.0, + "valid_tracks": int(scorable.sum()), + "per_track_pearson": [float(value) if np.isfinite(value) else None for value in pearson], + "values_per_track": self.n.tolist(), + } + + +def evaluate(args: argparse.Namespace) -> None: + device = _device(args.device) + payload = torch.load(args.checkpoint, map_location="cpu", weights_only=True) + config = ModelConfig(**payload["config"]) + model = CompactModel(config).to(device) + model.load_state_dict(payload["model"], strict=True) + model.eval() + full_dataset = AlphaGenomeDataset(args.input, args.split) + model.set_junction_track_halves(full_dataset.junction_track_halves) + size = min(len(full_dataset), args.max_examples or len(full_dataset)) + dataset = Subset(full_dataset, range(size)) + metrics = { + **{family: PerTrackMoments(config_value) for family, config_value in { + "atac": 167, "dnase": 305, "procap": 12, "cage": 546, + "rna_seq": 667, "chip_tf": 1617, "chip_histone": 1116, + }.items()}, + "splice_sites": PerTrackMoments(4), + "splice_usage": PerTrackMoments(config.splice_usage_tracks), + "junctions": PerTrackMoments(config.junction_tracks), + "contact_maps": PerTrackMoments(config.contact_tracks), + } + with torch.inference_mode(): + for index, sample in enumerate(_loader(dataset, workers=args.workers, shuffle=False)): + sample = _to_device(sample, device) + outputs, junction_target, junction_valid, junction_missing = ( + _validation_inference(model, sample) + ) + # Match predict.py: FP32 inference is serialized to FP16 Zarr values. + outputs = {name: value.half().float() for name, value in outputs.items()} + for family in DENSE_FAMILIES: + metrics[family].update( + outputs[family][0], + torch.asinh(sample["dense"][family]["values"]), + sample["dense"][family]["mask"], + ) + metrics["splice_sites"].update(outputs["splice_sites"][0], sample["splice_sites"]) + metrics["splice_usage"].update( + outputs["splice_usage"][0], + torch.asinh(sample["splice_usage"]["values"]), + sample["splice_usage"]["mask"], + ) + junction_track = sample["junction_scoring"]["track"] + metrics["junctions"].update_indexed( + outputs["junctions"][0][junction_valid], + junction_target[0][junction_valid], + junction_track[junction_valid], + ) + metrics["contact_maps"].update( + outputs["contact_maps"][0], + torch.asinh(sample["contact_maps"]), + sample["contact_map_mask"], + ) + if index == 0 or (index + 1) % args.log_every == 0 or index + 1 == size: + print(json.dumps({"event": "evaluate_progress", "examples": index + 1, "total": size}), flush=True) + families = {name: accumulator.report() for name, accumulator in metrics.items()} + valid_family_scores = [item["mean_pearson"] for item in families.values()] + report = { + "schema": "alphagenome-50m-evaluation/v3", + "split": args.split, + "examples": size, + "metric_space": ( + "asinh continuous targets and binary splice-site targets; " + "junction Pearson uses hidden coordinate queries and predicted endpoints; " + "FP32 inference rounded to FP16 prediction values" + ), + "families": families, + "overall_mean_family_pearson": float(np.mean(valid_family_scores)), + } + args.report.parent.mkdir(parents=True, exist_ok=True) + args.report.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n") + print(json.dumps({"event": "evaluate_complete", "overall_mean_family_pearson": report["overall_mean_family_pearson"]}), flush=True) + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + common = argparse.ArgumentParser(add_help=False) + common.add_argument("--input", required=True, type=Path) + common.add_argument("--config", type=Path, default=DEFAULT_CONFIG) + common.add_argument("--workers", type=int, default=2) + common.add_argument("--max-examples", type=int, default=0) + common.add_argument("--device", choices=("auto", "cuda", "cpu"), default="auto") + common.add_argument("--log-every", type=int, default=25) + subparsers = parser.add_subparsers(dest="command", required=True) + train_parser = subparsers.add_parser("train", parents=[common]) + train_parser.add_argument("--output", required=True, type=Path) + train_parser.add_argument("--epochs", type=int, default=10) + train_parser.add_argument("--stop-after-epoch", type=int) + train_parser.add_argument("--learning-rate", type=float, default=2e-4) + train_parser.add_argument("--min-learning-rate", type=float, default=2e-5) + train_parser.add_argument("--weight-decay", type=float, default=1e-4) + train_parser.add_argument("--max-gradient-norm", type=float, default=5.0) + train_parser.add_argument("--save-every", type=int, default=250) + train_parser.add_argument("--seed", type=int, default=17) + train_parser.add_argument("--resume", type=Path) + train_parser.set_defaults(run=train) + evaluate_parser = subparsers.add_parser("evaluate", parents=[common]) + evaluate_parser.add_argument("--checkpoint", required=True, type=Path) + evaluate_parser.add_argument("--split", choices=("train", "valid", "test"), required=True) + evaluate_parser.add_argument("--report", required=True, type=Path) + evaluate_parser.set_defaults(run=evaluate) + args = parser.parse_args() + if args.workers < 0 or args.max_examples < 0: + parser.error("workers/max-examples must be nonnegative") + if args.command == "train" and ( + args.epochs < 1 + or (args.stop_after_epoch is not None + and not 1 <= args.stop_after_epoch <= args.epochs) + ): + parser.error("epochs must be positive and stop-after-epoch must be within the schedule") + args.run(args) + + +if __name__ == "__main__": + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/single_prediction.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/single_prediction.py new file mode 100644 index 0000000000000000000000000000000000000000..a47022029461e5c8097b3a653af8b3407196d67d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/single_prediction.py @@ -0,0 +1,97 @@ +"""The parent single-orientation decoder, preserving float32 until ensemble averaging.""" +import numpy as np +import torch +from prediction_contract import DENSE_TRACKS, MAX_JUNCTION_PAIRS, MAX_USAGE_SITES +from splice_output import splice_output_scores + +COMPRESSOR = None + +class ArrayCollector: + def __init__(self): + self.arrays = {} + def create_dataset(self, name, *, data, **kwargs): + self.arrays[name] = np.asarray(data) + +def single_prediction(model, dna): + config = model.config + group = ArrayCollector() + trunk = model._trunk(dna) + dense_features = model._dense_features(trunk) + for family in DENSE_TRACKS: + values = model.dense_heads[family](dense_features)[0] + group.create_dataset(family, data=values.cpu().numpy(), compressor=COMPRESSOR) + + all_site_logits = model._splice_site_logits(trunk) + target_start = (config.input_length - config.target_length) // 2 + group.create_dataset( + "splice_sites", + data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy(), + chunks=(4096, 4), compressor=COMPRESSOR, + ) + predicted_positions = model.select_junction_positions(all_site_logits)[0] + + # Usage at the model's own predicted sites (v4 is sparse; unlisted hidden sites score 0). + usage_bins = model.splice_usage_head(dense_features)[0].float() + sites = predicted_positions.cpu().numpy().astype(np.int64).reshape(-1) + sites = np.unique(sites[(sites >= target_start) & (sites < target_start + config.target_length)] + - target_start)[:MAX_USAGE_SITES] + group.create_dataset("splice_usage_positions", data=sites.astype(np.int32), compressor=COMPRESSOR) + usage = usage_bins[torch.as_tensor(sites // 128, device=usage_bins.device)] if len(sites) \ + else usage_bins[:0] + group.create_dataset("splice_usage", data=usage.cpu().numpy(), + chunks=(1024, 734), compressor=COMPRESSOR) + + # Junctions: score every donor x acceptor pair of each strand, keep the strongest pairs. + positions_batch = predicted_positions[None] + site_factors = model.junction_site_factors(trunk, positions_batch) + halves = model.junction_track_half + candidates = [] + for row_start in range(0, 512, 128): + for column_start in range(0, 512, 128): + block = torch.cat([ + model.junction_block( + trunk, positions_batch, row_start, row_start + 128, + column_start, column_start + 128, track_start, + min(track_start + 64, config.junction_tracks), site_factors, + )[0] + for track_start in range(0, config.junction_tracks, 64) + ], dim=-1).float() # [128, 128, 734] + for half in (0, 1): + donors = predicted_positions[2 * half, row_start:row_start + 128] + acceptors = predicted_positions[2 * half + 1, column_start:column_start + 128] + valid = (donors[:, None] >= 0) & (acceptors[None, :] >= 0) + # Transcript direction is determined by the public strand half. + forward = donors[:, None] < acceptors[None, :] if half == 0 else donors[:, None] > acceptors[None, :] + valid = valid & forward + strength = block[..., halves == half].amax(dim=-1).masked_fill(~valid, -torch.inf) + keep = min(MAX_JUNCTION_PAIRS // 2, int(valid.sum())) + if not keep: + continue + best = strength.flatten().topk(keep) + rows, columns = best.indices // 128, best.indices % 128 + candidates.append((best.values, torch.full_like(rows, half), donors[rows], + acceptors[columns], block[rows, columns])) + if candidates: + score, half, donor, acceptor, values = (torch.cat(x) for x in zip(*candidates)) + order = score.topk(min(MAX_JUNCTION_PAIRS, len(score))).indices + half, donor, acceptor, values = half[order], donor[order], acceptor[order], values[order] + else: + half = donor = acceptor = torch.zeros(0, dtype=torch.long) + values = torch.zeros(0, config.junction_tracks) + group.create_dataset("junction_half", data=half.cpu().numpy().astype(np.int8), compressor=COMPRESSOR) + group.create_dataset("junction_donor", data=donor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junction_acceptor", data=acceptor.cpu().numpy().astype(np.int32), compressor=COMPRESSOR) + group.create_dataset("junctions", data=values.cpu().numpy(), + chunks=(4096, 734), compressor=COMPRESSOR) + + contact = trunk.reshape( + 1, config.contact_bins, -1, config.hidden_dim + ).mean(dim=2) + q, k = model.contact_q(contact), model.contact_k(contact) + row, column = model.contact_row, model.contact_column + contact_values = 0.5 * (q[:, row] * k[:, column] + q[:, column] * k[:, row]) + group.create_dataset( + "contact_maps", data=contact_values[0].cpu().numpy(), + chunks=(8192, 28), compressor=COMPRESSOR, + ) + return group.arrays diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_calibration.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_calibration.json new file mode 100644 index 0000000000000000000000000000000000000000..470734a574a4b9b9f2f34fd3bcae0007db3737f6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_calibration.json @@ -0,0 +1,33 @@ +{ + "training_examples": 256, + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "training_positive_counts": [ + 3968.0, + 4160.0, + 3874.0, + 3921.0 + ], + "training_prior": [ + 7.883707682291667e-05, + 8.265177408854167e-05, + 7.696946461995442e-05, + 7.790327072143555e-05 + ], + "effective_balanced_loss_prior": [ + 0.3203125, + 0.3125, + 0.30859375, + 0.30859375 + ], + "logit_offset": [ + -8.695712089538574, + -8.612334251403809, + -8.665323257446289, + -8.653263092041016 + ], + "caveat": "marginal prior correction; balanced loss is normalized separately in each window", + "link": "sigmoid(logits + training_prior_offset)", + "training_split": "train", + "seed": 1729, + "development_labels_used_to_fit_offsets": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_output.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_output.py new file mode 100644 index 0000000000000000000000000000000000000000..1081264d9cf4b4473f4fe53c97ff02b2c3b249e3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/splice_output.py @@ -0,0 +1,11 @@ +"""Shared splice output link; constants estimated only from training targets.""" +import json +from pathlib import Path +import torch + +_OFFSETS = json.loads((Path(__file__).with_name('splice_calibration.json')).read_text())['logit_offset'] + +def splice_output_scores(logits): + logits = logits.float() + offset = torch.tensor(_OFFSETS, device=logits.device, dtype=logits.dtype) + return torch.sigmoid(logits + offset) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/strand_permutations.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/strand_permutations.json new file mode 100644 index 0000000000000000000000000000000000000000..044c090773f20347601dcd4f0983c4dfd4c0ecc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/strand_permutations.json @@ -0,0 +1,5959 @@ +{ + "schema": "seqfunc-strand-mapping/v1", + "source": "public track metadata; no target values", + "source_sha256": "23450e6ff9463c64d271d7014f8a798875bc95186c36aeb34609c7dc1d62cd5f", + "permutations": { + "atac": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166 + ], + "dnase": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304 + ], + "procap": [ + 6, + 7, + 8, + 9, + 10, + 11, + 0, + 1, + 2, + 3, + 4, + 5 + ], + "cage": [ + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272 + ], + "rna_seq": [ + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666 + ], + "chip_tf": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115, + 1116, + 1117, + 1118, + 1119, + 1120, + 1121, + 1122, + 1123, + 1124, + 1125, + 1126, + 1127, + 1128, + 1129, + 1130, + 1131, + 1132, + 1133, + 1134, + 1135, + 1136, + 1137, + 1138, + 1139, + 1140, + 1141, + 1142, + 1143, + 1144, + 1145, + 1146, + 1147, + 1148, + 1149, + 1150, + 1151, + 1152, + 1153, + 1154, + 1155, + 1156, + 1157, + 1158, + 1159, + 1160, + 1161, + 1162, + 1163, + 1164, + 1165, + 1166, + 1167, + 1168, + 1169, + 1170, + 1171, + 1172, + 1173, + 1174, + 1175, + 1176, + 1177, + 1178, + 1179, + 1180, + 1181, + 1182, + 1183, + 1184, + 1185, + 1186, + 1187, + 1188, + 1189, + 1190, + 1191, + 1192, + 1193, + 1194, + 1195, + 1196, + 1197, + 1198, + 1199, + 1200, + 1201, + 1202, + 1203, + 1204, + 1205, + 1206, + 1207, + 1208, + 1209, + 1210, + 1211, + 1212, + 1213, + 1214, + 1215, + 1216, + 1217, + 1218, + 1219, + 1220, + 1221, + 1222, + 1223, + 1224, + 1225, + 1226, + 1227, + 1228, + 1229, + 1230, + 1231, + 1232, + 1233, + 1234, + 1235, + 1236, + 1237, + 1238, + 1239, + 1240, + 1241, + 1242, + 1243, + 1244, + 1245, + 1246, + 1247, + 1248, + 1249, + 1250, + 1251, + 1252, + 1253, + 1254, + 1255, + 1256, + 1257, + 1258, + 1259, + 1260, + 1261, + 1262, + 1263, + 1264, + 1265, + 1266, + 1267, + 1268, + 1269, + 1270, + 1271, + 1272, + 1273, + 1274, + 1275, + 1276, + 1277, + 1278, + 1279, + 1280, + 1281, + 1282, + 1283, + 1284, + 1285, + 1286, + 1287, + 1288, + 1289, + 1290, + 1291, + 1292, + 1293, + 1294, + 1295, + 1296, + 1297, + 1298, + 1299, + 1300, + 1301, + 1302, + 1303, + 1304, + 1305, + 1306, + 1307, + 1308, + 1309, + 1310, + 1311, + 1312, + 1313, + 1314, + 1315, + 1316, + 1317, + 1318, + 1319, + 1320, + 1321, + 1322, + 1323, + 1324, + 1325, + 1326, + 1327, + 1328, + 1329, + 1330, + 1331, + 1332, + 1333, + 1334, + 1335, + 1336, + 1337, + 1338, + 1339, + 1340, + 1341, + 1342, + 1343, + 1344, + 1345, + 1346, + 1347, + 1348, + 1349, + 1350, + 1351, + 1352, + 1353, + 1354, + 1355, + 1356, + 1357, + 1358, + 1359, + 1360, + 1361, + 1362, + 1363, + 1364, + 1365, + 1366, + 1367, + 1368, + 1369, + 1370, + 1371, + 1372, + 1373, + 1374, + 1375, + 1376, + 1377, + 1378, + 1379, + 1380, + 1381, + 1382, + 1383, + 1384, + 1385, + 1386, + 1387, + 1388, + 1389, + 1390, + 1391, + 1392, + 1393, + 1394, + 1395, + 1396, + 1397, + 1398, + 1399, + 1400, + 1401, + 1402, + 1403, + 1404, + 1405, + 1406, + 1407, + 1408, + 1409, + 1410, + 1411, + 1412, + 1413, + 1414, + 1415, + 1416, + 1417, + 1418, + 1419, + 1420, + 1421, + 1422, + 1423, + 1424, + 1425, + 1426, + 1427, + 1428, + 1429, + 1430, + 1431, + 1432, + 1433, + 1434, + 1435, + 1436, + 1437, + 1438, + 1439, + 1440, + 1441, + 1442, + 1443, + 1444, + 1445, + 1446, + 1447, + 1448, + 1449, + 1450, + 1451, + 1452, + 1453, + 1454, + 1455, + 1456, + 1457, + 1458, + 1459, + 1460, + 1461, + 1462, + 1463, + 1464, + 1465, + 1466, + 1467, + 1468, + 1469, + 1470, + 1471, + 1472, + 1473, + 1474, + 1475, + 1476, + 1477, + 1478, + 1479, + 1480, + 1481, + 1482, + 1483, + 1484, + 1485, + 1486, + 1487, + 1488, + 1489, + 1490, + 1491, + 1492, + 1493, + 1494, + 1495, + 1496, + 1497, + 1498, + 1499, + 1500, + 1501, + 1502, + 1503, + 1504, + 1505, + 1506, + 1507, + 1508, + 1509, + 1510, + 1511, + 1512, + 1513, + 1514, + 1515, + 1516, + 1517, + 1518, + 1519, + 1520, + 1521, + 1522, + 1523, + 1524, + 1525, + 1526, + 1527, + 1528, + 1529, + 1530, + 1531, + 1532, + 1533, + 1534, + 1535, + 1536, + 1537, + 1538, + 1539, + 1540, + 1541, + 1542, + 1543, + 1544, + 1545, + 1546, + 1547, + 1548, + 1549, + 1550, + 1551, + 1552, + 1553, + 1554, + 1555, + 1556, + 1557, + 1558, + 1559, + 1560, + 1561, + 1562, + 1563, + 1564, + 1565, + 1566, + 1567, + 1568, + 1569, + 1570, + 1571, + 1572, + 1573, + 1574, + 1575, + 1576, + 1577, + 1578, + 1579, + 1580, + 1581, + 1582, + 1583, + 1584, + 1585, + 1586, + 1587, + 1588, + 1589, + 1590, + 1591, + 1592, + 1593, + 1594, + 1595, + 1596, + 1597, + 1598, + 1599, + 1600, + 1601, + 1602, + 1603, + 1604, + 1605, + 1606, + 1607, + 1608, + 1609, + 1610, + 1611, + 1612, + 1613, + 1614, + 1615, + 1616 + ], + "chip_histone": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366, + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 734, + 735, + 736, + 737, + 738, + 739, + 740, + 741, + 742, + 743, + 744, + 745, + 746, + 747, + 748, + 749, + 750, + 751, + 752, + 753, + 754, + 755, + 756, + 757, + 758, + 759, + 760, + 761, + 762, + 763, + 764, + 765, + 766, + 767, + 768, + 769, + 770, + 771, + 772, + 773, + 774, + 775, + 776, + 777, + 778, + 779, + 780, + 781, + 782, + 783, + 784, + 785, + 786, + 787, + 788, + 789, + 790, + 791, + 792, + 793, + 794, + 795, + 796, + 797, + 798, + 799, + 800, + 801, + 802, + 803, + 804, + 805, + 806, + 807, + 808, + 809, + 810, + 811, + 812, + 813, + 814, + 815, + 816, + 817, + 818, + 819, + 820, + 821, + 822, + 823, + 824, + 825, + 826, + 827, + 828, + 829, + 830, + 831, + 832, + 833, + 834, + 835, + 836, + 837, + 838, + 839, + 840, + 841, + 842, + 843, + 844, + 845, + 846, + 847, + 848, + 849, + 850, + 851, + 852, + 853, + 854, + 855, + 856, + 857, + 858, + 859, + 860, + 861, + 862, + 863, + 864, + 865, + 866, + 867, + 868, + 869, + 870, + 871, + 872, + 873, + 874, + 875, + 876, + 877, + 878, + 879, + 880, + 881, + 882, + 883, + 884, + 885, + 886, + 887, + 888, + 889, + 890, + 891, + 892, + 893, + 894, + 895, + 896, + 897, + 898, + 899, + 900, + 901, + 902, + 903, + 904, + 905, + 906, + 907, + 908, + 909, + 910, + 911, + 912, + 913, + 914, + 915, + 916, + 917, + 918, + 919, + 920, + 921, + 922, + 923, + 924, + 925, + 926, + 927, + 928, + 929, + 930, + 931, + 932, + 933, + 934, + 935, + 936, + 937, + 938, + 939, + 940, + 941, + 942, + 943, + 944, + 945, + 946, + 947, + 948, + 949, + 950, + 951, + 952, + 953, + 954, + 955, + 956, + 957, + 958, + 959, + 960, + 961, + 962, + 963, + 964, + 965, + 966, + 967, + 968, + 969, + 970, + 971, + 972, + 973, + 974, + 975, + 976, + 977, + 978, + 979, + 980, + 981, + 982, + 983, + 984, + 985, + 986, + 987, + 988, + 989, + 990, + 991, + 992, + 993, + 994, + 995, + 996, + 997, + 998, + 999, + 1000, + 1001, + 1002, + 1003, + 1004, + 1005, + 1006, + 1007, + 1008, + 1009, + 1010, + 1011, + 1012, + 1013, + 1014, + 1015, + 1016, + 1017, + 1018, + 1019, + 1020, + 1021, + 1022, + 1023, + 1024, + 1025, + 1026, + 1027, + 1028, + 1029, + 1030, + 1031, + 1032, + 1033, + 1034, + 1035, + 1036, + 1037, + 1038, + 1039, + 1040, + 1041, + 1042, + 1043, + 1044, + 1045, + 1046, + 1047, + 1048, + 1049, + 1050, + 1051, + 1052, + 1053, + 1054, + 1055, + 1056, + 1057, + 1058, + 1059, + 1060, + 1061, + 1062, + 1063, + 1064, + 1065, + 1066, + 1067, + 1068, + 1069, + 1070, + 1071, + 1072, + 1073, + 1074, + 1075, + 1076, + 1077, + 1078, + 1079, + 1080, + 1081, + 1082, + 1083, + 1084, + 1085, + 1086, + 1087, + 1088, + 1089, + 1090, + 1091, + 1092, + 1093, + 1094, + 1095, + 1096, + 1097, + 1098, + 1099, + 1100, + 1101, + 1102, + 1103, + 1104, + 1105, + 1106, + 1107, + 1108, + 1109, + 1110, + 1111, + 1112, + 1113, + 1114, + 1115 + ], + "contact_maps": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27 + ], + "splice_sites": [ + 2, + 3, + 0, + 1 + ], + "splice_usage": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ], + "junctions": [ + 367, + 368, + 369, + 370, + 371, + 372, + 373, + 374, + 375, + 376, + 377, + 378, + 379, + 380, + 381, + 382, + 383, + 384, + 385, + 386, + 387, + 388, + 389, + 390, + 391, + 392, + 393, + 394, + 395, + 396, + 397, + 398, + 399, + 400, + 401, + 402, + 403, + 404, + 405, + 406, + 407, + 408, + 409, + 410, + 411, + 412, + 413, + 414, + 415, + 416, + 417, + 418, + 419, + 420, + 421, + 422, + 423, + 424, + 425, + 426, + 427, + 428, + 429, + 430, + 431, + 432, + 433, + 434, + 435, + 436, + 437, + 438, + 439, + 440, + 441, + 442, + 443, + 444, + 445, + 446, + 447, + 448, + 449, + 450, + 451, + 452, + 453, + 454, + 455, + 456, + 457, + 458, + 459, + 460, + 461, + 462, + 463, + 464, + 465, + 466, + 467, + 468, + 469, + 470, + 471, + 472, + 473, + 474, + 475, + 476, + 477, + 478, + 479, + 480, + 481, + 482, + 483, + 484, + 485, + 486, + 487, + 488, + 489, + 490, + 491, + 492, + 493, + 494, + 495, + 496, + 497, + 498, + 499, + 500, + 501, + 502, + 503, + 504, + 505, + 506, + 507, + 508, + 509, + 510, + 511, + 512, + 513, + 514, + 515, + 516, + 517, + 518, + 519, + 520, + 521, + 522, + 523, + 524, + 525, + 526, + 527, + 528, + 529, + 530, + 531, + 532, + 533, + 534, + 535, + 536, + 537, + 538, + 539, + 540, + 541, + 542, + 543, + 544, + 545, + 546, + 547, + 548, + 549, + 550, + 551, + 552, + 553, + 554, + 555, + 556, + 557, + 558, + 559, + 560, + 561, + 562, + 563, + 564, + 565, + 566, + 567, + 568, + 569, + 570, + 571, + 572, + 573, + 574, + 575, + 576, + 577, + 578, + 579, + 580, + 581, + 582, + 583, + 584, + 585, + 586, + 587, + 588, + 589, + 590, + 591, + 592, + 593, + 594, + 595, + 596, + 597, + 598, + 599, + 600, + 601, + 602, + 603, + 604, + 605, + 606, + 607, + 608, + 609, + 610, + 611, + 612, + 613, + 614, + 615, + 616, + 617, + 618, + 619, + 620, + 621, + 622, + 623, + 624, + 625, + 626, + 627, + 628, + 629, + 630, + 631, + 632, + 633, + 634, + 635, + 636, + 637, + 638, + 639, + 640, + 641, + 642, + 643, + 644, + 645, + 646, + 647, + 648, + 649, + 650, + 651, + 652, + 653, + 654, + 655, + 656, + 657, + 658, + 659, + 660, + 661, + 662, + 663, + 664, + 665, + 666, + 667, + 668, + 669, + 670, + 671, + 672, + 673, + 674, + 675, + 676, + 677, + 678, + 679, + 680, + 681, + 682, + 683, + 684, + 685, + 686, + 687, + 688, + 689, + 690, + 691, + 692, + 693, + 694, + 695, + 696, + 697, + 698, + 699, + 700, + 701, + 702, + 703, + 704, + 705, + 706, + 707, + 708, + 709, + 710, + 711, + 712, + 713, + 714, + 715, + 716, + 717, + 718, + 719, + 720, + 721, + 722, + 723, + 724, + 725, + 726, + 727, + 728, + 729, + 730, + 731, + 732, + 733, + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9, + 10, + 11, + 12, + 13, + 14, + 15, + 16, + 17, + 18, + 19, + 20, + 21, + 22, + 23, + 24, + 25, + 26, + 27, + 28, + 29, + 30, + 31, + 32, + 33, + 34, + 35, + 36, + 37, + 38, + 39, + 40, + 41, + 42, + 43, + 44, + 45, + 46, + 47, + 48, + 49, + 50, + 51, + 52, + 53, + 54, + 55, + 56, + 57, + 58, + 59, + 60, + 61, + 62, + 63, + 64, + 65, + 66, + 67, + 68, + 69, + 70, + 71, + 72, + 73, + 74, + 75, + 76, + 77, + 78, + 79, + 80, + 81, + 82, + 83, + 84, + 85, + 86, + 87, + 88, + 89, + 90, + 91, + 92, + 93, + 94, + 95, + 96, + 97, + 98, + 99, + 100, + 101, + 102, + 103, + 104, + 105, + 106, + 107, + 108, + 109, + 110, + 111, + 112, + 113, + 114, + 115, + 116, + 117, + 118, + 119, + 120, + 121, + 122, + 123, + 124, + 125, + 126, + 127, + 128, + 129, + 130, + 131, + 132, + 133, + 134, + 135, + 136, + 137, + 138, + 139, + 140, + 141, + 142, + 143, + 144, + 145, + 146, + 147, + 148, + 149, + 150, + 151, + 152, + 153, + 154, + 155, + 156, + 157, + 158, + 159, + 160, + 161, + 162, + 163, + 164, + 165, + 166, + 167, + 168, + 169, + 170, + 171, + 172, + 173, + 174, + 175, + 176, + 177, + 178, + 179, + 180, + 181, + 182, + 183, + 184, + 185, + 186, + 187, + 188, + 189, + 190, + 191, + 192, + 193, + 194, + 195, + 196, + 197, + 198, + 199, + 200, + 201, + 202, + 203, + 204, + 205, + 206, + 207, + 208, + 209, + 210, + 211, + 212, + 213, + 214, + 215, + 216, + 217, + 218, + 219, + 220, + 221, + 222, + 223, + 224, + 225, + 226, + 227, + 228, + 229, + 230, + 231, + 232, + 233, + 234, + 235, + 236, + 237, + 238, + 239, + 240, + 241, + 242, + 243, + 244, + 245, + 246, + 247, + 248, + 249, + 250, + 251, + 252, + 253, + 254, + 255, + 256, + 257, + 258, + 259, + 260, + 261, + 262, + 263, + 264, + 265, + 266, + 267, + 268, + 269, + 270, + 271, + 272, + 273, + 274, + 275, + 276, + 277, + 278, + 279, + 280, + 281, + 282, + 283, + 284, + 285, + 286, + 287, + 288, + 289, + 290, + 291, + 292, + 293, + 294, + 295, + 296, + 297, + 298, + 299, + 300, + 301, + 302, + 303, + 304, + 305, + 306, + 307, + 308, + 309, + 310, + 311, + 312, + 313, + 314, + 315, + 316, + 317, + 318, + 319, + 320, + 321, + 322, + 323, + 324, + 325, + 326, + 327, + 328, + 329, + 330, + 331, + 332, + 333, + 334, + 335, + 336, + 337, + 338, + 339, + 340, + 341, + 342, + 343, + 344, + 345, + 346, + 347, + 348, + 349, + 350, + 351, + 352, + 353, + 354, + 355, + 356, + 357, + 358, + 359, + 360, + 361, + 362, + 363, + 364, + 365, + 366 + ] + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_batched.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_batched.py new file mode 100644 index 0000000000000000000000000000000000000000..df00b798c655555704cd6c8abbd7e4eacedd0ba3 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_batched.py @@ -0,0 +1,168 @@ +"""Resumable single-H100 training with equal-per-example batched losses. + +global_step counts training examples, optimizer_steps counts updates. Every epoch +uses the complete deterministic permutation, including its final partial batch. +""" +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset +from reference.alphagenome_50m.dataset import AlphaGenomeDataset +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _to_device +from batched_training import list_collate, prepare_batch, equal_example_losses + +def main(): + p=argparse.ArgumentParser(description=__doc__) + p.add_argument('--input',type=Path,default=Path('/workspace/data')) + p.add_argument('--output',type=Path,required=True) + p.add_argument('--config',type=Path,required=True) + p.add_argument('--resume',type=Path) + p.add_argument('--batch-size',type=int,default=8) + p.add_argument('--epochs',type=int,default=14) + p.add_argument('--stop-at',type=int,default=0) + p.add_argument('--workers',type=int,default=6) + p.add_argument('--threads',type=int,default=2) + p.add_argument('--seed',type=int,default=17) + p.add_argument('--learning-rate',type=float,default=2e-4) + p.add_argument('--min-learning-rate',type=float,default=2e-5) + p.add_argument('--warmup',type=int,default=4000,help='Warmup in training examples') + p.add_argument('--objective',choices=['balanced_splice','variance_balanced'],default='balanced_splice') + p.add_argument('--loss-scales',type=Path) + p.add_argument('--save-every',type=int,default=10000,help='Training examples between progress checkpoints') + p.add_argument('--keep-every',type=int,default=50000) + p.add_argument('--log-every',type=int,default=50,help='Optimizer updates between logs') + args=p.parse_args() + if min(args.batch_size,args.epochs,args.save_every,args.log_every) < 1: + raise ValueError('batch, epoch, checkpoint and log intervals must be positive') + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(args.threads);torch.set_num_interop_threads(1);blosc.set_nthreads(1) + torch.manual_seed(args.seed) + device=torch.device('cuda:0') + config=ModelConfig.from_yaml(args.config) + model=CompactModel(config).to(device) + optimizer=torch.optim.AdamW(model.parameters(),lr=args.learning_rate,weight_decay=1e-4) + dataset=AlphaGenomeDataset(args.input,'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True,exist_ok=True) + total=len(dataset)*args.epochs + step,updates,elapsed_before=0,0,0. + lineage,scales={},{} + if args.loss_scales: + scale_source=json.loads(args.loss_scales.read_text()) + assert scale_source['split']=='train' + scales=scale_source['family_loss_scales'] + lineage['loss_scale_provenance']={k:v for k,v in scale_source.items() if k!='junction_moments'} + if args.resume: + state=torch.load(args.resume,map_location='cpu',weights_only=True) + if state['schema'] not in ['seqfunc-single/v1','seqfunc-batched/v1']: + raise ValueError('only locally trained checkpoints may resume') + if ModelConfig(**state['config']).to_dict()!=config.to_dict() or state['examples']!=len(dataset): + raise ValueError('resume architecture or dataset differs') + if state['schedule_epochs']!=args.epochs or state['objective']!=args.objective or state['seed']!=args.seed: + raise ValueError('resume schedule, objective or seed differs') + previous=state.get('training_arguments',{}) + for field in ['learning_rate','min_learning_rate','warmup']: + if field in previous and previous[field]!=getattr(args,field): + raise ValueError(f'resume {field} differs') + if state['schema']=='seqfunc-batched/v1' and previous['batch_size']!=args.batch_size: + raise ValueError('batched resume must keep its batch size') + if scales and scales!=state.get('family_loss_scales',{}): + raise ValueError('resume family scales differ') + scales=state.get('family_loss_scales',{}) + lineage.update({k:state[k] for k in ['intervention','interventions','loss_scale_provenance'] if k in state}) + model.load_state_dict(state['model'],strict=True) + optimizer.load_state_dict(state['optimizer']) + step=state['global_step'];updates=state.get('optimizer_steps',step) + elapsed_before=state.get('elapsed_seconds',0.) + torch.set_rng_state(state['rng_cpu']);torch.cuda.set_rng_state(state['rng_cuda'],device) + if state['schema']=='seqfunc-single/v1': + with args.resume.open('rb') as f: + digest=hashlib.file_digest(f,'sha256').hexdigest() + history=list(lineage.get('interventions',[])) + if lineage.get('intervention') and lineage['intervention'] not in history: + history.append(lineage['intervention']) + change={'kind':'batch_size_change','from':1,'to':args.batch_size,'at_example':step, + 'parent_sha256':digest,'parameters_optimizer_rng_unchanged_at_start':True} + history.append(change);lineage.update(intervention=change,interventions=history) + if args.objective=='variance_balanced': + expected=set(model.dense_heads)|{'splice_usage','junctions','contact_maps'} + if set(scales)!=expected or not all(math.isfinite(v) and v>0 for v in scales.values()): + raise ValueError('fixed positive training-derived scales required') + elif scales: + raise ValueError('unscaled objective cannot carry family scales') + start=time.monotonic() + def emit(event,**values): + print(json.dumps({'event':event,'step':step,'optimizer_steps':updates, + 'elapsed_seconds':elapsed_before+time.monotonic()-start,**values}),flush=True) + def save(name): + _atomic_save({**lineage,'schema':'seqfunc-batched/v1','model':model.state_dict(), + 'optimizer':optimizer.state_dict(),'config':config.to_dict(), + 'global_step':step,'optimizer_steps':updates,'epoch':step//len(dataset),'offset':step%len(dataset), + 'examples':len(dataset),'schedule_epochs':args.epochs,'objective':args.objective,'seed':args.seed, + 'family_loss_scales':scales, + 'training_arguments':{k:str(v) if isinstance(v,Path) else v for k,v in vars(args).items()}, + 'rng_cpu':torch.get_rng_state(),'rng_cuda':torch.cuda.get_rng_state(device), + 'elapsed_seconds':elapsed_before+time.monotonic()-start, + 'provenance':{'data':str(args.input),'split':'train','pretrained':False,'world_size':1, + 'device':torch.cuda.get_device_name(0),'batch_reduction':'equal per example', + 'train_manifest_sha256':hashlib.sha256((args.input/'public/train.parquet').read_bytes()).hexdigest()}, + },args.output/name) + emit('train_start',parameters=count_parameters(model),examples=len(dataset),batch_size=args.batch_size, + objective=args.objective,config=config.to_dict(),schedule_examples=total,family_loss_scales=scales) + if args.stop_at and step>=args.stop_at: + raise ValueError('stop-at must exceed the resumed example count') + model.train() + for epoch in range(step//len(dataset),args.epochs): + order=torch.randperm(len(dataset),generator=torch.Generator().manual_seed(args.seed+epoch)).tolist() + offset=step%len(dataset) + emit('epoch_order',epoch=epoch+1,offset=offset,count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order,dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader=DataLoader(Subset(dataset,order[offset:]),batch_size=args.batch_size,shuffle=False, + num_workers=args.workers,pin_memory=True,collate_fn=list_collate, + persistent_workers=False,prefetch_factor=2 if args.workers else None) + for cpu_samples in loader: + if args.stop_at: + cpu_samples=cpu_samples[:args.stop_at-step] + inputs,targets,valid=prepare_batch(cpu_samples) + inputs=_to_device(inputs,device) + targets=[x.to(device,non_blocking=True) for x in targets] + valid=[x.to(device,non_blocking=True) for x in valid] + samples=[_to_device(s,device) for s in cpu_samples] + progress=min(max(step-args.warmup,0)/max(total-args.warmup,1),1) + lr=args.min_learning_rate+.5*(args.learning_rate-args.min_learning_rate)*(1+math.cos(math.pi*progress)) + lr*=min((step+len(samples))/max(args.warmup,1),1) + for group in optimizer.param_groups: group['lr']=lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda',dtype=torch.bfloat16): + outputs=model(**inputs) + losses,raw=equal_example_losses(outputs,samples,targets,valid,scales=scales) + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm=torch.nn.utils.clip_grad_norm_(model.parameters(),5,error_if_nonfinite=True) + optimizer.step() + before=step;step+=len(samples);updates+=1 + if updates==1 or updates%args.log_every==0: + emit('train_step',epoch=epoch+1,example_ids=[s['example_id'] for s in samples], + learning_rate=lr,gradient_norm=float(norm),actual_batch_size=len(samples), + losses={k:float(v.detach()) for k,v in losses.items()}, + raw_losses={k:float(v.detach()) for k,v in raw.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step//args.save_every>before//args.save_every: save('progress.pt') + if args.keep_every and step//args.keep_every>before//args.keep_every: save(f'step-{step:06d}.pt') + if step%len(dataset)==0: + save('last.pt');save(f'epoch-{epoch+1:02d}.pt') + emit('epoch_complete',completed_epochs=epoch+1) + if args.stop_at and step>=args.stop_at: + save('last.pt');emit('train_segment_complete',completed_epochs=step//len(dataset),offset=step%len(dataset)) + return + emit('train_complete') + +if __name__=='__main__': main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_single.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_single.py new file mode 100644 index 0000000000000000000000000000000000000000..03024393206b918eac3ae23a87c09173848b0d15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/train_single.py @@ -0,0 +1,158 @@ +"""Resumable single-H100 training; every epoch permutes every eligible training interval.""" +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from pathlib import Path +import time + +import torch +from numcodecs import blosc +from torch.utils.data import DataLoader, Subset + +from reference.alphagenome_50m.dataset import AlphaGenomeDataset, single_example_collate +from reference.alphagenome_50m.model import CompactModel, ModelConfig, count_parameters +from reference.alphagenome_50m.train_eval import _atomic_save, _losses, _model_inputs, _to_device + + +def balanced_splice_loss(logits, target): + """Equal positive/negative contributions per type, using only training labels. + + An absent class contributes its negative loss rather than dividing by zero or being dropped. + """ + dims = tuple(range(target.ndim - 1)) + positive = target.sum(dims) + negative = (1 - target).sum(dims) + pos_loss = (torch.nn.functional.softplus(-logits) * target).sum(dims) / positive.clamp_min(1) + neg_loss = (torch.nn.functional.softplus(logits) * (1 - target)).sum(dims) / negative.clamp_min(1) + return torch.where(positive > 0, 0.5 * (pos_loss + neg_loss), neg_loss).mean() + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('--input', type=Path, default=Path('/workspace/data')) + p.add_argument('--output', type=Path, required=True) + p.add_argument('--config', type=Path, default=Path(__file__).parent / 'reference/alphagenome_50m/alphagenome_50m.yaml') + p.add_argument('--resume', type=Path) + p.add_argument('--objective', choices=['reference', 'balanced_splice'], default='balanced_splice') + p.add_argument('--epochs', type=int, default=4) + p.add_argument('--stop-at', type=int, default=0, help='Stop after this total number of examples, preserving exact sampler position') + p.add_argument('--workers', type=int, default=8) + p.add_argument('--threads', type=int, default=4) + p.add_argument('--seed', type=int, default=17) + p.add_argument('--learning-rate', type=float, default=2e-4) + p.add_argument('--min-learning-rate', type=float, default=2e-5) + p.add_argument('--warmup', type=int, default=500) + p.add_argument('--save-every', type=int, default=2000) + p.add_argument('--keep-every', type=int, default=10000, help='Keep immutable intermediate checkpoints') + p.add_argument('--log-every', type=int, default=100) + args = p.parse_args() + torch.set_num_threads(args.threads) + torch.set_num_interop_threads(1) + blosc.set_nthreads(1) + if hasattr(os, 'sched_getaffinity'): + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.manual_seed(args.seed) + device = torch.device('cuda:0') + config = ModelConfig.from_yaml(args.config) + model = CompactModel(config).to(device) + optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=1e-4) + dataset = AlphaGenomeDataset(args.input, 'train') + model.set_junction_track_halves(dataset.junction_track_halves) + args.output.mkdir(parents=True, exist_ok=True) + total = len(dataset) * args.epochs + step, elapsed_before = 0, 0.0 + lineage = {} + if args.resume: + state = torch.load(args.resume, map_location='cpu', weights_only=True) + lineage = {k: state[k] for k in ['intervention', 'interventions'] if k in state} + if state['config'] != config.to_dict() or state['examples'] != len(dataset): + raise ValueError('resume config or dataset size differs') + if state['schedule_epochs'] != args.epochs or state['objective'] != args.objective: + raise ValueError('resume schedule or objective differs') + if state['seed'] != args.seed: + raise ValueError('resume shuffle seed differs') + previous = state.get('training_arguments', {}) + for field in ['learning_rate', 'min_learning_rate', 'warmup']: + if field in previous and previous[field] != getattr(args, field): + raise ValueError(f'resume {field} differs') + model.load_state_dict(state['model'], strict=True) + optimizer.load_state_dict(state['optimizer']) + step = state['global_step'] + elapsed_before = state.get('elapsed_seconds', 0.0) + torch.set_rng_state(state['rng_cpu']) + torch.cuda.set_rng_state(state['rng_cuda'], device) + start = time.monotonic() + def emit(event, **values): + record = {'event': event, 'step': step, 'elapsed_seconds': elapsed_before + time.monotonic() - start, **values} + print(json.dumps(record), flush=True) + def save(name): + _atomic_save({ + **lineage, + 'schema': 'seqfunc-single/v1', 'model': model.state_dict(), + 'optimizer': optimizer.state_dict(), 'config': config.to_dict(), + 'global_step': step, 'epoch': step // len(dataset), 'offset': step % len(dataset), + 'examples': len(dataset), 'schedule_epochs': args.epochs, + 'objective': args.objective, 'seed': args.seed, + 'training_arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(args).items()}, + 'rng_cpu': torch.get_rng_state(), 'rng_cuda': torch.cuda.get_rng_state(device), + 'elapsed_seconds': elapsed_before + time.monotonic() - start, + 'provenance': {'data': str(args.input), 'split': 'train', 'pretrained': False, + 'device': torch.cuda.get_device_name(0), 'world_size': 1}, + }, args.output / name) + emit('train_start', parameters=count_parameters(model), examples=len(dataset), objective=args.objective, + config=config.to_dict(), schedule_examples=total, device=torch.cuda.get_device_name(0)) + model.train() + for epoch in range(step // len(dataset), args.epochs): + generator = torch.Generator().manual_seed(args.seed + epoch) + order = torch.randperm(len(dataset), generator=generator).tolist() + offset = step % len(dataset) + emit('epoch_order', epoch=epoch + 1, offset=offset, count=len(order), + permutation_sha256=hashlib.sha256(torch.tensor(order, dtype=torch.int64).numpy().tobytes()).hexdigest()) + loader = DataLoader(Subset(dataset, order[offset:]), batch_size=1, shuffle=False, + num_workers=args.workers, pin_memory=True, collate_fn=single_example_collate, + persistent_workers=False, prefetch_factor=2 if args.workers else None) + for sample in loader: + sample = _to_device(sample, device) + inputs, target, valid = _model_inputs(sample) + progress = min(max(step - args.warmup, 0) / max(total - args.warmup, 1), 1) + lr = args.min_learning_rate + 0.5 * (args.learning_rate - args.min_learning_rate) * (1 + math.cos(math.pi * progress)) + lr *= min((step + 1) / max(args.warmup, 1), 1) + for group in optimizer.param_groups: + group['lr'] = lr + optimizer.zero_grad(set_to_none=True) + with torch.autocast('cuda', dtype=torch.bfloat16): + outputs = model(**inputs) + losses = _losses(outputs, sample, target, valid) + if args.objective == 'balanced_splice': + losses['splice_sites'] = balanced_splice_loss(outputs['splice_sites'], sample['splice_sites'][None]) + losses['total'] = torch.stack([v for k, v in losses.items() if k != 'total']).mean() + if not torch.isfinite(losses['total']): + raise FloatingPointError(f'nonfinite loss at {step}') + losses['total'].backward() + norm = torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0, error_if_nonfinite=True) + optimizer.step() + step += 1 + if step == 1 or step % args.log_every == 0: + emit('train_step', epoch=epoch + 1, example_id=sample['example_id'], learning_rate=lr, + gradient_norm=float(norm), losses={k: float(v.detach()) for k, v in losses.items()}, + peak_gpu_bytes=torch.cuda.max_memory_allocated()) + if step % args.save_every == 0: + save('progress.pt') + if args.keep_every and step % args.keep_every == 0: + save(f'step-{step:06d}.pt') + if args.stop_at and step >= args.stop_at: + save('last.pt') + emit('train_segment_complete', completed_epochs=step // len(dataset), offset=step % len(dataset)) + return + save('last.pt') + save(f'epoch-{epoch + 1:02d}.pt') + emit('epoch_complete', completed_epochs=epoch + 1) + emit('train_complete') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_assets_manifest.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_assets_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..8e314c842cca039da1d2be8da7fff6a146178651 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_assets_manifest.json @@ -0,0 +1,21 @@ +{ + "schema": "seqfunc-training-assets/v1", + "files": { + "training_history/block-checkpoint-cpu-check.json": "d099b0a4d4d6cdd1aab33a6d8efaf3b4a5bddb2be4db0563eed83b3f849a31ef", + "training_history/block-checkpoint-gpu-check.json": "b2e598359fad26fd63c4e7487997432987cfd8c836680726458f62d137df04ce", + "training_history/bytecache-integration_check.json": "a712f359807b414956a87f2486bbc3c6e9288727d1f79e5d0b6fdf321fd9a67b", + "training_history/bytecache-io-amendment.json": "0cb0d61d2ed7de26df12a62c7d4e4df2b77c6704b786c400eb216ad3735a96ac", + "training_history/bytecache-migration-plan.json": "e90379ec449255094a5789321b14ec9cec7692d61fba6900759510a34cd13035", + "training_history/bytecache-preparation.json": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "training_history/bytecache-timing_report.json": "25e6f24f940b815f4765255a8ebb81ad63eda28a4e7c9c412d81b32d78cf1000", + "training_history/execution_change.json": "5035b72c3dbc0e8d73eba0133232d21180cb03325630a1d95f94d35ab90fcdd0", + "training_history/memory-check-interruption.json": "895de55a7ff7871b2e6a053f6033aee02d6f96fa98eb525c80e3a1eeae407d16", + "training_history/migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "training_history/model_before_block_checkpoint.py": "2a185c9eea285b0dd6e2e05a00e8181c62887c6e26ab9e0e1bd1ca7c8eef66cf", + "training_history/parallel-io-amendment.json": "97b44cda7f5d3f2b70a91f55cce8772ccc46f4a24a537f07f7aa12cc22977a79", + "training_history/parallel-io-integration-check.json": "aa807f5f087b698e1132eadc6aa8dee82ecffc82d06971734836f6896de0174f", + "training_history/parallel-loader4x4-train256.json": "3ef8e42e6e90fee4320add09c54ea8f83fa426f5e9c8b76c2abb00ab5ae62412", + "training_history/reader-before-parallel-io.py": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "training_history/run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_config.yaml b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3de85396b2b503e0e87c323a7d09b18f3cf7cec1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_config.yaml @@ -0,0 +1,32 @@ +attention_head_dim: 64 +channel_growth: 96 +contact_bin_size: 2048 +contact_tracks: 28 +conv_kernel_size: 5 +dense_bin_size: 128 +dna_kernel_size: 15 +dropout: 0.0 +enable_pairwise: false +encoder_width: 96 +gradient_checkpointing: true +head_hidden_size: 768 +hidden_dim: 1024 +input_length: 1048576 +junction_rank: 32 +junction_tracks: 734 +local_context_blocks: 6 +local_to_trunk: true +local_width: 64 +mlp_expansion: 4.0 +num_attention_heads: 16 +num_downsampling_blocks: 7 +num_sequence_blocks: 16 +pair_downsample: 16 +pair_update_interval: 2 +pair_width: 32 +residual_rms_cap: 0.1 +splice_context_mode: offset +splice_usage_tracks: 734 +stem_stride: 8 +target_length: 196608 +trunk_bin_size: 1024 diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-cpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-cpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..588d9318734995508b1ece2948b790e2d0ce5172 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-cpu-check.json @@ -0,0 +1,10 @@ +{ + "all_11_training_and_evaluation_outputs_bit_exact": true, + "max_absolute_gradient_error": 0.0, + "nonzero_local_heads_and_bridge": true, + "six_local_blocks": true, + "active_branch_caps": true, + "unequal_and_empty_event_queries": true, + "precision": "float32", + "device": "CPU" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-gpu-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-gpu-check.json new file mode 100644 index 0000000000000000000000000000000000000000..144a03aa8b8a1b0624d424c2cf577788896fdca1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/block-checkpoint-gpu-check.json @@ -0,0 +1,198 @@ +{ + "weights_discarded": true, + "source_training_examples": 1000, + "same_nonzero_trained_parameters": true, + "precision": "bfloat16 training", + "batch_size": 4, + "input_bp": 1048576, + "training_indices": [ + 37958, + 33184, + 21022, + 30377 + ], + "objective": "variance_balanced", + "benchmarks": [ + { + "policy": "whole_path", + "step_seconds": [ + 1.681253916991409, + 0.45803058601450175, + 0.4365358139912132, + 0.439815385005204, + 0.4368974999815691, + 0.4549180749745574 + ], + "steady_seconds_per_example": 0.11130986799835227, + "peak_allocated_GiB": 40.11026668548584, + "all_11_outputs_and_losses_bit_exact": null, + "all_parameter_gradients_bit_exact": null, + "max_absolute_gradient_error": null, + "gradient_relative_L2_error": null, + "gradient_cosine": null, + "gradient_difference_top10": [] + }, + { + "policy": "whole_path_repeat", + "step_seconds": [ + 2.481021937011974, + 0.4393464530003257, + 0.48240915301721543, + 0.43610669201007113, + 0.4453651010117028, + 0.44543533099931665 + ], + "steady_seconds_per_example": 0.11243313650193158, + "peak_allocated_GiB": 40.10498094558716, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009534162545958643, + "gradient_cosine": 0.999999545498656, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011708050442353522, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.stem.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0011145246032800173, + "reference_L2": 0.015316175233840698 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0008722401804490078, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007519333551147762, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0029496872658462915, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00357048619228398, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.002108935810380534, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0008152801893268156, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.1.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0007619146919563889, + "reference_L2": 0.02294275974276798 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0004888368419478168, + "reference_L2": 0.01560724126448807 + } + ] + }, + { + "policy": "per_block", + "step_seconds": [ + 2.334278088004794, + 0.4152768689964432, + 0.41614379500970244, + 0.4367661690048408, + 0.41646649700123817, + 0.4161676469957456 + ], + "steady_seconds_per_example": 0.10504104885039851, + "peak_allocated_GiB": 15.99086332321167, + "all_11_outputs_and_losses_bit_exact": true, + "all_parameter_gradients_bit_exact": false, + "max_absolute_gradient_error": 1.52587890625e-05, + "gradient_relative_L2_error": 0.0009547321603675331, + "gradient_cosine": 0.9999995442418171, + "gradient_difference_top10": [ + { + "parameter": "local_path.stem.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0010200615858954154, + "reference_L2": 0.038250174700771404 + }, + { + "parameter": "local_path.blocks.0.contract.bias", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.0007522781995975294, + "reference_L2": 0.020936495583310583 + }, + { + "parameter": "local_projection.weight", + "max_absolute_error": 1.52587890625e-05, + "relative_L2_error": 0.002942487910648785, + "reference_L2": 0.1874866973862457 + }, + { + "parameter": "encoder.stem.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0036607697988758504, + "reference_L2": 0.02370183607081833 + }, + { + "parameter": "encoder.stem.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.00198984390485536, + "reference_L2": 0.009654671844638604 + }, + { + "parameter": "local_path.blocks.0.expand.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0009106435365869254, + "reference_L2": 0.01960642126710496 + }, + { + "parameter": "local_path.blocks.0.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006495164608475437, + "reference_L2": 0.03117670960545457 + }, + { + "parameter": "local_path.blocks.1.contract.bias", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0005038814832610538, + "reference_L2": 0.01560724126448807 + }, + { + "parameter": "local_path.blocks.2.contract.weight", + "max_absolute_error": 7.62939453125e-06, + "relative_L2_error": 0.0006085661912956853, + "reference_L2": 0.02301358982961947 + }, + { + "parameter": "encoder.blocks.0.conv.weight", + "max_absolute_error": 4.121946403756738e-06, + "relative_L2_error": 0.00431087128397906, + "reference_L2": 0.027548808788221385 + } + ] + } + ], + "additional_gpu_consumers": [], + "complete": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-integration_check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-integration_check.json new file mode 100644 index 0000000000000000000000000000000000000000..1377348921feb42310e66b133df8ebee9f0d591c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-integration_check.json @@ -0,0 +1,166 @@ +{ + "created_utc": "2026-10-25T03:48:07.725319+00:00", + "preparation_sha256": "aeb7f326fdddd2dfce6e9df162fd214565b11bb03049d982f0253290d7217caa", + "source_sha256": "b0921bfd3d3dc47f3cb82debc0b492d8291693c40e0cef028764c713c1257e85", + "original_full256_report_sha256": "e6620ff13b00a36857dad22fa0c9d1ad5c2b266eca70773650d280998ba20005", + "examples": 16, + "worker_processes": 4, + "ordered_indices": [ + 38509, + 39126, + 38926, + 17432, + 6551, + 22599, + 24749, + 10790, + 7022, + 30789, + 17340, + 40258, + 26222, + 22674, + 23658, + 17167 + ], + "all_returned_training_ids_and_tensor_bytes_exact": true, + "reference_and_cache_both_match_original_report": true, + "public_inference_and_validation_cache_disabled_despite_environment": true, + "default_cache_disabled": true, + "models_and_trainers_unchanged": true, + "rows": { + "reference": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ], + "cached": [ + { + "id": "agv2-8556145a51c1c60ee1ccc11e", + "sha256": "3793afb7f6108df3c7483c045a40be7bf1dc65dd49fa5cd7eaf6ce88ab83cec7" + }, + { + "id": "agv2-d6dcfb9c16994d850e7f7602", + "sha256": "bf0d4a3ecd1fa488d961dcd1ceb3b311c915512d8decece7c5426dd49714b06f" + }, + { + "id": "agv2-618847475be4ec5e05f1f4f9", + "sha256": "da8a54e2f8c77bbe46053a7d4cf009cbd52aae638c68faa05b1f11e8bed10977" + }, + { + "id": "agv2-043fd8c479dd35aa36eaa03b", + "sha256": "10f5df0a669885f382939a29b545124c6606905a6fc302a6bb5cffab31c3b4c5" + }, + { + "id": "agv2-e402a204ca2ba47b2d67412d", + "sha256": "272ea0ea2267a059e535b027edaedb866d39ffb3bded7059be0d28df59446aaf" + }, + { + "id": "agv2-83b57c27b623d818719d9a3c", + "sha256": "d9a208c403b8715bb61f3739d32db0d8e5d367549e1cce65f632fb25adf311c3" + }, + { + "id": "agv2-9f4bc2e53b583656ea30af1c", + "sha256": "283e30c384714c96dea332bbdbe7a36a9496ffe830bdff766092692f97a05aef" + }, + { + "id": "agv2-50625982f9cf30a65da05e6a", + "sha256": "18f71a8f272728f6b492eb238d5509527774ace00708f018d81e2543a1159056" + }, + { + "id": "agv2-c2ac73f0cfe05dd0667f3ae3", + "sha256": "964c067f998b523ed7e2e5e077e71f3dbbe20816fa060f977ac8ad733eda80bb" + }, + { + "id": "agv2-93075cb0a1390d16d5e3ff5f", + "sha256": "54bb0017a5f439f16b9bd771963f4d4db975215ac682952409b6127ef60d6423" + }, + { + "id": "agv2-6d48b3a065397dd93057abd7", + "sha256": "3785743dda56f1e2c4a75c38ee69923a0cf43202846cb37a7503727ff332b199" + }, + { + "id": "agv2-8776ebdc2068929446d1d127", + "sha256": "f2a1d3aa93269a1480deeeb7dcbb07d1c13b04722a5b97456f2dafd4867875f0" + }, + { + "id": "agv2-346477626eeb3a049727a743", + "sha256": "c97666f9ee8f56ff62818ddcab1ed806f7a7deb6f06c5a9dc537561a7a8a0e80" + }, + { + "id": "agv2-00537d263d837f5824cc44ee", + "sha256": "8d8cc2d1a3f69383d73a17391e65360e07fb11b4bd5f0cd1dd6ef28c2f00e46d" + }, + { + "id": "agv2-ae12b59b048cc5fc83b77de5", + "sha256": "dc90aa36afd13521189d81265ee4cdfbb7bf2aee75e24e5acba3224a8c672ef2" + }, + { + "id": "agv2-a85d0f912d8551638f7074cf", + "sha256": "db7758c1aa4b30fe6dce5d8911217063cee3a6f163ec23f458601ad355d29768" + } + ] + }, + "no_active_source_or_training_changes": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..6b231a5e416976f48c6cca5b572937b4761cdf31 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-io-amendment.json @@ -0,0 +1,41 @@ +{ + "adopted_utc": "2026-10-25T04:07:41.689004+00:00", + "change": "Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.", + "source_before": "/workspace/research/studies/scale13/source-parallel-io", + "source_after": "/workspace/research/studies/loader_io/cached-integration/source", + "checkpoints": { + "scale13-small": { + "step": 96000, + "path": "/workspace/research/experiments/scale13-small/bytecache-resume-096000.pt", + "sha256": "62c322bf950af508fafbed8267a7036cd22cedb784ac3226830f4b1111774fba", + "single_inode_copy": true + }, + "scale13-large": { + "step": 78000, + "path": "/workspace/research/experiments/scale13-large/bytecache-resume-078000.pt", + "sha256": "bcac37454809ef1ba3bf6013722ae4569bd6c712729b6d86f507b31a1dd0b034", + "single_inode_copy": true + } + }, + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "model_optimizer_rng_exact_at_each_resume": true, + "model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged": true, + "different_resume_boundaries_are_IO_only": true, + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "timing_cache_over_reference": 0.23811392449245397, + "timing_caveat": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset.", + "cache_not_an_inference_asset": true, + "adopted_evaluations_without_interruption": { + "scale13-small-eval02": { + "pid": 12624, + "starttime_ticks": 21982067, + "cmdline_sha256": "7c6eb145cc8f7c825796945e00d17a6c41868e152f337b404477a8ad6e1b6c73" + } + }, + "old_monitor_retained_stopped_until_adopted_evaluation_finishes": { + "pid": 12248, + "starttime_ticks": 21911693, + "cmdline_sha256": "fa05426c3f41db24f960797d8171d2d369bf0ad15c957c24fcb8fb307282e14a" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-migration-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-migration-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..f66141e5eaa8ffa0691022fb3e727f2d75e7542a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-migration-plan.json @@ -0,0 +1,26 @@ +{ + "created_utc": "2026-10-25T03:51:39.643333+00:00", + "scope": "Execution-only checkpoint-safe cache adoption for both original scale trajectories; no scientific or evaluation change", + "original_study": "/workspace/research/studies/scale13/study.json", + "original_study_sha256": "61c9c4c2025b3c83067a27a35fdd58f00efef51ee3d9beb06711425d30cc8200", + "preparation": "/workspace/research/studies/loader_io/cached-integration/preparation.json", + "integration_check": "/workspace/research/studies/loader_io/cached-integration/integration-check.json", + "timing_report": "/workspace/research/analysis/cached-training-recheck.json", + "cache_root": "/tmp/seqfunc-bytecache-train256-20261025T0148", + "old_controller_pid": 12248, + "boundaries": { + "scale13-small": 96000, + "scale13-large": 78000 + }, + "helper_sha256": { + "migrate_scale_bytecache.py": "8c0a181c44fc6ad480dc0d74f2a803dc41d860b16007753bccc35dad17f78c0a", + "run_scale_study_cached_io.py": "adc1bf9c5d3054f63425b605e7146d2b83a24f933a7f9e795083fde0ad98174b" + }, + "controller_adoption_check": "/workspace/research/studies/loader_io/cached-integration/adoption-controller-check.json", + "decision": "Adopt verified exact-data optional reader after complete ABBA training probe; ratio.2381under shared load is evidence of current stalls, not a promised sustained gain. Cold population cost and full-working-set performance remain to measure.", + "unchanged_runs": [ + "qknorm35" + ], + "preserved_evaluation": "scale13-small-epoch02", + "old_monitor_policy": "Suspend old monitor after captures; retain its session while the adopted evaluation completes; then retire. Stop only original trainers/their data workers and idle future checkpoint waiters." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..9ce7b02b6f484c9b2d4f61c5bad46e4ed719abf6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-preparation.json @@ -0,0 +1,22 @@ +{ + "created_utc": "2026-10-25T03:47:11.178799+00:00", + "status": "Prepared only, active sources untouched; no migration decision before complete timing and integrated equality check", + "old_source": "/workspace/research/studies/scale13/source-parallel-io", + "new_source": "/workspace/research/studies/loader_io/cached-integration/source", + "only_modified_existing_source": [ + "reference/alphagenome_50m/dataset.py" + ], + "new_files": [ + "reference/alphagenome_50m/direct_read_store.py", + "reference/alphagenome_50m/cached_read_store.py" + ], + "model_trainer_objective_sources_exact": true, + "cache_activation": "Explicit SEQFUNC_TRAIN_BYTECACHE environment path and split=train and include_targets=True; otherwise exact original reader. Zipstore path remains original.", + "namespace": "SHA256ofpublic+labels training manifest digests and original absolute store path/key; every hit checks original bytes SHA/length/key", + "old_dataset_sha256": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f", + "new_files_sha256": { + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "reference/alphagenome_50m/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "reference/alphagenome_50m/cached_read_store.py": "53beb273318ad3603c1a7eec0a66f82887e6b05cabc523cbc33c7a80aa58e688" + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-timing_report.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-timing_report.json new file mode 100644 index 0000000000000000000000000000000000000000..f65930cac3b7e9dbbaeb4ba5001dcb5488f5bca0 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/bytecache-timing_report.json @@ -0,0 +1,85 @@ +{ + "created_utc": "2026-10-25T03:47:08.875281+00:00", + "plan_sha256": "6e0c89b326b3b5d254a8206167e4ec067a5c6cf5921489a4fa15a55816020e46", + "checkpoint_sha256": "31b67fe7d1b3b691f3d74876ad49e6fc8d096858c646df27aa27ff51c8999cb5", + "modes": [ + { + "repeat": 0, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 139.6908002400014, + "seconds_per_example": 0.5820450010000059, + "timed_dataloader_wait_seconds": 79.04790870507713, + "last_weighted_loss": 0.6381173133850098, + "peak_allocated_GiB": 6.530080318450928, + "peak_reserved_GiB": 9.171875 + }, + { + "repeat": 1, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 32.168527343979804, + "seconds_per_example": 0.13403553059991585, + "timed_dataloader_wait_seconds": 0.005411982972873375, + "last_weighted_loss": 0.6376754641532898, + "peak_allocated_GiB": 6.532010555267334, + "peak_reserved_GiB": 8.953125 + }, + { + "repeat": 2, + "mode": "existing_cache", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 17.468506193021312, + "seconds_per_example": 0.07278544247092214, + "timed_dataloader_wait_seconds": 0.005700364970834926, + "last_weighted_loss": 0.63838130235672, + "peak_allocated_GiB": 6.534998416900635, + "peak_reserved_GiB": 8.9453125 + }, + { + "repeat": 3, + "mode": "reference", + "fresh_parent_loaded": true, + "initial_optimizer_step_min": 10255.0, + "initial_optimizer_step_max": 10255.0, + "ordered_training_ids_exact": true, + "warmup_examples": 16, + "timed_examples": 240, + "timed_seconds": 68.76838014100213, + "seconds_per_example": 0.2865349172541755, + "timed_dataloader_wait_seconds": 5.523634409008082, + "last_weighted_loss": 0.6376829743385315, + "peak_allocated_GiB": 6.534746170043945, + "peak_reserved_GiB": 8.470703125 + } + ], + "total_timed_seconds": { + "reference": 208.45918038100353, + "existing_cache": 49.637033537001116 + }, + "cache_over_reference": 0.23811392449245397, + "all_probe_updates_discarded": true, + "checkpoint_or_weights_written": false, + "actual_training_runs_changed": false, + "split": "train only", + "precision": { + "autocast": "bfloat16", + "allow_tf32": false + }, + "caveats": "Shared GPU load, fixed256preselected training examples and an already populated byte cache; reference/cache/cache/reference order. Includes data delivery, transfer, forward/backward and AdamW after16warmup examples per mode. Excludes source loading/setup, tensor-digest testing and cache population. Does not measure long-tail storage outages or guarantee gains on the full training dataset." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/check_decoder111_parent.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/check_decoder111_parent.py new file mode 100644 index 0000000000000000000000000000000000000000..b0b893ff12f3727492b2a2f2b7908a702df34fcc --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/check_decoder111_parent.py @@ -0,0 +1,57 @@ +"""Compare the six unchanged native site/contact arrays with the sole eight-pass parent.""" +import argparse +from datetime import datetime, timezone +import json +import os +from pathlib import Path +import zarr +from numcodecs import blosc +from direct_read_store import DirectReadDirectoryStore +from verify_package import inference_signature +from prepare_decoder111 import digest + + +def main(): + parser=argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan',type=Path) + args=parser.parse_args();plan=json.loads(args.plan.read_text());study=args.plan.parent + assert not (study/'native-parent-check.json').exists() + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:2]);blosc.set_nthreads(1) + for name,expected in plan['native_public_sha256'].items(): + assert digest(Path(plan['native_public_input'])/name)==expected,name + package=Path(plan['output'])/'package' + candidate=study/'native-recovery';parent=Path(plan['parent_native']) + proof=json.loads((candidate/'report.json').read_text()) + previous=json.loads((parent/'report.json').read_text()) + assert inference_signature(package)==proof['package_inference_signature']==plan['candidate_inference_signature'] + assert previous['package_inference_signature']==plan['parent_inference_signature'] + assert proof['all_arrays_bit_identical'] and proof['arrays']=={'tracks':30,'perturbations':22} + count=0;observations=[] + for kind,filename,attribute in [('tracks','predictions.zarr','example_ids'), + ('perturbations','perturbations.zarr','perturbation_ids')]: + a=zarr.open_group(DirectReadDirectoryStore(str(candidate/('native-'+kind)/filename)),mode='r') + b=zarr.open_group(DirectReadDirectoryStore(str(parent/('native-'+kind)/filename)),mode='r') + assert a.attrs[attribute]==b.attrs[attribute] + arrays={} + def collect(name,obj): + if isinstance(obj,zarr.Array): + arrays[name]=obj + a.visititems(collect) + for name,value in arrays.items(): + if name.rsplit('/',1)[-1] not in ['splice_sites','contact_maps']: + continue + other=b[name] + assert value.shape==other.shape and value.dtype==other.dtype,name + assert value[:].tobytes()==other[:].tobytes(),name + count+=1;observations.append(dict(kind=kind,key=name,shape=list(value.shape))) + assert count==6,observations + report=dict(utc=datetime.now(timezone.utc).isoformat(),plan_sha256=digest(args.plan), + unchanged_native_arrays_exact=count,observations=observations, + native_report_sha256=digest(candidate/'report.json'),parent_native_report_sha256=digest(parent/'report.json'), + labels_read=False,new_GPU_work=False) + path=study/'native-parent-check.json';temporary=path.with_suffix('.tmp') + temporary.write_text(json.dumps(report,indent=2)+'\n');temporary.replace(path) + print(json.dumps(report),flush=True) + + +if __name__=='__main__':main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-CPU-construction-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-CPU-construction-check.json new file mode 100644 index 0000000000000000000000000000000000000000..69856d0cb69dbe3b69311a6a104b90fb77dc3777 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-CPU-construction-check.json @@ -0,0 +1,35 @@ +{ + "utc": "2026-10-26T10:58:18.161441+00:00", + "candidate_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "all_learned_tensors_exact": true, + "parameters": 215626632, + "synthetic_batch": 4, + "synthetic_input_bp": 8192, + "actual_allele_and_N_edits": true, + "both_forward_and_RC_synthetic_inputs": true, + "all11_outputs_finite": true, + "classifier_contacts_trunk_and_selected_positions_exact_parent": true, + "full_central_and_sparse_features_exact": true, + "source_exact_completed96": true, + "source_exact_completed113": true, + "weights_exact_large114": true, + "orientation_wrappers_and_policy_exact_RC114": true, + "CUDA_used": false, + "training_updates": 0, + "labels_read": false, + "accepted111_CPU_proof_sha256": "004cf4925d682937f80cb4ffcc5838d0bae0508a5801a7c325ad3640e9c46ed1", + "caveat": "Short synthetic base-network check plus unchanged accepted orientation wrapper. Own full1Mb native recovery, six exact parent output arrays and complete accuracy are still required." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-admission.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-admission.json new file mode 100644 index 0000000000000000000000000000000000000000..8c515714128636530291f78bb7a6a56e6cd6038f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-admission.json @@ -0,0 +1,20 @@ +{ + "utc": "2026-10-26T11:38:09.575005+00:00", + "admitted": true, + "additional_SGD": 0, + "sole_parent": "/workspace/research/experiments/rc_large114", + "scientific_reason": "The completed small-model113 decoder comparison improves reward0.0022182677 and six dense families under both paired track schemes, with nine positive changed-family points. Published largeRC114 has higher reward0.5222012904 but CAGE0.35921416 remains weak. Test whether the identical fixed decoder operation transfers to the larger network; no gain is assumed.", + "contrary_evidence": "Small113 reduces Mendelian AP by0.0105883003 with paired interval[-0.0251803684,-0.0003448354]; its other12 variant intervals includezero. Large114 already has supported caQTL regression versusitsforwardparent and weak eQTL/caQTL/SPI1/enhancer effects. Interpolation can further attenuate or distortalleleeffects. This candidate must report all13,including every regression, and exactfiveunchangedsplice reports; no broadvariantrescue claim.", + "previous115_all_full_native_runtime_paired_complete": true, + "previous115_runtime_passed": true, + "scope": "Allpredeclared115checks complete; oneinference-only large-decoder transfer, noadditionaltrainer. Small113Mendelianregression remains contraryevidence.", + "candidate_policy": "Replace RC114 nearest global-feature decoding for dense, usage and junction heads with the exact selective linear-bin-center operation from completed111/96. Keep its original nearest splice classifier, contacts, learned tensors, four training-derived offsets and equal forward/RC views.", + "no_tuning_or_new_parameters": true, + "requirements": "Own52native/recovery, six exact RC114 native arrays, all200/2697/all13, exact twofamily/fivevariant reports, own624native/runtime with allthree4h estimates, one all11/all13 paired observer against RC114. Atomic monotonic publication only after full scores and runtime.", + "resource_review_sha256": "0946f74de81094208f216e9a8172e5e46ab87be027d4d11299089232bba9ce00", + "operator_recovery_context": "/workspace/research/studies/operator_recovery_20261026T104431/recovery.json", + "operator_recovery_context_sha256": "0b193405bfb9dcf75c92f5e6eff530835dc6989ace5dbc94469d3b2276bc12be", + "research_deadline_utc": "2026-10-28T08:25:17+00:00", + "no_training_cadence_or_allocation_change": true, + "zero_interval_is_not_equivalence": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-controller-reuse.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-controller-reuse.json new file mode 100644 index 0000000000000000000000000000000000000000..d106356767571a2078e672411fd79e69642ecab9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-controller-reuse.json @@ -0,0 +1,10 @@ +{ + "utc": "2026-10-26T11:38:09.575005+00:00", + "controller": "run_decoder111.py", + "controller_sha256": "0c88342f7a15deb2d112139b99200fea34cdca69d9bcdce8239f18e8ef5e64b7", + "parent_checker": "check_decoder111_parent.py", + "parent_checker_sha256": "053a3f6a6e84bd81d8103833b748b325833de0090887be82e1cb6f7c943a73be", + "sources_unchanged": true, + "scope": "Reuse completed111 plan-driven evaluator/checker without code edits. Paths, parent signature, native parent and model hash are supplied by the new117 plan.", + "accepted_controller_completion_sha256": "dcc21750b7079db2b31349b45d828c40a1aad2414ed9cc1ff54506b3299f433f" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-fixture-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-fixture-check.json new file mode 100644 index 0000000000000000000000000000000000000000..17050e6ac690369c39f2be5823c8df16931c9c3c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-fixture-check.json @@ -0,0 +1,23 @@ +{ + "utc": "2026-10-26T11:38:09.575005+00:00", + "timing_backing": "/workspace/research/studies/native_latency_input_recovery/frozen-validation-public-input", + "timing_rows": [ + 16, + 96 + ], + "timing_rows_and_manifest_hashes_exact": true, + "native_rows": [ + 2, + 5 + ], + "native_public_sha256": { + "public/test.parquet": "ff36ae00a21f7d84bf8f436f6236a3a4da1d71be9e096d4106485097d9a62099", + "public/perturbations.parquet": "fcd349372a6363dc7a7e73f37fdccc05956f5596b488611146a8a3a8fea53e7c", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4", + "reference/GRCh38.p13.genome.fa.fai": "9293fb33f63b7f09d8fadc055d78e401d07c76de0211243fba778780ff836ff9" + }, + "FAI_bound": true, + "full_reference_hash_claim": false, + "labels_read": false, + "GPU_used": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-preparation.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-preparation.json new file mode 100644 index 0000000000000000000000000000000000000000..64524a6f1e27645830ce58b8de0f46cbb92e3e0e --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-preparation.json @@ -0,0 +1,56 @@ +{ + "utc": "2026-10-26T10:58:18.164939+00:00", + "package": "/workspace/research/experiments/rc_large_decoder117/package", + "package_verification": { + "files_verified": 79, + "bytes_verified": 4442382926 + }, + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "candidate_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "evidence_sha256": { + "/workspace/research/experiments/rc_large114/evaluation.json": "061bac5023384900265cdb37766dbf9cdf2cb816b5f2b2ca8eac7e3af5f6bdf5", + "/workspace/research/experiments/rc_large114/valid-score/report.json": "1a13fcf3aaef93bf63eb258c72da5b13fa8acdf2ec040bd135e362b094340a96", + "/workspace/research/experiments/rc_large114/valid-variant-score/report.json": "4803180c390f3cc66033f6a5a2641ea40025e30145237fa9b55598102fa41822", + "/workspace/research/studies/rc_large114/completion-observation.json": "cbe63fefd627c6df4d37860afb99f4f131ee9cedcd8377c09efa31893f49bab7", + "/workspace/research/studies/rc_large114/diagnostics/parent/summary.json": "f0bb95af4f95c643fb4281e48b62a9eae39ce317730b270ebd133fe8cf5a2fe8", + "/workspace/research/studies/rc_decoder113/completion-observation.json": "bc1ffaae63c65c6dd8b6c4eafda7f46cacb2856cbf3fbed37352f5056dcae87d", + "/workspace/research/studies/rc_decoder113/diagnostics/summary.json": "7730015e9e4f945c79e1bd4799eafc6ae0dc4456afcc9c6cb2ccd309660682d9", + "/workspace/research/studies/decoder111/CPU-construction-check.json": "004cf4925d682937f80cb4ffcc5838d0bae0508a5801a7c325ad3640e9c46ed1", + "/workspace/research/studies/rc_large_decoder117/proposed-recipe.json": "b95ffe4953d0c11d4510002bda739e3e7a0f20893faa11d987d8c5548cb0e71b", + "/workspace/research/studies/rc_large_decoder117/CPU-construction-check.json": "174ec3758eae1f0ca89d9cf24eb2c8ec9f9915beb937266030412eeff5f13603", + "/workspace/research/seqfunc/prepare_decoder111.py": "afa19189b21fa7780f5f801df3ef0216b1d004bca9e17a034b8606c28420989e", + "/workspace/research/seqfunc/make_linear_upsample83.py": "2048f81143e187051b20471e3fed1b2068e36ef9f4bbc92b66b51c040fb5da0f", + "/workspace/research/seqfunc/make_splice_grid96.py": "2ae38b29e6c055b724b156240ba38ed22cd93ebdb1bf6da490d427f82c6577ec", + "/workspace/research/seqfunc/prepare_rc_large_decoder117.py": "74b74d9513ae1a6a021926b7cffbc440875e986cd71f4065dbf5609c5f4feb06" + }, + "evaluation_admitted": false, + "controller_created": false, + "GPU_used": false, + "next": "Separate resource/admission decision, wait115 full/native/runtime/paired complete then freeze native/full/all13/runtime/paired gates versus RC114; no automatic GPU launch." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-proposed-recipe.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-proposed-recipe.json new file mode 100644 index 0000000000000000000000000000000000000000..b492eea7fcc2e25ff9fd54d9adc04126dc71a494 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-proposed-recipe.json @@ -0,0 +1,45 @@ +{ + "utc": "2026-10-26T10:57:56.711123+00:00", + "parent": "/workspace/research/experiments/rc_large114", + "parent_package": "/workspace/research/studies/rc_large114/paired-documentation/package", + "output": "/workspace/research/experiments/rc_large_decoder117", + "parameters": 215626632, + "inherited_examples": 328160, + "inherited_optimizer_steps": 82040, + "inherited_complete_passes": 8, + "additional_training_updates": 0, + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "candidate_model_sha256": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "only_change": "Replace RC114 nearest global-feature decoding for dense, usage and junction heads with the exact selective linear-bin-center operation from completed111/96. Keep its original nearest splice classifier, contacts, learned tensors, four training-derived offsets and equal forward/RC views.", + "expected_exact_families": [ + "splice_sites", + "contact_maps" + ], + "expected_exact_variants": [ + "sqtl_causality", + "clinvar_noncoding", + "clinvar_splice_region", + "clinvar_missense", + "mfass" + ], + "motivation": "The completed small-model113 decoder comparison improves reward0.0022182677 and six dense families under both paired track schemes, with nine positive changed-family points. Published largeRC114 has higher reward0.5222012904 but CAGE0.35921416 remains weak. Test whether the identical fixed decoder operation transfers to the larger network; no gain is assumed.", + "contrary_evidence": "Small113 reduces Mendelian AP by0.0105883003 with paired interval[-0.0251803684,-0.0003448354]; its other12 variant intervals includezero. Large114 already has supported caQTL regression versusitsforwardparent and weak eQTL/caQTL/SPI1/enhancer effects. Interpolation can further attenuate or distortalleleeffects. This candidate must report all13,including every regression, and exactfiveunchangedsplice reports; no broadvariantrescue claim.", + "scope": "One fixed network and the same two orientation views serve every family and both entrypoints. No learned or fitted mixture, routing, new splice prior, motif mask, SWA, extra flanking DNA, variant-label training or additional SGD.", + "limitations": "Exploratory one-seed postselection. Interaction can worsen any changed readout. Full native replay, runtime and accuracy must be measured independently; unchanged paths require exact complete reports.", + "execution_gate": "Wait RC115 original full/native/runtime/all8paired completion, then fresh resources preserving mainlarge14,seed110 and sixhours evaluation reserves. CPU preparation only now; no thirdtrainer.", + "evaluation_admitted": false, + "GPU_used": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-resource-review.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-resource-review.json new file mode 100644 index 0000000000000000000000000000000000000000..d435c7a06c676e2917029b9843da3c2fa3c69e96 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-resource-review.json @@ -0,0 +1,104 @@ +{ + "utc": "2026-10-26T11:38:09.620693+00:00", + "trainers": [ + { + "pid": 1058, + "output": "/workspace/research/experiments/seed110-small" + }, + { + "pid": 1067, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1418, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1420, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1422, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1424, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1426, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1428, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1430, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1432, + "output": "/workspace/research/experiments/scale13-large" + }, + { + "pid": 1711, + "output": "/workspace/research/experiments/seed110-small" + }, + { + "pid": 1713, + "output": "/workspace/research/experiments/seed110-small" + } + ], + "main_rates": { + "small": { + "completed": true, + "examples": 574280, + "remaining_main14_hours": 0 + }, + "large": { + "step": 422800, + "optimizer_steps": 105700, + "examples_per_second": 5.689341856821747, + "remaining_main14_hours": 7.3958954896206235, + "last_log_mtime": 1793014680.7741828, + "caveat": "Recent shared-load observations; not isolated rates or upper bounds." + } + }, + "gpu": "32935, 48156, 100", + "memory_current_bytes": 248728301568, + "memory_events": { + "low": "0", + "high": "0", + "max": "20518", + "oom": "0", + "oom_kill": "0" + }, + "research_deadline_utc": "2026-10-28T08:25:17+00:00", + "hours_remaining": 44.78539583194445, + "active_training_rates": { + "scale13-large": { + "step": 422800, + "target": 574280, + "samples": 11, + "current_examples_per_second": 5.689341856821747, + "planning_examples_per_second": 5.689341856821747, + "remaining_hours": 7.3958954896206235, + "latest_resumed_segment_only": true + }, + "seed110-small": { + "step": 90000, + "target": 328160, + "samples": 11, + "current_examples_per_second": 5.307232227132246, + "planning_examples_per_second": 4.2616295612937565, + "remaining_hours": 15.523534977421145, + "latest_resumed_segment_only": true + } + }, + "planning_slowdown_factor": 1.5, + "large14_evaluation_reserve_hours": 3, + "additional_evaluation_planning_margin_hours": 3, + "concurrent_slowdown_adjusted_training_plus_reserves_hours": 29.285302466131718, + "scope": "One inference-only candidate on existing full/bounded locks after all RC115 checks. Large14 and seed110 retain source, workers and schedules. Rates/reserves are shared-load estimates, not bounds; cgroup max events include reclaim pressure." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-training-replay.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-training-replay.md new file mode 100644 index 0000000000000000000000000000000000000000..756d9e750ce29ead3bfaa22c25a0ef26b2b18a5c --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/decoder117-training-replay.md @@ -0,0 +1,5 @@ +All eight inherited training passes used pre-decoder117-model.py at reference/alphagenome_50m/model.py. +Restore that source for the inherited training recipe. Training remains forward-only with the original +objective, seed17, batch4, data order, AdamW state and fourteen-pass learning-rate schedule. +RC114 introduced equal forward/RC inference without SGD. Decoder117 adds only the exact model.py +from completed111/96/113 via preserve_classifier(interpolate(original_source)); no weights are changed. diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/execution_change.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/execution_change.json new file mode 100644 index 0000000000000000000000000000000000000000..93f72a0c4f5f51e9192005535fc45e6aa3bbd671 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/execution_change.json @@ -0,0 +1,10 @@ +{ + "adopted_utc": "2026-10-24T21:51:59.762636+00:00", + "change": "Activation checkpoint boundaries only; all model operations, parameters, objective, data and schedule unchanged.", + "example_boundary": 1000, + "before": "Recompute whole local path during backward.", + "after": "Recompute each local residual block during backward.", + "evidence": "CPU gradients and all eleven outputs exactly match. Full-shape BF16 outputs/losses are bit-exact; gradient relative L2 discrepancy .0009547 versus .0009534 for an unchanged repeat, consistent with existing reduction variability. Peak memory40.11 to15.99GiB; isolated compute .11131 to.10504seconds/example.", + "initial_source_manifest_sha256": "18329ef35e89fc5aeb50ccc045b08db690e6c8ef40d661b01ddb74c7f32b6f0b", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/freeze_rc_large114_execution.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/freeze_rc_large114_execution.py new file mode 100644 index 0000000000000000000000000000000000000000..5e2ea168f8b491cc56b1a0721db80f9d7fdd39a7 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/freeze_rc_large114_execution.py @@ -0,0 +1,144 @@ +"""Freeze large114 evaluation after completed parent, accepted CPU proof and fresh resources.""" +from datetime import datetime, timezone, timedelta +import json +from pathlib import Path +import shutil + +import pandas as pd + +from run_seed110_pilot import digest,read,save,rows,resource_snapshot +from verify_package import inference_signature,verify_package + + +def main(): + root=Path('/workspace/research');here=root/'seqfunc';study=root/'studies/rc_large114' + assert not (study/'study.json').exists() and not (study/'admission.json').exists() + recipe=read(study/'proposed-recipe.json');prep=read(study/'preparation.json') + for path,expected in prep['evidence_sha256'].items():assert digest(path)==expected,path + proof=read(study/'CPU-construction-check.json') + assert proof['parameters']==215626632 and proof['all_learned_tensors_and_CPU_RNG_unchanged'] + assert proof['actual_native_track_arrays_exact_manual_fixed_mean']==15 + assert proof['paired_dense_and_splice_arrays_exact_manual_fixed_mean']==8 and not proof['CUDA_used'] + parent=Path(recipe['parent']);parent_study=root/'studies/late_fixed_links81/large-epoch08' + assert read(parent_study/'complete.json')['phase']=='complete' + completion=read(parent_study/'completion-observation.json') + assert completion['full_all13_scored'] and completion['native52_passed'] and completion['native36_derivation_accepted'] + decision=read(parent_study/'publication-decision.json') + assert decision['eligible_at_review'] is False and decision['runtime_passed'] is None + for phase in ['tracks','remaining-variants']: + assert read(parent_study/f'paired-{phase}-complete.json')['output_sha256']==digest(parent_study/f'paired-{phase}.json') + assert read(root/'studies/rc108/diagnostics/summary.json')['complete_metrics']==13 + assert read(root/'studies/rc108/state.json')['phase']=='complete' + package=Path(prep['package']);signature=inference_signature(package) + assert signature==prep['candidate_inference_signature'] + assert inference_signature(parent/'package')==prep['parent_inference_signature'] + context=root/'studies/operator_recovery_20261026T064822/recovery.json' + recovery=read(context);now=datetime.now(timezone.utc);deadline=datetime.fromisoformat(recovery['new_deadline']) + assert deadline==datetime.fromisoformat(recovery['recovery_utc'])+timedelta(seconds=recovery['remaining_seconds_at_recovery']) + resource=resource_snapshot({name:str(root/'experiments'/('scale13-'+name)) for name in ['small','large']}) + allowed={str(root/'experiments/scale13-large'),str(root/'experiments/seed110-small')} + assert resource['trainers'] and all(x['output'] in allowed for x in resource['trainers']) + assert resource['memory_events']['oom']=='0' and resource['memory_events']['oom_kill']=='0' + assert int(resource['gpu'].split(',')[1])>=24*1024 + rates={} + for name,target in [('scale13-large',574280),('seed110-small',328160)]: + events=rows(root/'experiments'/name/'train.log') + start=max(i for i,r in enumerate(events) if r.get('event')=='train_start') + tail=[r for r in events[start:] if r.get('event')=='train_step'][-11:] + assert len(tail)>=3 and all(a['step']= 24 * 1024 + rates = {} + for name, target in [('scale13-large', 574280), ('seed110-small', 328160)]: + events = rows(root / 'experiments' / name / 'train.log') + start = max(i for i, row in enumerate(events) if row.get('event') == 'train_start') + tail = [row for row in events[start:] if row.get('event') == 'train_step'][-11:] + assert len(tail) >= 3 and all(a['step'] < b['step'] for a, b in zip(tail, tail[1:])) + measured = (tail[-1]['step'] - tail[0]['step']) / (tail[-1]['elapsed_seconds'] - tail[0]['elapsed_seconds']) + rate = measured + if name == 'seed110-small': + rate = min(rate, read(root / 'studies/seed110/continuation-resource-review.json')['admission_rate_examples_per_second']) + rates[name] = dict(step=tail[-1]['step'], target=target, samples=len(tail), + current_examples_per_second=measured, planning_examples_per_second=rate, + remaining_hours=max(0, target - tail[-1]['step']) / rate / 3600, + latest_resumed_segment_only=True) + hours = (deadline - now).total_seconds() / 3600 + required = max(row['remaining_hours'] for row in rates.values()) * 1.5 + 3 + 3 + assert required < hours + resources.update(research_deadline_utc=deadline.isoformat(), hours_remaining=hours, active_training_rates=rates, + planning_slowdown_factor=1.5, large14_evaluation_reserve_hours=3, + additional_evaluation_planning_margin_hours=3, + concurrent_slowdown_adjusted_training_plus_reserves_hours=required, + scope='One inference-only candidate on existing full/bounded locks after all RC115 checks. Large14 and seed110 retain source, workers and schedules. Rates/reserves are shared-load estimates, not bounds; cgroup max events include reclaim pressure.') + resources['main_rates']['small'] = dict(completed=True, examples=574280, remaining_main14_hours=0) + save(study / 'resource-review.json', resources) + admission=dict(utc=now.isoformat(),admitted=True,additional_SGD=0,sole_parent=str(parent), + scientific_reason=recipe['motivation'],contrary_evidence=recipe['contrary_evidence'], + previous115_all_full_native_runtime_paired_complete=True, + previous115_runtime_passed=previous_gate['passed'], + scope='Allpredeclared115checks complete; oneinference-only large-decoder transfer, noadditionaltrainer. Small113Mendelianregression remains contraryevidence.', + candidate_policy=recipe['only_change'],no_tuning_or_new_parameters=True, + requirements='Own52native/recovery, six exact RC114 native arrays, all200/2697/all13, exact twofamily/fivevariant reports, own624native/runtime with allthree4h estimates, one all11/all13 paired observer against RC114. Atomic monotonic publication only after full scores and runtime.', + resource_review_sha256=digest(study/'resource-review.json'),operator_recovery_context=str(args.recovery_context), + operator_recovery_context_sha256=digest(args.recovery_context),research_deadline_utc=deadline.isoformat(), + no_training_cadence_or_allocation_change=True,zero_interval_is_not_equivalence=True) + save(study/'admission.json',admission) + old=read(root/'studies/decoder111/plan.json') + timing=json.loads(json.dumps(old['native_latency_template'])) + timing.update(created_utc=now.isoformat(),minimum_free_GiB=24, + memory_evidence='215626632parameters withtwo sequentialorientations;24GiBfree schedulingguard based oncompletedlargeRC114,notapeak-memorybound.',scope='RCdecoder117 own16validationwindows/96developmentqueries and624exactnative arrays after full200/2697/all13. Originalthree4h gates; estimates are not bounds.') + backing=root/'studies/native_latency_input_recovery/frozen-validation-public-input' + public=Path(timing['original_public_input']) + for name,expected in timing['public_inputs_sha256'].items():assert digest(backing/name)==expected,name + for name,expected in timing['original_manifests_sha256'].items():assert digest(public/name)==expected,name + for name,key in [('test','track_indices'),('perturbations','query_indices')]: + selected=pd.read_parquet(public/'public'/f'{name}.parquet').iloc[timing[key]].reset_index(drop=True) + pd.testing.assert_frame_equal(selected,pd.read_parquet(backing/'public'/f'{name}.parquet')) + assert (backing/'reference').resolve()==(public/'reference').resolve() + latency=study/'native-latency';latency.mkdir();(latency/'public-input').symlink_to(backing,target_is_directory=True) + native_public=root/'analysis/recovery-check' + native_hashes={name:digest(native_public/name) for name in ['public/test.parquet','public/perturbations.parquet','public/junction_tracks.parquet','reference/GRCh38.p13.genome.fa.fai']} + assert len(pd.read_parquet(native_public/'public/test.parquet'))==2 and len(pd.read_parquet(native_public/'public/perturbations.parquet'))==5 + save(study/'fixture-check.json',dict(utc=now.isoformat(),timing_backing=str(backing),timing_rows=[16,96], + timing_rows_and_manifest_hashes_exact=True,native_rows=[2,5],native_public_sha256=native_hashes, + FAI_bound=True,full_reference_hash_claim=False,labels_read=False,GPU_used=False)) + helpers=['run_decoder111.py','check_decoder111_parent.py','prepare_rc_large_decoder117.py','prepare_decoder111.py', + 'make_linear_upsample83.py','make_splice_grid96.py','direct_read_store.py','run_seed110_pilot.py', + 'check_native_recovery.py','evaluate_package.py','resume_tracks.py','resume_perturbations.py', + 'verify_package.py','measure_native_latency.py','promote_if_better.py','finalize_package.py','publish_atomic.py',Path(__file__).name] + for name in helpers:ast.parse((here/name).read_text()) + assert digest(here/'run_decoder111.py')==old['helper_sha256']['run_decoder111.py'] + assert digest(here/'check_decoder111_parent.py')==old['helper_sha256']['check_decoder111_parent.py'] + save(study/'controller-reuse.json',dict(utc=now.isoformat(),controller='run_decoder111.py', + controller_sha256=digest(here/'run_decoder111.py'),parent_checker='check_decoder111_parent.py', + parent_checker_sha256=digest(here/'check_decoder111_parent.py'),sources_unchanged=True, + scope='Reuse completed111 plan-driven evaluator/checker without code edits. Paths, parent signature, native parent and model hash are supplied by the new117 plan.', + accepted_controller_completion_sha256=digest(root/'studies/decoder111/completion-observation.json'))) + paths=[study/name for name in ['preparation.json','proposed-recipe.json','CPU-construction-check.json','resource-review.json','admission.json','fixture-check.json','controller-reuse.json']] + evidence_paths=paths+completed115 + plan=dict(schema='seqfunc-fixed-decoder117/v1',created_utc=now.isoformat(),parent=str(parent),output=str(output), + parameters=215626632,additional_training_updates=0,inherited_examples=328160,inherited_optimizer_steps=82040,inherited_complete_passes=8, + parent_inference_signature=preparation['parent_inference_signature'],candidate_inference_signature=preparation['candidate_inference_signature'], + candidate_model_sha256=recipe['candidate_model_sha256'],evaluation_lock=old['evaluation_lock'],minimum_free_GiB=24,native_timeout_seconds=1800, + parent_native=str(root/'studies/rc_large114/native-recovery'),native_public_input=str(native_public),native_public_sha256=native_hashes, + expected_exact_families=recipe['expected_exact_families'],expected_exact_variants=recipe['expected_exact_variants'],native_latency_template=timing, + helper_sha256={name:digest(here/name) for name in helpers}, + evidence_sha256={**preparation['evidence_sha256'],**{str(p):digest(p) for p in evidence_paths},str(args.recovery_context):digest(args.recovery_context)}, + only_change=recipe['only_change'],limitations=recipe['limitations']) + paired=read(root/'studies/decoder111/diagnostics/plan.json') + text=json.dumps(paired).replace('experiments/decoder111','experiments/rc_large_decoder117').replace('studies/decoder111','studies/rc_large_decoder117') + text=text.replace('decoder111-vs-calib81-small8','rc-large-decoder117-vs-RC114').replace('experiments/calib81-small-epoch08','experiments/rc_large114') + paired=json.loads(text);paired['created_utc']=now.isoformat() + paired['dependencies']=[str(output/'evaluation.json'),str(study/'native-parent-check.json'),str(study/'parent-comparison.json'),str(study/'runtime-gate.json')] + paired['scope']='All11/all13 paired117 versus soleRC114; fixed2000replicates/seed1729 and unchanged verifier parity. One observer; no fitting.' + paired['caveat']='One fixed decoder change within the same equal orientation mean. Inherited weights and all priors unchanged. Exploratory postselection/unadjusted; intervals are not independent training replications or test results. Exact twofamily/fivevariant reports are required, not broader equivalence.' + assert len(paired['phases'])==4 and 'calib81-small' not in json.dumps(paired) + for name,expected in paired['source_sha256'].items():assert digest(here/name)==expected,name + save(study/'diagnostics/plan.json',paired);plan['paired_plan_sha256']=digest(study/'diagnostics/plan.json') + history=package/'training_history' + for path in paths:shutil.copy2(path,history/('decoder117-'+path.name)) + for name in ['run_decoder111.py','check_decoder111_parent.py',Path(__file__).name]:shutil.copy2(here/name,history/name) + manifest=read(package/'package_manifest.json') + for path in history.iterdir(): + if path.is_file():manifest['files'][str(path.relative_to(package))]=digest(path) + (package/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + verification=verify_package(package);assert inference_signature(package)==plan['candidate_inference_signature'] + save(study/'package-check.json',dict(utc=now.isoformat(),verification=verification,inference_signature=plan['candidate_inference_signature'],metadata_only_since_CPU_proof=True)) + plan['evidence_sha256'][str(study/'package-check.json')]=digest(study/'package-check.json') + save(study/'plan.json',plan) + receipt=dict(utc=datetime.now(timezone.utc).isoformat(),plan_sha256=digest(study/'plan.json'),admission_sha256=digest(study/'admission.json'), + paired_plan_sha256=plan['paired_plan_sha256'],resource_sha256=digest(study/'resource-review.json'), + fixture_sha256=digest(study/'fixture-check.json'),package_check_sha256=digest(study/'package-check.json'), + controller='run_decoder111.py',controller_sha256=digest(here/'run_decoder111.py'),package=verification,controller_launched=False) + save(study/'execution.json',receipt);print(json.dumps(receipt),flush=True) + + +if __name__=='__main__':main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-master-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-master-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..60f920df90c63e6614b09745bb36dedab5f40cc2 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-master-plan.json @@ -0,0 +1,562 @@ +{ + "created_utc": "2026-10-25T16:20:50.709567+00:00", + "members": { + "large-epoch04": { + "capacity": "large", + "epoch": 4, + "previous_epoch": 2, + "parent": "/workspace/research/experiments/scale13-large-epoch04", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch02", + "output": "/workspace/research/experiments/calib81-large-epoch04", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch04" + }, + "small-epoch08": { + "capacity": "small", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-small-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch04", + "output": "/workspace/research/experiments/calib81-small-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch08" + }, + "large-epoch08": { + "capacity": "large", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-large-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch04", + "output": "/workspace/research/experiments/calib81-large-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch08" + }, + "small-epoch14": { + "capacity": "small", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-small-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-small-epoch08", + "output": "/workspace/research/experiments/calib81-small-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/small-epoch14" + }, + "large-epoch14": { + "capacity": "large", + "epoch": 14, + "previous_epoch": 8, + "parent": "/workspace/research/experiments/scale13-large-epoch14", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch08", + "output": "/workspace/research/experiments/calib81-large-epoch14", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch14" + } + }, + "capacities": { + "small": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 32, + "channel_growth": 32, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 384, + "num_sequence_blocks": 6, + "num_attention_heads": 6, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 256, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 32, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 13290504 + }, + "large": { + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 96, + "channel_growth": 96, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 1024, + "num_sequence_blocks": 16, + "num_attention_heads": 16, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 768, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 64, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "parameters": 215626632 + } + }, + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "training_prior": "/workspace/research/analysis/prior-calibration-14000.json", + "training_prior_sha256": "dc2fbb82e37768540b6a3d527f73a8071e98c81fe3e0e5b4bff643037a43a235", + "calibration_reference": "/workspace/research/experiments/calib57/package/splice_calibration.json", + "helper_sha256": { + "/run-state/workspace/research/seqfunc/check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "/run-state/workspace/research/seqfunc/check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "/run-state/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "/run-state/workspace/research/seqfunc/diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "/run-state/workspace/research/seqfunc/direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "/run-state/workspace/research/seqfunc/evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "/run-state/workspace/research/seqfunc/finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "/run-state/workspace/research/seqfunc/late_fixed_links81.py": "a31130ce8f8a1a98449ab4e89ba689c7ea537d5ef601b8d72998c0aaed443999", + "/run-state/workspace/research/seqfunc/make_splice_probability_candidate.py": "42c0502bc0734ff4ca7fcf69a67e723b8fa83035c91083112845a2bd5758a0c6", + "/run-state/workspace/research/seqfunc/materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906", + "/run-state/workspace/research/seqfunc/measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "/run-state/workspace/research/seqfunc/promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "/run-state/workspace/research/seqfunc/publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "/run-state/workspace/research/seqfunc/resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "/run-state/workspace/research/seqfunc/resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "/run-state/workspace/research/seqfunc/run_fixed_link_candidate.py": "b63e8f914833a5afd5d4a9fcb76682ba2d23f7fb78bcef65dedff04d256665f4", + "/run-state/workspace/research/seqfunc/run_fixed_link_followup69.py": "3234a647219aae3df72afee9e5d9ed671d9db8b08adafff58bbffcbebcafb40d", + "/run-state/workspace/research/seqfunc/run_late_fixed_links81.py": "a06662cd6cfe340e03eb3faade4fbdd0b78e05deeb8a1ca9a3c9968dae033e87", + "/run-state/workspace/research/seqfunc/verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "evidence_sha256": { + "/workspace/research/studies/fixed_link69/plan.json": "45541635f549af8ce6c3db3fc7c857720970a5ee35e9de3c6a57f41dbe1f56bc", + "/workspace/research/studies/fixed_link61/plan.json": "100e4da9bf4faa715e901a0aa928dab65fc19f60970312e3e0eb548fb5b70860", + "/workspace/research/studies/fixed_link57/complete-summary.json": "5bba2a23162a77a523777f19de0be819d44cb44246c8c5e2f4bef913f5ee7a87", + "/workspace/research/studies/scale13/study-cached-io.json": "88aa6d95ef93776ed3551d68d98b1a31d726aa4b0270edbfc16b3f16ceeb937f", + "/workspace/research/studies/scale13/loss_scales.json": "add4711457568308ec64ccc77c113ed5ba9afaeddebd7a9cb276d6465784fb42", + "/workspace/data/public/train.parquet": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64", + "/workspace/research/experiments/calib57/package/splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5", + "/workspace/research/experiments/scale13-large-epoch04/package/training_provenance.json": "6589e14349cbd5419c814aa3d292c81f5328f1441bc75bc3ea7cae2b114dcdac" + }, + "materialization": { + "schema": "seqfunc-exact-fixed-link-evaluation-reuse/v1", + "public_queries": "/workspace/research/public-valid/public/perturbations.parquet", + "public_queries_sha256": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public_input": "/workspace/research/public-valid", + "complete_native_proof": "/workspace/research/analysis/splice-link-reuse-calib19-proof.json", + "complete_native_proof_sha256": "3361b646308f3af9c77e82ca8ef72cec9cb12be5da5273c826bc958233a04844", + "proof_origin": "Full original package-model evaluation via its recovery runner, already mechanically checked against native entrypoints. All12486arrays match exactly. Every actual reuse candidate additionally gets a fresh standalone vendored perturb.sh fixture check.", + "native_query_indices": [ + 0, + 567, + 1548, + 1746, + 2106, + 2696 + ], + "native_minimum_free_GiB": 12, + "native_timeout_seconds": 300, + "helper_sha256": { + "check_splice_link_reuse.py": "a9c92658c3a1a9bcc80b594f8b7d2d0e7f93313e811f361d7bbae9cb5e2add17", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "materialize_splice_link_outputs_v2.py": "d6ffb21a81b4394700adbcebcba1adf95029cc06c09452ed34a4d65e03cbb906" + }, + "change": "Apply the unchanged verified fixed-link output derivation to this newly declared parent/candidate only. Full track inference and original official scoring remain required.", + "installation": "Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.", + "labels_read": false, + "not_an_inference_asset": true + }, + "native_latency_template": { + "created_utc": "2026-10-25T09:54:57.944795+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.", + "scope": "Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "The rawlarge2 parent timing is contextual only. Require fresh candidate replay and allthree estimates <=14400seconds; no estimate is a guaranteed upper bound." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "bounded_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 16, + "native_timeout_seconds": 1800, + "materialization_timeout_seconds": 1800, + "parent_minimum_reward_gain": 1e-06, + "publication_minimum_reward_gain": 1e-06, + "hypothesis": "Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.", + "admission": "Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.", + "validation": "Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.", + "diagnostics": "After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.", + "nonimprovement_policy": "A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.", + "limitations": [ + "The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.", + "The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.", + "Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.", + "No test labels, variant-label training, external parameters or inference caches are used." + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-parent-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-parent-check.json new file mode 100644 index 0000000000000000000000000000000000000000..4e85ff1231bb0b51a26679743837f33f5732f829 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-parent-check.json @@ -0,0 +1,53 @@ +{ + "created_utc": "2026-10-25T16:21:07.054587+00:00", + "plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "actual_completed_parent_checks": [ + { + "experiment": "/workspace/research/experiments/scale13-small-epoch04", + "capacity": "small", + "epoch": 4, + "inference_signature": { + "last.pt": "ae6a15c21e8d9b5900f986f28f91b00962c0030ffc952340b1b3264eef1a6659", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.44610804541871724, + "sha256": { + "/workspace/research/experiments/scale13-small-epoch04/evaluation.json": "3193c974189a7ad77a06d0abf64a2c3e002953470660ca20f30d97271e3ffd0a", + "/workspace/research/experiments/scale13-small-epoch04/evaluation-package.json": "e06f0c731873e9ef1a94722d29651c180eefc6c1b54650b52622bad525dd5da7", + "/workspace/research/experiments/scale13-small-epoch04/valid-score/report.json": "3a45ae11ac8ad3897b926283073b5e755b3013a7796aa9837954dec842d7a942", + "/workspace/research/experiments/scale13-small-epoch04/valid-variant-score/report.json": "19224eae63273b62b0421fc48c32704b86418a2bfb21adda2a90449cfa0988d8", + "/workspace/research/experiments/scale13-small-epoch04/package/training_provenance.json": "30c05badb228e6d94b3e70113c4ea671ac2cac370c380d4e4e245ca2a00b00b5" + } + }, + { + "experiment": "/workspace/research/experiments/scale13-large-epoch02", + "capacity": "large", + "epoch": 2, + "inference_signature": { + "last.pt": "6e7c605f2683881b9f59571f042cb7df93a3a0f224d8f2cb4e37be0a123cdb27", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.4231233468773853, + "sha256": { + "/workspace/research/experiments/scale13-large-epoch02/evaluation.json": "d2336bfba9e6788cb2ab7dcf01834d825846920c21db43b763f3f5a4de056afb", + "/workspace/research/experiments/scale13-large-epoch02/evaluation-package.json": "cad26f6c8167a550d9a11057e69f2feca299cb091bfce64a95b403a627271610", + "/workspace/research/experiments/scale13-large-epoch02/valid-score/report.json": "eeec21adce86c14356ed29a5bc0e76e84f847bceba41c0faefa8e3a8d6793d4d", + "/workspace/research/experiments/scale13-large-epoch02/valid-variant-score/report.json": "f806730a18850e289cb95020db0ac6f69d4203c8ab53c2ef31845fd798a36f28", + "/workspace/research/experiments/scale13-large-epoch02/package/training_provenance.json": "1eb6eafaaec3f950708cabd7dc43a3ed22e67be602d7c036fe0fbbe0fbbff802" + } + } + ], + "gpu_used": false, + "packages_changed": false, + "no_new_predictions_or_parameter_fitting": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-resolved-parents.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-resolved-parents.json new file mode 100644 index 0000000000000000000000000000000000000000..bb11e638ada93a10dc258a86d05668c1d47052af --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late-fixed-links81-resolved-parents.json @@ -0,0 +1,58 @@ +{ + "master_plan_sha256": "39844848f0b239ada3f89e52a5c1e1e49b473ed6fc4746f20bbb329f194d2bd6", + "member": { + "capacity": "large", + "epoch": 8, + "previous_epoch": 4, + "parent": "/workspace/research/experiments/scale13-large-epoch08", + "previous_raw": "/workspace/research/experiments/scale13-large-epoch04", + "output": "/workspace/research/experiments/calib81-large-epoch08", + "study": "/workspace/research/studies/late_fixed_links81/large-epoch08" + }, + "receipts": { + "parent": { + "experiment": "/workspace/research/experiments/scale13-large-epoch08", + "capacity": "large", + "epoch": 8, + "inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.47738247727682986, + "sha256": { + "/workspace/research/experiments/scale13-large-epoch08/evaluation.json": "3b164a241ad64a50e02557217a9169d91ee0b89d0a3d5317fd0558f0ac60f80d", + "/workspace/research/experiments/scale13-large-epoch08/evaluation-package.json": "0591a542536b2a428332b5976a6ee00b451f3a35fdfe939b4fa1a3b3df513f08", + "/workspace/research/experiments/scale13-large-epoch08/valid-score/report.json": "e3ad0b0bc55f8492f1a256e6b289a87c7a456b49023685f4e0702fb28515e6a2", + "/workspace/research/experiments/scale13-large-epoch08/valid-variant-score/report.json": "99ab33b4d0053b7f6ee5eb1b7bfa22855308ab764b3fddb31dd59594c633906e", + "/workspace/research/experiments/scale13-large-epoch08/package/training_provenance.json": "0391b8c72aa2a3696450422c31dca049dc915d58ac9acb210467af2b5425aba6" + } + }, + "previous": { + "experiment": "/workspace/research/experiments/scale13-large-epoch04", + "capacity": "large", + "epoch": 4, + "inference_signature": { + "last.pt": "62ed23f6e35dff94567b128b4e55e10d7eb81752448acdea521275b3fc7531ae", + "predict.py": "c257bd031d41b946053e37d5a5759c6c1416d22d9b4d3f403e1fd13e0f63968b", + "perturb.py": "ba37a6f9f8b8eccb052903618bcb4c91d3f4cbed62dfe678ee362644dc80801e", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919" + }, + "reward": 0.457463750328102, + "sha256": { + "/workspace/research/experiments/scale13-large-epoch04/evaluation.json": "d50911309435bd38ec64f0928429b0258701111191cae754f4c0177ee123f794", + "/workspace/research/experiments/scale13-large-epoch04/evaluation-package.json": "e169b3f8604747d147e4108992c4bfe9e3314cc637eface507bdc915a55e56c2", + "/workspace/research/experiments/scale13-large-epoch04/valid-score/report.json": "fa3ff5a189a658dc449e8d511d04ce0f59ab48b395039c011a273dc52ed1ee51", + "/workspace/research/experiments/scale13-large-epoch04/valid-variant-score/report.json": "acc6f6901c41a5ee88176a5094d16d89714f9cd2d9195410222f161613d23d74", + "/workspace/research/experiments/scale13-large-epoch04/package/training_provenance.json": "6589e14349cbd5419c814aa3d292c81f5328f1441bc75bc3ea7cae2b114dcdac" + } + } + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..28c50d9e3a33f6accbd9ce3cdeca8bdbb9b481a6 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/late_fixed_links81.py @@ -0,0 +1,145 @@ +"""Predeclare fixed-prior output links for the remaining main training checkpoints.""" +import argparse +from datetime import datetime, timezone +import hashlib +import json +import math +from pathlib import Path + +from verify_package import inference_signature, verify_package + +ROOT = Path('/workspace/research') +STUDY = ROOT / 'studies/late_fixed_links81' +HERE = Path(__file__).resolve().parent + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def write_json(path, value): + path = Path(path) + temporary = path.with_suffix(path.suffix + '.tmp') + temporary.write_text(json.dumps(value, indent=2, allow_nan=False) + '\n') + temporary.replace(path) + + +def parent_ready(path): + path = Path(path) + try: + return (json.loads((path / 'pipeline-state.json').read_text())['phase'] == 'complete' + and (path / 'evaluation.json').exists()) + except (FileNotFoundError, json.JSONDecodeError, KeyError): + return False + + +def parent_receipt(path, capacity, epoch, plan): + path = Path(path) + assert parent_ready(path), str(path) + package = path / 'package' + verify_package(package) + signature = inference_signature(package) + assert json.loads((path / 'evaluation.json').read_text())['package_inference_signature'] == signature + provenance = json.loads((package / 'training_provenance.json').read_text()) + assert provenance['config'] == plan['capacities'][capacity]['config'] + assert provenance['global_step'] == epoch * 41020 and provenance['optimizer_steps'] == epoch * 10255 + assert provenance['epoch'] == epoch and provenance['offset'] == 0 + assert provenance['seed'] == 17 and provenance['schedule_epochs'] == 14 + assert provenance['objective'] == 'variance_balanced' and provenance['examples'] == 41020 + assert provenance['training_arguments']['batch_size'] == 4 + assert provenance['family_loss_scales'] == plan['family_loss_scales'] + source = provenance['provenance'] + assert source['split'] == 'train' and source['pretrained'] is False and source['world_size'] == 1 + assert source['train_manifest_sha256'] == plan['train_manifest_sha256'] + manifest = json.loads((package / 'package_manifest.json').read_text()) + assert manifest['parameters'] == plan['capacities'][capacity]['parameters'] + assert not any((package / name).exists() for name in [ + 'splice_calibration.json', 'checkpoint_ensemble_policy.json', 'rc_ensemble_policy.json']) + tracks = json.loads((path / 'valid-score/report.json').read_text()) + variants = json.loads((path / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + assert all(math.isfinite(value['model']) for value in variants.values()) + return dict(experiment=str(path), capacity=capacity, epoch=epoch, inference_signature=signature, + reward=tracks['overall_mean_family_pearson'], + sha256={str(p): digest(p) for p in [path / 'evaluation.json', path / 'evaluation-package.json', + path / 'valid-score/report.json', path / 'valid-variant-score/report.json', + package / 'training_provenance.json']}) + + +def prepare(): + STUDY.mkdir(exist_ok=True) + assert not (STUDY / 'plan.json').exists() + link69_path = ROOT / 'studies/fixed_link69/plan.json' + link69 = json.loads(link69_path.read_text()) + link61_path = ROOT / 'studies/fixed_link61/plan.json' + link61 = json.loads(link61_path.read_text()) + members = {} + for capacity, epoch, previous in [('large', 4, 2), ('small', 8, 4), ('large', 8, 4), + ('small', 14, 8), ('large', 14, 8)]: + key = f'{capacity}-epoch{epoch:02d}' + members[key] = dict(capacity=capacity, epoch=epoch, previous_epoch=previous, + parent=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{epoch:02d}'), + previous_raw=str(ROOT / 'experiments' / f'scale13-{capacity}-epoch{previous:02d}'), + output=str(ROOT / 'experiments' / f'calib81-{key}'), study=str(STUDY / key)) + (STUDY / key).mkdir(exist_ok=True) + capacities, evidence = {}, [link69_path, link61_path, ROOT / 'studies/fixed_link57/complete-summary.json', + ROOT / 'studies/scale13/study-cached-io.json', ROOT / 'studies/scale13/loss_scales.json', + Path('/workspace/data/public/train.parquet'), ROOT / 'experiments/calib57/package/splice_calibration.json'] + for capacity in ['small', 'large']: + package = ROOT / 'experiments' / f'scale13-{capacity}-epoch04/package' + provenance_path = package / 'training_provenance.json' + provenance = json.loads(provenance_path.read_text()) + capacities[capacity] = dict(config=provenance['config'], parameters=13290504 if capacity == 'small' else 215626632) + evidence.append(provenance_path) + helpers = set(link69['helper_sha256']) | { + 'late_fixed_links81.py', 'run_late_fixed_links81.py', 'diagnose_track_uncertainty.py', + 'diagnose_remaining_variant_uncertainty.py'} + materialization = dict(link69['materialization']) + materialization['installation'] = 'Atomic research-only alias to prepared-{candidate-name} within each member study; preserve while referenced. Never an inference asset.' + latency = dict(link61['native_latency_template']) + latency['scope'] = 'Native16-window/96-query replay for a completed late fixed-link candidate; all624arrays must equal complete evaluation and allthree4h estimates pass before any publication.' + latency['memory_evidence'] = 'One unchanged small or large main model plus the original four-offset elementwise splice link; require16GiB free for native probes. This is a scheduling guard, not a guaranteed peak bound.' + latency['minimum_free_GiB'] = 16 + plan = dict(created_utc=datetime.now(timezone.utc).isoformat(), members=members, capacities=capacities, + train_manifest_sha256=digest('/workspace/data/public/train.parquet'), + family_loss_scales=json.loads((ROOT / 'studies/scale13/loss_scales.json').read_text())['family_loss_scales'], + training_prior=link69['training_prior'], training_prior_sha256=link69['training_prior_sha256'], + calibration_reference=str(ROOT / 'experiments/calib57/package/splice_calibration.json'), + helper_sha256={str(HERE / name): digest(HERE / name) for name in sorted(helpers)}, + evidence_sha256={str(path): digest(path) for path in evidence}, materialization=materialization, + native_latency_template=latency, evaluation_lock=str(ROOT / 'studies/scale13/evaluation.lock'), + bounded_lock=str(ROOT / 'studies/junction_discovery39/bounded-diagnostic.lock'), + minimum_free_GiB=16, native_timeout_seconds=1800, materialization_timeout_seconds=1800, + parent_minimum_reward_gain=1e-6, publication_minimum_reward_gain=1e-6, + hypothesis='Transfer the same preselected training-prior sigmoid to the remaining main raw checkpoints. No new network updates, constants, orientation, site selection or sparse decoder changes. Prior link gains do not guarantee later track or variant improvements.', + admission='Wait for the declared raw parent and its preceding same-capacity raw checkpoint to finish200windows/2697queries/all13metrics. Require a raw reward gain>1e-6. Parent metadata must match exact whole epochs, all41020intervals, originalseed17/B4/14epochobjective and frozen capacity configs. Freeze actual parent signatures/reports only after completion; do not substitute checkpoints.', + validation='Original builder and verifier helpers; fresh native52/recovery, exact link-only core guard, original full12486array proof and36fresh native arrays, then fresh200track inference and full13officialmetrics using verified research-only perturbation reuse. Require10unaffectedfamily/eightvariant reports exact. Reward improvements additionally require624native arrays/allthree4h estimates before atomic promotion.', + diagnostics='After full evaluation, use unchanged paired track and remaining-seven variant diagnostics against each own raw parent,2000draws/seed1729. Five splice metrics may change; ca/SPI and eight other official reports are exact. This is not anall13bootstrap. Matched learning/capacity comparisons can be declared separately after both candidates complete.', + nonimprovement_policy='A fully scored candidate below the current publication skips optional native latency/publication, but still finishes its declared paired diagnostics. Original raw/other candidate selection pools and GPU schedules are unchanged.', + limitations=['The fixed marginal correction approximates per-window balanced BCE; no new prior fitting.', + 'The verifier applies its unchanged sigmoid to emitted splice probabilities; report all resulting tradeoffs.', + 'Intervals are exploratory/unadjusted and one training seed. Runtime extrapolations are shared-load estimates, not guaranteed bounds.', + 'No test labels, variant-label training, external parameters or inference caches are used.']) + write_json(STUDY / 'plan.json', plan) + print(json.dumps(dict(plan=str(STUDY / 'plan.json'), sha256=digest(STUDY / 'plan.json'), members=list(members))), flush=True) + + +def audit(): + assert not (STUDY / 'parent-check.json').exists() + plan = json.loads((STUDY / 'plan.json').read_text()) + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + receipts = [parent_receipt(ROOT / 'experiments' / name, capacity, epoch, plan) + for name, capacity, epoch in [('scale13-small-epoch04', 'small', 4), ('scale13-large-epoch02', 'large', 2)]] + write_json(STUDY / 'parent-check.json', dict(created_utc=datetime.now(timezone.utc).isoformat(), + plan_sha256=digest(STUDY / 'plan.json'), actual_completed_parent_checks=receipts, + gpu_used=False, packages_changed=False, no_new_predictions_or_parameter_fitting=True)) + print(json.dumps(dict(phase='parent_audit_complete', report_sha256=digest(STUDY / 'parent-check.json'))), flush=True) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('action', choices=['prepare', 'audit']) + globals()[parser.parse_args().action]() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_linear_upsample83.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_linear_upsample83.py new file mode 100644 index 0000000000000000000000000000000000000000..110da0ed7b9336e709813bfe66c4f889692d3b46 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_linear_upsample83.py @@ -0,0 +1,136 @@ +"""One fixed inference intervention: linearly upsample decoded global features.""" +import argparse +import ast +from datetime import datetime, timezone +import hashlib +import json +from pathlib import Path +import shutil + +from verify_package import inference_signature, verify_package + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def changed_source(source): + replacements = { + '_dense_features': ''' def _global_dense_features(self, trunk: Tensor) -> Tensor: + # Coarse features are centered in 1-kb bins; align_corners=False maps their + # centers onto the aligned 128-bp grid and extends boundary values constantly. + decoded = self.dense_decoder(trunk) + return F.interpolate(decoded.transpose(1, 2), scale_factor=self.config.dense_upsample, + mode='linear', align_corners=False).transpose(1, 2) + + def _dense_features(self, trunk: Tensor) -> Tensor: + first = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = self._global_dense_features(trunk)[:, first:first + self.config.dense_bins] + if self.config.local_width: + features = features + self._local_features[:, first:first + self.config.dense_bins] + return features''', + '_all_dense_features': ''' def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self._global_dense_features(trunk) + return features + self._local_features if self.config.local_width else features''', + '_position_features': ''' def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self._gather_bins(self._global_dense_features(trunk), positions, self.config.dense_bin_size) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features''', + } + tree = ast.parse(source) + cls = next(n for n in tree.body if isinstance(n, ast.ClassDef) and n.name == 'CompactModel') + assert not any(isinstance(n, ast.FunctionDef) and n.name == '_global_dense_features' for n in cls.body) + methods = {n.name:n for n in cls.body if isinstance(n, ast.FunctionDef)} + lines = source.splitlines(keepends=True) + for name in sorted(replacements, key=lambda k:methods[k].lineno, reverse=True): + node = methods[name] + lines[node.lineno-1:node.end_lineno] = [replacements[name]+'\n'] + changed = ''.join(lines) + new_tree = ast.parse(changed) + new_cls = next(n for n in new_tree.body if isinstance(n, ast.ClassDef) and n.name == 'CompactModel') + # The module is identical after removing the three replaced methods and new helper. + cls.body = [n for n in cls.body if not isinstance(n, ast.FunctionDef) or n.name not in replacements] + new_cls.body = [n for n in new_cls.body if not isinstance(n, ast.FunctionDef) or n.name not in {*replacements, '_global_dense_features'}] + assert ast.dump(tree, include_attributes=False) == ast.dump(new_tree, include_attributes=False) + return changed + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan', type=Path) + args = parser.parse_args() + plan, study = json.loads(args.plan.read_text()), args.plan.parent + parent, output = Path(plan['parent'])/'package', Path(plan['output'])/'package' + assert not output.exists(), output + assert inference_signature(parent) == plan['parent_inference_signature'] + verify_package(parent) + source = changed_source((parent/'reference/alphagenome_50m/model.py').read_text()) + assert hashlib.sha256(source.encode()).hexdigest() == plan['candidate_model_sha256'] + shutil.copytree(parent, output, ignore=shutil.ignore_patterns('__pycache__')) + (output/'reference/alphagenome_50m/model.py').write_text(source) + policy = dict(created_utc=datetime.now(timezone.utc).isoformat(), + parent_inference_signature=plan['parent_inference_signature'], + parameters=plan['parameters'], additional_training_updates=0, + interpolation='Linear decoded global features, align_corners=False, full1024->8192bin grid before central crop or position gathering.', + changed_methods=['_dense_features','_all_dense_features','_position_features'], + new_helper='_global_dense_features', + learned_weights_unchanged=True, local_path_unchanged=True, + both_entrypoints_share_model=True, contact_path_unchanged=True, + scope='Inference-only architectural intervention after the inherited training. No refitting, auxiliary input, label or target access.') + (output/'global_upsampling_policy.json').write_text(json.dumps(policy,indent=2)+'\n') + provenance_path=output/'training_provenance.json' + provenance=json.loads(provenance_path.read_text());provenance['global_upsampling_intervention']=policy + provenance_path.write_text(json.dumps(provenance,indent=2)+'\n') + readme=output/'README.md';body=readme.read_text().split('\n## Measured validation results')[0].split('\n',1)[1] + body=body.replace('In preliminary\ndiagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed.', + "The complete reports below measure this checkpoint's track and variant tradeoffs.") + note=''' +This candidate changes one architectural operation after training: decoded global features +at 1 kb are interpolated linearly onto the 128 bp grid instead of repeated eight times. +Interpolation uses bin centers (align_corners=False) over the full input before cropping. +Dense outputs, splice logits, usage and junction position queries share this feature grid. +The local base-resolution path, transformer, output heads, four fixed splice offsets and +all learned weights remain inherited. Contacts use the inherited global trunk directly. + +The intervention adds no parameters or optimizer updates; the inherited training used +repeated global features. A public-DNA-only coordinate-aligned probe motivated the test: +128 bp shifts changed effects much more than 1 kb shifts. Encoder aliasing and changed +edge context can also contribute, so smoother upsampling does not guarantee accuracy or +shift invariance. This package requires its own full track and development variant scores, +native replay and representative runtime checks. No values are fit to development labels. +''' + readme.write_text('# SeqFunc small3 with linear global upsampling\n'+body+note) + history=output/'training_history';history.mkdir(exist_ok=True) + (history/'linear-upsample83-plan.json').write_bytes(args.plan.read_bytes()) + (history/'make_linear_upsample83.py').write_bytes(Path(__file__).read_bytes()) + (history/'pre-linear-upsample-model.py').write_bytes((parent/'reference/alphagenome_50m/model.py').read_bytes()) + for name in ['run_linear_upsample83.py','check_linear_upsample83.py']: + (history/name).write_bytes((Path(__file__).parent/name).read_bytes()) + (history/'linear-upsample-replay.md').write_text( + 'The inherited training used pre-linear-upsample-model.py at reference/alphagenome_50m/model.py.\n' + 'Restore that source when reproducing the inherited training trajectory; all training configuration,\n' + 'objective, interval order and optimizer provenance remain inherited. The installed model.py is\n' + 'the post-training inference intervention, generated by make_linear_upsample83.py.\n' + 'No optimizer update produced the intervention. Native inference uses the installed source.\n') + manifest=json.loads((output/'package_manifest.json').read_text()) + for name in ['validation_tracks.json','validation_variants.json']: + (output/name).unlink(missing_ok=True);manifest['files'].pop(name,None) + changed=['reference/alphagenome_50m/model.py','global_upsampling_policy.json', + 'training_provenance.json','README.md','training_history/linear-upsample83-plan.json', + 'training_history/make_linear_upsample83.py','training_history/pre-linear-upsample-model.py', + 'training_history/run_linear_upsample83.py','training_history/check_linear_upsample83.py', + 'training_history/linear-upsample-replay.md'] + for name in changed:manifest['files'][name]=digest(output/name) + manifest['files']={k:v for k,v in manifest['files'].items() if '__pycache__' not in Path(k).parts} + (output/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + actual=inference_signature(output) + assert {k for k in actual if actual[k]!=plan['parent_inference_signature'][k]} == {'reference/alphagenome_50m/model.py'} + for name in ['last.pt','predict.py','perturb.py','infer.sh','perturb.sh','runtime.tar.gz']: + assert digest(output/name)==digest(parent/name),name + print(json.dumps(dict(event='linear_upsample_candidate_built',**verify_package(output))),flush=True) + + +if __name__=='__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_rc_candidate.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_rc_candidate.py new file mode 100644 index 0000000000000000000000000000000000000000..e9d24e695a507789457828af4251e6fb468de827 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_rc_candidate.py @@ -0,0 +1,100 @@ +"""Build the verified RC17 two-view decoder around a compatible local checkpoint. + +Strict parent-decoder hashes prevent silently applying this extraction to an +incompatible output head or sparse-selection policy. No predictions are cached. +""" +import argparse +from datetime import datetime, timezone +import hashlib +import json +from pathlib import Path +import shutil + +from verify_package import inference_signature, verify_package + + +def digest(path): + with path.open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('parent', type=Path) + parser.add_argument('output', type=Path) + parser.add_argument('--name', default='SeqFunc forward/reverse-complement mean') + parser.add_argument('--decoder-study', type=Path, + default=Path('/workspace/research/studies/rc17')) + args = parser.parse_args() + if args.output.exists(): + raise FileExistsError(args.output) + if (args.parent/'rc_ensemble.py').exists(): + raise ValueError('Parent already has orientation averaging') + verify_package(args.parent) + plan = json.loads((args.decoder_study/'study.json').read_text()) + source = args.decoder_study/'source' + for name in ['predict.py', 'perturb.py']: + expected = plan['source_files_sha256'][name.replace('.py', '-parent.py')] + if digest(args.parent/name) != expected: + raise ValueError(f'Parent {name} differs from the verified extraction; review the decoder before adapting it') + assets = ['predict.py', 'perturb.py', 'rc_ensemble.py', 'single_prediction.py', + 'strand_permutations.json'] + for name in assets: + assert digest(source/name) == plan['source_files_sha256'][name], name + shutil.copytree(args.parent, args.output, ignore=shutil.ignore_patterns('__pycache__')) + manifest = json.loads((args.output/'package_manifest.json').read_text()) + for name in ['validation_tracks.json', 'validation_variants.json']: + (args.output/name).unlink(missing_ok=True) + manifest['files'].pop(name, None) + for name in assets: + shutil.copyfile(source/name, args.output/name) + policy = {key: plan[key] for key in ['method', 'orientation_views', 'fitted_ensemble_weights', + 'weights', 'dna_input', 'site_scores', 'sparse_outputs', 'outputs']} + policy.update(created_utc=datetime.now(timezone.utc).isoformat(), + parent_package=str(args.parent.resolve()), + parent_inference_signature=inference_signature(args.parent), + unique_learned_parameters=manifest['parameters'], + decoder_study=str(args.decoder_study.resolve()), + decoder_assets_sha256={name: digest(source/name) for name in assets}) + (args.output/'rc_ensemble_policy.json').write_text(json.dumps(policy, indent=2)+'\n') + provenance_file = args.output/'training_provenance.json' + provenance = json.loads(provenance_file.read_text()) + provenance['inference_ensemble'] = policy + provenance_file.write_text(json.dumps(provenance, indent=2)+'\n') + readme = args.output/'README.md' + text = readme.read_text().split('\n## Measured validation results')[0] + _, body = text.split('\n', 1) + text = '# '+args.name+'\n'+body + text += f''' +Every output family and both entry points average this same checkpoint on the provided DNA +and its reverse complement, mapped back using vendored public-metadata strand permutations. +There are {manifest['parameters']:,} unique learned parameters and two deterministic views, +weighted equally without fitting development labels. No additional flanking DNA is read. +Dense and contact values average in the contract's asinh space; splice scores average the +training-prior-corrected probabilities. This approximately doubles inference work. + +Each view selects sparse coordinates independently. Their union averages missing-view values +as zero. Usage retains at most4,096 positions; junctions retain16,384 pairs by averaged own-strand +strength, with deterministic coordinate tie handling. Track arrays are averaged in float32 and +then stored in float16; perturbation arrays remain float32. Sparse recall and score dilution +can both change. The frozen decoder passed the RC17 synthetic, native offline and partial-resume +checks; its exact parent-extraction and source fingerprints are enforced by the builder. +This candidate requires its own complete validation track and development variant reports +before publication. Architecture, learned weights and training history are inherited unchanged. +''' + readme.write_text(text) + changed = [*assets, 'rc_ensemble_policy.json', 'training_provenance.json', 'README.md'] + for name in changed: + manifest['files'][name] = digest(args.output/name) + for name in list(manifest['files']): + if '__pycache__' in Path(name).parts: + manifest['files'].pop(name) + (args.output/'package_manifest.json').write_text(json.dumps(manifest, indent=2)+'\n') + result = verify_package(args.output) + assert digest(args.output/'last.pt') == digest(args.parent/'last.pt') + print(json.dumps({'event': 'rc_candidate_built', 'output': str(args.output), + 'weights_unchanged': True, **result}), flush=True) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_splice_grid96.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_splice_grid96.py new file mode 100644 index 0000000000000000000000000000000000000000..c4e7cb81bb064023689cc3fdc5b8edf79de8ee21 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/make_splice_grid96.py @@ -0,0 +1,132 @@ +"""Restore the trained nearest-neighbor splice-classifier grid in one Linear83 network.""" +import argparse +from datetime import datetime, timezone +import hashlib +import json +from pathlib import Path +import shutil + +from verify_package import inference_signature, verify_package + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def changed_source(source): + before = ' logits = self.splice_site_head(self._all_dense_features(trunk))\n' + after = ''' # The splice classifier retains the feature grid used during training. + # Dense, usage and junction readouts keep the separate interpolated grid. + features = self.dense_decoder(trunk.repeat_interleave(self.config.dense_upsample, dim=1)) + if self.config.local_width: + features = features + self._local_features + logits = self.splice_site_head(features) +''' + assert source.count(before) == 1 + changed = source.replace(before, after) + assert changed.replace(after, before) == source + return changed + + +def main(): + parser=argparse.ArgumentParser(description=__doc__);parser.add_argument('plan',type=Path) + args=parser.parse_args();plan=json.loads(args.plan.read_text());study=args.plan.parent + parent=Path(plan['parent'])/'package';output=Path(plan['output'])/'package' + assert not output.exists() + assert inference_signature(parent)==plan['parent_inference_signature'] + verify_package(parent) + source=changed_source((parent/'reference/alphagenome_50m/model.py').read_text()) + assert hashlib.sha256(source.encode()).hexdigest()==plan['candidate_model_sha256'] + shutil.copytree(parent,output,ignore=shutil.ignore_patterns('__pycache__','*.pyc')) + (output/'reference/alphagenome_50m/model.py').write_text(source) + policy=dict(created_utc=datetime.now(timezone.utc).isoformat(), + parent_inference_signature=plan['parent_inference_signature'],parameters=plan['parameters'], + additional_training_updates=0,learned_weights_unchanged=True, + only_changed_method='_splice_site_logits',splice_classifier_grid='nearest-neighbor repetition, as trained', + dense_usage_junction_grid='inherited bin-center linear interpolation', + local_and_global_backbone_unchanged=True,contacts_unchanged=True, + both_entrypoints_share_one_fixed_model=True, + scope='One fixed internal readout path, independent of input manifest or evaluation identity. ' + 'All outputs are computed from public DNA by the same network. No prediction mixing, fitting or caches.') + (output/'splice_grid_policy.json').write_text(json.dumps(policy,indent=2)+'\n') + provenance=json.loads((output/'training_provenance.json').read_text()) + provenance['splice_classifier_grid_intervention']=policy + (output/'training_provenance.json').write_text(json.dumps(provenance,indent=2)+'\n') + history=output/'training_history' + (history/'pre-splice-grid96-model.py').write_bytes((parent/'reference/alphagenome_50m/model.py').read_bytes()) + (history/'splice-grid96-plan.json').write_bytes(args.plan.read_bytes()) + for name in ['make_splice_grid96.py','check_splice_grid96.py','run_splice_grid96.py']: + (history/name).write_bytes((Path(__file__).parent/name).read_bytes()) + old_global=json.loads((output/'global_upsampling_policy.json').read_text()) + (history/'linear83-global-upsampling-policy.json').write_text(json.dumps(old_global,indent=2)+'\n') + old_global['splice_classifier_grid_refinement']='Restored nearest-neighbor classifier input by splice_grid_policy.json; dense/usage/junction interpolation remains inherited.' + (output/'global_upsampling_policy.json').write_text(json.dumps(old_global,indent=2)+'\n') + (output/'README.md').write_text('''# SeqFunc with linear readouts and the trained splice-classifier grid + +One **13,290,504-parameter** network reads the complete1,048,576-bp DNA window and serves +all eleven output families and both entrypoints. A convolutional encoder and six width384 +transformer blocks process1,024 global tokens. A width32 base-resolution path has a21-bp +stem and six dilated residual blocks, with10,941-bp receptive field. Its pooled features +also enter the global trunk through a learned projection. Transformer updates retain the +inherited per-token RMS cap0.1. + +Decoded global features are interpolated from1kb to128bp at bin centers for the seven +dense families and the usage/junction readouts. The splice classifier uses the original +repeated global features, with its learned128-offset decoder and local base-resolution +logits. These are fixed internal paths of the same backbone and learned heads. Contact +maps retain the symmetric global-feature product. Model-selected splice coordinates and +strand-consistent junction pairs are used; inference receives no splice coordinates. + +Weights come from local random initialization, seed17: **123,060 training examples, +30,765 optimizer updates, three complete passes** over all41,020 eligible FOLD_0 training +intervals (folds2–7). Train8-bit targets are decoded per example by the supplied reader. +Batch4 BF16 AdamW uses weight decay1e-4, clipping5 and the original14-epoch cosine schedule +from2e-4 to2e-5 with4,000-example warmup. Fixed variance-balanced family scales use256 +training windows; splice BCE is balanced. Accumulated inherited training-loop time is +494.6minutes on one shared H10080GB. Setup/evaluation and other experiments are additional +research compute. Training source, configuration, history and provenance are packaged. + +All three training passes used nearest-neighbor global upsampling. Linear83 introduced +interpolation after training. This candidate restores only the splice classifier's trained +grid; it adds **zero parameters and zero training updates**. Four original training-derived +splice offsets and the fixed probability link remain unchanged. They were not refit. +Global usage/junction features still interpolate, but predicted site selection can change; +their track accuracy therefore needs a fresh full evaluation. + +This experiment follows Linear83's higher track reward and lower ClinVar non-coding AP. +It is selected using validation evidence, with one training seed and no multiplicity +adjustment. Its own validation measurements are appended only after complete scoring. +No test labels, externally pretrained weights or variant-label training were used. + +Both executable entrypoints take INPUT_ROOT and OUTPUT_ROOT and run offline from public +manifests and the pinned reference FASTA. CPython3.11,PyTorch2.7.1/CUDA12.6 and dependencies +are vendored in runtime.tar.gz; the host supplies Linux,bash/tar/gzip and an NVIDIA driver. +Track outputs use float16 and perturbations float32. No prediction/target cache is included. +The same fixed model computes every requested output from DNA in both entrypoints. + +Limitations include three-pass training, pooled usage/junction features, low-rank contacts, +the finite local receptive field and uneven variant performance. The54 masked GTEx RNA +tracks have no training targets and are excluded from scored RNA means. Functional readouts +do not supply the conservation information used by strong TraitGym anchors. Native replay, +complete200-window/2,697-query development scoring and representative four-hour runtime +gates are required before publication; runtime estimates are not guaranteed bounds. +''') + manifest=json.loads((output/'package_manifest.json').read_text()) + for name in ['validation_tracks.json','validation_variants.json']: + (output/name).unlink(missing_ok=True);manifest['files'].pop(name,None) + updated=['reference/alphagenome_50m/model.py','splice_grid_policy.json','global_upsampling_policy.json', + 'training_provenance.json','README.md','training_history/pre-splice-grid96-model.py', + 'training_history/splice-grid96-plan.json','training_history/linear83-global-upsampling-policy.json', + *['training_history/'+name for name in ['make_splice_grid96.py','check_splice_grid96.py','run_splice_grid96.py']]] + for name in updated:manifest['files'][name]=digest(output/name) + manifest['files']={k:v for k,v in manifest['files'].items() if '__pycache__' not in Path(k).parts} + (output/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + signature=inference_signature(output) + assert {k for k in signature if signature[k]!=plan['parent_inference_signature'][k]}=={'reference/alphagenome_50m/model.py'} + for name in ['last.pt','predict.py','perturb.py','infer.sh','perturb.sh','runtime.tar.gz','splice_calibration.json','splice_output.py']: + assert digest(output/name)==digest(parent/name),name + print(json.dumps(dict(event='splice_grid96_built',**verify_package(output))),flush=True) + + +if __name__=='__main__':main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/memory-check-interruption.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/memory-check-interruption.json new file mode 100644 index 0000000000000000000000000000000000000000..19bc02cd2784b11f5f3e658efa28017639c2efeb --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/memory-check-interruption.json @@ -0,0 +1,19 @@ +{ + "time_utc": "2026-10-24T21:49:06.154124+00:00", + "reason": "Mechanical activation-checkpoint memory/equivalence check; optimizer, data order and model math preserved.", + "process_group": 4169, + "checkpoints": [ + { + "run": "scale13-small", + "step": 1000, + "path": "/workspace/research/experiments/scale13-small/step-001000.pt", + "sha256": "a53ea6d92c21c06576c734dc05eac6e3f21d33701f25bfc335817a6e6ec197fc" + }, + { + "run": "scale13-large", + "step": 1000, + "path": "/workspace/research/experiments/scale13-large/step-001000.pt", + "sha256": "7398dd28297d6cc3134b8d3b0cea703fb0bb0a8e60b4c0c2d81abe04c1d7dc43" + } + ] +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/migrate_scale_bytecache.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/migrate_scale_bytecache.py new file mode 100644 index 0000000000000000000000000000000000000000..05b927345326f5b660fa5b28118ce867eee97c88 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/migrate_scale_bytecache.py @@ -0,0 +1,235 @@ +"""Checkpoint-safe opt-in byte-cache amendment, preserving an active evaluation.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import os +from pathlib import Path +import shutil +import signal +import sys +import time + +import torch + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def identity(pid): + p = Path(f'/proc/{pid}') + stat = (p/'stat').read_text().split(') ', 1)[1].split() + return {'pid': pid, 'starttime_ticks': int(stat[19]), + 'cmdline_sha256': hashlib.sha256((p/'cmdline').read_bytes()).hexdigest()} + + +def command(pid): + return Path(f'/proc/{pid}/cmdline').read_bytes().replace(b'\0', b' ').decode() + + +def alive(pid): + try: + return Path(f'/proc/{pid}/stat').read_text().split(') ', 1)[1][0] not in 'ZX' + except FileNotFoundError: + return False + + +def main(): + p = argparse.ArgumentParser(description=__doc__) + p.add_argument('plan', type=Path) + args = p.parse_args() + plan = json.loads(args.plan.read_text()) + r = Path('/workspace/research') + root = args.plan.parent + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + torch.set_num_threads(1) + for name, value in plan['helper_sha256'].items(): + assert digest(r/'seqfunc'/name) == value, name + assert digest(plan['original_study']) == plan['original_study_sha256'] + original = json.loads(Path(plan['original_study']).read_text()) + prepared = json.loads(Path(plan['preparation']).read_text()) + integrated = json.loads(Path(plan['integration_check']).read_text()) + timing = json.loads(Path(plan['timing_report']).read_text()) + assert integrated['all_returned_training_ids_and_tensor_bytes_exact'] and integrated['examples'] == 16 + assert integrated['public_inference_and_validation_cache_disabled_despite_environment'] + assert timing['cache_over_reference'] < .8 and timing['all_probe_updates_discarded'] + source = Path(prepared['new_source']) + for name, value in prepared['new_files_sha256'].items(): + assert digest(source/name) == value, name + for path in Path(original['source']).rglob('*'): + if path.is_file() and '__pycache__' not in path.parts and '.pytest_cache' not in path.parts: + relative = path.relative_to(original['source']) + if str(relative) != 'reference/alphagenome_50m/dataset.py': + assert digest(path) == digest(source/relative), str(relative) + old_controller = plan['old_controller_pid'] + assert 'run_scale_study.py' in command(old_controller) + assert os.getpgid(old_controller) != os.getpgrp() + processes = json.loads((root/'study-state.json').read_text()) + for name, target in plan['boundaries'].items(): + assert target % 1000 == 0 and target % 4 == 0 + pid = processes[name+'-train']['pid'] + assert str(Path(original['source'])/'train_batched.py') in command(pid) + assert str(r/'experiments'/name) in command(pid) + stopped, snapshots, seen = set(), {}, {} + killed = False + controller_stopped = False + begun = time.monotonic() + + def status(phase, **extra): + record = {'phase': phase, 'utc': datetime.now(timezone.utc).isoformat(), + 'boundaries': plan['boundaries'], 'checkpoints': snapshots, **extra} + temporary = root/'cached-io-migration-state.tmp' + temporary.write_text(json.dumps(record, indent=2)+'\n') + temporary.replace(root/'cached-io-migration-state.json') + print(json.dumps(record), flush=True) + + def interrupted(signum, frame): + raise SystemExit(f'Migration interrupted by signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + status('waiting_for_saved_boundaries') + try: + while len(snapshots) < len(plan['boundaries']): + assert time.monotonic()-begun < 3600, 'Boundary wait exceeded one hour' + for name, target in plan['boundaries'].items(): + if name in snapshots: + continue + pid = processes[name+'-train']['pid'] + assert alive(pid), f'Trainer {name} exited before capture' + progress = r/'experiments'/name/'progress.pt' + with progress.open('rb') as stream: + stat = os.fstat(stream.fileno()) + signature = (stat.st_ino, stat.st_mtime_ns, stat.st_size) + if signature == seen.get(name): + continue + state = torch.load(f'/proc/self/fd/{stream.fileno()}', mmap=True, + map_location='cpu', weights_only=True) + step = state['global_step'] + assert step <= target, f'{name} missed boundary at {step}' + seen[name] = signature + if step == target: + assert state['optimizer_steps'] == step//4 + assert state['objective'] == 'variance_balanced' + os.kill(pid, signal.SIGSTOP) + stopped.add(pid) + destination = progress.with_name(f'bytecache-resume-{step:06d}.pt') + assert not destination.exists() + temporary = destination.with_suffix('.tmp') + stream.seek(0) + with temporary.open('wb') as output: + shutil.copyfileobj(stream, output, length=8*1024*1024) + temporary.replace(destination) + snapshots[name] = {'step': step, 'path': str(destination), + 'sha256': digest(destination), 'single_inode_copy': True} + status('captured_saved_boundary', trainer=name) + del state + time.sleep(1) + # All pending later evaluations still wait for checkpoints. Preserve the + # already staged small epoch2 evaluation, which belongs to the old reader. + keep = 'scale13-small-eval02' + adopted = {} + keep_directory = r/'experiments/scale13-small-epoch02' + if json.loads((keep_directory/'pipeline-state.json').read_text())['phase'] != 'complete': + adopted[keep] = identity(processes[keep]['pid']) + assert 'run_candidate.py' in command(processes[keep]['pid']) + terminate_waiters = [] + for name, record in processes.items(): + if '-eval' not in name or name in adopted or record['returncode'] is not None: + continue + pid = record['pid'] + directory = r/'experiments'/name.replace('-eval', '-epoch') + if json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'complete': + continue + assert json.loads((directory/'pipeline-state.json').read_text())['phase'] == 'waiting_for_checkpoint' + assert not Path(f'/proc/{pid}/task/{pid}/children').read_text().strip() + terminate_waiters.append(pid) + # Keep this old session alive until its adopted evaluation finishes. + # Its monitor is stopped, so only the new controller owns training status. + old_identity = identity(old_controller) + os.kill(old_controller, signal.SIGSTOP) + controller_stopped = True + worker_pids = [] + for name in plan['boundaries']: + pid = processes[name+'-train']['pid'] + worker_pids += [int(p) for p in Path(f'/proc/{pid}/task/{pid}/children').read_text().split()] + terminated = terminate_waiters+worker_pids+[processes[name+'-train']['pid'] for name in plan['boundaries']] + status('replacing_trainers_and_idle_waiters', preserved_evaluations=adopted, terminated_pids=terminated) + for pid in terminated: + if alive(pid): + os.kill(pid, signal.SIGTERM) + os.kill(pid, signal.SIGCONT) + killed = True + stopped.clear() + deadline = time.monotonic()+60 + while any(alive(pid) for pid in terminated): + assert time.monotonic() < deadline, 'Old training or idle waiter did not exit' + time.sleep(1) + for name, record in snapshots.items(): + assert digest(r/'experiments'/name/'progress.pt') == record['sha256'] + amendment = {'adopted_utc': datetime.now(timezone.utc).isoformat(), + 'change': 'Optional verified compressed-byte cache, training only; every hit checks source-byte digest, key and length.', + 'source_before': original['source'], 'source_after': str(source), + 'checkpoints': snapshots, 'cache_root': plan['cache_root'], + 'model_optimizer_rng_exact_at_each_resume': True, + 'model_trainer_loss_sources_data_order_objective_batch_schedule_unchanged': True, + 'different_resume_boundaries_are_IO_only': True, + 'integration_check': plan['integration_check'], 'timing_report': plan['timing_report'], + 'timing_cache_over_reference': timing['cache_over_reference'], + 'timing_caveat': timing['caveats'], + 'cache_not_an_inference_asset': True, + 'adopted_evaluations_without_interruption': adopted, + 'old_monitor_retained_stopped_until_adopted_evaluation_finishes': old_identity} + new = dict(original) + new.update(source=str(source), training_bytecache_root=plan['cache_root'], + cache_execution_amendment=amendment, adopted_evaluations=adopted, + retired_monitor_identity=old_identity) + new['execution_amendments'] = list(original.get('execution_amendments', []))+[amendment] + new['provenance'] += (' At the saved small96k/large78k boundaries, an optional training-only byte cache ' + 'was enabled after exact256example and integrated16example tensor/ID checks. ' + 'Source compressed bytes are verified on every cache hit, with read-only source fallback. ' + 'Model/trainer/loss code, complete epoch permutations, batch and schedule are unchanged. ' + 'The cache is disposable and is disabled for validation/public inference.') + history = source/'training_history' + (history/'bytecache-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + for name in ['preparation', 'integration_check', 'timing_report']: + path = Path(plan[name]);shutil.copyfile(path, history/('bytecache-'+name+'.json')) + shutil.copyfile(args.plan, history/'bytecache-migration-plan.json') + shutil.copyfile(Path(__file__), history/Path(__file__).name) + shutil.copyfile(r/'seqfunc/run_scale_study_cached_io.py', history/'run_scale_study_cached_io.py') + new_path = root/'study-cached-io.json' + new_path.write_text(json.dumps(new, indent=2)+'\n') + manifest = {str(p.relative_to(source)): digest(p) for p in sorted(source.rglob('*')) + if p.is_file() and '__pycache__' not in p.parts and '.pytest_cache' not in p.parts} + (root/'source_manifest-cached-io.json').write_text(json.dumps(manifest, indent=2)+'\n') + (root/'cached-io-amendment.json').write_text(json.dumps(amendment, indent=2)+'\n') + with Path('/workspace/.submission-promotion.lock').open('a') as lock: + fcntl.flock(lock, fcntl.LOCK_EX) + p = r/'STATE.json'; state = json.loads(p.read_text()) + state['scale_study'].update(plan=str(new_path), source=str(source), + controller_pid=os.getpid(), status='Resuming original trajectories with verified opt-in training byte cache', + cached_reader_amendment=str(root/'cached-io-amendment.json'), + previous_controller_retained_stopped=old_controller) + state['scale_study']['session'] = state['scale_study']['bytecache_migration']['session'] + state['scale_study']['bytecache_migration']['status'] = 'Checkpoints captured; replacement controller starting' + temporary=p.with_suffix('.tmp');temporary.write_text(json.dumps(state,indent=2)+'\n');temporary.replace(p) + with (r/'LEDGER.md').open('a') as ledger: + ledger.write('\n'+amendment['adopted_utc']+': saved small96k/large78k trajectories resumed with optional verified training byte cache; original scientific plans unchanged. Active small epoch2 evaluation preserved. Original study/source/checkpoints retained; full amendment at studies/scale13/cached-io-amendment.json.\n') + status('replacement_controller_starting', source=str(source), new_study=str(new_path)) + log = (root/'study-cached-io.log').open('a') + os.dup2(log.fileno(), 1);os.dup2(log.fileno(), 2) + os.execv(sys.executable,[sys.executable,'-u',str(r/'seqfunc/run_scale_study_cached_io.py'),str(new_path)]) + except BaseException as error: + status('failed_or_interrupted', error=repr(error), old_trainers_terminated=killed) + raise + finally: + if not killed: + for pid in stopped: + if alive(pid):os.kill(pid,signal.SIGCONT) + if controller_stopped and alive(old_controller):os.kill(old_controller,signal.SIGCONT) + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/model_before_block_checkpoint.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/model_before_block_checkpoint.py new file mode 100644 index 0000000000000000000000000000000000000000..31f17eb68837a9caa6492ec862feda39d068a43f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/model_before_block_checkpoint.py @@ -0,0 +1,457 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + base = F.gelu(self.norm(self.blocks(base)), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + if self.config.gradient_checkpointing and self.training: + sites, pooled = checkpoint(self.local_path, dna.transpose(1, 2), use_reentrant=False) + else: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-amendment.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-amendment.json new file mode 100644 index 0000000000000000000000000000000000000000..48ca3e0d57ff557a5fd2bd2bc517508c6de05b0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-amendment.json @@ -0,0 +1,27 @@ +{ + "adopted_utc": "2026-10-25T00:42:46.303812+00:00", + "example_boundary": 40000, + "change": "I/O only: direct read and four bounded chunk-read threads per loader process.", + "source_before": "/workspace/research/studies/scale13/source-block-checkpoint", + "source_after": "/workspace/research/studies/scale13/source-parallel-io", + "model_parameters_optimizer_RNG_data_order_objective_schedule_unchanged": true, + "loader_worker_counts_unchanged": true, + "common_immutable_checkpoints": { + "scale13-small": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-small/io-resume-040000.pt", + "sha256": "62daa49a4dc329d0b7cf3ecea0a30834346f714079d975bc636bdd81102f38f2", + "single_inode_copy": true + }, + "scale13-large": { + "step": 40000, + "path": "/workspace/research/experiments/scale13-large/io-resume-040000.pt", + "sha256": "db6c7394de11aba07278c907f328bcf18200c01ab1fb23578b12d293eb7219f4", + "single_inode_copy": true + } + }, + "evidence": "256 full examples through four forked workers and16 integrated-source examples, every tensor byte and ordered ID identical; missing-chunk fixtures pass.", + "loader_parallel_over_reference_time": 0.5340049503280426, + "timing_caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included.", + "no_benchmark_weights_retained": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-integration-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-integration-check.json new file mode 100644 index 0000000000000000000000000000000000000000..652abba42a037aa1738675ff6c33ac0a769ffe0f --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-io-integration-check.json @@ -0,0 +1,92 @@ +{ + "created_utc": "2026-10-25T00:29:21.312924+00:00", + "all_returned_training_tensor_bytes_exact": true, + "source_sha256": { + "original": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "parallel": "dc4263a10fecca979811161da97e731c471468465fe9480bbf2a9aacca68fb7f" + }, + "seed": 8585, + "examples": [ + { + "index": 12814, + "id": "agv2-b546f0182590dcfe5eec816c", + "sha256": "7a9dc9eca02761b7f5e14c311863c3974971cddf58a1b6549a2347d0c26bc001" + }, + { + "index": 1383, + "id": "agv2-b9a6e9555a2ac818d1dac07a", + "sha256": "a7e4e9b7dd20a97e630331bd939013f30ae9de72db0df646d95b02f9705fb942" + }, + { + "index": 12116, + "id": "agv2-86084f0f2fbb8fa0b4041b84", + "sha256": "8174fc2d6b2cff62aa531ad79326852cc5dbc8913b6d22ec6e7bfe9f7b7d54a9" + }, + { + "index": 39401, + "id": "agv2-4de6a38a66c26f774a8f36d0", + "sha256": "aa37d4aae2583af45d8f10fcac8e15c6ad024734f8a4c03874e899ef7d6d7ee3" + }, + { + "index": 6056, + "id": "agv2-aae7027f20edc50c08c46283", + "sha256": "8786d8b2399f4da03b49af989ab7f253bfcb2c8e07c149c5bfed2b37c579d1a9" + }, + { + "index": 35710, + "id": "agv2-8e59f96b89d66502382ebc3a", + "sha256": "796e4a560f99b950abc2564592a867dfef8a871806169240d665ba7d0bec374f" + }, + { + "index": 5687, + "id": "agv2-6ae7660428f809240c6a66f4", + "sha256": "d861dcefd6decc06327587f75654b696346bb601cd5fa1e78ec29627cf1c991c" + }, + { + "index": 23822, + "id": "agv2-4080f99a4ff1fa78c353b790", + "sha256": "b1298cbebf6c42815f4417b71d1004af18106f47bcf246e4f49ee9eab7767f7c" + }, + { + "index": 7090, + "id": "agv2-06072490dcad81dd885092ef", + "sha256": "bfc34315814bdfd19c554d4cffeae8176b6954783ad5439871add7920d571dfa" + }, + { + "index": 26573, + "id": "agv2-7f9d85bd9de0885faef7310f", + "sha256": "dbebfb62d397d713aab8a75f663beafce0aaf8ba6365f44d419e490032dc591e" + }, + { + "index": 12702, + "id": "agv2-11309128e469c8ca46cb1146", + "sha256": "64ab77e7ca11eeff0d631928cb71fda4ee3491557c277c183f1f60c8672c13c8" + }, + { + "index": 28025, + "id": "agv2-ac4fef8454a821046718c5a0", + "sha256": "d81d676be913e405949c5b92317dbb512d37133255f3ff071975c08415e23356" + }, + { + "index": 10066, + "id": "agv2-53a75e497596ef94806b0c63", + "sha256": "9778bf8f8bad19354943fa481cb227f79817056ca026b085e805e78fe192f7b9" + }, + { + "index": 18162, + "id": "agv2-b1accce145a7ed155cec1929", + "sha256": "c16cb2f6005b45ac88f05891044db668944f0fd9dcff2527b8beeaef4f8f65d6" + }, + { + "index": 24160, + "id": "agv2-a0f157f236929d108c1b7cf9", + "sha256": "0711ba4727d71ffcc86bd336ddf1b250c4fde37ed74cc4b0067f75d4ddd20f23" + }, + { + "index": 37802, + "id": "agv2-7bc22c9b204aac797187f069", + "sha256": "7b7b9b848661f6840139affae8c60fe1748a17a74e4bcb9871fb7ecf1a767a56" + } + ], + "active_trainers_unchanged": true +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-loader4x4-train256.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-loader4x4-train256.json new file mode 100644 index 0000000000000000000000000000000000000000..4b7189856ab6ebfd8e1f1fd3cbfb8ef78517a429 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/parallel-loader4x4-train256.json @@ -0,0 +1,2403 @@ +{ + "created_utc": "2026-10-25T00:28:12.867145+00:00", + "split": "train", + "no_training_or_inference_changes": true, + "all_ordered_ids_and_tensor_bytes_exact": true, + "loader_workers": 4, + "io_workers_per_worker_process": 4, + "threads_bounded_per_process_and_cpu_affinity_22": true, + "rounds": [ + { + "round": 0, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 38676, + 39399, + 35909, + 24618, + 421, + 22423, + 14151, + 2127, + 9300, + 12706, + 18005, + 17114, + 16691, + 18591, + 8219, + 37349, + 25569, + 37577, + 32169, + 40794, + 3729, + 6012, + 13718, + 27980, + 24199, + 37627, + 29258, + 6870, + 11257, + 32770, + 15891, + 26903, + 13569, + 34719, + 4817, + 25612, + 40663, + 9353, + 16565, + 29973, + 20423, + 12428, + 13055, + 26663, + 22489, + 6628, + 33720, + 6237, + 17185, + 34264, + 8196, + 37504, + 27822, + 5892, + 5753, + 27082, + 19966, + 23445, + 2827, + 39710, + 25082, + 18336, + 26856, + 16513 + ], + "seconds": { + "reference": 30.35943350600428, + "parallel": 5.833638833020814 + }, + "results": { + "reference": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ], + "parallel": [ + { + "id": "agv2-b4691d69dbaeb6e36464ee7c", + "sha256": "32e66dc611076f377d7c7a083ee93c3ef44cd0fd3085a15db410448a2ca156b0" + }, + { + "id": "agv2-14596c105cd65f7747cd2c8f", + "sha256": "9059a0bc2845a76baee98167b84e623b3e719cf34da1cbe196b887293f3a3998" + }, + { + "id": "agv2-c3fdf7bccfa5fb3293a46f59", + "sha256": "7eacbf86baec97ff355e6e81ae139f8b18bd86bddcce982323ba4a693d7338d4" + }, + { + "id": "agv2-77c0fd45cf43895fd896b6fe", + "sha256": "370f45f1e2efa7c50d829927767ce46e6090dae1b4ed7b50de82ee8d27c9c39b" + }, + { + "id": "agv2-789af8442d42d80c9dff9e1e", + "sha256": "e190a24e6ad16e9c7373a862d316baa02acb6db182aee170eca91a131deab8d3" + }, + { + "id": "agv2-338c9a0a12ea37409f2ed107", + "sha256": "c4a29557bb63aaabdc05e03044b8bb61e3198b98a6578449b923affca84de8d1" + }, + { + "id": "agv2-c7663ed2a1740cc43cfdca5d", + "sha256": "da8ea82c37e60b37f8efe227c9505810953150e29af6473637e7fcf6d0b1e14a" + }, + { + "id": "agv2-38ead9368d61b444d59696c0", + "sha256": "eaba603f6b6b27f26915da0ed1e1a64011ee06d1e4c4b38174cb5bab818097d1" + }, + { + "id": "agv2-2c93b9ae09fd0b86d13b3029", + "sha256": "f536a066b8e1e383c9e008a7495cd81918bead48cd3e5b21d9903dfec8e26905" + }, + { + "id": "agv2-9d7c699f81d6195ca15bea55", + "sha256": "af159319dd9e676e6d77950b2a371031a8dd16c44c23bdd236548eafac086fdc" + }, + { + "id": "agv2-0ece154213376c7aee71c6a2", + "sha256": "30d052e3d5197aae6c53635b5ed40803bf9cd26ee98ede17e5ccc723707109ca" + }, + { + "id": "agv2-c5fead053df72392aac93318", + "sha256": "1bff522be878c667bf8e822e92dbcb03fccfcda585b379d411536f0c34e75b0e" + }, + { + "id": "agv2-a054999d674c50131cee4060", + "sha256": "49c3a58e889c0b11be69b95407aaf1a2233698b1752ba1ffdc9dfa8753f21eb6" + }, + { + "id": "agv2-4a0d8218493abfbf843950d4", + "sha256": "95b5122f6f2a57f50f4d086d381f972ba0bd3e996162645cf86c18e6035aec29" + }, + { + "id": "agv2-ada09197c4c8516e231a561c", + "sha256": "c54e43f9bd7726fc1eddf89949900ebb05000b763b1e37bc5d1010d424af9861" + }, + { + "id": "agv2-8cceb76ee01a2f7b70d14ae5", + "sha256": "fadf1c5cdaea8ca9af40245075015f17f45eecabc8215d31945862cc876c7bbe" + }, + { + "id": "agv2-2974679b2a0b14867cfb10c5", + "sha256": "2165861252e8db4aea9245fbec4c39e1ba7a46123f31b3fe7c9c06871c6e605a" + }, + { + "id": "agv2-d7dcdabb10acef2fac4afb61", + "sha256": "8bde4465c42dffbd4336b7f353872ab0573946fbe885e536ed8a2be2b1dab7ef" + }, + { + "id": "agv2-386f6a3a3a4519e670fb7913", + "sha256": "ddf51e253f5ea6b40c30a6d28a38fc74fc4e2a1ce5a752bc948dd20e71c50ba9" + }, + { + "id": "agv2-b7ce535b9b2a0a0f446eaf13", + "sha256": "6550ef49a79166bb32a9a5b540ad07214d4b474a112673fd8724c85ce41889fc" + }, + { + "id": "agv2-d77389638a6b26845f83cc48", + "sha256": "60342725762e7922b1205f3fb673e6e65c355a49b0efd990aea7b62dc53bc7a9" + }, + { + "id": "agv2-6db4311339341c28d93dd1af", + "sha256": "b2e2e12f5c375ad0e06cc9afceb0e1ae5c2dc9915943b21b1f70c08bf5bcdc60" + }, + { + "id": "agv2-9bfec476494147f162c56ebd", + "sha256": "ac79b614ee14dc55bed2733901eff8b917c0369e72761531b1e864582ef24756" + }, + { + "id": "agv2-fa3e8acef407bb586dc6555e", + "sha256": "519af082aeee970bf3d01e783efc52e6671d99adf39676ed6e76ab300a13460d" + }, + { + "id": "agv2-25dcfa6dc84cd418f312a724", + "sha256": "9061e326cbb620c867abc843fc8c8f8b9643f5eeeb855c7c0a37cdf464e3e999" + }, + { + "id": "agv2-9553a04b08312a3942385462", + "sha256": "02ee30c8b9322d88e3a2beb6b57522a297082384fc644e52a8ca218eff49a267" + }, + { + "id": "agv2-ad3891675273881cf50e28b3", + "sha256": "fd591f6ffce637a72e4312e6c718899dafb0c2ca6074dc2516e6654cb1dfb07a" + }, + { + "id": "agv2-01c815eadaa69a18f7a1043c", + "sha256": "6ba56641cdfc3a709d0d78fa7b9ee7f997c822932e6b5033ecd26d4e1f078482" + }, + { + "id": "agv2-6c81ae5818c7ffba230b68d9", + "sha256": "a037126287112727556e6995bba3ec7b56fe9528fe52e3fa07c70014cc883e67" + }, + { + "id": "agv2-b40a2785824a7a61cae9dd9f", + "sha256": "8753af137830b6e4f6f50b4bcf346f988fbe3f7b45c60eb1a010a69c94722d9e" + }, + { + "id": "agv2-37e84417e209430f60c554bf", + "sha256": "c3077939b9c81c7a049594051cbd27ab7d27f041e8c42dcff07c0e468c589897" + }, + { + "id": "agv2-92531b2f338c7100784e4798", + "sha256": "9bbe7144f9a6f8e8a6f949513da9cb6a57e0cdbb2d8efcb8114bbfc3f8d9cb22" + }, + { + "id": "agv2-9efffa4bae19edd9cfc19e02", + "sha256": "06a4723067732184c98a30bd80bef2a7baba015e1433957ed7cbb3aeea0ca057" + }, + { + "id": "agv2-80deb8fe6bde171c8e6f0438", + "sha256": "869a58c03209dec7865eb193e250ed3381434668e7ed684ca74ed30866900268" + }, + { + "id": "agv2-597169dba9c573509a022a18", + "sha256": "bfff3c5dbafa9688331626e3fb4269fe176cefa53397ed41f0e77902572d9aef" + }, + { + "id": "agv2-7505c650005f470efe084190", + "sha256": "e286cd3859621f9aa203a8011a03016a916f6b5f2d9421fd8aaf50283ace774d" + }, + { + "id": "agv2-333e926957e20a5775f338ca", + "sha256": "170acea5fdfb58a31be54f362f13157d8afbb9d7f4247f448a099f42f2f609ea" + }, + { + "id": "agv2-53d361a84cd9c77248237acf", + "sha256": "270766b8171777f1e22ab4fb990a47c8fd6a73a6911878263675000dbb6ee967" + }, + { + "id": "agv2-ab0c598731f84ba550a886ea", + "sha256": "135b20a08a73fa0b7a91df7e7a6d33723781766e83782cc3a4a880f480ef1c38" + }, + { + "id": "agv2-1a1e7fabbc37d846e41d5edd", + "sha256": "7b6133af5c4824b3b3e5eef2faf429043b3f6bc4d412df893e584d4318c4d6c6" + }, + { + "id": "agv2-9b1f770ceb55cb74985f392a", + "sha256": "05c1d330c34d27482a07cc235f9b13755fab73027a2836d2c21983d6d959faaf" + }, + { + "id": "agv2-dcfa74904ce08aceb50b0030", + "sha256": "9955d102ba42f6d18a4500d05ad340ea329d97aced7af765c285cc28f83e75a8" + }, + { + "id": "agv2-4b2f5d4214ccbf4452cd2fc5", + "sha256": "8daecce92d7e2b6cedf7cae2cb7de8fe536c162df511bc0db974ea00e105144b" + }, + { + "id": "agv2-cd20902dab3f987dcfb9c591", + "sha256": "75ce3f5ae39e2c88a942f3d99194ffd0ed5321ba183071b5188a92cba5990f19" + }, + { + "id": "agv2-cdc2ad93b6b97e017cb658c4", + "sha256": "fa80ede8b5b273aac06cded6ac06dcff7464744d4ca90f4440e4ab20aa092773" + }, + { + "id": "agv2-4cc129cb990807f9dcd872d5", + "sha256": "37a741f89b3b90305c4f725192dc7c3d6d87e2299a3ac3c160e71a759cc19fa7" + }, + { + "id": "agv2-577b1de6f310d5b1aca5f29c", + "sha256": "969303aa08852e1331888e66d2ce7a96b9495b84c4a248609c33d580865af9eb" + }, + { + "id": "agv2-709388f719e03b4bf26bcba0", + "sha256": "8a372129649cf3cf1cd8c1bde74977f1f14e8bd6ae5234c2ae753230e3e11242" + }, + { + "id": "agv2-07533de7388248c0ddf23744", + "sha256": "4c0dfba5eb0a52f30d3511fdd607fcecd3c901326ce701d32e73db96d0deecb2" + }, + { + "id": "agv2-bd9c9a67cd38c9365a3f5de5", + "sha256": "ee523e84846df9f20ea9b7fb9454fdf993b364ec73aed4fc63bf9241a159b9b2" + }, + { + "id": "agv2-9e7ead09ddb87ebae4737250", + "sha256": "3e578803196fcf30bb0045a6581fc1782e025c2423a234cb5ad80464161eddf9" + }, + { + "id": "agv2-e435d48ff335c4536b565e12", + "sha256": "64a18c70197f6b5f9f64678a5aaaa0b0aa699abfe22fa8a881e0b9c96622b269" + }, + { + "id": "agv2-dd229d58c074b6042ff27bb5", + "sha256": "6148efd6523ceaa18a9b01ac181244b5b9983ca895026b4f9765d3a64e12ebc7" + }, + { + "id": "agv2-75cad3a9b6363dbf106a067c", + "sha256": "370614ea26b99905cf0443e60a0e5c2b86f0738aae259b78d2b1a9d4c5c3bf31" + }, + { + "id": "agv2-587fb525049b345fe455c5c5", + "sha256": "ca0d354285ab53c79aaf6c87fabedb7c3f9386b612e24829978e6951e2318939" + }, + { + "id": "agv2-af73041af1383b63113cfcd6", + "sha256": "a8b8ef171d5c592486414cb5350b8d4ebc29b5a6c1bfa5907e6427c0199125a8" + }, + { + "id": "agv2-178809d352005e3ae221ae4d", + "sha256": "ad41a5b8c2c605cca7f04c5ae570e0a027fba9049fc857375488916a4e4a7419" + }, + { + "id": "agv2-6d5405729c3a34b645e3ba60", + "sha256": "5d2c5c2f9fdf70f8a5934d263ded76a7a159f96dbfdc5e6d97c82fbb877dfdce" + }, + { + "id": "agv2-684a20cf42bb222618482fa7", + "sha256": "3432809de33acf83d9222d7dfea9545ea34f49f9a695b366eefb309f2b0183fd" + }, + { + "id": "agv2-255712e7cab7bd84c1c658f8", + "sha256": "e76e681dbc6bbda9e126c30fd6e07e697ee820c5c363d1f3ac04f35e59f1f861" + }, + { + "id": "agv2-3fad0c95986bd6b66ceca6f4", + "sha256": "803d0cc5c4e77168d901bbd34bb143322862b102a81e6a26c382e1664bfa6fc1" + }, + { + "id": "agv2-f8074cffeba29e7bb0ac0694", + "sha256": "8bdaa93d9bd211cf301fd2bb1b7f5aaf70458248a4b7a5924cd60f89cb6c70cf" + }, + { + "id": "agv2-d4a028505f29adabacd5264c", + "sha256": "5147fd74c9c25ce22757789dea0313d1496b0ae95c54b3c2c0f3a0093798ba14" + }, + { + "id": "agv2-a2715143c9550aa79b7de1ee", + "sha256": "8c98c04d71ac507ef13bbf322408876a683720f11e3efef051792d57cf49b63e" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 1, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 17521, + 7285, + 34525, + 17370, + 1520, + 8606, + 26811, + 30634, + 13050, + 10968, + 23208, + 32427, + 32010, + 39600, + 25585, + 2334, + 4590, + 33396, + 9343, + 3898, + 11684, + 2689, + 8432, + 39762, + 41012, + 10543, + 2515, + 33816, + 33241, + 7079, + 37500, + 6534, + 28705, + 29263, + 15112, + 36324, + 37798, + 22328, + 39440, + 2080, + 1952, + 13410, + 7486, + 20487, + 25663, + 25023, + 33230, + 16288, + 19261, + 4998, + 3916, + 4717, + 15672, + 7487, + 10744, + 21673, + 37057, + 9564, + 15289, + 29633, + 12071, + 18889, + 34852, + 19718 + ], + "seconds": { + "parallel": 13.774986350996187, + "reference": 7.188194855989423 + }, + "results": { + "parallel": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ], + "reference": [ + { + "id": "agv2-28ffe0b6f9cc60856ea5430a", + "sha256": "dcc72d0c27ff57f8775c24a2b458a24d9ab2814e4d6149a94f0ee065661149e8" + }, + { + "id": "agv2-79670779e5e4ea2eb194393b", + "sha256": "a560f8542acd50b9576552d2b7133722ad2c73e128b1669e079ea92ab5e05c6c" + }, + { + "id": "agv2-70fc2eaedf6be73d1ec1bbcb", + "sha256": "f46a673254b593e79b52d0198f2f443cbc0d82c795ebbf4b882031a4c3341827" + }, + { + "id": "agv2-27e48bbd68a4a954d555b400", + "sha256": "013c5e64c4759c05ab5c767a006b7580a02e505a16b96a180beeab9f89c2b762" + }, + { + "id": "agv2-de0e3da6f13934a4e1eaffe4", + "sha256": "75c45c27e517d7505bbba2ef40e3789321dd75a38614a2ccc1a75b4820267e1e" + }, + { + "id": "agv2-6f4054294015d9e79d59aa98", + "sha256": "1d75fd69b795c5a7f61ab8dbdc66b905ee17f743f20e07e1e3c8a7bdbbb1a451" + }, + { + "id": "agv2-301eeed96bc9086edbdd3fd7", + "sha256": "9dd1f41103991ea01268ddd3d411cf834bf0aacd78760dc1ea9fca36b02fe92c" + }, + { + "id": "agv2-53741c9325d07a61fa4385d1", + "sha256": "950aabee902f4c0d377b1bd8ffa93bc97f53e52b06b07b9bb74436cd93e778e3" + }, + { + "id": "agv2-c115653ba7b50d0535877224", + "sha256": "bcebbdcf407eb3029584a199d616f9bbc4c748fe093f960b4f1c50c2f51bd5e2" + }, + { + "id": "agv2-2996e5d1055327d11fc63ca6", + "sha256": "67bddb5b3b7cbfc07ff818d3648a473a6da3237d42112f593b15c570f5a685e4" + }, + { + "id": "agv2-1e5ef0a22a916c6724cec75f", + "sha256": "54f4abaf2eb0e2a7f6f17d43803fade86f6f8841afe08082f5a5e36f01b735ca" + }, + { + "id": "agv2-6eb27f8e4ae4cb96e8672884", + "sha256": "6b0f896946339cab1724b5f9c26408939f7324a2cb10c2197a628e439c755f41" + }, + { + "id": "agv2-d087ad3b991d7ba65e2f37b0", + "sha256": "e5de2e872aa5d39218faef3b55a6466f2f33e321fe62eabb83c02cab5327201e" + }, + { + "id": "agv2-40854f49f6307213fa227dbd", + "sha256": "6162627beed40ef19f6cbc16c52ec99af2dc5a3d6bd6e10401b1a910dd0f53f7" + }, + { + "id": "agv2-a770b0b6f4334a39da337f0c", + "sha256": "cf1dc0818a34ec90a85419d442ed9f32a6282d0deaf30b4d0e119dff412bba85" + }, + { + "id": "agv2-e787620be5ffc4e0243e6560", + "sha256": "58afee8bb8b20ffcb2f1d2dd4ae70675aec033a1ac985274fd7b5a632e94fd62" + }, + { + "id": "agv2-734a82493392a2c385963efc", + "sha256": "9c86c50da4a36ae2a609bfd13a32568317772dbaa28d60b717bbf5e23853cd69" + }, + { + "id": "agv2-c9d6d3c0fac28960edb28137", + "sha256": "47bcd99344503abb1d7302faacfebb7ca11b2b401caec65d00afac74acbe27b7" + }, + { + "id": "agv2-1e5907008c71aa714b2fcb26", + "sha256": "ed496871ff892ed9b817d1682e11f2f7dda7bf93392c18313eb603c884d47378" + }, + { + "id": "agv2-f25c34a3637bb6c7ec00117e", + "sha256": "3ac5268047cc20dae0f772bda0458b396b8b072b0e7713037522fd9e1611742f" + }, + { + "id": "agv2-4aa6ca732fadba86dfc7eac0", + "sha256": "c8b0e2a2dfe5dd842b70c060bd406d9ed3ec291928405d8df280195a79e4fa17" + }, + { + "id": "agv2-922ac64a33f5a1904d7e4b7f", + "sha256": "931c146537c09ebcadb4acf45858f1c345aeac9e3eb309a06f190c9d26335d43" + }, + { + "id": "agv2-c066e04afdc64050ab129082", + "sha256": "60e9d22ef2965aa4834e09c8c5cce4189c67ee787608906ec968ec507b564ad5" + }, + { + "id": "agv2-d46f3d8b0bef0e5ab12b3d56", + "sha256": "888739bdd746cea3d86b3c9384e5dce6b725a2448cbedf4ebef9359c9f4b5b95" + }, + { + "id": "agv2-668dda35595c07513ab78701", + "sha256": "c58ffb04d1a69e9f271c711a14d48c8c099e442e5fe852313f2f1db5c39a9c98" + }, + { + "id": "agv2-6c2cec3dc4cbde2ec222afd9", + "sha256": "4b9f802486c9241f60878b6a27600df25821884661cda5f02429348e643e31bf" + }, + { + "id": "agv2-b51b13b3a595559bd156d09a", + "sha256": "4be8ef7792ca69735c941535788b1b676038a97679e8e38e8faabad053f32f65" + }, + { + "id": "agv2-d537361a39b9df70bfec1592", + "sha256": "301bc18db5a34ca32f360ee9dfbea561b240dafde316166893fe5e6156ba2114" + }, + { + "id": "agv2-9fb56dc75c643f7b8b9f6a05", + "sha256": "d2730d23653e16db81a035f71fa3b9dfdb93f65f56e81976ae2db445e8b320f3" + }, + { + "id": "agv2-875c013df689b9d697fb52d5", + "sha256": "ba6fcb63f843be8b45fff1d3e01916dede2227bfec8666630ba529f483d32e84" + }, + { + "id": "agv2-259f98aaad422ff5774caa71", + "sha256": "65d5a356d7adce2e90a062b09f88c37e3bb6bcb66fd5cd90d6bc456d8953da9a" + }, + { + "id": "agv2-667687dc4100e8c4aaeb32a8", + "sha256": "e37edd897b9ba9da3c07bec30fb2148fbef672530aa184baa776475c9e12d302" + }, + { + "id": "agv2-61cb5285b3c61e425dd9d591", + "sha256": "c391f6a1373c9e9792ba9364241c3891c32444528237ef226b209e286f75d8ee" + }, + { + "id": "agv2-576826f7a8d414e48bd2a872", + "sha256": "0a53d0e4ad078d0b2982c33f7fcc0984e165f77fdeac74222a30bcf13e19e369" + }, + { + "id": "agv2-31a39d17f65dcbe956ed834c", + "sha256": "cbdbc482989bd60480c15d2cafb6845163dfda66d57ba98e69f5c210de0b706b" + }, + { + "id": "agv2-4bfadda33154c7ee7db1766b", + "sha256": "d755cd68ca103e57ac7ba972306298ff52f5cdffe4dd68ccdc998b0a828e4643" + }, + { + "id": "agv2-fcf2fc89292ad7e74d12058d", + "sha256": "72f86dea47d6d2dc2ed3542b9db75485573aa3f615c9fba1125f61f39efbd7d1" + }, + { + "id": "agv2-1b7009d824e30ca9692b6ea6", + "sha256": "fd8ffdb9b54e743d9262885e4436c121d36d1991dc3e0fc246a8bd89465e1193" + }, + { + "id": "agv2-3cbb8dec299fdf1c89393a40", + "sha256": "2a65e0072f400dbcd681199f77e28a32f056cad34e1e940f61ec205e4547cf4a" + }, + { + "id": "agv2-d2cb274a4eec75ae1e3724ef", + "sha256": "908cafb17cb408ac114ac618a6e6936f4794373585fd04e6aee1f5192add6638" + }, + { + "id": "agv2-71afd76a1251f2e86180e2c6", + "sha256": "ab7d1ae0f4baabc22f2150f2168202d706ed7a8b2882e5c9d10517acac794766" + }, + { + "id": "agv2-bee064e5ed5ccf724ef84791", + "sha256": "c7bc13f12c2bc2bf89f658696941bbf72294485db25a828af5a24ec06a53b596" + }, + { + "id": "agv2-f2a7d5645ef0a36abe2a6b44", + "sha256": "b9a4b8189133681f7ce40a2e73c276f7a12bc021867832a1eeb5b89f8195581d" + }, + { + "id": "agv2-e984fa445867e3ab81530ce2", + "sha256": "11e13bcc15f8d1fbda64ec678bdb7c1b22aae1b085936c6871e0bfbf05ea4220" + }, + { + "id": "agv2-a6f58ce94c4977b67f79268b", + "sha256": "6ef92b3dbc132b748b0aa8284934d7b83218d2dc23db1811da14a76c1ebcaaef" + }, + { + "id": "agv2-46cd0664be46f2c08c400876", + "sha256": "61ded70edceb6bf9b929983fa8b58b1b98da2f485104d440a077f614553f5ef9" + }, + { + "id": "agv2-416346ce60cc677a093eb89e", + "sha256": "c5d7b5af4cd7f499015c7d921a3e1edbc95ffdb14e23ed322432eb785a3cfd9f" + }, + { + "id": "agv2-8d7387236d8bf938888a1d08", + "sha256": "7d3d79fa4ae7d0b6b1e6faeea6e6384217e7e8544ff55f6c9cb260c2f54b5682" + }, + { + "id": "agv2-05a888c3646048563d9ea3d6", + "sha256": "314202a27be418fbf230adfb542a6836612b7092e8a93d1d00d1ff54ef4c1a2a" + }, + { + "id": "agv2-c24a971274b25afa192d77c8", + "sha256": "75a9679e0b259f0abdbf3c78809ea8c51327309ee1900e5b0ca0811bd8d9b3ab" + }, + { + "id": "agv2-12e5689fccc7db327f48f904", + "sha256": "a9d99944d686df8366e79667a18998dc99488a4ee27272567807ca893a81e71b" + }, + { + "id": "agv2-126200954af7054e0444840b", + "sha256": "91a1cc80f15ac1b2d884a746f3986cacd07a2fa736c9d50af202d1fb3e2d6cb3" + }, + { + "id": "agv2-a02e157aad24625a5920e4f8", + "sha256": "aaa6925d913e2f6ac3258df48b897ffc3c165a25a404d2a647bf7120fef0551f" + }, + { + "id": "agv2-fff8e6084a27cc1a2d0548ff", + "sha256": "ddd1b410148c7dc833958a873a7a9ad373a6b7e67f3e80da3c2c132d8c81cd9d" + }, + { + "id": "agv2-d81fcac98b39be2fdff192f7", + "sha256": "7403c214b0e1f2a5421eb0bcee52bc7adf239c94fda144c8eadf7c6063b5c918" + }, + { + "id": "agv2-e1574b1dc8d592386e81a87b", + "sha256": "b389d9c8d6e24cf1acd6dde37cbb8787f0e5600d75c38cd844b3477cfb257343" + }, + { + "id": "agv2-bf3044f5832e749e9786f2ac", + "sha256": "e062d13c5c5262b773c4f203d6d86dc958992b51b62e4d1e8e3d7cdcb30c0849" + }, + { + "id": "agv2-9adf8ae25193552134a6960b", + "sha256": "13d961d52357ac45fed10ec5cbdb084fb72d34bc53b43f8f0e91a449d1a25c7f" + }, + { + "id": "agv2-87c9ed70bbf13eacf4e2e5c0", + "sha256": "63ab7d57ac902c50bc4167994aed2e3d591377b4a5795481a7710cfed58ff933" + }, + { + "id": "agv2-0342c753d9120f57738eac80", + "sha256": "e8aab53c2cf64334ea1e2d42f996a21a138a8675a83f7c676990d9a583226e12" + }, + { + "id": "agv2-227dff43fb16e93613bd3c2e", + "sha256": "f676a65d10ec6c4b6b5052e51503e4dad62b0eb3f2179a579a3738035e6ebf17" + }, + { + "id": "agv2-0959437462e369114e9e8bed", + "sha256": "8654521f45bbcff69ef0bfe718a50eeb5708b64653d710f1b32dfd965ed71bcf" + }, + { + "id": "agv2-860970ef57dba3e5b344285d", + "sha256": "1bd2b046e5acb718056339675fedb01ea8332d38584f21b58d94c141ef449494" + }, + { + "id": "agv2-9e18ce3a12ab3a2651003fd1", + "sha256": "9c52f92d965ddd334d112ee45def09f2eee5d11d2e89ec7cdc6fad5d9f066b94" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 2, + "order": [ + "reference", + "parallel" + ], + "indices": [ + 23359, + 35426, + 19359, + 17818, + 19728, + 7666, + 40741, + 2219, + 26064, + 33769, + 35102, + 969, + 20570, + 2517, + 35929, + 18553, + 24738, + 5249, + 13739, + 559, + 21213, + 17476, + 904, + 8826, + 1803, + 19820, + 32640, + 11976, + 12556, + 11970, + 21033, + 27025, + 3966, + 1359, + 26067, + 6974, + 24701, + 30395, + 36002, + 7805, + 10720, + 34553, + 35890, + 26717, + 24912, + 809, + 24455, + 7932, + 13149, + 7092, + 29946, + 4918, + 18094, + 38333, + 35239, + 12607, + 16149, + 37138, + 19202, + 2032, + 22905, + 40874, + 30759, + 31780 + ], + "seconds": { + "reference": 25.228934401995502, + "parallel": 5.8488235650002025 + }, + "results": { + "reference": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ], + "parallel": [ + { + "id": "agv2-da1d9e1f7c95906243deda22", + "sha256": "eecf5e0408310aa012b2ead3d90507d864a91e0d3198364c247935e505acd75d" + }, + { + "id": "agv2-67dc27b873e1b57dc9bccedd", + "sha256": "d049b7fe770d13cb59e8f3077d9db209c5263b28ff83d97f7a7bff1182d56105" + }, + { + "id": "agv2-a284008f899829498f5dc6b6", + "sha256": "5950f4b6c89ad7c2012d691f4fe86bcfef9494438d9fdc9d1194b92ab27db81e" + }, + { + "id": "agv2-c8c3f4d55bb2a9645091cbe9", + "sha256": "f8dd074987646d20165d0a7701780dd5c9757af75651e04a2548e3ade9755944" + }, + { + "id": "agv2-b1c8e4b54f44731f57fcb396", + "sha256": "dc11be98920a1a5c6897be6d9abed7bfef110d0b01c583cd6da90a2f2556ec7a" + }, + { + "id": "agv2-b0598f51b16e2c3e4693ed1d", + "sha256": "538567bfce7b5e78280bae69a10997dffef02d9dc34ec9cb913d2deedb94ed45" + }, + { + "id": "agv2-57dde540ef00bebd71d1a8c9", + "sha256": "1365c616e878093a0e7b9acc9984043429b356717b5590506ae4cf885e04973f" + }, + { + "id": "agv2-f0ff4cd8e504052cc071cc4c", + "sha256": "ece3d30d1db5b91e687d3924c0e6741cd81ce35d619abe63484a40bf92e3be38" + }, + { + "id": "agv2-2725a27de495c42c506a431f", + "sha256": "79e343a3622f10b992598541ee59649529cc3f1a475dcff2a950b5fb87a51fda" + }, + { + "id": "agv2-99bde78c518511b6e147b5e2", + "sha256": "68fcbd76b9e2613a6e0582bab6742fdc2112ce8fc7b06b3213eedc4bf4b11ffc" + }, + { + "id": "agv2-51a77a25b2a7b196573e3f6c", + "sha256": "3a99a5ae258d51cc5a3a8e0a578289dbac2dbadb6d70ef65849fdb88e1bf5073" + }, + { + "id": "agv2-c95cc71f1528e1d7622e8849", + "sha256": "6ec9daec11f7332a3517662be9a0304078fd10a4ef1b1531d67ae0f97c52efd8" + }, + { + "id": "agv2-e217622786bc5e5bb23c4d20", + "sha256": "9f0b324e54ab7d547a1fc94c8d58be88b97077c2c304bbd7d301741a36f49e3a" + }, + { + "id": "agv2-eaa2220134b6b4f01eb0960d", + "sha256": "9b79dce3fea961e2ab30722393c91bcbbc1da0098f781706a0014453f52fa416" + }, + { + "id": "agv2-032c227fb97984d06f94816a", + "sha256": "3c47de0c98c901c3c0606302b2d9ec5c2646fd9c62b2cfeea0b7fe58e2dec5e7" + }, + { + "id": "agv2-9fd00737a954ab323e87c6cc", + "sha256": "d92664cbed39449f484e315090383a45b2accf9a1743aad6564af7d24ffd7305" + }, + { + "id": "agv2-90e6a5d22cc6f8be59ac5467", + "sha256": "565009593e0f468213be2e59a4790a1531a02534d4e847963c8e158d95e31332" + }, + { + "id": "agv2-674244a74f64165aebc2f582", + "sha256": "f9eef5cdd5f330bf30c61bb32f70116bbcf6748db13fa8f7b2b064301b4966bb" + }, + { + "id": "agv2-5433d4c86b9941e602355ae6", + "sha256": "ebb3fa76896b0a50e553c01ecdeaa67ffa9a015fdfb5dc91d28efce094baa69b" + }, + { + "id": "agv2-07bb33a309dd4e44c220ed9c", + "sha256": "25ef8e10a8b12e02aba13c126b07c91e1fd6d94c99d5fe65263a47765bedb8ab" + }, + { + "id": "agv2-949680b43853346c1d6fbdad", + "sha256": "107fc7b71e9156746073654212821e78cad77381d0c582a7c06cdb64cdba50d9" + }, + { + "id": "agv2-0a0c79731a4dfe5c875f0caf", + "sha256": "4ce196beaa224db98584ccde82e925b18e1a7a2e0e13b8c61f796c634a1d9326" + }, + { + "id": "agv2-1c708cba6068bb5b9d676883", + "sha256": "6055b9cfd34695023f861cae53cbb637971bd5e9894e6d21974ff4bec3dea339" + }, + { + "id": "agv2-2a92175e43bfd8e33ed4f8a4", + "sha256": "4628a26e62ae0db8e3f56b570499787f31af96eef601a29a58f17d88a5d57840" + }, + { + "id": "agv2-5e38f133c222ad0db3d4aafc", + "sha256": "6a6b1fce1407e7a4323a3ae003077138f7b1aa3e28b8892ae7dd12bcd0e573ca" + }, + { + "id": "agv2-585a8dada77bb885e1c13a17", + "sha256": "df0e6c42383b07a8aceebee25dad195fb34f2b848ddf126ff97df8e88eac3024" + }, + { + "id": "agv2-ebe8ea77f9f8ca26589ca7ab", + "sha256": "8c289381e7365f70c00b3609b8012f2df80eb3155a4f3eed198616c68fd3bae3" + }, + { + "id": "agv2-4681a08f325f76114bf07732", + "sha256": "30f988077cfb536a3c72e27724734d04ac6c533ba8b0a0399b44f3268faf7703" + }, + { + "id": "agv2-0c6c621c80b8e3a384227194", + "sha256": "b557d4fdeb7f0105240e7a58b30451e066780d7b4c80aa9c51abaa1851f62bff" + }, + { + "id": "agv2-3b352ac02147df781a0a36ea", + "sha256": "72c21ec8560299fb9c9d3fabcd2a8e5b01f56ba52e8e5d8c53138e21b89f2c97" + }, + { + "id": "agv2-ecfdc6fa4116cf41c60e9d2a", + "sha256": "09b61f0401b11c96d3381f4590b9a7b6a104fdd2619ceb5e24d2e26a563cf0a8" + }, + { + "id": "agv2-bd70f9d0f0ccffd8263ea790", + "sha256": "637724a96562cb453129a7eace0bc453907f90d3f65e9fc563cce1251e08801b" + }, + { + "id": "agv2-2a4602e42a60c62c29403e76", + "sha256": "cea6ba4c97ebed1d5aaa9487bf28b3edc59c7bd96fda67d6d66fb8837590f01a" + }, + { + "id": "agv2-00686bd56d2466bd49ac2842", + "sha256": "84d1141299c9e34befab791237c734c759f91b5b51ab2a30c03cf8d5da320f6c" + }, + { + "id": "agv2-f66e01025d23f357dcec8726", + "sha256": "f292a8b68ae1d8ef0b486a7f027b5e8d35615011f27059b5e032f7643709e195" + }, + { + "id": "agv2-e0453694d07a39d8706a82f6", + "sha256": "9af7b27848b58a08fae99bc9f05bbc5f6108136c16f07ed9b9b1bd39decbe098" + }, + { + "id": "agv2-5e11ad7ab51f1d9cac3f439b", + "sha256": "7205402029a2d6c024f232409e24091cda73c5da65949b6d314039311b5ed34e" + }, + { + "id": "agv2-129eea8ba08a244d2646a19c", + "sha256": "1123c56b3ad1985138fd33d91e40aba98b250b51228e0163b193b2ead3eac7d3" + }, + { + "id": "agv2-9c7cad1583f3de19f1ef10cc", + "sha256": "43376c61eb0be350fa44df7976d46128d52b178ca9dc6233661eca520ea46806" + }, + { + "id": "agv2-612794a016932b945e4a4aa6", + "sha256": "b18c70c0882a0bbd3342a52310ee9893561b7ce9da3fc7f9d4649aaceb26240f" + }, + { + "id": "agv2-edbd51a9e5e0c2df6cbe0762", + "sha256": "4aeab3b5ca3894acc54d5e8fea9e2c1ce50cf4160a59c49a89f9847b05f1b2ae" + }, + { + "id": "agv2-4864b3bd5e334bcd8832fab3", + "sha256": "47729d9bfd8d3f032195457ff9ee1187cd6b3d8e9604d423ef195f5ddbae74eb" + }, + { + "id": "agv2-43f67caab9cbf280764c46c6", + "sha256": "5fc579bfc44ac48750b6e5f7a424212d8e52f8a487f316e3663e9a0e857c9dd6" + }, + { + "id": "agv2-495a7358175ff5775e8af8b4", + "sha256": "54335dfe55c019d256cf56c0b9d83b1d1653233ead3917163c8de85e91eeba94" + }, + { + "id": "agv2-bad741cbe87b020410be7373", + "sha256": "f28246aeaf6f92e3a91d0fb7ecc913cd65483b4322d5b7898d5db42f5303dd4c" + }, + { + "id": "agv2-8f26ecfee7f6646431937936", + "sha256": "f5084305d6061e7a61032624705bdd922fc563d6b1c8437ee68edb69a786fbfd" + }, + { + "id": "agv2-e07facf7e1e8c530c76f799e", + "sha256": "7650c6bda8325add93f7a0c76a99417780d07c31862fc8e5c874e5bf348fd22d" + }, + { + "id": "agv2-2a8b8f2bd17ae9d063220d0f", + "sha256": "c739f09c5443b88105e29c341ac3922772867ab278345054bb02acb27a97f905" + }, + { + "id": "agv2-13f2a29a209e8d1fdc5e8b43", + "sha256": "9b529442b4ae37beb72f5c6cd82cf825836951394093ebf9fb6b332a7d09e7de" + }, + { + "id": "agv2-d9916a5c557227144f4accd9", + "sha256": "a5cf4e9d3f271c6728b7b72cf224e81ffbd8bfd1ab2ffa4ff7ecd2f447991db0" + }, + { + "id": "agv2-2f22744a15fbc154b5a66285", + "sha256": "7b1ae059abb1e778fe3008f24351515df740ec8fd59269d66c563f4f9db9a54f" + }, + { + "id": "agv2-84ae7d1f6b7b4c85b4eb03f3", + "sha256": "2e33041312c34b553e5666415488673a86694033698e9d68169a37af0979386f" + }, + { + "id": "agv2-1d418efc3f9aa5fad359fd2c", + "sha256": "e75ff4aef0b9c388eb0b54f48ab10e904aa36712f39c94d726c5c94f1fbef3d2" + }, + { + "id": "agv2-d7777b81bdf240692c22aa87", + "sha256": "f2959ebd74d78ff75a9f6c6fbf1a896b16e8f4ce1225e0d030f35ede6038ca29" + }, + { + "id": "agv2-125afac12832937c5538162a", + "sha256": "198843e44897e416b1d4c91640fdac7a4bf12b3b6758198fdec917d0347fbc33" + }, + { + "id": "agv2-e441b1e7c2153a2ef2d8e607", + "sha256": "89719c60555e7877a6dcd8144f9bf6bffb3c7044757914488982ba28bc77dd97" + }, + { + "id": "agv2-1dc80e26529a2de047d54cf0", + "sha256": "09c4f4e27f5844d9dac3f91372fc885ed125ed3c9eb8d26a96118b1bba387848" + }, + { + "id": "agv2-9c7aaaf8bb9f48a2510df871", + "sha256": "8f8b7c92e542b74f9b04b27ebbeff0bd14e557b9e97830e5c8457b9231d33d28" + }, + { + "id": "agv2-2074831ff12bd10f4593a0af", + "sha256": "80bfcf204a8621029a99fe74ec252217571ed3e2dc26f21be2482792d4b83fe5" + }, + { + "id": "agv2-3e1279d114a82346b7259d7f", + "sha256": "8ada207a6e7334a67a19bc1d14fc97aa5e7161117ed87d21edaf64c88efe591f" + }, + { + "id": "agv2-cd8203b3af8676ba3ea29ac1", + "sha256": "7994fa34b2a6acd385080c05e4a5944bfed21a95fb4abdeaa7d6c8eebc94368d" + }, + { + "id": "agv2-e84e1a040576690432731408", + "sha256": "3281dfa1a15c1877f13ed46559b6804b4a17d25016c25a44f6dedbe6c1c81843" + }, + { + "id": "agv2-1e47ec9bb3d89e363fdaf4ce", + "sha256": "f27381c902a678c19ebd88e092d12d760d8161b5d87373459713c5a1a5b149b6" + }, + { + "id": "agv2-c7696f7a8aa5e2cc2d75b397", + "sha256": "094c6695197188f99aca1dc5c0452fd8ec0ba54574d7aab4d769aca98ebbcd07" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + }, + { + "round": 3, + "order": [ + "parallel", + "reference" + ], + "indices": [ + 4949, + 21235, + 10227, + 2090, + 8037, + 36623, + 20093, + 9410, + 39920, + 29736, + 32166, + 32057, + 18211, + 36220, + 21369, + 23192, + 21102, + 18105, + 9007, + 2879, + 10163, + 18261, + 26249, + 37719, + 14270, + 15553, + 12203, + 13474, + 22944, + 32607, + 89, + 33180, + 24261, + 19506, + 35055, + 31136, + 4245, + 9226, + 34043, + 41004, + 24819, + 5609, + 856, + 832, + 947, + 25281, + 31633, + 7493, + 3400, + 2091, + 40318, + 302, + 24855, + 25783, + 30629, + 36465, + 6239, + 38943, + 35333, + 4584, + 2925, + 18024, + 11616, + 37446 + ], + "seconds": { + "parallel": 11.824136001989245, + "reference": 7.038491812010761 + }, + "results": { + "parallel": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ], + "reference": [ + { + "id": "agv2-15920059467f2ba8ffd0bc86", + "sha256": "13588b3310aba0bc37d4abf807a38a9bface6ac15459040f2ccd2d85ac7d7332" + }, + { + "id": "agv2-1a6496b1f9f5ed8e674fc608", + "sha256": "8bc8aa93c7a6efedb0582ea80d3e58b48465ee656b9adaae85dfcb41c2fec7a7" + }, + { + "id": "agv2-8bf867534b26f983fdc3201c", + "sha256": "8da71771c2844ab7cbbedf3701d9059e67123229d3168cee6e439fbdaf1757cc" + }, + { + "id": "agv2-3c222fc3f9a31e7181469c9c", + "sha256": "3f844e4c3bd1769ec64cbc835f07af2f3111ded06231bcbdf33c3c459f870467" + }, + { + "id": "agv2-836f25c4bf145422fc941397", + "sha256": "48b0063d7ae4ac62e21d732fdb91a224a4255760f3bd2409e4585b306f0f6a7d" + }, + { + "id": "agv2-f5c4c578abcab44b79b252f3", + "sha256": "731dec17335884ad7def6f4ac05ce4d50423bd43b2322e98e1c6b8b9fdd20dd1" + }, + { + "id": "agv2-f2e3ba5d4556bf04c579334e", + "sha256": "09291f3f2cefd62bdfdd76ad657251d1284b260a36eded534d9aa376be546355" + }, + { + "id": "agv2-62f5bddf422125a7ab25f27d", + "sha256": "a67b62174a9659bd6bdb1a1648744ce99edab1f6b940d4ec604ea6b3935ee26b" + }, + { + "id": "agv2-98205fb33b3e6be57644c2a1", + "sha256": "b3ab5691b44bd8d8695e0e1b318165974376c84ac7533e68b57db4ccc471ede9" + }, + { + "id": "agv2-4fcc2186da3c0ee7321d6e56", + "sha256": "c972dabd957ccce40fc0076c1402eb12763b6f9a68ded4bbe5d9d917cc87b9c6" + }, + { + "id": "agv2-6caf77cecb0c9a8c7c6f1e3e", + "sha256": "7c7eecf97c245a84df863703a02b28920324cd38c5071708ba09530f599e88cf" + }, + { + "id": "agv2-67d8e70b5db48bbe8acba935", + "sha256": "b25e492f35ef225c909eed9429670eb7d2d07d5ea2422acee3f8d52f1b7431fa" + }, + { + "id": "agv2-8936d2d72e37dad67460b6d5", + "sha256": "8e0ecf5817a04d6fe97a5d1edf0b3faa7445f4e9563529ca3d3c1e66a88c01cc" + }, + { + "id": "agv2-90dceee8177aab0baec60c02", + "sha256": "9953429965a75df058a51e7b8f795047b10160af94c514394134123aa91363e7" + }, + { + "id": "agv2-0f01b8e28525c45e8b1a200c", + "sha256": "40373729c234b7b20230339046820b442beb91a56b1e3401983f7eecbdd9ceaa" + }, + { + "id": "agv2-7e15252fc37c2fd8ed54fe46", + "sha256": "3b0134bf917e0441a787a26cb44f45996cf40e67e48c59ac70df8c03977a958f" + }, + { + "id": "agv2-3cfcb023ecb52a8a7e27c867", + "sha256": "c84ac6d5cca83bfae13f4e5a6c6ef9195a375ad42555846d30f4c66782253dc0" + }, + { + "id": "agv2-ee842e0fe4b5acbdd38b8989", + "sha256": "177d4b65eee92bcaeb745d9ccca5cc19c23cad62f06219c0db109914c69319f0" + }, + { + "id": "agv2-4bb3d65986c4b73aced6ade4", + "sha256": "149c44ef3104f6ca850293e4bed99bf89e6066f11ff7ba10a0f03d865fcca9f9" + }, + { + "id": "agv2-7ca2d7eeedfab250a44e0352", + "sha256": "a41cf9a5d4f4ed72040e0deaa2f9e998e489b027e95518171372d3341203bd22" + }, + { + "id": "agv2-45b194d186dd501167810565", + "sha256": "e4fef00208f6451ff3ab1ec2a811c6afd1ba33ef93659a23c4e83e4cbcc2663b" + }, + { + "id": "agv2-3ab19a19d3f5daa0cb84a989", + "sha256": "7c912e478ee14ae0e42641f8d52311da8755424f059384618060acd2864ee68e" + }, + { + "id": "agv2-6aa5a949339d188d8f787663", + "sha256": "5fbc88e044e7d15750bd14d39957dce78400de15596f5e2d1e245bb5b43fd851" + }, + { + "id": "agv2-74f67a96910c6c4cbac05b47", + "sha256": "bfb10b17e61110fa4d53e4dd6d14b1ef79ee5284c372b9a525afc952930c016a" + }, + { + "id": "agv2-13d493c6cbd3d14ecdfcf27b", + "sha256": "f2d76da831e121d6dfa91388ec8ca781e9634e5e440db2639860645efba4aee9" + }, + { + "id": "agv2-c2958294491decee4ed47258", + "sha256": "a1e276656cb618deb0597490f3cbd045160648f98207e574b6f47e69c2b46f65" + }, + { + "id": "agv2-094548937e1bf35429666180", + "sha256": "be41ebe8d4ab4fc8955950226b6c56d84d7794d1641ff483a2fdc9889fc82dc6" + }, + { + "id": "agv2-3da48d152db9b9b24da8ac47", + "sha256": "c3a08b54330bf0e5d1c8254e161080c43c22ec0c257f17f56d10ad467b342b9e" + }, + { + "id": "agv2-0deb09ebd53e0ef6e5675582", + "sha256": "c50f276fa67d458d6b6d6c572331bc2ad6b2f2c59a3a1660019f3e1168dce601" + }, + { + "id": "agv2-e95bdd2a62b1107fc4f36dca", + "sha256": "688609037273ceca2b5f106787f39089952c68a2d6d2c7ae610f5bc40985fc23" + }, + { + "id": "agv2-ea9940ccf7eca9bda488021c", + "sha256": "5da9791edb96fec7ea191f024adb08cf3ebe89e4cd910613cb9f74f0dad76ff6" + }, + { + "id": "agv2-f32acec82583c04eca8ca8bc", + "sha256": "6b078af647f02c1c88d70a76c65b7e3bacc685483e7494c98c300a516e15ad74" + }, + { + "id": "agv2-e94cda58cd544b44beffe084", + "sha256": "8bc81995de9244f60f010aab36c35f96957d50e9133e53cc7109efa639b52625" + }, + { + "id": "agv2-86549f2d231821ae0397b323", + "sha256": "bb5156d465e6ca966c151480dffae630eae40f56e4b4446f7af48190de8d4a68" + }, + { + "id": "agv2-d4df9abee1943c5af35638d0", + "sha256": "000e86732e15a1993ed9f3d95cf9af73706a691f84c8aa809b06c0db41398b80" + }, + { + "id": "agv2-e29de52792ab96ce817ffcf9", + "sha256": "9526cbd83b80c31a70f563258d0feb035c3d34947a427057b6029b12aa5a69ed" + }, + { + "id": "agv2-4f9a382588a8762477676d5e", + "sha256": "f9fa358a855fa9d2d55d83e830fa41439abdd3d291bf856402bd836ef9186479" + }, + { + "id": "agv2-69f9f34592bb0fbfb4edd003", + "sha256": "a23f3487f198e1ff43d1b00ebdd3bef057f07316a5a4ebb7d13de68ddf43b73a" + }, + { + "id": "agv2-87d4be39594cece645ae3042", + "sha256": "fbd8552c6715d4d633e5ce809cd0c5f939762c097e2d4a8b78d9c9f4f9d27d12" + }, + { + "id": "agv2-e0662497a1a458ace773bd63", + "sha256": "e891b7f25ab9db0c5982bea78d00420a3647fee029c74990fd8e32ae30e65145" + }, + { + "id": "agv2-b5c2afeacc70e5f1624847ca", + "sha256": "0e6340921b348d55e96b5b7e78744f4d9d0f4f070416c6a279bbfe3d295fa4a5" + }, + { + "id": "agv2-6327dafb07a6b70fd40e19b3", + "sha256": "743ee12c75b0a0f71d272867a5fdfc274eea7e6b721a28ed2e7b21349ad3db5a" + }, + { + "id": "agv2-379bb60651142e8351be89ac", + "sha256": "cd56311bed4f4d47d3be8852050aab55713d2043cbafe6815374b119bf8f6021" + }, + { + "id": "agv2-1104e68259f1e809111e07f7", + "sha256": "76d6a238b78670ff25cfbfd60f4ed86b4c2c31c34eaa8fda4c1073c9b8d4c331" + }, + { + "id": "agv2-7c83bbb2ed8e2ead6741a831", + "sha256": "8bf95eb4432ac9fc2d7b37c6458193afef1b24274ed050bb3e3e8528d0df3123" + }, + { + "id": "agv2-e81e882a673d3ca0872ddcee", + "sha256": "2f70e32cb01d7bc60477680a0b299ad0aea6578f38c8c2c0f1a05874e004a268" + }, + { + "id": "agv2-88a8685c2d321be26142a3c6", + "sha256": "a920246fbdb44755a9fe5dec8b33d3456217930e1fd83140b4adc5a01e59db18" + }, + { + "id": "agv2-3c0c23a75cf0aa636a32fc95", + "sha256": "4ec0ad8f57274de280e8deaf2ac6e46a8ecfe7279afcd7017e86f27a89faff71" + }, + { + "id": "agv2-8ec6413a02e6240cc37b403f", + "sha256": "0f164f9bf769a60bfdc2e14420cb66707112f860ffad6bf4113b99e87abb6596" + }, + { + "id": "agv2-30d2cbe80c53fca2b9fcc6ea", + "sha256": "9a9d2b4ec2931ec0f75c15d2171ca9d6eadb0425a230864264bbe6879394a513" + }, + { + "id": "agv2-c3322a2447456cc71219de00", + "sha256": "5e4a8f2030012a181c7557b34f787bb9dac18cfa99c64728dc7e7392168a482f" + }, + { + "id": "agv2-d0493463c51474a9e1574533", + "sha256": "78aba4dba6b944d0543cdbb8f5cc08b4f083bdf0f26faa1df798f75f6e745bb6" + }, + { + "id": "agv2-3b7a144b012cf764ac1e054e", + "sha256": "b205ce7cf1f5058e040df4ebbc50db17f42b4d6623fe75d538573b1d24ad687b" + }, + { + "id": "agv2-67ad7ec15c9cd55576147643", + "sha256": "2bc435e142a589e3a48ab1b7ae5a5bca2de23bbd59cc486da2861d71d144114c" + }, + { + "id": "agv2-b6f9aefeaf6405e80b845272", + "sha256": "bbcd48a046dcda7b2de993ff99408e4683f617b4569c12cd6bf9e1ddfa53ad9f" + }, + { + "id": "agv2-f84bb5dce429a718b8bb9c14", + "sha256": "ba3f479104f17a7940a25edbdfcb570cde34fedc833e51d578457c2ec8cc1e07" + }, + { + "id": "agv2-79d7ee5938b859895d232b45", + "sha256": "376e444289681ea59d33784398e724d46129d418a13bff9a6b969bf30e96bd33" + }, + { + "id": "agv2-fa1370e9dc8c96819e7efaf6", + "sha256": "ecc32558b0f072b439b254986bede96db7303eb62409c40182f0e8287612d68d" + }, + { + "id": "agv2-d726cd8e33d2f009ca243f34", + "sha256": "300b0ddf267199bde5245bb4a608533cb4d6d33e1ca8c4401c75c33fd0f533ad" + }, + { + "id": "agv2-6975c46b9cc1d0e6750a501c", + "sha256": "75643424d4185af210ef246031c8a0f6f3242580fd705426f09e160b720716c2" + }, + { + "id": "agv2-63d40af6b755f1d0b2a7e952", + "sha256": "a3e66bef396d3faec8e20a19cdc3f5983b8dab5e801b0f17537103c0de8e0b61" + }, + { + "id": "agv2-8e2c526ba23a47129b05b2b2", + "sha256": "822a2f90d2b6b92b405f3646ffc635e57e1463f1361e18283a163ce55c969d46" + }, + { + "id": "agv2-74feff525e7d71c2bd696b47", + "sha256": "8e5c7d52e70c0a8fb987f264ae1e59934386f98a4c79742116d94c6114380cec" + }, + { + "id": "agv2-f1e210d33993c57ba2886f6e", + "sha256": "6f2d7c601ca45b6ecc0cec798f8ebf46e64e2ca0d474c396a1f3063134bdce5d" + } + ] + }, + "all_ordered_ids_and_tensor_bytes_exact": true + } + ], + "total_read_hash_and_transfer_seconds": { + "reference": 69.81505457599997, + "parallel": 37.28158475100645 + }, + "parallel_over_reference_time": 0.5340049503280426, + "reader_sha256": "ad6670f742f4de9898064becc525d0aea98b402abe7b210bb1b26a9eee6402b9", + "adapter_sha256": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "caveat": "Alternating paired readers under shared training load, including worker startup, full per-example hashing and pinned tensor IPC. File caches are shared; no GPU computation included." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/pre-decoder117-model.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/pre-decoder117-model.py new file mode 100644 index 0000000000000000000000000000000000000000..6f211fc29de3fa559db45ce6c2ee3d40816f4a72 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/pre-decoder117-model.py @@ -0,0 +1,460 @@ +"""Affordable multi-resolution baseline for the compact AlphaGenome task.""" + +from __future__ import annotations + +from dataclasses import asdict, dataclass, fields +import math +from pathlib import Path +from typing import Any, Iterable + +import torch +from torch import Tensor, nn +from torch.nn import functional as F +from torch.utils.checkpoint import checkpoint +import yaml + +from .layers import ChannelRMSNorm, ConvBlock, SequenceBlock + + +DENSE_TRACKS = { + "atac": 167, + "dnase": 305, + "procap": 12, + "cage": 546, + "rna_seq": 667, + "chip_tf": 1_617, + "chip_histone": 1_116, +} +ALL_FAMILIES = (*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions", "contact_maps") + + +@dataclass(frozen=True) +class ModelConfig: + input_length: int = 1_048_576 + target_length: int = 196_608 + dense_bin_size: int = 128 + trunk_bin_size: int = 1_024 + contact_bin_size: int = 2_048 + stem_stride: int = 8 + encoder_width: int = 32 + channel_growth: int = 32 + num_downsampling_blocks: int = 7 + conv_kernel_size: int = 5 + dna_kernel_size: int = 15 + hidden_dim: int = 384 + num_sequence_blocks: int = 6 + num_attention_heads: int = 6 + attention_head_dim: int = 64 + mlp_expansion: float = 4.0 + dropout: float = 0.0 + head_hidden_size: int = 256 + junction_rank: int = 32 + splice_usage_tracks: int = 734 + junction_tracks: int = 734 + contact_tracks: int = 28 + gradient_checkpointing: bool = True + enable_pairwise: bool = False + pair_width: int = 32 + pair_downsample: int = 16 + pair_update_interval: int = 2 + local_width: int = 0 + local_context_blocks: int = 4 + splice_context_mode: str = 'offset' + local_to_trunk: bool = False + residual_rms_cap: float = 0.0 + + def __post_init__(self) -> None: + if not 1 <= self.local_context_blocks <= 7: + raise ValueError('local_context_blocks must be between 1 and 7') + if not math.isfinite(self.residual_rms_cap) or self.residual_rms_cap < 0: + raise ValueError('residual_rms_cap must be finite and nonnegative') + if self.local_to_trunk and not self.local_width: + raise ValueError('local_to_trunk requires local features') + if self.splice_context_mode not in ('offset', 'uniform'): + raise ValueError('unknown splice_context_mode') + if self.splice_context_mode == 'uniform' and not self.local_width: + raise ValueError('uniform splice context requires a base-resolution path') + downsample = self.stem_stride * 2**self.num_downsampling_blocks + if downsample != self.trunk_bin_size: + raise ValueError("stem_stride * 2**num_downsampling_blocks must equal trunk_bin_size") + for length in (self.input_length, self.target_length): + if length % self.trunk_bin_size: + raise ValueError("input and target lengths must divide by trunk_bin_size") + if self.trunk_bin_size % self.dense_bin_size: + raise ValueError("trunk_bin_size must divide by dense_bin_size") + if self.input_length % self.contact_bin_size: + raise ValueError("input_length must divide by contact_bin_size") + if self.hidden_dim <= 0 or self.num_attention_heads <= 0: + raise ValueError("hidden dimensions and attention heads must be positive") + + @property + def trunk_bins(self) -> int: + return self.input_length // self.trunk_bin_size + + @property + def target_trunk_bins(self) -> int: + return self.target_length // self.trunk_bin_size + + @property + def dense_bins(self) -> int: + return self.target_length // self.dense_bin_size + + @property + def dense_upsample(self) -> int: + return self.trunk_bin_size // self.dense_bin_size + + @property + def contact_bins(self) -> int: + return self.input_length // self.contact_bin_size + + @property + def encoder_output_width(self) -> int: + return self.encoder_width + self.channel_growth * self.num_downsampling_blocks + + @classmethod + def from_yaml(cls, path: str | Path) -> "ModelConfig": + values: dict[str, Any] = yaml.safe_load(Path(path).read_text()) or {} + known = {field.name for field in fields(cls)} + unknown = sorted(values.keys() - known) + if unknown: + raise ValueError(f"Unknown model configuration keys: {unknown}") + return cls(**values) + + def to_dict(self) -> dict[str, Any]: + return asdict(self) + + +class SequenceEncoder(nn.Module): + def __init__(self, config: ModelConfig): + super().__init__() + self.stem = nn.Conv1d( + 4, + config.encoder_width, + config.dna_kernel_size, + stride=config.stem_stride, + padding=config.dna_kernel_size // 2, + ) + blocks = [] + width = config.encoder_width + for _ in range(config.num_downsampling_blocks): + output_width = width + config.channel_growth + blocks.append(ConvBlock(width, output_width, config.conv_kernel_size)) + width = output_width + self.blocks = nn.ModuleList(blocks) + + def forward(self, dna: Tensor) -> Tensor: + x = F.gelu(self.stem(dna.transpose(1, 2)), approximate="tanh") + for block in self.blocks: + x = F.max_pool1d(block(x), kernel_size=2, stride=2) + return x.transpose(1, 2) + + +class LocalResidualBlock(nn.Module): + def __init__(self, width: int, dilation: int): + super().__init__() + self.norm = ChannelRMSNorm(width) + self.depthwise = nn.Conv1d(width, width, 9, dilation=dilation, + padding=4 * dilation, groups=width) + self.expand = nn.Conv1d(width, 2 * width, 1) + self.contract = nn.Conv1d(2 * width, width, 1) + + def forward(self, x: Tensor) -> Tensor: + update = self.depthwise(self.norm(x)) + update = self.contract(F.gelu(self.expand(update), approximate='tanh')) + return x + update + + +class LocalSequencePath(nn.Module): + """Base-resolution motif path with powers-of-four dilation and aligned pooling.""" + def __init__(self, config: ModelConfig): + super().__init__() + width = config.local_width + self.stem = nn.Conv1d(4, width, 21, padding=10) + self.blocks = nn.Sequential(*(LocalResidualBlock(width, 4**i) + for i in range(config.local_context_blocks))) + self.norm = ChannelRMSNorm(width) + self.sites = nn.Conv1d(width, 4, 1) + self.bin_size = config.dense_bin_size + self.gradient_checkpointing = config.gradient_checkpointing + + def forward(self, dna: Tensor) -> tuple[Tensor, Tensor]: + base = F.gelu(self.stem(dna), approximate='tanh') + for block in self.blocks: + if self.gradient_checkpointing and self.training: + base = checkpoint(block, base, use_reentrant=False) + else: + base = block(base) + base = F.gelu(self.norm(base), approximate='tanh') + sites = self.sites(base).transpose(1, 2) + pooled = torch.cat((F.avg_pool1d(base, self.bin_size), + F.max_pool1d(base, self.bin_size)), dim=1).transpose(1, 2) + return sites, pooled + + +class CompactModel(nn.Module): + """One shared 1-kb trunk with representation-specific output heads.""" + + def __init__(self, config: ModelConfig | None = None): + super().__init__() + self.config = config or ModelConfig() + self.encoder = SequenceEncoder(self.config) + self.projection = nn.Linear(self.config.encoder_output_width, self.config.hidden_dim) + self.blocks = nn.ModuleList( + SequenceBlock(self.config) for _ in range(self.config.num_sequence_blocks) + ) + self.final_norm = nn.RMSNorm(self.config.hidden_dim) + self.dense_decoder = nn.Sequential( + nn.RMSNorm(self.config.hidden_dim), + nn.Linear(self.config.hidden_dim, self.config.head_hidden_size), + nn.GELU(approximate="tanh"), + ) + self.dense_heads = nn.ModuleDict( + { + family: nn.Linear(self.config.head_hidden_size, tracks) + for family, tracks in DENSE_TRACKS.items() + } + ) + self.splice_site_head = nn.Linear( + self.config.head_hidden_size, self.config.dense_bin_size * 4 + ) + self.splice_usage_head = nn.Linear( + self.config.head_hidden_size, self.config.splice_usage_tracks + ) + rank = self.config.junction_rank + self.junction_source = nn.Linear(self.config.head_hidden_size, rank) + self.junction_destination = nn.Linear(self.config.head_hidden_size, rank) + self.junction_track = nn.Embedding(self.config.junction_tracks, rank) + if self.config.junction_tracks % 2: + raise ValueError("junction tracks must split evenly into strand halves") + self.register_buffer( + "junction_track_half", + torch.arange(self.config.junction_tracks) // (self.config.junction_tracks // 2), + persistent=False, + ) + self.contact_q = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + self.contact_k = nn.Linear(self.config.hidden_dim, self.config.contact_tracks) + row, column = torch.triu_indices( + self.config.contact_bins, self.config.contact_bins + ) + self.register_buffer("contact_row", row, persistent=False) + self.register_buffer("contact_column", column, persistent=False) + if self.config.local_width: + self.local_path = LocalSequencePath(self.config) + self.local_projection = nn.Linear(2 * self.config.local_width, self.config.head_hidden_size) + # A new local path can be attached to our own checkpoint without changing its initial outputs. + nn.init.zeros_(self.local_path.sites.weight) + nn.init.zeros_(self.local_path.sites.bias) + nn.init.zeros_(self.local_projection.weight) + nn.init.zeros_(self.local_projection.bias) + if self.config.local_to_trunk: + self.local_trunk_projection = nn.Linear(2 * self.config.local_width, self.config.hidden_dim) + nn.init.zeros_(self.local_trunk_projection.weight) + nn.init.zeros_(self.local_trunk_projection.bias) + self._local_sites = None + self._local_features = None + + def set_junction_track_halves(self, halves: Tensor) -> None: + halves = torch.as_tensor(halves, dtype=torch.long, device=self.junction_track_half.device) + if halves.shape != (self.config.junction_tracks,) or not torch.isin( + halves, torch.tensor([0, 1], device=halves.device) + ).all(): + raise ValueError("junction track halves must be a 734-element 0/1 mapping") + self.junction_track_half.copy_(halves) + + def _trunk(self, dna: Tensor) -> Tensor: + if dna.ndim != 3 or dna.shape[1:] != (self.config.input_length, 4): + raise ValueError( + f"Expected DNA [B,{self.config.input_length},4], got {tuple(dna.shape)}" + ) + if self.config.local_width: + sites, pooled = self.local_path(dna.transpose(1, 2)) + self._local_sites = sites + self._local_features = self.local_projection(pooled) + x = self.projection(self.encoder(dna)) + if self.config.local_to_trunk: + local_tokens = pooled.reshape(len(dna), self.config.trunk_bins, + self.config.dense_upsample, -1).mean(dim=2) + x = x + self.local_trunk_projection(local_tokens) + for block in self.blocks: + if self.config.gradient_checkpointing and self.training and x.requires_grad: + x = checkpoint(block, x, None, use_reentrant=False) + else: + x = block(x, None) + return self.final_norm(x) + + def _dense_features(self, trunk: Tensor) -> Tensor: + start = (self.config.trunk_bins - self.config.target_trunk_bins) // 2 + target = trunk[:, start : start + self.config.target_trunk_bins] + target = target.repeat_interleave(self.config.dense_upsample, dim=1) + features = self.dense_decoder(target) + if self.config.local_width: + first_bin = (self.config.input_length - self.config.target_length) // (2 * self.config.dense_bin_size) + features = features + self._local_features[:, first_bin:first_bin + self.config.dense_bins] + return features + + def _all_dense_features(self, trunk: Tensor) -> Tensor: + features = self.dense_decoder( + trunk.repeat_interleave(self.config.dense_upsample, dim=1) + ) + return features + self._local_features if self.config.local_width else features + + def _splice_site_logits(self, trunk: Tensor) -> Tensor: + if self.config.splice_context_mode == 'uniform': + # The global trunk supplies one gene-context bias per class and kb. Base-resolution + # localization comes from the local path, with no arbitrary 128-bp offset decoder + # and no pooled dense-feature contribution to site logits. + context = self.splice_site_head(self.dense_decoder(trunk)) + context = context.view(len(trunk), self.config.trunk_bins, self.config.dense_bin_size, 4).mean(dim=2) + logits = context.repeat_interleave(self.config.trunk_bin_size, dim=1) + self._local_sites + return logits.float() + logits = self.splice_site_head(self._all_dense_features(trunk)) + logits = logits.view(len(trunk), self.config.input_length, 4) + if self.config.local_width: + logits = logits + self._local_sites + return logits.float() + + def _position_features(self, trunk: Tensor, positions: Tensor) -> Tensor: + features = self.dense_decoder(self._gather_bins(trunk, positions, self.config.trunk_bin_size)) + if self.config.local_width: + features = features + self._gather_bins(self._local_features, positions, self.config.dense_bin_size) + return features + + @staticmethod + def select_junction_positions(site_logits: Tensor, limit: int = 512) -> Tensor: + """Select each site type from model scores alone, then sort by coordinate. + + This follows the four independent site classes in Google DeepMind's + AlphaGenome splicing implementation (pinned in docs/v3-author/PROTOCOL.md). Padding is + -1 when an input is shorter than ``limit``. + """ + count = min(limit, site_logits.shape[1]) + selected = [] + for site_type in range(4): + coordinates = torch.topk(site_logits[..., site_type], count, dim=1).indices + coordinates = coordinates.sort(dim=1).values + if count < limit: + coordinates = F.pad(coordinates, (0, limit - count), value=-1) + selected.append(coordinates) + return torch.stack(selected, dim=1) + + @staticmethod + def _gather_bins(features: Tensor, positions: Tensor, bin_size: int) -> Tensor: + indices = torch.div(positions, bin_size, rounding_mode="floor") + indices = indices.clamp(0, features.shape[1] - 1) + return torch.gather( + features, + 1, + indices[..., None].expand(-1, -1, features.shape[-1]), + ) + + def _junction_predictions( + self, trunk: Tensor, site_positions: Tensor, queries: Tensor + ) -> Tensor: + row, column, track = queries.unbind(dim=-1) + half = self.junction_track_half[track] + batch = torch.arange(len(trunk), device=trunk.device)[:, None] + donor = site_positions[batch, 2 * half, row] + acceptor = site_positions[batch, 2 * half + 1, column] + source = self._position_features(trunk, donor.clamp(0, self.config.input_length - 1)) + destination = self._position_features(trunk, acceptor.clamp(0, self.config.input_length - 1)) + interaction = self.junction_source(source) * self.junction_destination(destination) + track_embedding = self.junction_track(track) + return (interaction * track_embedding).sum(dim=-1) / math.sqrt( + self.config.junction_rank + ) + + def junction_block( + self, + trunk: Tensor, + site_positions: Tensor, + row_start: int, + row_end: int, + column_start: int, + column_end: int, + track_start: int, + track_end: int, + site_factors: tuple[Tensor, Tensor] | None = None, + ) -> Tensor: + """Compute a bounded block of the full predicted-position junction cube.""" + tracks = torch.arange(track_start, track_end, device=trunk.device) + halves = self.junction_track_half[tracks] + if site_factors is None: + site_factors = self.junction_site_factors(trunk, site_positions) + source_all, destination_all = site_factors + source = source_all[:, halves, row_start:row_end] + destination = destination_all[:, halves, column_start:column_end] + embedding = self.junction_track(tracks)[None, :, None, None, :] + values = ( + source[:, :, :, None, :] * destination[:, :, None, :, :] * embedding + ).sum(dim=-1) / math.sqrt(self.config.junction_rank) + return values.permute(0, 2, 3, 1).float() + + def junction_site_factors( + self, trunk: Tensor, site_positions: Tensor + ) -> tuple[Tensor, Tensor]: + """Precompute the two strand-specific factors reused by all cube blocks.""" + donor = torch.stack((site_positions[:, 0], site_positions[:, 2]), dim=1).clamp_min(0) + acceptor = torch.stack((site_positions[:, 1], site_positions[:, 3]), dim=1).clamp_min(0) + donor_features = self._position_features(trunk, donor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, donor.shape[-1], -1) + acceptor_features = self._position_features(trunk, acceptor.reshape(len(trunk), -1)).reshape( + len(trunk), 2, acceptor.shape[-1], -1) + return self.junction_source(donor_features), self.junction_destination(acceptor_features) + + def forward( + self, + dna: Tensor, + *, + families: Iterable[str] | None = None, + usage_positions: Tensor | None = None, + junction_site_positions: Tensor | None = None, + junction_queries: Tensor | None = None, + ) -> dict[str, Tensor]: + requested = set(families or ALL_FAMILIES) + unknown = requested - set(ALL_FAMILIES) + if unknown: + raise ValueError(f"Unknown output families: {sorted(unknown)}") + trunk = self._trunk(dna) + needs_dense = bool( + requested.intersection((*DENSE_TRACKS, "splice_sites", "splice_usage", "junctions")) + ) + features = self._dense_features(trunk) if needs_dense else None + outputs: dict[str, Tensor] = {} + for family in DENSE_TRACKS: + if family in requested: + outputs[family] = self.dense_heads[family](features).float() + if "splice_sites" in requested: + all_logits = self._splice_site_logits(trunk) + start = (self.config.input_length - self.config.target_length) // 2 + outputs["splice_sites"] = all_logits[:, start:start + self.config.target_length] + if "splice_usage" in requested: + if usage_positions is None: + raise ValueError("splice_usage requires usage_positions") + gathered = self._gather_bins( + features, usage_positions, self.config.dense_bin_size + ) + outputs["splice_usage"] = self.splice_usage_head(gathered).float() + if "junctions" in requested: + if junction_site_positions is None or junction_queries is None: + raise ValueError("junctions require site positions and query triples") + outputs["junctions"] = self._junction_predictions( + trunk, junction_site_positions, junction_queries + ).float() + if "contact_maps" in requested: + contact = trunk.reshape( + len(dna), self.config.contact_bins, -1, self.config.hidden_dim + ).mean(dim=2) + q, k = self.contact_q(contact), self.contact_k(contact) + row, column = self.contact_row, self.contact_column + outputs["contact_maps"] = ( + 0.5 + * ( + q[:, row] * k[:, column] + + q[:, column] * k[:, row] + ) + ).float() + return outputs + + +def count_parameters(model: nn.Module) -> int: + return sum(parameter.numel() for parameter in model.parameters() if parameter.requires_grad) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_decoder111.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_decoder111.py new file mode 100644 index 0000000000000000000000000000000000000000..1a5b8c2dd386ef19447380be11a7b9018f673595 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_decoder111.py @@ -0,0 +1,226 @@ +"""CPU-only transfer of the exact grid96 decoder to the fixed eight-pass parent.""" +from datetime import datetime, timezone +import hashlib +import importlib +import importlib.util +import json +import os +from pathlib import Path +import shutil +import sys + +from make_linear_upsample83 import changed_source as interpolate +from make_splice_grid96 import changed_source as preserve_classifier +from verify_package import inference_signature, verify_package + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def read(path): + return json.loads(Path(path).read_text()) + + +def save(path, value): + assert not path.exists(), path + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(value, indent=2, allow_nan=False)+'\n') + + +def module(name, path): + spec = importlib.util.spec_from_file_location(name, path) + value = importlib.util.module_from_spec(spec) + sys.modules[name] = value + spec.loader.exec_module(value) + return value + + +def main(): + root = Path('/workspace/research') + here = root/'seqfunc' + study, output = root/'studies/decoder111', root/'experiments/decoder111' + parent = root/'experiments/calib81-small-epoch08' + previous = root/'experiments/splice_grid96' + assert not study.exists() and not output.exists() + assert read(root/'studies/splice_grid96/state.json')['phase'] == 'complete' + assert read(root/'studies/splice_grid96/diagnostics/summary/report.json')['complete_metrics'] == 13 + assert read(root/'studies/late_fixed_links81/small-epoch08/state.json')['phase'] == 'complete' + study.mkdir(); output.mkdir() + package, old = output/'package', parent/'package' + signature = inference_signature(old) + verify_package(old) + assert signature == read(parent/'evaluation.json')['package_inference_signature'] + original_source = (old/'reference/alphagenome_50m/model.py').read_text() + source = preserve_classifier(interpolate(original_source)) + assert source == (previous/'package/reference/alphagenome_50m/model.py').read_text() + old_three = inference_signature(root/'experiments/calib57/package') + assert all(signature[name] == old_three[name] for name in signature if name != 'last.pt') + assert digest(old/'last.pt') == '4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd' + now = datetime.now(timezone.utc).isoformat() + specification = dict(utc=now, parent=str(parent), output=str(output), parameters=13290504, + inherited_examples=328160, inherited_optimizer_steps=82040, inherited_complete_passes=8, + additional_training_updates=0, parent_inference_signature=signature, + candidate_model_sha256=hashlib.sha256(source.encode()).hexdigest(), + only_change='Apply the exact grid96 fixed decoder: linear bin-center global features for dense/usage/junction readouts, retaining the original nearest-neighbor splice classifier. Same backbone, heads, weights, four prior offsets and forward view.', + exact_existing_model_source=str(previous/'package/reference/alphagenome_50m/model.py'), + expected_exact_families=['splice_sites','contact_maps'], + expected_exact_variants=['sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'], + motivation='Grid96 restores the Linear83 ClinVar noncoding regression with positive paired intervals while retaining eight exact83 family reports. Its own reward difference from83 is uncertain. Test this fixed decoder on the stronger eight-pass parent without assuming transfer.', + scope='One fixed network for every family and both entrypoints; no routing among predictions, variant labels, fitted mixture, new offset, RC, SWA or motif mask.', + limitations='Exploratory validation-guided selection, one seed and shared development data. RNA/eQTL and other dense effects can change or regress. No accuracy or runtime benefit presumed.', + evaluation_admitted=False, GPU_used=False) + save(study/'proposed-recipe.json', specification) + shutil.copytree(old, package, ignore=shutil.ignore_patterns('__pycache__','*.pyc')) + (package/'reference/alphagenome_50m/model.py').write_text(source) + policy = {key:specification[key] for key in ['utc','parent_inference_signature','candidate_model_sha256', + 'parameters','inherited_examples','inherited_optimizer_steps','inherited_complete_passes', + 'additional_training_updates','only_change','scope','limitations']} + save(package/'decoder_policy.json', policy) + provenance = read(package/'training_provenance.json') + provenance['fixed_decoder111_intervention'] = policy + (package/'training_provenance.json').write_text(json.dumps(provenance,indent=2)+'\n') + history = package/'training_history'; history.mkdir(exist_ok=True) + (history/'pre-decoder111-model.py').write_text(original_source) + shutil.copy2(study/'proposed-recipe.json',history/'decoder111-proposed-recipe.json') + for name in ['prepare_decoder111.py','make_linear_upsample83.py','make_splice_grid96.py']: + shutil.copy2(here/name,history/name) + (history/'decoder111-training-replay.md').write_text( + 'The eight inherited training passes used pre-decoder111-model.py at reference/alphagenome_50m/model.py.\n' + 'Restore that source to replay the inherited training recipe. All model weights, training order, objective,\n' + 'batch and LR schedule are inherited. The inference-only change is the exact model source from grid96,\n' + 'generated by preserve_classifier(interpolate(original_source)); it adds zero optimizer updates.\n') + (package/'README.md').write_text('''# SeqFunc eight-pass model with the fixed grid96 decoder + +One **13,290,504-parameter** network reads 1,048,576 bp and serves all eleven families +and both entrypoints. A convolutional encoder and six width-384 transformer blocks +process 1,024 global tokens. A width-32 base-resolution path with six dilated blocks +has a 10,941-bp receptive field; pooled local features also enter the global trunk. +Attention and MLP residual updates retain the per-token RMS cap of 0.1. + +This candidate transfers the exact decoder source from experiment96 to the unchanged +eight-pass weights. Decoded global features are interpolated linearly from 1 kb to +128 bp at bin centers for dense, usage and junction outputs. The splice classifier +retains its original repeated global features, learned offset decoder and local logits. +Contacts retain their symmetric global-feature product. All are fixed internal paths +of one network. Both entrypoints use the same model and one forward orientation. +Model scores select splice coordinates; junctions follow transcript direction. + +All weights originate from local random initialization, seed17: **328,160 training +examples, 82,040 optimizer updates and eight complete passes** over all 41,020 eligible +FOLD_0 training intervals (folds2–7). The supplied reader decodes quantized targets. +Batch4 BF16 AdamW uses weight decay1e-4, clipping5 and the original14-pass cosine LR +schedule from2e-4 to2e-5 with4,000-example warmup. Fixed family variance scales use +256 training windows; splice BCE is balanced. Accumulated training-loop wall time is +1231.6 minutes on one shared H10080GB, with at most22CPUs, four loader workers and two +torch threads. Setup and evaluations are additional research compute. Complete inherited +configuration, training history and provenance are included. Recorded recovery preserves +checkpoint state and order; floating-point training replay is not claimed bit-exact. + +The decoder adds **zero parameters and zero training updates**. The original four +training-derived splice offsets and sigmoid output link remain fixed. They were not +refit. The inherited eight training passes used nearest-neighbor global features; +training_history/pre-decoder111-model.py records that source for training replay. + +Grid96 restored an earlier ClinVar noncoding regression at three passes; its small +reward difference from the fully interpolated decoder was uncertain. Transfer to eight +passes is unproven. This is exploratory validation-guided architecture selection on +the same development split and one training seed. No test labels, pretrained external +weights, variant-label fitting, conservation input or prediction caches are used. + +Both executable entrypoints take INPUT_ROOT and OUTPUT_ROOT and run offline using public +manifests and the pinned reference FASTA. runtime.tar.gz vendors CPython3.11, +PyTorch2.7.1/CUDA12.6 and dependencies; the host supplies Linux, bash/tar/gzip and an +NVIDIA driver. Tracks use float16 and perturbations float32. No targets or supplied +splice coordinates are used at inference. + +Own native/recovery checks, all200 validation windows, all2,697 development perturbations, +all13 metrics with same-row published anchors, representative native runtime and paired +comparisons to the sole eight-pass parent are required. Candidate measurements are +appended after scoring. Splice-site/contact family reports and the five splicing variant +reports must exactly match the parent. Other effects can change or regress. Sampled +runtime estimates are not guaranteed bounds. Limitations include finite local context, +pooled usage/junction features, low-rank contacts and uneven variant effects. The54 masked +GTEx RNA tracks receive no direct target supervision; they are excluded from scored RNA +means. Functional readouts do not supply the conservation information used by strong +TraitGym anchors. +''') + manifest = read(package/'package_manifest.json') + for name in list(manifest['files']): + if name.startswith('validation_'): + (package/name).unlink(missing_ok=True); manifest['files'].pop(name) + for path in package.rglob('*'): + if path.is_file() and path.name!='package_manifest.json' and '__pycache__' not in path.parts: + relative=str(path.relative_to(package)) + if relative not in manifest['files'] or relative in ['README.md','training_provenance.json','reference/alphagenome_50m/model.py']: + manifest['files'][relative]=digest(path) + (package/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + actual = inference_signature(package) + assert {name for name in actual if actual[name]!=signature[name]} == {'reference/alphagenome_50m/model.py'} + verification = verify_package(package) + # Actual eight-pass learned tensors, short synthetic DNA, CPU only. + import torch + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:2]) + torch.set_num_threads(2); torch.set_num_interop_threads(1); torch.manual_seed(1729) + sys.path.insert(0,str(package)) + candidate = importlib.import_module('reference.alphagenome_50m.model') + nearest = module('reference.alphagenome_50m.nearest111',old/'reference/alphagenome_50m/model.py') + linear = module('reference.alphagenome_50m.linear111',root/'experiments/linear_upsample83/package/reference/alphagenome_50m/model.py') + payload = torch.load(package/'last.pt',map_location='cpu',weights_only=True) + config = dict(payload['config'],input_length=8192,target_length=4096,gradient_checkpointing=False) + models = [m.CompactModel(m.ModelConfig(**config)).eval() for m in [candidate,nearest,linear]] + for model in models: + model.load_state_dict(payload['model'],strict=True) + assert candidate.count_parameters(model)==13290504 + assert all(torch.equal(value,payload['model'][key]) for key,value in model.state_dict().items()) + dna = torch.nn.functional.one_hot(torch.randint(0,4,(2,8192)),4).float() + dna[1] = dna[0].clone(); dna[1,4096] = torch.roll(dna[1,4096],1); dna[1,6000:6016]=0 + usage = torch.tensor([[0,127,128,2048,4095]]).repeat(2,1) + sites = torch.tensor([[[0,1023,4096],[1024,6144,8191],[128,4095,8191],[0,4096,6144]]]).repeat(2,1,1) + junctions = torch.tensor([[[0,1,0],[1,0,733],[2,2,367]]]).repeat(2,1,1) + with torch.no_grad(): + outputs=[m(dna,usage_positions=usage,junction_site_positions=sites,junction_queries=junctions) for m in models] + assert len(outputs[0])==11 + for name,value in outputs[0].items(): + expected=outputs[1][name] if name in ['splice_sites','contact_maps'] else outputs[2][name] + assert torch.equal(value,expected),name + assert torch.isfinite(value).all(),name + assert any(not torch.equal(outputs[0][name],outputs[1][name]) for name in ['atac','rna_seq','dnase']) + trunks=[m._trunk(dna) for m in models[:2]] + assert torch.equal(trunks[0],trunks[1]) + selected=[m.select_junction_positions(m._splice_site_logits(t),limit=32) for m,t in zip(models[:2],trunks)] + assert torch.equal(*selected) + full=models[0]._all_dense_features(trunks[0]) + assert torch.equal(models[0]._dense_features(trunks[0]),full[:,16:48]) + positions=torch.tensor([[0,127,128,1023,1024,4095,4096,8191]]).repeat(2,1) + assert torch.equal(models[0]._position_features(trunks[0],positions),models[0]._gather_bins(full,positions,128)) + assert not torch.cuda.is_initialized() + cpu=dict(utc=datetime.now(timezone.utc).isoformat(),candidate_inference_signature=actual, + all_learned_tensors_exact=True,parameters=13290504,synthetic_batch=2,synthetic_input_bp=8192, + actual_allele_and_N_edits=True,all11_outputs_finite=True, + classifier_contacts_trunk_and_selected_positions_exact_parent=True, + other_nine_fixed_query_outputs_exact_linear_operator_with_same8pass_weights=True, + full_central_and_sparse_features_exact=True,source_exact_completed96=True, + CUDA_used=False,training_updates=0,labels_read=False, + caveat='Mechanical short synthetic input; own full1Mb native/recovery and full accuracy remain required.') + save(study/'CPU-construction-check.json',cpu) + evidence=[parent/'evaluation.json',parent/'valid-score/report.json',parent/'valid-variant-score/report.json', + root/'studies/late_fixed_links81/small-epoch08/pipeline-completion-observation.json', + root/'studies/splice_grid96/completion-observation.json', + root/'studies/splice_grid96/diagnostics/summary/report.json', + root/'studies/splice_grid96/construction-check.json',study/'proposed-recipe.json',study/'CPU-construction-check.json', + here/'make_linear_upsample83.py',here/'make_splice_grid96.py',Path(__file__)] + save(study/'preparation.json',dict(utc=datetime.now(timezone.utc).isoformat(), + package=str(package),package_verification=verification, + parent_inference_signature=signature,candidate_inference_signature=actual, + evidence_sha256={str(p):digest(p) for p in evidence}, + evaluation_admitted=False,controller_created=False,GPU_used=False, + next='Freeze independent native52, six exact parent arrays, full200/2697/all13, exact twofamily/fivevariant reports, native624/runtime and one all11/all13 paired observer after fresh resource review.')) + print(json.dumps(dict(preparation_sha256=digest(study/'preparation.json'), + CPU_check_sha256=digest(study/'CPU-construction-check.json'),package=verification, + candidate_model_sha256=actual['reference/alphagenome_50m/model.py'],GPU_used=False)),flush=True) + + +if __name__=='__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large114.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large114.py new file mode 100644 index 0000000000000000000000000000000000000000..da903609218bc69e8fd933953513deba2c8e9fde --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large114.py @@ -0,0 +1,222 @@ +"""CPU-only preparation of unchanged fixed RC inference on actual large8 weights.""" +from datetime import datetime, timezone +import importlib +import json +import os +from pathlib import Path +import shutil +import subprocess +import sys + +from make_rc_candidate import digest +from verify_package import inference_signature, verify_package + +ROOT=Path('/workspace/research') +HERE=ROOT/'seqfunc' + + +def read(path): + return json.loads(Path(path).read_text()) + + +def save(path,value): + path=Path(path) + assert not path.exists(),path + path.parent.mkdir(parents=True,exist_ok=True) + path.write_text(json.dumps(value,indent=2,allow_nan=False)+'\n') + + +def main(): + study=ROOT/'studies/rc_large114'; output=ROOT/'experiments/rc_large114' + recipe=read(study/'proposed-recipe.json') + assert recipe['parameters']==215626632 and recipe['orientation_weights']==[.5,.5] + assert recipe['additional_training_updates']==0 and not recipe['GPU_evaluation_admitted'] + assert not output.exists() and not (study/'preparation.json').exists() + parent=Path(recipe['parent']); parent_package=parent/'package' + parent_study=ROOT/'studies/late_fixed_links81/large-epoch08' + signature=inference_signature(parent_package) + assert read(parent/'evaluation.json')['package_inference_signature']==signature + tracks=read(parent/'valid-score/report.json'); variants=read(parent/'valid-variant-score/report.json') + assert tracks['examples']==200 and tracks['invalid_reason'] is None and len(tracks['families'])==11 + assert len(variants)==13 + native=read(parent_study/'native-recovery/report.json') + assert native['package_inference_signature']==signature and native['all_arrays_bit_identical'] + assert native['arrays']=={'tracks':30,'perturbations':22} + small_signature=inference_signature(ROOT/'experiments/calib81-small-epoch08/package') + assert set(signature)==set(small_signature) + assert all(v==small_signature[k] for k,v in signature.items() if k!='last.pt') + assert signature['last.pt']=='91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb' + provenance=read(parent_package/'training_provenance.json') + assert [provenance[k] for k in ['global_step','optimizer_steps','epoch','offset','seed']]==[328160,82040,8,0,17] + assert read(parent_package/'package_manifest.json')['parameters']==215626632 + assert provenance['config']['hidden_dim']==1024 and provenance['config']['num_sequence_blocks']==16 + assert provenance['config']['local_width']==64 and provenance['config']['head_hidden_size']==768 + assert digest(parent_package/'splice_calibration.json')==digest(ROOT/'experiments/rc108/package/splice_calibration.json') + output.mkdir(); package=output/'package' + environment=dict(os.environ,OPENBLAS_NUM_THREADS='1',OMP_NUM_THREADS='2',MKL_NUM_THREADS='2', + PYTHONDONTWRITEBYTECODE='1',PYTHONNOUSERSITE='1') + environment.pop('SEQFUNC_TRAIN_BYTECACHE',None) + with (study/'build.log').open('w') as log: + subprocess.run([sys.executable,str(HERE/'make_rc_candidate.py'),str(parent_package),str(package), + '--name','SeqFunc large eight-pass model with fixed forward/reverse-complement inference'], + env=environment,stdout=log,stderr=subprocess.STDOUT,check=True) + policy=read(package/'rc_ensemble_policy.json') + assert policy['parent_inference_signature']==signature and policy['unique_learned_parameters']==215626632 + assert policy['weights']==[.5,.5] and policy['fitted_ensemble_weights'] is False + assets=['predict.py','perturb.py','rc_ensemble.py','single_prediction.py','strand_permutations.json'] + for name in assets:assert digest(package/name)==digest(ROOT/'experiments/rc108/package'/name),name + assert digest(package/'last.pt')==signature['last.pt'] + assert digest(package/'reference/alphagenome_50m/model.py')==signature['reference/alphagenome_50m/model.py'] + manifest=read(package/'package_manifest.json') + for name in list(manifest['files']): + if name.startswith('validation_'): + (package/name).unlink(missing_ok=True); manifest['files'].pop(name) + history=package/'training_history';history.mkdir(exist_ok=True) + shutil.copy2(study/'proposed-recipe.json',history/'rc-large114-proposed-recipe.json') + for name in [Path(__file__).name,'make_rc_candidate.py']: + shutil.copy2(HERE/name,history/name) + (history/'rc-large114-training-replay.md').write_text( + 'All eight inherited training passes used the unchanged forward model source and original\n' + 'B4 seed17 fourteen-pass schedule. This candidate changes only inference wrappers via\n' + 'make_rc_candidate.py, retaining all model tensors, four training-derived splice offsets\n' + 'and vendored training assets. It adds no optimizer updates or learned parameters.\n') + (package/'README.md').write_text('''# SeqFunc large model with fixed forward/reverse-complement inference + +One model with **215,626,632 unique learned parameters** reads 1,048,576 bp and predicts +all eleven families and 5,930 tracks through both offline entrypoints. It averages the +same learned weights on the supplied DNA and its reverse complement, with fixed equal +weights and vendored public strand permutations. No additional flanking sequence is read. + +The model uses a strided convolutional encoder (initial width96, growth96), sixteen +width1024 transformer blocks over1,024 tokens, and width768 output features. A width64 +base-resolution path has six dilated residual blocks and a10,941-bp receptive field; +its pooled features feed both the heads and the global trunk. Attention and MLP updates +use a per-token RMS cap0.1. Dense outputs cover the central196,608bp; contacts and junctions +use factorized readouts. The original nearest global-feature decoder is retained. + +Each orientation applies the same four training-prior splice offsets before averaging +probabilities. Dense and contact readouts average in asinh space. Sparse positions are +selected independently per view, merged with zero for an unlisted view, and capped at +4,096 usage positions and16,384 junction pairs. The orientation code is exactly the +verified RC17/85/108 implementation. This candidate adds no interpolation, motif filter, +parameter averaging, fitted mixing weights or learned parameters. + +All weights originate from local random initialization with seed17. Training used eight +complete passes through all41,020 eligible FOLD_0 intervals in folds2–7: **328,160 examples +and82,040 AdamW updates**. The supplied reader decodes quantized training targets per +example. BF16 batch4 training uses fixed family-variance scales estimated from256 training +windows, balanced splice BCE, AdamW weight decay1e-4, gradient clipping5, and the original +fourteen-pass cosine schedule2e-4 to2e-5 with4,000 warmup examples. The saved checkpoint +records108,193.2591 accumulated training-loop seconds (30.05h) on one shared H10080GB, +within22CPUs. Setup and evaluation add cost; this is shared wall time rather than exclusive +GPU compute. Numerical training recovery is not bit-exact. This inference change adds zero SGD. + +Run `infer.sh INPUT_ROOT OUTPUT_ROOT` or `perturb.sh INPUT_ROOT OUTPUT_ROOT` offline. +The package vendors CPython3.11, PyTorch2.7.1/CUDA12.6 and dependencies; the host provides +Linux, bash/tar/gzip and the NVIDIA driver. Inputs are public manifests and the pinned +reference only. No targets, supplied splice positions, prediction cache or network are used. +Track arrays are stored as float16 after float32 averaging; perturbation arrays stay float32. + +Candidate validation is pending. Required checks are its own52-array native/offline/recovery +test,200 validation windows, all2,697 development queries and13 metrics with published anchors, +624 representative native timing arrays, and all three combined runtime estimates within4h. +All11/all13 paired comparisons are fixed against the calibrated large forward parent and +RC108. The latter has equal exposure, offsets and views but smaller global/local capacity +and different compute; it is not an independent training replicate. Parent scores are not +candidate scores. The strongest fully gated package remains installed until an improvement. + +Limitations include one seed per size, repeated development selection, pooled local features, +finite local context, low-rank contacts, sparse-union dilution and uneven variant prediction. +Fifty-four GTEx RNA heads have no direct training targets and are excluded from scored RNA +means; eQTL scoring uses supervised ENCODE RNA tracks. There is no conservation input. +Zero-inclusive intervals do not establish equivalence; sampled runtimes are not bounds. +''') + for path in package.rglob('*'): + if path.is_file() and path.name!='package_manifest.json' and '__pycache__' not in path.parts: + name=str(path.relative_to(package)) + if name not in manifest['files'] or name=='README.md':manifest['files'][name]=digest(path) + (package/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + verified=verify_package(package) + actual=inference_signature(package) + save(study/'package-before-CPU-check.json',dict(verification=verified,candidate_inference_signature=actual)) + + import numpy as np + import pandas as pd + import torch + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:2]) + torch.set_num_threads(2);torch.set_num_interop_threads(1);torch.manual_seed(1729) + sys.path.insert(0,str(package)) + module=importlib.import_module('reference.alphagenome_50m.model') + rc=importlib.import_module('rc_ensemble') + single=importlib.import_module('single_prediction').single_prediction + splice=importlib.import_module('splice_output').splice_output_scores + payload=torch.load(package/'last.pt',map_location='cpu',weights_only=True) + config=dict(payload['config'],input_length=8192,target_length=4096,gradient_checkpointing=False) + model=module.CompactModel(module.ModelConfig(**config)).eval() + model.load_state_dict(payload['model'],strict=True) + assert module.count_parameters(model)==215626632 + metadata_path=Path('/workspace/data/public/junction_tracks.parquet') + metadata=pd.read_parquet(metadata_path).sort_values('tensor_index') + halves=metadata.junction_channel_half.to_numpy(np.int64) + model.set_junction_track_halves(halves) + engine=rc.engine_for(model) + forward=torch.nn.functional.one_hot(torch.randint(0,4,(2,8192)),4).float() + forward[1]=forward[0].clone();forward[1,4096]=torch.roll(forward[1,4096],1);forward[1,6000:6016]=0 + four=torch.cat((forward,forward.flip((1,2))),dim=0) + usage=torch.tensor([[0,127,128,2048,4095]]).repeat(4,1) + sites=torch.tensor([[[0,1023,4096],[1024,6144,8191],[128,4095,8191],[0,4096,6144]]]).repeat(4,1,1) + queries=torch.tensor([[[0,1,0],[1,0,733],[2,2,367]]]).repeat(4,1,1) + rng=torch.get_rng_state().clone() + with torch.no_grad(): + outputs=model(four,usage_positions=usage,junction_site_positions=sites,junction_queries=queries) + assert len(outputs)==11 and all(torch.isfinite(v).all() for v in outputs.values()) + del outputs + original=single(model,forward[:1]); reverse=single(model,forward[:1].flip((1,2))) + mapped=rc.reverse_readout(reverse,engine.permutations,8192,4096,model.config.contact_bins) + expected=rc.combine_readouts(original,mapped,halves) + observed=engine.tracks(forward[:1]) + assert set(observed)==set(expected) and len(observed)==15 + for name in observed: + assert np.isfinite(observed[name]).all() and np.array_equal(observed[name],expected[name]),name + # Round-trip the actual native sparse/dense arrays through the frozen coordinate map. + back=rc.reverse_readout(rc.reverse_readout(original,engine.permutations,8192,4096,model.config.contact_bins),engine.permutations,8192,4096,model.config.contact_bins) + for name in original:assert np.array_equal(back[name],original[name]),name + families=[*rc.DENSE_TRACKS,'splice_sites'] + requested=engine.dense_and_sites(forward,families) + views=[] + for reverse_view in [False,True]: + trunk=model._trunk(forward.flip((1,2)) if reverse_view else forward) + features=model._dense_features(trunk) + arrays={name:model.dense_heads[name](features).float().numpy() for name in rc.DENSE_TRACKS} + arrays['splice_sites']=splice(model._splice_site_logits(trunk)[:,2048:6144]).float().numpy() + if reverse_view:arrays={name:value[:,::-1][:,:,engine.permutations[name]].copy() for name,value in arrays.items()} + views.append(arrays) + for name in families: + assert np.isfinite(requested[name]).all() + assert np.array_equal(requested[name],(views[0][name]+views[1][name])*np.float32(.5)),name + assert torch.equal(rng,torch.get_rng_state()) + assert all(torch.equal(value,payload['model'][name]) for name,value in model.state_dict().items()) + assert not torch.cuda.is_initialized() + proof=dict(utc=datetime.now(timezone.utc).isoformat(),parameters=215626632, + actual_large8_weights_sha256=digest(package/'last.pt'),candidate_inference_signature=actual, + synthetic_input_bp=8192,synthetic_target_bp=4096,base_model_batch=4, + actual_allele_and_N_edits=True,forward_and_reverse_inputs=True,all11_base_outputs_finite=True, + actual_native_track_arrays_exact_manual_fixed_mean=15, + actual_native_coordinate_and_strand_roundtrip_exact=True, + paired_dense_and_splice_arrays_exact_manual_fixed_mean=8, + all_learned_tensors_and_CPU_RNG_unchanged=True,public_junction_metadata_sha256=digest(metadata_path), + exact_RC108_wrapper_and_permutation_assets=True,CUDA_used=False,training_updates=0,labels_read=False, + caveat='CPU short-input compatibility of the actual large model and unchanged RC operation. It does not establish full1Mb GPU memory/runtime or accuracy; all candidate native/full/runtime gates remain required.') + save(study/'CPU-construction-check.json',proof) + evidence=[study/'proposed-recipe.json',study/'package-before-CPU-check.json',study/'CPU-construction-check.json', + parent/'evaluation.json',parent/'valid-score/report.json',parent/'valid-variant-score/report.json', + parent_study/'native-recovery/report.json',HERE/'make_rc_candidate.py',Path(__file__)] + save(study/'preparation.json',dict(utc=datetime.now(timezone.utc).isoformat(),package=str(package), + package_verification=verified,parent_inference_signature=signature,candidate_inference_signature=actual, + evidence_sha256={str(p):digest(p) for p in evidence},GPU_evaluation_admitted=False, + controller_created=False,GPU_used=False,next='Wait parent original paired completion; review resources and freeze candidate native/full/runtime/dual-control evaluation before execution.')) + print(json.dumps(dict(preparation_sha256=digest(study/'preparation.json'),CPU_proof_sha256=digest(study/'CPU-construction-check.json'), + package=verified,additional_training_updates=0,GPU_used=False)),flush=True) + + +if __name__=='__main__':main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large_decoder117.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large_decoder117.py new file mode 100644 index 0000000000000000000000000000000000000000..6457c65e742e35bbd5b7f51b021f5efe9f6f4b7b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/prepare_rc_large_decoder117.py @@ -0,0 +1,201 @@ +"""CPU-only fixed decoder111 transfer into the completed RC114 large model; no evaluation admission.""" +from datetime import datetime, timezone +import importlib +import json +import os +from pathlib import Path +import shutil +import sys + +from prepare_decoder111 import digest, read, save, module +from make_linear_upsample83 import changed_source as interpolate +from make_splice_grid96 import changed_source as preserve_classifier +from verify_package import inference_signature, verify_package + + +def main(): + root = Path('/workspace/research') + here = root/'seqfunc' + study, output = root/'studies/rc_large_decoder117', root/'experiments/rc_large_decoder117' + parent = root/'experiments/rc_large114' + old = root/'studies/rc_large114/paired-documentation/package' + decoder = root/'experiments/rc_decoder113/package' + assert not study.exists() and not output.exists() + assert read(root/'studies/rc_large114/completion-observation.json')['reward'] == .5222012903969038 + assert read(root/'studies/rc_decoder113/completion-observation.json')['reward'] == .5125703796142637 + assert read(root/'studies/rc_large114/diagnostics/parent/summary.json')['complete_metrics'] == 13 + assert read(root/'studies/rc_decoder113/diagnostics/summary.json')['complete_metrics'] == 13 + assert read(root/'studies/rc_large114/state.json')['phase'] == 'complete' + assert read(root/'studies/rc_decoder113/state.json')['phase'] == 'complete' + verify_package(old) + signature = inference_signature(old) + assert signature == inference_signature(parent/'package') == read(parent/'evaluation.json')['package_inference_signature'] + original_source = (old/'reference/alphagenome_50m/model.py').read_text() + changed = preserve_classifier(interpolate(original_source)) + assert changed == (decoder/'reference/alphagenome_50m/model.py').read_text() + assert digest(old/'last.pt') == '91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb' + assert digest(decoder/'last.pt') == '4e81120dd5424c175e43684c0827b8a4f21e5be8b0b554f17df615ac16a027dd' + assert digest(old/'last.pt') != digest(decoder/'last.pt') + assert digest(old/'splice_calibration.json') == digest(decoder/'splice_calibration.json') + orientation = read(old/'rc_ensemble_policy.json') + assert orientation['orientation_views'] == 2 and orientation['weights'] == [.5,.5] + assert orientation['fitted_ensemble_weights'] is False + study.mkdir(); output.mkdir() + package = output/'package' + now = datetime.now(timezone.utc).isoformat() + recipe = dict(utc=now,parent=str(parent),parent_package=str(old),output=str(output),parameters=215626632, + inherited_examples=328160,inherited_optimizer_steps=82040,inherited_complete_passes=8, + additional_training_updates=0,parent_inference_signature=signature, + candidate_model_sha256=digest(decoder/'reference/alphagenome_50m/model.py'), + only_change='Replace RC114 nearest global-feature decoding for dense, usage and junction heads with the exact selective linear-bin-center operation from completed111/96. Keep its original nearest splice classifier, contacts, learned tensors, four training-derived offsets and equal forward/RC views.', + expected_exact_families=['splice_sites','contact_maps'], + expected_exact_variants=['sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'], + motivation='The completed small-model113 decoder comparison improves reward0.0022182677 and six dense families under both paired track schemes, with nine positive changed-family points. Published largeRC114 has higher reward0.5222012904 but CAGE0.35921416 remains weak. Test whether the identical fixed decoder operation transfers to the larger network; no gain is assumed.', + contrary_evidence='Small113 reduces Mendelian AP by0.0105883003 with paired interval[-0.0251803684,-0.0003448354]; its other12 variant intervals includezero. Large114 already has supported caQTL regression versusitsforwardparent and weak eQTL/caQTL/SPI1/enhancer effects. Interpolation can further attenuate or distortalleleeffects. This candidate must report all13,including every regression, and exactfiveunchangedsplice reports; no broadvariantrescue claim.', + scope='One fixed network and the same two orientation views serve every family and both entrypoints. No learned or fitted mixture, routing, new splice prior, motif mask, SWA, extra flanking DNA, variant-label training or additional SGD.', + limitations='Exploratory one-seed postselection. Interaction can worsen any changed readout. Full native replay, runtime and accuracy must be measured independently; unchanged paths require exact complete reports.', + execution_gate='Wait RC115 original full/native/runtime/all8paired completion, then fresh resources preserving mainlarge14,seed110 and sixhours evaluation reserves. CPU preparation only now; no thirdtrainer.', + evaluation_admitted=False,GPU_used=False) + save(study/'proposed-recipe.json',recipe) + shutil.copytree(old,package,ignore=shutil.ignore_patterns('__pycache__','*.pyc')) + (package/'reference/alphagenome_50m/model.py').write_text(changed) + policy={k:recipe[k] for k in ['utc','parent_inference_signature','candidate_model_sha256','parameters', + 'inherited_examples','inherited_optimizer_steps','inherited_complete_passes','additional_training_updates', + 'only_change','scope','limitations']} + save(package/'decoder117_policy.json',policy) + provenance=read(package/'training_provenance.json') + provenance['fixed_decoder117_intervention']=policy + (package/'training_provenance.json').write_text(json.dumps(provenance,indent=2)+'\n') + history=package/'training_history';history.mkdir(exist_ok=True) + (history/'pre-decoder117-model.py').write_text(original_source) + shutil.copy2(study/'proposed-recipe.json',history/'decoder117-proposed-recipe.json') + for name in [Path(__file__).name,'prepare_decoder111.py','make_linear_upsample83.py','make_splice_grid96.py']: + shutil.copy2(here/name,history/name) + (history/'decoder117-training-replay.md').write_text( + 'All eight inherited training passes used pre-decoder117-model.py at reference/alphagenome_50m/model.py.\n' + 'Restore that source for the inherited training recipe. Training remains forward-only with the original\n' + 'objective, seed17, batch4, data order, AdamW state and fourteen-pass learning-rate schedule.\n' + 'RC114 introduced equal forward/RC inference without SGD. Decoder117 adds only the exact model.py\n' + 'from completed111/96/113 via preserve_classifier(interpolate(original_source)); no weights are changed.\n') + (package/'README.md').write_text('''# SeqFunc RC model with selective linear feature decoding + +One model with **215,626,632 unique learned parameters** reads 1,048,576 bp and serves +all eleven families, 5,930 tracks and both offline entrypoints. It uses the same fixed +equal forward/reverse-complement views as RC114. The only new operation is the exact +decoder from experiments111/96/113: global features are interpolated at bin centers from +1 kb to128 bp for dense, splice-usage and junction outputs. The original nearest +splice classifier and contact path remain unchanged. No weights or constants are fitted. + +The architecture has a strided convolutional encoder (initial width96, growth96), sixteen width1024 transformer blocks +over1,024 global tokens, a width64 base-resolution path with a21-bp stem and six dilated +residual blocks (10,941-bp receptive field), and a local-to-global bridge. Attention and +MLP updates use RMS cap0.1. Dense outputs cover the central196,608bp; contacts and junctions +use low-rank factors. Each orientation applies the four inherited training-prior splice +offsets before averaging probabilities. Dense and contact values average in asinh space. +Sparse coordinates are predicted independently in each view; the union treats an unlisted +view as zero, keeping at most4,096 usage positions and16,384 junction pairs. + +All weights originate from local random initialization, seed17, with **328,160 examples / +82,040 AdamW updates**, eight complete passes through all41,020 FOLD_0 training intervals +in folds2–7. The supplied reader decodes quantized targets per example. Batch4 BF16 training +uses fixed family-variance scales estimated from256 training windows and balanced splice +BCE. AdamW uses weight decay1e-4, clipping5, and the original14-pass cosine schedule from +2e-4 to2e-5 with4,000 warmup examples. The source checkpoint records108,193.2591 accumulated +training-loop seconds (30.05h) on one shared H10080GB within22CPUs. Setup/evaluation cost is +additional. Training recovery preserves checked saved state/order but is not bit-identical. +This candidate adds zero parameters and zero training updates. Restore the recorded +pre-decoder model source to replay the inherited training recipe. + +Run `infer.sh INPUT_ROOT OUTPUT_ROOT` or `perturb.sh INPUT_ROOT OUTPUT_ROOT` offline. +The runtime archive vendors CPython3.11, PyTorch2.7.1/CUDA12.6 and dependencies; the host +supplies Linux, bash/tar/gzip and the NVIDIA driver. Both entrypoints use public manifests +and the pinned reference only, with no targets, supplied splice coordinates, prediction +cache, extra flanking sequence or network access. Track arrays use float16 after float32 +orientation averaging; perturbation arrays remain float32. + +Candidate validation is pending. Its own52-array native/recovery check, six unchanged +native site/contact arrays versus RC114, full200 validation windows, all2,697 development +queries/all13 metrics with same-row anchors,624 native timing arrays, and all three4h +runtime estimates are required. Splice-site/contact family reports and five splice-based +variant reports must exactly match RC114. An all11/all13 paired comparison to RC114 is +required. Parent scores are not candidate scores. The smaller113 comparison reduced MendelianAP despite +track gains; all newvariant gains andregressions must be measured. GPU evaluation waits for +RC115 full/native/runtime/allpaired completion and a fresh single-H100 resource review. + +Limitations include one-seed exploratory development selection, pooled local features, +finite local context, sparse-union dilution, low-rank contacts and uneven variant effects. +Fifty-four GTEx RNA heads lack direct target supervision and are excluded from scored RNA +means; eQTL scoring uses supervised ENCODE RNA tracks. The model has no conservation input. +Interval overlap with zero does not show equivalence; sampled runtimes are not upper bounds. +''') + manifest=read(package/'package_manifest.json') + for name in list(manifest['files']): + if name.startswith('validation_'): + (package/name).unlink(missing_ok=True);manifest['files'].pop(name) + for path in package.rglob('*'): + if path.is_file() and path.name!='package_manifest.json' and '__pycache__' not in path.parts: + relative=str(path.relative_to(package)) + if relative not in manifest['files'] or relative in ['README.md','training_provenance.json','reference/alphagenome_50m/model.py']: + manifest['files'][relative]=digest(path) + (package/'package_manifest.json').write_text(json.dumps(manifest,indent=2)+'\n') + actual=inference_signature(package) + assert set(actual)==set(signature) + assert {name for name in actual if actual[name]!=signature[name]}=={'reference/alphagenome_50m/model.py'} + for name in ['rc_ensemble.py','single_prediction.py','strand_permutations.json','rc_ensemble_policy.json','splice_calibration.json']: + assert digest(package/name)==digest(old/name),name + verification=verify_package(package) + + import torch + os.sched_setaffinity(0,sorted(os.sched_getaffinity(0))[:2]) + torch.set_num_threads(2);torch.set_num_interop_threads(1);torch.manual_seed(1729) + sys.path.insert(0,str(package)) + candidate=importlib.import_module('reference.alphagenome_50m.model') + nearest=module('reference.alphagenome_50m.nearest117',old/'reference/alphagenome_50m/model.py') + payload=torch.load(package/'last.pt',map_location='cpu',weights_only=True) + config=dict(payload['config'],input_length=8192,target_length=4096,gradient_checkpointing=False) + models=[m.CompactModel(m.ModelConfig(**config)).eval() for m in [candidate,nearest]] + for model in models: + model.load_state_dict(payload['model'],strict=True) + assert candidate.count_parameters(model)==215626632 + assert all(torch.equal(v,payload['model'][k]) for k,v in model.state_dict().items()) + forward=torch.nn.functional.one_hot(torch.randint(0,4,(2,8192)),4).float() + forward[1]=forward[0].clone();forward[1,4096]=torch.roll(forward[1,4096],1);forward[1,6000:6016]=0 + dna=torch.cat((forward,torch.flip(forward,dims=[1,2])),dim=0) + usage=torch.tensor([[0,127,128,2048,4095]]).repeat(4,1) + sites=torch.tensor([[[0,1023,4096],[1024,6144,8191],[128,4095,8191],[0,4096,6144]]]).repeat(4,1,1) + junctions=torch.tensor([[[0,1,0],[1,0,733],[2,2,367]]]).repeat(4,1,1) + with torch.no_grad(): + outputs=[m(dna,usage_positions=usage,junction_site_positions=sites,junction_queries=junctions) for m in models] + assert len(outputs[0])==11 and all(torch.isfinite(v).all() for v in outputs[0].values()) + for name in ['splice_sites','contact_maps']:assert torch.equal(outputs[0][name],outputs[1][name]),name + assert any(not torch.equal(outputs[0][name],outputs[1][name]) for name in ['atac','dnase','rna_seq']) + trunks=[m._trunk(dna) for m in models];assert torch.equal(*trunks) + selected=[m.select_junction_positions(m._splice_site_logits(t),limit=32) for m,t in zip(models,trunks)] + assert torch.equal(*selected) + full=models[0]._all_dense_features(trunks[0]);assert torch.equal(models[0]._dense_features(trunks[0]),full[:,16:48]) + positions=torch.tensor([[0,127,128,1023,1024,4095,4096,8191]]).repeat(4,1) + assert torch.equal(models[0]._position_features(trunks[0],positions),models[0]._gather_bins(full,positions,128)) + assert not torch.cuda.is_initialized() + proof=dict(utc=datetime.now(timezone.utc).isoformat(),candidate_inference_signature=actual, + all_learned_tensors_exact=True,parameters=215626632,synthetic_batch=4,synthetic_input_bp=8192, + actual_allele_and_N_edits=True,both_forward_and_RC_synthetic_inputs=True,all11_outputs_finite=True, + classifier_contacts_trunk_and_selected_positions_exact_parent=True,full_central_and_sparse_features_exact=True, + source_exact_completed96=True,source_exact_completed113=True,weights_exact_large114=True, + orientation_wrappers_and_policy_exact_RC114=True,CUDA_used=False,training_updates=0,labels_read=False, + accepted111_CPU_proof_sha256=digest(root/'studies/decoder111/CPU-construction-check.json'), + caveat='Short synthetic base-network check plus unchanged accepted orientation wrapper. Own full1Mb native recovery, six exact parent output arrays and complete accuracy are still required.') + save(study/'CPU-construction-check.json',proof) + evidence=[parent/'evaluation.json',parent/'valid-score/report.json',parent/'valid-variant-score/report.json', + root/'studies/rc_large114/completion-observation.json',root/'studies/rc_large114/diagnostics/parent/summary.json', + root/'studies/rc_decoder113/completion-observation.json',root/'studies/rc_decoder113/diagnostics/summary.json', + root/'studies/decoder111/CPU-construction-check.json',study/'proposed-recipe.json',study/'CPU-construction-check.json', + here/'prepare_decoder111.py',here/'make_linear_upsample83.py',here/'make_splice_grid96.py',Path(__file__)] + save(study/'preparation.json',dict(utc=datetime.now(timezone.utc).isoformat(),package=str(package),package_verification=verification, + parent_inference_signature=signature,candidate_inference_signature=actual,evidence_sha256={str(p):digest(p) for p in evidence}, + evaluation_admitted=False,controller_created=False,GPU_used=False, + next='Separate resource/admission decision, wait115 full/native/runtime/paired complete then freeze native/full/all13/runtime/paired gates versus RC114; no automatic GPU launch.')) + print(json.dumps(dict(preparation_sha256=digest(study/'preparation.json'),CPU_check_sha256=digest(study/'CPU-construction-check.json'),package=verification,GPU_used=False)),flush=True) + + +if __name__=='__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-CPU-construction-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-CPU-construction-check.json new file mode 100644 index 0000000000000000000000000000000000000000..a9d30ed4d23cd2ca9bb0d4e0953bcc1531f4553b --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-CPU-construction-check.json @@ -0,0 +1,35 @@ +{ + "utc": "2026-10-26T07:37:39.350375+00:00", + "parameters": 215626632, + "actual_large8_weights_sha256": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "candidate_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "synthetic_input_bp": 8192, + "synthetic_target_bp": 4096, + "base_model_batch": 4, + "actual_allele_and_N_edits": true, + "forward_and_reverse_inputs": true, + "all11_base_outputs_finite": true, + "actual_native_track_arrays_exact_manual_fixed_mean": 15, + "actual_native_coordinate_and_strand_roundtrip_exact": true, + "paired_dense_and_splice_arrays_exact_manual_fixed_mean": 8, + "all_learned_tensors_and_CPU_RNG_unchanged": true, + "public_junction_metadata_sha256": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4", + "exact_RC108_wrapper_and_permutation_assets": true, + "CUDA_used": false, + "training_updates": 0, + "labels_read": false, + "caveat": "CPU short-input compatibility of the actual large model and unchanged RC operation. It does not establish full1Mb GPU memory/runtime or accuracy; all candidate native/full/runtime gates remain required." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-admission.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-admission.json new file mode 100644 index 0000000000000000000000000000000000000000..aadd5377fcba533d47a3ff0074b204dea71f675d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-admission.json @@ -0,0 +1,24 @@ +{ + "utc": "2026-10-26T07:40:05.079170+00:00", + "admitted": true, + "training_admitted": false, + "additional_training_updates": 0, + "sole_parent": "/workspace/research/experiments/calib81-large-epoch08", + "parameters": 215626632, + "weights": [ + 0.5, + 0.5 + ], + "resource_review_sha256": "e900e71bbbe7239e766f8a25bea4b18eb9d8f6e59b3de068017841eca7e88f26", + "parent_completion_sha256": "3a684e6d6014cd9c6d3bdb0f8f89865b703df3a5d1279ceb3411626ffb82b666", + "parent_native624_runtime_skipped_nonwinner": true, + "own_native624_runtime_mandatory": true, + "CPU_proof_sha256": "5d0ddbd0993914889b1b2ea5bd83576b5ca0c4459a82e2c0b694ac812e8c999e", + "operator_recovery_context": "/workspace/research/studies/operator_recovery_20261026T064822/recovery.json", + "operator_recovery_sha256": "e3edb6bc383d74cc1f6d8d08c48c4d5265c719a5e86d26c51431539c6e18a0d3", + "research_deadline_utc": "2026-10-28T04:29:08+00:00", + "motivation": "RC108 improves reward and10families against its forward eight-pass small parent, with all13 variant intervals includingzero. At equal eight-pass exposure, rawlarge8 improves missense and worsens enhancer linking versus rawsmall8. Test whether the same fixed orientation averaging helps the large model and report every variant outcome. Neither additivity nor variant rescue is assumed.", + "primary_comparison": "All11/all13 versus the exact calibrated large eight-pass forward parent; isolates the fixed orientation operation on identical weights.", + "secondary_comparison": "All11/all13 versus RC108. Both use eightpasses/seed17/originalschedule/alltrainingintervals, four identical prior offsets and equalF/RC inference. Architecture includes globalcapacity andlocalwidth32-to64 differences; compute and numerical histories differ. One trajectory per size, not independent replication or equalcompute.", + "limitations": "One fixed transfer of equal orientation averaging, not a new trained model or presumed rescue of variant effects. Parent runtime skip is not a pass for this candidate. All own native/full/runtime checks and both serial paired analyses remain required." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-fixture-check.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-fixture-check.json new file mode 100644 index 0000000000000000000000000000000000000000..310874df6776af8f76cf5e49d34d114785c064e9 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-fixture-check.json @@ -0,0 +1,11 @@ +{ + "utc": "2026-10-26T07:40:05.079170+00:00", + "rows": [ + 16, + 96 + ], + "all_three_manifest_hashes_and_rows_exact": true, + "reference_directory_exact": true, + "labels_read": false, + "GPU_used": false +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-paired-plan.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-paired-plan.json new file mode 100644 index 0000000000000000000000000000000000000000..1b97d4216f63aec62c1362157337659bfb497c11 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-paired-plan.json @@ -0,0 +1,142 @@ +{ + "created_utc": "2026-10-26T07:40:05.079170+00:00", + "dependencies": [ + "/workspace/research/experiments/rc_large114/evaluation.json", + "/workspace/research/studies/rc_large114/parent-comparison.json", + "/workspace/research/studies/rc_large114/runtime-gate.json" + ], + "source_sha256": { + "run_planned_diagnostics.py": "9c0fc164753231538d83d502a505958f05874c916ca6060c00e1c471a316013c", + "diagnose_track_uncertainty.py": "1685ed591c99f56780926a27263c7d46391c5a6c5a2824d11a4828e71178922d", + "diagnose_variant_uncertainty_weighted_io.py": "575a3b22b61b23965032dca0e07e56c21b50b41dcac2322efea49c4f7fcd1271", + "diagnose_remaining_variant_uncertainty.py": "8328cd323b62bf318d4d6705eafece60be878cddb26d129cfa6e38a3a30db317", + "weighted_variant_metrics.py": "28fe99e06ea545dcb6e28173e35b59600d9af8a7a11f1c48a11e7477f930e7f6", + "direct_read_store.py": "ab182b3b687a3b470ff1bbc311a9130cc791974366bdc0a400df5e0e28602c9c", + "summarize_paired_diagnostics.py": "8575ce9df387d7c12b6037b35a2bf9ee133e4d99a8e8fff4f50a1c181c09d12d" + }, + "phases": [ + { + "name": "parent-tracks", + "arguments": [ + "/workspace/research/seqfunc/diagnose_track_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json" + }, + { + "name": "parent-dense-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_variant_uncertainty_weighted_io.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json", + "--replicates", + "2000", + "--seed", + "1729", + "--evaluations", + "eqtl_sign", + "eqtl_causality", + "eqtl_coefficient", + "enhancer_gene", + "traitgym_mendelian", + "traitgym_complex" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json" + }, + { + "name": "parent-remaining-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/calib81-large-epoch08", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json" + }, + { + "name": "parent-summary", + "arguments": [ + "/workspace/research/seqfunc/summarize_paired_diagnostics.py", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-tracks.json", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-dense-variants.json", + "/workspace/research/analysis/rc-large114-vs-calib81-large8-remaining-variants.json", + "/workspace/research/studies/rc_large114/diagnostics/parent/summary.json" + ], + "output": "/workspace/research/studies/rc_large114/diagnostics/parent/summary.json" + }, + { + "name": "smallRC108-tracks", + "arguments": [ + "/workspace/research/seqfunc/diagnose_track_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json" + }, + { + "name": "smallRC108-dense-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_variant_uncertainty_weighted_io.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json", + "--replicates", + "2000", + "--seed", + "1729", + "--evaluations", + "eqtl_sign", + "eqtl_causality", + "eqtl_coefficient", + "enhancer_gene", + "traitgym_mendelian", + "traitgym_complex" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json" + }, + { + "name": "smallRC108-remaining-variants", + "arguments": [ + "/workspace/research/seqfunc/diagnose_remaining_variant_uncertainty.py", + "/workspace/research/experiments/rc_large114", + "/workspace/research/experiments/rc108", + "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json", + "--replicates", + "2000", + "--seed", + "1729" + ], + "output": "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json" + }, + { + "name": "smallRC108-summary", + "arguments": [ + "/workspace/research/seqfunc/summarize_paired_diagnostics.py", + "/workspace/research/analysis/rc-large114-vs-smallRC108-tracks.json", + "/workspace/research/analysis/rc-large114-vs-smallRC108-dense-variants.json", + "/workspace/research/analysis/rc-large114-vs-smallRC108-remaining-variants.json", + "/workspace/research/studies/rc_large114/diagnostics/smallRC108/summary.json" + ], + "output": "/workspace/research/studies/rc_large114/diagnostics/smallRC108/summary.json" + } + ], + "method": "2000 paired clustered draws,seed1729; original-point and first32-replicate verifier parity required by existing helpers. No fitting, label-derived inference assets or GPU work.", + "scope": "Serial original four-phase all11/all13 analyses against two fixed controls,2000/1729. Shared candidate moments collected once; no second concurrent observer.", + "caveat": "Exploratory postselection on repeatedly inspected development data. Resampling schemes are not independent replications; zero-inclusive intervals are not equivalence. No variant-label fitting, pretrained external parameters, test data, new sequence flank or inference cache. One GPU and unchanged existing trainers. All11/all13 versus RC108. Both use eightpasses/seed17/originalschedule/alltrainingintervals, four identical prior offsets and equalF/RC inference. Architecture includes globalcapacity andlocalwidth32-to64 differences; compute and numerical histories differ. One trajectory per size, not independent replication or equalcompute." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-proposed-recipe.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-proposed-recipe.json new file mode 100644 index 0000000000000000000000000000000000000000..f607d8fd875571d0970ed2260059a37ce6ac1497 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-proposed-recipe.json @@ -0,0 +1,64 @@ +{ + "schema": "seqfunc-large-eight-pass-fixed-RC114-proposal/v1", + "created_utc": "2026-10-26T07:33:19.178799+00:00", + "name": "Fixed equal forward/reverse-complement inference on the eight-pass large model", + "parent": "/workspace/research/experiments/calib81-large-epoch08", + "primary_control": "/workspace/research/experiments/calib81-large-epoch08", + "secondary_control": "/workspace/research/experiments/rc108", + "parameters": 215626632, + "additional_training_updates": 0, + "inherited_complete_passes": 8, + "inherited_examples": 328160, + "inherited_optimizer_updates": 82040, + "seed": 17, + "orientation_weights": [ + 0.5, + 0.5 + ], + "change": "Apply the exact unchanged RC17/85/108 inference assets to sole calibrated81-large8. Every family and both entrypoints use the same two deterministic orientations; retain learned weights, four training-derived prior offsets, nearest global-feature decoder and original per-view sparse selection. No interpolation, SWA, joint-position or motif-mask composition.", + "architecture": "Global width1024/16transformerblocks, localwidth64/sixlocalblocks, local-to-trunk bridge, residualRMS cap0.1; same1,048,576bp input and all11families.", + "motivation": "RC108 improves reward and10families against its forward eight-pass small parent, with all13 variant intervals includingzero. At equal eight-pass exposure, rawlarge8 improves missense and worsens enhancer linking versus rawsmall8. Test whether the same fixed orientation averaging helps the large model and report every variant outcome. Neither additivity nor variant rescue is assumed.", + "primary_comparison": "All11/all13 versus the exact calibrated large eight-pass forward parent; isolates the fixed orientation operation on identical weights.", + "secondary_comparison": "All11/all13 versus RC108. Both use eightpasses/seed17/originalschedule/alltrainingintervals, four identical prior offsets and equalF/RC inference. Architecture includes globalcapacity andlocalwidth32-to64 differences; compute and numerical histories differ. One trajectory per size, not independent replication or equalcompute.", + "required_before_execution": [ + "Complete parent200track/all2697variant reports and parent own52/36derivation checks", + "Complete original parent fixed-link paired checks and exact unaffected family/variant reports", + "Accept either parent measured runtime pass or its explicit nonwinner runtime skip; this cannot stand in for the new candidate runtime check", + "CPU actual-large-weight finite/compatibility checks, exact decoder asset and parent-weight fingerprints, replayable package and correct README", + "Fresh one-H100 memory/time/CPU resource review including large14 andseed110, current hashed operator recovery deadline and evaluation reserve" + ], + "required_candidate_gates": [ + "Own52native/offline/partial-recovery arrays", + "Full200trackwindows/2697developmentqueries/all11/all13 and same-row publishedanchors", + "Own624native arrays on the existing fixed16window/96query fixture", + "Allthree sampled combined1000window/6587query runtime estimates<=14400s", + "One serial eight-phase all11/all13 paired observer against both fixed controls,2000replicates/seed1729", + "Atomic monotonic publication only after complete score/native/runtime gates; retain strongest existing package otherwise" + ], + "training_admitted": false, + "GPU_evaluation_admitted": false, + "controller_launched": false, + "parent_report_status_at_proposal": { + "evaluation.json": { + "exists": true, + "sha256": "bd0f0a864278e1086930bed08cf19b1ae19b61b68188d1cf2bc3d289d44a9618" + }, + "valid-score/report.json": { + "exists": true, + "sha256": "8cdb4e7ad48dc7b2c4c06622b1aa170557164d5777cf81b7481ce06f7bc3cf0c" + }, + "valid-variant-score/report.json": { + "exists": true, + "sha256": "fbcc4d8eca03106bb29703effdf2d5278f3a2f513341be2b2aab3ff41cff0965" + } + }, + "evidence_sha256": { + "/workspace/research/studies/rc108/completion-observation.json": "457ae0d8c35be775f898a3f19ef2d97df407290d423f57fb2a625bbe5f086490", + "/workspace/research/studies/rc108/diagnostics/summary.json": "0c754eac92f646930be8c29c6c4dc170713fbc1f17fcab10bf57c74b048a2486", + "/workspace/research/studies/scale13/epoch08-learning-report/report.json": "c13f184b573712c2dad47bca6366c353a46455e93ada6e0c6806c34949baf580", + "/workspace/research/studies/linear_training94/completion-observation.json": "6e47294a7710fad979349792d4c144a44c0eb75386e3e5a78b1ef9bd67e8924b", + "/workspace/research/seqfunc/make_rc_candidate.py": "522b03f803136ca230cea9675b902d57fbc88718dad1bfdf9ec0b88568b4e637", + "/workspace/research/seqfunc/run_fixed_rc_with_runtime.py": "5c123ec89d4c0bd354170b17077533cd87e42e09ee703d812053b1645001c7e9" + }, + "limitations": "Exploratory postselection on repeatedly inspected development data. Resampling schemes are not independent replications; zero-inclusive intervals are not equivalence. No variant-label fitting, pretrained external parameters, test data, new sequence flank or inference cache. One GPU and unchanged existing trainers." +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-study.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-study.json new file mode 100644 index 0000000000000000000000000000000000000000..0516002875d1eb1268cf08d98ed9359ffe51efb1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-study.json @@ -0,0 +1,419 @@ +{ + "schema": "seqfunc-fixed-large-RC114/v1", + "created_utc": "2026-10-26T07:40:05.079170+00:00", + "name": "SeqFunc large eight-pass model with fixed forward/reverse-complement inference", + "parent": "/workspace/research/experiments/calib81-large-epoch08", + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "output": "/workspace/research/experiments/rc_large114", + "weights": [ + 0.5, + 0.5 + ], + "parameters": 215626632, + "additional_training_updates": 0, + "evaluation_lock": "/workspace/research/studies/scale13/evaluation.lock", + "minimum_free_GiB": 24, + "native_timeout_seconds": 1800, + "helper_sha256": { + "run_fixed_rc_with_runtime.py": "5c123ec89d4c0bd354170b17077533cd87e42e09ee703d812053b1645001c7e9", + "make_rc_candidate.py": "522b03f803136ca230cea9675b902d57fbc88718dad1bfdf9ec0b88568b4e637", + "check_native_recovery.py": "47adc6fd7695f8d7e5e05fb9450b41194ce3695ba369dda07dbbd43188ff6965", + "evaluate_package.py": "38885699310241dc84b4d412f14abf240d9708bb8b2de97bc365754df65c5a50", + "resume_tracks.py": "e539554f7962456d0b06c8d6a1c72b279639a0acd393a015147b7b555e747e3a", + "resume_perturbations.py": "a4c87b25563e08eae1c8486338a45ed5b7d7e3f7930fd7165b44d623bce1d174", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4", + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "promote_if_better.py": "fb863e77b3da094f24de62dcfd08ed8e384b253ee884759babddbab2cb5f283d", + "finalize_package.py": "d8c40bdf02e44e8ad33a22ca93b718fa2f1f53a0e5f51cb228287446d15bbb54", + "publish_atomic.py": "34f4f44e21e9130070d1818eeddffdf9210efcc2e034b5be7b41011fe375f289", + "freeze_rc_large114_execution.py": "1da2453683f36685296f4653af5e6bb1042f051a27c02b33b200d664577b2566", + "prepare_rc_large114.py": "b942dde822add66bdac0686411cc54c2cbdbd31315a1bd62b8e7c6bb1127f608" + }, + "evidence_sha256": { + "/workspace/research/studies/rc_large114/proposed-recipe.json": "6d068b2dfd9f8cafbd42cb23ec05505f06cfb9c08aed677038b9d7e6853f82ca", + "/workspace/research/studies/rc_large114/preparation.json": "ba8467419e37b02d8a78de0c6ee954456480eec983c65eec8dc4a9b6827e3397", + "/workspace/research/studies/rc_large114/CPU-construction-check.json": "5d0ddbd0993914889b1b2ea5bd83576b5ca0c4459a82e2c0b694ac812e8c999e", + "/workspace/research/studies/rc_large114/admission.json": "d48ea57465a7578e1f16b380ca3d2221024a957eb61368fbd089bb1b281cc986", + "/workspace/research/studies/rc_large114/resource-review.json": "e900e71bbbe7239e766f8a25bea4b18eb9d8f6e59b3de068017841eca7e88f26", + "/workspace/research/studies/rc_large114/fixture-check.json": "12b06b35e555e3b78f9bee40acf9ae887714832dc52428ea29cddf7871784691", + "/workspace/research/studies/rc_large114/diagnostics/plan.json": "81c750c4e1302d17a897d3474ee6b62c2274a65f89cf1fa5376510ea387c6332", + "/workspace/research/studies/late_fixed_links81/large-epoch08/complete.json": "bfcdf9356df76c541ba6d9bbdff5e9eab3e93b8c1b78e7a29b9dc8efb6c387e5", + "/workspace/research/studies/late_fixed_links81/large-epoch08/completion-observation.json": "3a684e6d6014cd9c6d3bdb0f8f89865b703df3a5d1279ceb3411626ffb82b666", + "/workspace/research/studies/late_fixed_links81/large-epoch08/publication-decision.json": "d5e70a46d08036e3496c3f435994d9941b6a8afa95dec2bb9bc923f16e1caf43", + "/workspace/research/studies/late_fixed_links81/large-epoch08/matched-comparison.json": "2799b39310336823871a0460f8d43a5f35b5319fdd97f0f1cd42e0be2e3a99ee", + "/workspace/research/studies/late_fixed_links81/large-epoch08/native-recovery/report.json": "c6b6352ef628bc372d5356d5950fa66ecc37c1da8bf9f0fcf397920012881f29", + "/workspace/research/studies/late_fixed_links81/large-epoch08/paired-tracks.json": "1f4f91699c3bd1c7c959f033446a509f124edd19b9cf9b5eb2f662f660fd1cbf", + "/workspace/research/studies/late_fixed_links81/large-epoch08/paired-remaining-variants.json": "53e0e0cbd2d62fecbdee033ddae9569555f9390644c8f64349db22fc79a90267", + "/workspace/research/studies/late_fixed_links81/large-epoch08/paired-tracks-complete.json": "7604809acb2cf354ec65690a65c8d1e461078f5906a57e8b803172da55137ee5", + "/workspace/research/studies/late_fixed_links81/large-epoch08/paired-remaining-variants-complete.json": "45718e6884f8b31ab92d90f22f3d60b4917b2da5d98aee6b41d8e90cf19d24da", + "/workspace/research/experiments/calib81-large-epoch08/evaluation.json": "bd0f0a864278e1086930bed08cf19b1ae19b61b68188d1cf2bc3d289d44a9618", + "/workspace/research/experiments/calib81-large-epoch08/valid-score/report.json": "8cdb4e7ad48dc7b2c4c06622b1aa170557164d5777cf81b7481ce06f7bc3cf0c", + "/workspace/research/experiments/calib81-large-epoch08/valid-variant-score/report.json": "fbcc4d8eca03106bb29703effdf2d5278f3a2f513341be2b2aab3ff41cff0965", + "/workspace/research/studies/rc108/completion-observation.json": "457ae0d8c35be775f898a3f19ef2d97df407290d423f57fb2a625bbe5f086490", + "/workspace/research/studies/rc108/diagnostics/summary.json": "0c754eac92f646930be8c29c6c4dc170713fbc1f17fcab10bf57c74b048a2486", + "/workspace/research/studies/operator_recovery_20261026T064822/recovery.json": "e3edb6bc383d74cc1f6d8d08c48c4d5265c719a5e86d26c51431539c6e18a0d3" + }, + "native_latency_template": { + "created_utc": "2026-10-26T07:40:05.079170+00:00", + "original_public_input": "/workspace/research/public-valid", + "track_indices": [ + 11, + 14, + 39, + 43, + 46, + 53, + 73, + 74, + 92, + 105, + 117, + 128, + 133, + 167, + 170, + 187 + ], + "query_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566, + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495, + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727, + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901, + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093, + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ], + "query_strata": [ + { + "stratum": { + "families": [ + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone" + ], + "edit": "snv", + "dense_bins": 17, + "splice_bases": 0 + }, + "development_count": 591, + "selected_indices": [ + 2111, + 2159, + 2163, + 2205, + 2334, + 2466, + 2469, + 2485, + 2511, + 2517, + 2526, + 2573, + 2617, + 2660, + 2667, + 2694 + ] + }, + { + "stratum": { + "families": [ + "chip_tf" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 160, + "selected_indices": [ + 1747, + 1805, + 1816, + 1820, + 1822, + 1835, + 1842, + 1851, + 1857, + 1861, + 1863, + 1868, + 1891, + 1892, + 1896, + 1901 + ] + }, + { + "stratum": { + "families": [ + "dnase" + ], + "edit": "snv", + "dense_bins": 4, + "splice_bases": 0 + }, + "development_count": 198, + "selected_indices": [ + 1575, + 1576, + 1584, + 1592, + 1602, + 1618, + 1622, + 1642, + 1648, + 1660, + 1665, + 1667, + 1680, + 1691, + 1713, + 1727 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "mask", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 200, + "selected_indices": [ + 1918, + 1920, + 1937, + 1950, + 1961, + 1978, + 1989, + 2000, + 2013, + 2018, + 2020, + 2031, + 2038, + 2054, + 2067, + 2093 + ] + }, + { + "stratum": { + "families": [ + "rna_seq" + ], + "edit": "snv", + "dense_bins": 1536, + "splice_bases": 0 + }, + "development_count": 567, + "selected_indices": [ + 68, + 101, + 121, + 163, + 179, + 264, + 269, + 437, + 455, + 462, + 476, + 478, + 520, + 522, + 558, + 566 + ] + }, + { + "stratum": { + "families": [ + "splice_sites" + ], + "edit": "snv", + "dense_bins": 0, + "splice_bases": 5001 + }, + "development_count": 981, + "selected_indices": [ + 586, + 599, + 753, + 841, + 854, + 863, + 1048, + 1053, + 1149, + 1191, + 1213, + 1215, + 1312, + 1320, + 1429, + 1495 + ] + } + ], + "selection_seed": 1729, + "helper_sha256": { + "measure_native_latency.py": "dd8ce9249d29a3814603d49cb3d3820d2a3a103e8f15f2266597a40e1e4bf363", + "verify_package.py": "23042e356bddbaa59f2abb1e094196b75154b7a59033284295c02885bb4991d4" + }, + "public_inputs_sha256": { + "public/test.parquet": "2d5242a0e61f3a99ccbcdcd0ce7d4f8c8ff0c9f7d0db00a00d2abe3a26ab08c8", + "public/perturbations.parquet": "0cb5ce07ffc6acf55d411acb8389f1dc59257ac89e5ce3af0bf0ba05d42478d8", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "original_manifests_sha256": { + "public/test.parquet": "a8f9cc119e8e6eeca63fe768f5b1b6a5b623942930ca654e4c0ba89b0bfc83f6", + "public/perturbations.parquet": "a95b37553157f1964d134b64eb30f149d8577f20d6d6e6e834c8b30fe7022aed", + "public/junction_tracks.parquet": "8b2f8888ae06779b487e086c5c49234e90452d03966debe0bea9767a555a00d4" + }, + "bounded_diagnostic_lock": "/workspace/research/studies/junction_discovery39/bounded-diagnostic.lock", + "minimum_free_GiB": 24, + "timeout_seconds_per_entrypoint": 1200, + "published_test_counts": { + "tracks": 1000, + "perturbations": 6587, + "source": "contract.md and data/README.md only; no test manifest or labels read" + }, + "memory_evidence": "215,626,632-parameter model with two sequential orientations;24GiB free-memory scheduling guard is not a measured peak or bound. Own native replay and timing required.", + "scope": "RC-large114 own16validationwindows/96developmentqueries/624nativearrays after ownfull200/2697/all13. Allthree4h estimates mandatory even though parent runtime was skipped.", + "limitations": [ + "Shared GPU, CPU and storage load; not isolated inference and not a guaranteed upper bound.", + "Random16 validation windows and16 queries in each of6 public request/edit/span strata; no test-manifest query mix or test targets accessed.", + "Startup plus first-example and post-last-example cleanup measured together; later rates include native model, IO and compression. No cost subtracted as free setup.", + "Extrapolations use documented1000/6587 counts, observed validation mixture, slowest sampled stratum mean and pooled95th percentile separately; none proves unseen-query latency.", + "File caches and concurrent load can differ across runs. Native accuracy replay remains separate from the runtime estimate.", + "Require allthree estimates <=14400s; shared-load estimates are not guaranteed upperbounds." + ], + "input_preparation_amendment": { + "created_utc": "2026-10-25T06:45:46.228017+00:00", + "prior_plan_sha256": "809e68426bb2b6e6e341d68387f818adc19d37792fafcc4f56d970f8ef3cac18", + "change": "Include unchanged public junction-track metadata required by the original dataset reader. No query, sequence, model, timing design or target/label change. Waiting controller stopped before any native execution." + } + }, + "only_change": "Apply the exact unchanged RC17/85/108 inference assets to sole calibrated81-large8. Every family and both entrypoints use the same two deterministic orientations; retain learned weights, four training-derived prior offsets, nearest global-feature decoder and original per-view sparse selection. No interpolation, SWA, joint-position or motif-mask composition.", + "motivation": "RC108 improves reward and10families against its forward eight-pass small parent, with all13 variant intervals includingzero. At equal eight-pass exposure, rawlarge8 improves missense and worsens enhancer linking versus rawsmall8. Test whether the same fixed orientation averaging helps the large model and report every variant outcome. Neither additivity nor variant rescue is assumed.", + "mandatory_checks": [ + "strict exactparentdecoder andRCassets unchanged", + "fresh52native/recoveryarrays", + "full200tracks/2697perturbations/13officialmetrics", + "same-row anchors/eligibility identical toparent", + "fresh16window96querynative replay and allthree runtimeestimates<=14400seconds", + "complete reward improvement before atomicpublication" + ], + "recovery": "Check controller.lock and processes before restart. Saved native/evaluation/timing markers must match signatures. Preserve partial output for normalresume; if packagebuild was interrupted, inspect and archive the incomplete package before restarting. No new training or model resume checkpoint.", + "caveat": "Exploratory postselection on repeatedly inspected development data. Resampling schemes are not independent replications; zero-inclusive intervals are not equivalence. No variant-label fitting, pretrained external parameters, test data, new sequence flank or inference cache. One GPU and unchanged existing trainers.", + "paired_plan_sha256": "81c750c4e1302d17a897d3474ee6b62c2274a65f89cf1fa5376510ea387c6332" +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-training-replay.md b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-training-replay.md new file mode 100644 index 0000000000000000000000000000000000000000..4a89eafb88068244b603723389ed9736a4fcfeb1 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/rc-large114-training-replay.md @@ -0,0 +1,4 @@ +All eight inherited training passes used the unchanged forward model source and original +B4 seed17 fourteen-pass schedule. This candidate changes only inference wrappers via +make_rc_candidate.py, retaining all model tensors, four training-derived splice offsets +and vendored training assets. It adds no optimizer updates or learned parameters. diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/reader-before-parallel-io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/reader-before-parallel-io.py new file mode 100644 index 0000000000000000000000000000000000000000..71cd39d15713f648e7d7e31aa1923fa4fa4800dd --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/reader-before-parallel-io.py @@ -0,0 +1,229 @@ +"""Streaming reader for the compact 1 Mb / eleven-family release.""" + +from __future__ import annotations + +from collections import OrderedDict +from pathlib import Path + +import numpy as np +import pandas as pd +import pysam +import torch +from torch.utils.data import Dataset +import zarr + + +def _open_part(path: Path): + if str(path).endswith(".zarr.zip"): + store = zarr.ZipStore(str(path), mode="r") + try: + return zarr.open_group(store=store, mode="r"), store + except BaseException: + store.close() + raise + return zarr.open_group(str(path), mode="r"), None + + +DENSE_FAMILIES = ( + "atac", + "dnase", + "procap", + "cage", + "rna_seq", + "chip_tf", + "chip_histone", +) +def normalize_splice_positions(value) -> np.ndarray: + """Normalize Arrow's nested-list representation to the public [4, 512] array.""" + + outer = np.asarray(value, dtype=object).reshape(-1).tolist() + positions = np.stack([np.asarray(item, dtype=np.int64) for item in outer]) + if positions.shape != (4, 512): + raise ValueError(f"splice_site_positions must have shape (4, 512), got {positions.shape}") + return positions + + +class AlphaGenomeDataset(Dataset): + """Resolve frozen coordinates to reference DNA and one Zarr example group.""" + + def __init__( + self, + data_root: str | Path, + split: str, + *, + max_open_parts: int = 8, + include_targets: bool = True, + ) -> None: + self.root = Path(data_root) + self.split = split + manifest = self.root / "public" / f"{split}.parquet" + if not manifest.is_file(): + raise FileNotFoundError(f"Missing materialized manifest: {manifest}") + self.rows = pd.read_parquet(manifest).reset_index(drop=True) + junction_tracks = pd.read_parquet(self.root / "public/junction_tracks.parquet") + junction_tracks = junction_tracks.sort_values("tensor_index") + self.junction_track_halves = torch.from_numpy( + junction_tracks.junction_channel_half.to_numpy(np.int64) + ) + if include_targets: + labels_path = self.root / "labels" / f"{split}.parquet" + if not labels_path.is_file(): + raise FileNotFoundError(f"Missing trusted labels manifest: {labels_path}") + labels = pd.read_parquet(labels_path).reset_index(drop=True) + if self.rows.example_id.astype(str).tolist() != labels.example_id.astype(str).tolist(): + raise ValueError("public and labels manifests have different IDs/order") + self.rows = self.rows.merge(labels, on="example_id", validate="one_to_one", sort=False) + self.fasta_path = self.root / "reference" / "GRCh38.p13.genome.fa" + self.max_open_parts = max_open_parts + self.include_targets = include_targets + self._fasta = None + self._parts: OrderedDict[str, tuple[zarr.Group, object | None]] = OrderedDict() + + def __len__(self) -> int: + return len(self.rows) + + def _reference(self): + if self._fasta is None: + self._fasta = pysam.FastaFile(str(self.fasta_path)) + return self._fasta + + def _part(self, relative: str): + if relative in self._parts: + opened = self._parts.pop(relative) + self._parts[relative] = opened + return opened[0] + group, store = _open_part(self.root / relative) + if len(self._parts) >= self.max_open_parts: + _, (_, old_store) = self._parts.popitem(last=False) + if old_store is not None: + old_store.close() + self._parts[relative] = (group, store) + return group + + @staticmethod + def _one_hot(sequence: str) -> np.ndarray: + encoded = np.frombuffer(sequence.upper().encode("ascii"), dtype=np.uint8) + output = np.zeros((len(encoded), 4), dtype=np.float32) + for index, letter in enumerate(b"ACGT"): + output[:, index] = encoded == letter + return output + + def _sequence(self, row) -> tuple[np.ndarray, np.ndarray]: + reference = self._reference() + chromosome_length = reference.get_reference_length(str(row.chromosome)) + start, end = int(row.input_start), int(row.input_end) + clipped_start, clipped_end = max(0, start), min(chromosome_length, end) + sequence = reference.fetch(str(row.chromosome), clipped_start, clipped_end) + left, right = clipped_start - start, end - clipped_end + if left or right: + sequence = "N" * left + sequence + "N" * right + valid = np.ones(end - start, dtype=bool) + if left: + valid[:left] = False + if right: + valid[-right:] = False + return self._one_hot(sequence), valid + + @staticmethod + def _decode(example, name: str, *, asinh: bool) -> np.ndarray: + """Train targets may be stored as 8-bit codes with per-track ranges; they are decoded + here, one example at a time, so the release stays compressed on disk.""" + array = example[name] + if array.dtype != np.uint8: + return np.asarray(array, dtype=np.float32) + lo = np.asarray(example[f"{name}_lo"], dtype=np.float32) + hi = np.asarray(example[f"{name}_hi"], dtype=np.float32) + x = lo + np.asarray(array, dtype=np.float32) * ((hi - lo) / 255.0) + return np.sinh(x) if asinh else x + + @staticmethod + def _dense_target(example, family: str) -> dict[str, torch.Tensor]: + values = AlphaGenomeDataset._decode(example, f"{family}/values", asinh=True) + mask = np.asarray(example[f"{family}/mask"], dtype=bool) + if mask.ndim == 1: + mask = np.broadcast_to(mask[None, :], values.shape).copy() + return { + "values": torch.from_numpy(values), + "mask": torch.from_numpy(mask), + } + + def __getitem__(self, item: int): + row = self.rows.iloc[item] + dna, input_mask = self._sequence(row) + result = { + "example_id": str(row.example_id), + "dna": torch.from_numpy(dna), + "input_mask": torch.from_numpy(input_mask), + "junction_track_halves": self.junction_track_halves, + } + if not self.include_targets: + return result + + part = self._part(str(row.target_shard)) + example = part[str(row.target_group)] + result["dense"] = { + family: self._dense_target(example, family) for family in DENSE_FAMILIES + } + result["splice_sites"] = torch.from_numpy( + np.asarray(example["splice_sites/values"], dtype=np.float32) + ) + result["splice_usage"] = { + "positions": torch.from_numpy( + np.asarray(example["splice_site_usage/positions"], dtype=np.int64) + ), + "values": torch.from_numpy( + np.asarray(example["splice_site_usage/values"], dtype=np.float32) + ), + "mask": torch.from_numpy( + np.asarray(example["splice_site_usage/mask"], dtype=bool) + ), + } + result["usage_positions"] = result["splice_usage"]["positions"] + result["junctions"] = { + name: torch.from_numpy( + np.asarray( + example[f"splice_junctions/{name}"], + dtype=np.float32 if name == "value" else np.int64, + ) + ) + for name in ("row", "column", "track", "value") + } + result["junctions"]["site_positions"] = torch.from_numpy( + np.asarray(example["splice_junctions/site_positions"], dtype=np.int64) + ) + site_positions = normalize_splice_positions(row.splice_site_positions) + result["site_positions"] = torch.from_numpy(site_positions) + if not torch.equal(result["junctions"]["site_positions"], result["site_positions"]): + raise ValueError(f"public/private splice positions differ for {row.example_id}") + scoring_path = self.root / "scoring" / self.split / f"part-{item:06d}.parquet" + scoring = pd.read_parquet(scoring_path) + if not (scoring.example_id.astype(str) == str(row.example_id)).all(): + raise ValueError(f"scoring part ID differs for {row.example_id}") + result["junction_scoring"] = { + "donor_position": torch.from_numpy( + scoring.donor_position.to_numpy(np.int64) + ), + "acceptor_position": torch.from_numpy( + scoring.acceptor_position.to_numpy(np.int64) + ), + "track": torch.from_numpy(scoring.track.to_numpy(np.int64)), + "target_value": torch.from_numpy( + scoring.target_value.to_numpy(np.float32) + ), + "valid": torch.from_numpy(scoring.valid.to_numpy(bool)), + } + result["contact_maps"] = torch.from_numpy( + self._decode(example, "contact_maps/upper_triangle", asinh=False) + ) + result["contact_map_mask"] = torch.from_numpy( + np.asarray(example["contact_maps/mask"], dtype=bool) + ) + return result + + +def single_example_collate(items): + """Keep variable-size event tensors intact; the reference currently trains at batch one.""" + + if len(items) != 1: + raise ValueError("The compact event loader currently requires batch_size=1") + return items[0] diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_decoder111.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_decoder111.py new file mode 100644 index 0000000000000000000000000000000000000000..f748f5a9cefa61d184223ef8899b1d2935b9d936 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_decoder111.py @@ -0,0 +1,170 @@ +"""Evaluate the fixed grid96 decoder transfer on eight-pass weights with complete gates.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from verify_package import inference_signature, verify_package + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def run(plan_path): + plan = json.loads(plan_path.read_text()) + study, here = plan_path.parent, Path(__file__).resolve().parent + for name, expected in plan['helper_sha256'].items(): + assert digest(here/name) == expected, name + for path, expected in plan['evidence_sha256'].items(): + assert digest(path) == expected, path + assert plan['additional_training_updates'] == 0 + parent, output = Path(plan['parent']), Path(plan['output']) + output.mkdir(exist_ok=True) + package = output/'package' + assert inference_signature(parent/'package') == plan['parent_inference_signature'] + parent_eval = json.loads((parent/'evaluation.json').read_text()) + assert parent_eval['package_inference_signature'] == plan['parent_inference_signature'] + parent_tracks = json.loads((parent/'valid-score/report.json').read_text()) + parent_variants = json.loads((parent/'valid-variant-score/report.json').read_text()) + assert parent_tracks['examples'] == 200 and parent_tracks['invalid_reason'] is None + assert len(parent_tracks['families']) == 11 and len(parent_variants) == 13 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', + MKL_NUM_THREADS='2', PYTHONDONTWRITEBYTECODE='1', PYTHONNOUSERSITE='1') + for key in ['PYTHONPATH', 'PYTHONHOME', 'SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def interrupted(signum, frame): + raise InterruptedError(f'Decoder111 supervisor received signal {signum}') + signal.signal(signal.SIGTERM, interrupted) + signal.signal(signal.SIGINT, interrupted) + + def status(phase, **extra): + row = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), **extra) + temporary = study/'state.tmp' + temporary.write_text(json.dumps(row, indent=2)+'\n') + temporary.replace(study/'state.json') + print(json.dumps(row), flush=True) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study/(phase+'.log')).open('a') as log: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=log, stderr=subprocess.STDOUT, start_new_session=True) + try: + assert process.wait(timeout=timeout) == 0, phase + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try: + process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL) + process.wait() + + assert package.is_dir(), 'Accepted CPU package must exist; no rebuild in this runner' + verify_package(package) + assert digest(package/'last.pt') == digest(parent/'package/last.pt') + construction = json.loads((study/'CPU-construction-check.json').read_text()) + assert construction['all_learned_tensors_exact'] and construction['source_exact_completed96'] + assert construction['classifier_contacts_trunk_and_selected_positions_exact_parent'] + assert construction['full_central_and_sparse_features_exact'] and not construction['CUDA_used'] + assert construction['candidate_inference_signature'] == plan['candidate_inference_signature'] + assert digest(package/'reference/alphagenome_50m/model.py') == plan['candidate_model_sha256'] + if (study/'state.json').exists() and json.loads((study/'state.json').read_text()).get('phase') in ['complete','complete_accuracy_runtime_gate_failed']: + print('Completed decoder111 pipeline preserved; no work repeated.', flush=True) + return + signature = inference_signature(package) + native = study/'native-recovery' + with Path(plan['evaluation_lock']).open('a') as lock: + status('waiting_for_native_and_full_evaluation_slot') + if not ((native/'report.json').exists() and (output/'evaluation.json').exists()): + fcntl.flock(lock, fcntl.LOCK_EX) + while int(subprocess.check_output(['nvidia-smi', '--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB']*1024: + time.sleep(5) + if not (native/'report.json').exists(): + execute('native', [here/'check_native_recovery.py', package, native], + timeout=plan['native_timeout_seconds']) + proof = json.loads((native/'report.json').read_text()) + assert proof['package_inference_signature'] == signature + assert proof['all_arrays_bit_identical'] and proof['native_vendored_entrypoints_passed'] + assert proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks': 30, 'perturbations': 22} + if not (study/'native-parent-check.json').exists(): + execute('native-parent-check', [here/'check_decoder111_parent.py', plan_path]) + parent_check = json.loads((study/'native-parent-check.json').read_text()) + assert parent_check['plan_sha256'] == digest(plan_path) and parent_check['unchanged_native_arrays_exact'] == 6 + if not (output/'evaluation.json').exists(): + execute('evaluate', [here/'evaluate_package.py', package, output, + '--resume-partial-tracks', '--resume-partial-perturbations']) + assert json.loads((output/'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output/'valid-score/report.json').read_text()) + variants = json.loads((output/'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + assert all(math.isfinite(v['model']) for v in variants.values()) + for name, value in variants.items(): + for key in ['rows', 'perturbations', 'metric', 'anchors']: + assert value[key] == parent_variants[name][key], (name, key) + for name in plan['expected_exact_families']: + assert tracks['families'][name] == parent_tracks['families'][name], name + for name in plan['expected_exact_variants']: + assert variants[name] == parent_variants[name], name + comparison = dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward_delta=tracks['overall_mean_family_pearson']-parent_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-parent_tracks['families'][k]['mean_pearson'] + for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-parent_variants[k]['model'] for k,v in variants.items()}) + if (study/'parent-comparison.json').exists(): + saved_comparison = json.loads((study/'parent-comparison.json').read_text()) + assert {k:v for k,v in saved_comparison.items() if k!='created_utc'} == {k:v for k,v in comparison.items() if k!='created_utc'} + comparison = saved_comparison + else: + (study/'parent-comparison.json').write_text(json.dumps(comparison, indent=2)+'\n') + latency = study/'native-latency' + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency/'plan.json').exists(): + assert json.loads((latency/'plan.json').read_text()) == timing_plan + else: + (latency/'plan.json').write_text(json.dumps(timing_plan, indent=2)+'\n') + if not (latency/'report.json').exists(): + execute('native-latency', [here/'measure_native_latency.py', latency/'plan.json']) + timing = json.loads((latency/'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + keys = ['development_mixture_seconds', 'slowest_observed_stratum_mean_seconds', 'pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in keys) + passed = max(estimates[k] for k in keys) <= 14400 + gate = dict(passed=passed, estimates=estimates, report_sha256=digest(latency/'report.json'), + scope='All three predeclared shared-load sample estimates must fit4h; none is a guaranteed upper bound.') + if (study/'runtime-gate.json').exists(): + assert json.loads((study/'runtime-gate.json').read_text()) == gate + else: + (study/'runtime-gate.json').write_text(json.dumps(gate, indent=2)+'\n') + if passed and not (study/'publication-check.json').exists(): + execute('publish', [here/'promote_if_better.py', package, output]) + (study/'publication-check.json').write_text(json.dumps(dict( + utc=datetime.now(timezone.utc).isoformat(), reward=tracks['overall_mean_family_pearson'], + package_inference_signature=signature, atomic_monotonic_promoter_completed=True), indent=2)+'\n') + status('complete' if passed else 'complete_accuracy_runtime_gate_failed', + runtime_passed=passed, reward=tracks['overall_mean_family_pearson'], **comparison) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan', type=Path) + args = parser.parse_args() + with (args.plan.parent/'controller.lock').open('a') as controller: + fcntl.flock(controller, fcntl.LOCK_EX | fcntl.LOCK_NB) + run(args.plan) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_fixed_rc_with_runtime.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_fixed_rc_with_runtime.py new file mode 100644 index 0000000000000000000000000000000000000000..1e3d2fcdeacf36475b6d78ef2f7cd04e2de4d654 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_fixed_rc_with_runtime.py @@ -0,0 +1,142 @@ +"""Evaluate one frozen RC parent with complete replay, accuracy and runtime gates.""" +import argparse +from datetime import datetime, timezone +import fcntl +import hashlib +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from verify_package import inference_signature, verify_package + + +def digest(path): + with Path(path).open('rb') as stream: + return hashlib.file_digest(stream, 'sha256').hexdigest() + + +def run(plan_path): + plan = json.loads(plan_path.read_text()) + study, here = plan_path.parent, Path(__file__).resolve().parent + for name, expected in plan['helper_sha256'].items(): + assert digest(here/name) == expected, name + for path, expected in plan['evidence_sha256'].items(): + assert digest(path) == expected, path + assert plan['weights'] == [.5, .5] and plan['additional_training_updates'] == 0 + parent, output = Path(plan['parent']), Path(plan['output']) + output.mkdir(exist_ok=True) + package = output/'package' + assert inference_signature(parent/'package') == plan['parent_inference_signature'] + parent_eval = json.loads((parent/'evaluation.json').read_text()) + assert parent_eval['package_inference_signature'] == plan['parent_inference_signature'] + parent_tracks = json.loads((parent/'valid-score/report.json').read_text()) + parent_variants = json.loads((parent/'valid-variant-score/report.json').read_text()) + assert parent_tracks['examples'] == 200 and parent_tracks['invalid_reason'] is None + assert len(parent_tracks['families']) == 11 and len(parent_variants) == 13 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', + MKL_NUM_THREADS='2', PYTHONDONTWRITEBYTECODE='1', PYTHONNOUSERSITE='1') + for key in ['PYTHONPATH', 'PYTHONHOME', 'SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def status(phase, **extra): + row = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), **extra) + temporary = study/'state.tmp' + temporary.write_text(json.dumps(row, indent=2)+'\n') + temporary.replace(study/'state.json') + print(json.dumps(row), flush=True) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study/(phase+'.log')).open('a') as log: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=log, stderr=subprocess.STDOUT, start_new_session=True) + try: + assert process.wait(timeout=timeout) == 0, phase + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try: + process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL) + process.wait() + + if not package.exists(): + execute('build', [here/'make_rc_candidate.py', parent/'package', package, + '--name', plan['name']]) + verify_package(package) + policy = json.loads((package/'rc_ensemble_policy.json').read_text()) + assert policy['parent_inference_signature'] == plan['parent_inference_signature'] + assert policy['weights'] == [.5, .5] + assert policy['unique_learned_parameters'] == plan['parameters'] + assert digest(package/'last.pt') == digest(parent/'package/last.pt') + signature = inference_signature(package) + native = study/'native-recovery' + with Path(plan['evaluation_lock']).open('a') as lock: + status('waiting_for_native_and_full_evaluation_slot') + fcntl.flock(lock, fcntl.LOCK_EX) + while int(subprocess.check_output(['nvidia-smi', '--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB']*1024: + time.sleep(5) + if not (native/'report.json').exists(): + execute('native', [here/'check_native_recovery.py', package, native], + timeout=plan['native_timeout_seconds']) + proof = json.loads((native/'report.json').read_text()) + assert proof['package_inference_signature'] == signature + assert proof['all_arrays_bit_identical'] and proof['native_vendored_entrypoints_passed'] + assert proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks': 30, 'perturbations': 22} + if not (output/'evaluation.json').exists(): + execute('evaluate', [here/'evaluate_package.py', package, output, + '--resume-partial-tracks', '--resume-partial-perturbations']) + assert json.loads((output/'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output/'valid-score/report.json').read_text()) + variants = json.loads((output/'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None + assert len(tracks['families']) == 11 and len(variants) == 13 + assert all(math.isfinite(v['model']) for v in variants.values()) + for name, value in variants.items(): + for key in ['rows', 'perturbations', 'metric', 'anchors']: + assert value[key] == parent_variants[name][key], (name, key) + comparison = dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward_delta=tracks['overall_mean_family_pearson']-parent_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-parent_tracks['families'][k]['mean_pearson'] + for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-parent_variants[k]['model'] for k,v in variants.items()}) + (study/'parent-comparison.json').write_text(json.dumps(comparison, indent=2)+'\n') + latency = study/'native-latency' + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency/'plan.json').exists(): + assert json.loads((latency/'plan.json').read_text()) == timing_plan + else: + (latency/'plan.json').write_text(json.dumps(timing_plan, indent=2)+'\n') + execute('native-latency', [here/'measure_native_latency.py', latency/'plan.json']) + timing = json.loads((latency/'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + keys = ['development_mixture_seconds', 'slowest_observed_stratum_mean_seconds', 'pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in keys) + passed = max(estimates[k] for k in keys) <= 14400 + (study/'runtime-gate.json').write_text(json.dumps(dict(passed=passed, estimates=estimates, + report_sha256=digest(latency/'report.json'), + scope='All three predeclared shared-load sample estimates must fit4h; none is a guaranteed upper bound.'), + indent=2)+'\n') + if passed: + execute('publish', [here/'promote_if_better.py', package, output]) + status('complete' if passed else 'complete_accuracy_runtime_gate_failed', + runtime_passed=passed, reward=tracks['overall_mean_family_pearson'], **comparison) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('plan', type=Path) + args = parser.parse_args() + with (args.plan.parent/'controller.lock').open('a') as controller: + fcntl.flock(controller, fcntl.LOCK_EX | fcntl.LOCK_NB) + run(args.plan) diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_late_fixed_links81.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_late_fixed_links81.py new file mode 100644 index 0000000000000000000000000000000000000000..3e3973883128f4aea458210d43790edce69d380a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_late_fixed_links81.py @@ -0,0 +1,268 @@ +"""Execute one predeclared late fixed-link follow-up after its complete raw parent.""" +import argparse +from datetime import datetime, timezone +import fcntl +import json +import math +import os +from pathlib import Path +import signal +import subprocess +import sys +import time + +from check_splice_link_reuse import check_link_only +from late_fixed_links81 import ROOT, STUDY, HERE, digest, write_json, parent_ready, parent_receipt +from verify_package import inference_signature, verify_package + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('member') + args = parser.parse_args() + plan_path = STUDY / 'plan.json' + plan = json.loads(plan_path.read_text()) + member = plan['members'][args.member] + study, parent, previous, output = [Path(member[k]) for k in ['study','parent','previous_raw','output']] + lock = (study / 'controller.lock').open('a') + fcntl.flock(lock, fcntl.LOCK_EX | fcntl.LOCK_NB) + complete = study / 'complete.json' + if complete.exists(): + assert json.loads(complete.read_text())['master_plan_sha256'] == digest(plan_path) + print(json.dumps(dict(phase='already_complete_no_rerun', member=args.member)), flush=True) + return + for path, expected in {**plan['helper_sha256'], **plan['evidence_sha256']}.items(): + assert digest(path) == expected, path + assert digest(plan['training_prior']) == plan['training_prior_sha256'] + audit = json.loads((STUDY / 'parent-check.json').read_text()) + assert audit['plan_sha256'] == digest(plan_path) and len(audit['actual_completed_parent_checks']) == 2 + os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) + environment = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2', + PYTHONNOUSERSITE='1', PYTHONDONTWRITEBYTECODE='1') + for key in ['PYTHONPATH','PYTHONHOME','SEQFUNC_TRAIN_BYTECACHE']: + environment.pop(key, None) + + def status(phase, **extra): + record = dict(phase=phase, utc=datetime.now(timezone.utc).isoformat(), member=args.member, **extra) + write_json(study / 'state.json', record) + print(json.dumps(record), flush=True) + + def finish(phase, **extra): + write_json(complete, dict(phase=phase, master_plan_sha256=digest(plan_path), + created_utc=datetime.now(timezone.utc).isoformat(), **extra)) + status(phase, **extra) + + def execute(phase, arguments, timeout=None): + status(phase) + with (study / f'{phase}.log').open('a') as stream: + process = subprocess.Popen([sys.executable, *map(str, arguments)], env=environment, + stdout=stream, stderr=subprocess.STDOUT, start_new_session=True) + try: + code = process.wait(timeout=timeout) + assert code == 0, (phase, code) + finally: + if process.poll() is None: + os.killpg(process.pid, signal.SIGTERM) + try:process.wait(timeout=10) + except subprocess.TimeoutExpired: + os.killpg(process.pid, signal.SIGKILL); process.wait() + + def wait_memory(): + while int(subprocess.check_output(['nvidia-smi','--query-gpu=memory.free', + '--format=csv,noheader,nounits'], text=True).strip()) < plan['minimum_free_GiB'] * 1024: + time.sleep(5) + + status('waiting_for_complete_declared_raw_parents') + while not (parent_ready(parent) and parent_ready(previous)): + time.sleep(10) + receipts = dict(parent=parent_receipt(parent, member['capacity'], member['epoch'], plan), + previous=parent_receipt(previous, member['capacity'], member['previous_epoch'], plan)) + resolved_path = study / 'resolved-parents.json' + resolved = dict(master_plan_sha256=digest(plan_path), member=member, receipts=receipts) + if resolved_path.exists():assert json.loads(resolved_path.read_text()) == resolved + else:write_json(resolved_path, resolved) + admitted = receipts['parent']['reward'] > receipts['previous']['reward'] + plan['parent_minimum_reward_gain'] + gate = dict(admitted=admitted, parent_reward=receipts['parent']['reward'], + previous_raw_reward=receipts['previous']['reward'], minimum_gain=plan['parent_minimum_reward_gain']) + if not (study / 'parent-gate.json').exists():write_json(study / 'parent-gate.json', gate) + else:assert json.loads((study / 'parent-gate.json').read_text()) == gate + if not admitted: + finish('rejected_at_parent_gain_gate', **gate) + return + output.mkdir(exist_ok=True) + package = output / 'package' + construction_path = study / 'construction-check.json' + if not construction_path.exists(): + if package.exists(): + try:verify_package(package);check_link_only(parent / 'package', package) + except (OSError,ValueError,KeyError,AssertionError): + package.rename(package.with_name('package-incomplete-' + str(time.time_ns()))) + if not package.exists(): + execute('build', [HERE / 'make_splice_probability_candidate.py', parent / 'package', package, + '--training-prior', plan['training_prior']]) + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] + assert json.loads((package / 'package_manifest.json').read_text())['parameters'] == plan['capacities'][member['capacity']]['parameters'] + before = (parent / 'package/predict.py').read_text() + marker = 'from reference.alphagenome_50m.model import CompactModel, ModelConfig\n' + old = 'data=all_site_logits[0, target_start:target_start + config.target_length].cpu().numpy().astype(np.float16),' + new = 'data=splice_output_scores(all_site_logits[0, target_start:target_start + config.target_length]).cpu().numpy().astype(np.float16),' + assert before.count(marker) == before.count(old) == 1 + expected = before.replace(marker, marker + 'from splice_output import splice_output_scores\n').replace(old, new) + assert (package / 'predict.py').read_text() == expected + for name in ['infer.sh','perturb.sh']: + assert digest(package / name) == digest(parent / 'package' / name) + assert json.loads((package / 'splice_calibration.json').read_text()) == json.loads(Path(plan['calibration_reference']).read_text()) + history = package / 'training_history'; history.mkdir(exist_ok=True) + additions = {'late-fixed-links81-master-plan.json': plan_path, + 'late-fixed-links81-resolved-parents.json': resolved_path, + 'late-fixed-links81-parent-check.json': STUDY / 'parent-check.json', + 'late_fixed_links81.py': HERE / 'late_fixed_links81.py', + 'run_late_fixed_links81.py': Path(__file__)} + for name, source in additions.items(): + (history / name).write_bytes(source.read_bytes()) + readme = package / 'README.md' + body = readme.read_text().replace( + 'In preliminary\ndiagnostics, MFASS and site Pearson improved while sQTL and some ClinVar metrics regressed.', + "The complete reports below determine this checkpoint's track and variant tradeoffs.") + note = (f'\nThis predeclared late fixed-link follow-up uses the complete {member["epoch"]}-epoch ' + f'{member["capacity"]} main model. It applies the original four training-only offsets ' + 'without refitting, SGD, orientation changes, or changes to site selection and sparse decoding. ' + 'Both entry points use the same link. All200 validation windows and all13 development metrics ' + 'are scored, with exact checks for ten unchanged family reports and eight unchanged variant reports. ' + 'A publication additionally requires native replay and three sampled runtime estimates below four hours.\n') + if note.strip() not in body:body += note + readme.write_text(body) + manifest = json.loads((package / 'package_manifest.json').read_text()) + for name in ['README.md', *['training_history/' + name for name in additions]]: + manifest['files'][name] = digest(package / name) + write_json(package / 'package_manifest.json', manifest) + verification = verify_package(package) + write_json(construction_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), resolved_parents_sha256=digest(resolved_path), + package_inference_signature=inference_signature(package), exact_link_only_checks=checks, + parameters=manifest['parameters'], network_updates=0, prior_refitted=False, package_verified=True)) + construction = json.loads(construction_path.read_text()) + assert construction['master_plan_sha256'] == digest(plan_path) + assert construction['resolved_parents_sha256'] == digest(resolved_path) + verify_package(package) + signature = inference_signature(package) + assert construction['package_inference_signature'] == signature + checks = check_link_only(parent / 'package', package) + assert checks['parent'] == receipts['parent']['inference_signature'] and checks['candidate'] == signature + native = study / 'native-recovery' + if not (native / 'report.json').exists() or not (output / 'valid-variants').exists(): + status('waiting_for_bounded_native_and_derivation_slot') + with Path(plan['bounded_lock']).open('a') as probe_lock: + fcntl.flock(probe_lock, fcntl.LOCK_EX);wait_memory() + if not (native / 'report.json').exists(): + execute('native', [HERE / 'check_native_recovery.py', package, native], plan['native_timeout_seconds']) + if not (output / 'valid-variants').exists(): + materialization = dict(plan['materialization'], candidates=[dict(parent=str(parent), candidate=str(output), + parent_inference_signature=checks['parent'], candidate_inference_signature=signature)]) + path = study / 'materialization-plan.json' + if path.exists():assert json.loads(path.read_text()) == materialization + else:write_json(path, materialization) + execute('materialize', [HERE / 'materialize_splice_link_outputs_v2.py', path], plan['materialization_timeout_seconds']) + proof = json.loads((native / 'report.json').read_text()) + assert proof['package_inference_signature'] == signature and proof['all_arrays_bit_identical'] + assert proof['native_vendored_entrypoints_passed'] and proof['partial_perturbation_recovery_passed'] + assert proof['arrays'] == {'tracks':30,'perturbations':22} + derivation = json.loads((output / 'valid-variants/perturb-link-derivation.json').read_text()) + assert derivation['checks']['candidate'] == signature and derivation['checks']['parent'] == checks['parent'] + assert derivation['native_arrays_bit_exact'] == 36 and derivation['queries'] == 2697 + if not (output / 'evaluation.json').exists(): + status('waiting_for_full_evaluation_slot') + with Path(plan['evaluation_lock']).open('a') as evaluation_lock: + fcntl.flock(evaluation_lock, fcntl.LOCK_EX);wait_memory() + execute('evaluate', [HERE / 'evaluate_package.py', package, output, + '--resume-partial-tracks','--resume-partial-perturbations']) + assert json.loads((output / 'evaluation.json').read_text())['package_inference_signature'] == signature + tracks = json.loads((output / 'valid-score/report.json').read_text()) + variants = json.loads((output / 'valid-variant-score/report.json').read_text()) + raw_tracks = json.loads((parent / 'valid-score/report.json').read_text()) + raw_variants = json.loads((parent / 'valid-variant-score/report.json').read_text()) + assert tracks['examples'] == 200 and tracks['invalid_reason'] is None and len(variants) == 13 + affected = {'sqtl_causality','clinvar_noncoding','clinvar_splice_region','clinvar_missense','mfass'} + for family, report in tracks['families'].items(): + if family != 'splice_sites':assert report == raw_tracks['families'][family],family + for name, report in variants.items(): + assert math.isfinite(report['model']) + for field in ['rows','perturbations','metric','anchors']:assert report[field] == raw_variants[name][field],(name,field) + if name not in affected:assert report == raw_variants[name],name + comparison_path = study / 'matched-comparison.json' + if not comparison_path.exists(): + write_json(comparison_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + master_plan_sha256=digest(plan_path), reward=tracks['overall_mean_family_pearson'], + reward_delta=tracks['overall_mean_family_pearson']-raw_tracks['overall_mean_family_pearson'], + family_deltas={k:v['mean_pearson']-raw_tracks['families'][k]['mean_pearson'] for k,v in tracks['families'].items()}, + variant_deltas={k:v['model']-raw_variants[k]['model'] for k,v in variants.items()}, + unaffected_family_reports_exact=10, unaffected_variant_reports_exact=8, package_inference_signature=signature)) + decision_path = study / 'publication-decision.json' + if not decision_path.exists(): + current = json.loads(Path('/workspace/submission/validation_tracks.json').read_text())['overall_mean_family_pearson'] + eligible = tracks['overall_mean_family_pearson'] > current + plan['publication_minimum_reward_gain'] + runtime_passed = None + if eligible: + latency = study / 'native-latency';latency.mkdir(exist_ok=True) + timing_plan = dict(plan['native_latency_template'], experiment=str(output), inference_signature=signature) + if (latency / 'plan.json').exists():assert json.loads((latency / 'plan.json').read_text()) == timing_plan + else:write_json(latency / 'plan.json', timing_plan) + if not (latency / 'report.json').exists(): + execute('native-latency', [HERE / 'measure_native_latency.py', latency / 'plan.json']) + timing = json.loads((latency / 'report.json').read_text()) + assert timing['all_arrays_bit_exact'] and timing['package_inference_signature'] == signature + estimates = timing['extrapolated_seconds'] + names = ['development_mixture_seconds','slowest_observed_stratum_mean_seconds','pooled_p95_seconds'] + assert all(math.isfinite(estimates[k]) for k in names) + runtime_passed = max(estimates[k] for k in names) <= 14400 + write_json(study / 'runtime-gate.json', dict(passed=runtime_passed, estimates=estimates, + report_sha256=digest(latency / 'report.json'), scope='Shared-load extrapolations, not guaranteed upper bounds.')) + if runtime_passed:execute('publish', [HERE / 'promote_if_better.py', package, output]) + write_json(decision_path, dict(created_utc=datetime.now(timezone.utc).isoformat(), + reward=tracks['overall_mean_family_pearson'], current_reward_at_review=current, + eligible_at_review=eligible, runtime_passed=runtime_passed, + publication='Atomic promoter rechecks the current strongest package; see publish.log if run.')) + for phase, helper in [('tracks','diagnose_track_uncertainty.py'), + ('remaining-variants','diagnose_remaining_variant_uncertainty.py')]: + report_path = study / f'paired-{phase}.json' + marker_path = study / f'paired-{phase}-complete.json' + if marker_path.exists(): + assert json.loads(marker_path.read_text())['output_sha256'] == digest(report_path) + continue + def diagnostic_complete(): + try: + report = json.loads(report_path.read_text()) + assert Path(report['candidate']).resolve() == output.resolve() + assert Path(report['control']).resolve() == parent.resolve() + assert report['replicates'] == 2000 and report['seed'] == 1729 + assert report['diagnostic_only_no_parameter_fitting'] + leaf = 'valid-score/report.json' if phase == 'tracks' else 'valid-variant-score/report.json' + assert report['report_sha256'] == {'candidate':digest(output / leaf),'control':digest(parent / leaf)} + if phase == 'tracks': + assert len(report['point_deltas']) == 12 and 'reward' in report['point_deltas'] + assert set(report['bootstrap']) == {'chromosome','chromosome_10Mb_blocks'} + for result in report['bootstrap'].values(): + assert set(result['deltas']) == set(report['point_deltas']) + else: + assert set(report['evaluations']) == affected | {'caqtl','bqtl_spi1'} + for item in report['evaluations'].values(): + assert set(item['bootstrap']) == {'variant_id','genomic_block'} + for result in item['bootstrap'].values(): + assert result['reference_replicates_checked'] >= 32 + assert result['maximum_absolute_metric_error'] < 1e-10 + return True + except (FileNotFoundError,json.JSONDecodeError,AssertionError,KeyError): + return False + if not diagnostic_complete(): + execute('paired-' + phase, [HERE / helper, output, parent, report_path, + '--replicates','2000','--seed','1729']) + assert diagnostic_complete(), ('incomplete diagnostic', phase) + write_json(marker_path, dict(output_sha256=digest(report_path), master_plan_sha256=digest(plan_path))) + finish('complete', reward=tracks['overall_mean_family_pearson'], + matched_comparison_sha256=digest(comparison_path), publication_decision=json.loads(decision_path.read_text()), + diagnostics='All11families/sevenvariants, with eight unaffected official variant reports exact; not anall13bootstrap.') + + +if __name__ == '__main__': + main() diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_scale_study_cached_io.py b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_scale_study_cached_io.py new file mode 100644 index 0000000000000000000000000000000000000000..3a64b3a086363780b355fe0c827bab67bc691b68 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_history/run_scale_study_cached_io.py @@ -0,0 +1,151 @@ +"""Train two locally initialized scales on an identical recipe and evaluate selected full epochs. + +This script requires a pre-reviewed study.json; it never chooses scientific options itself. +It resumes checkpoints and preserves an independently replayable package for every evaluation. +""" +import argparse +import hashlib +import json +import os +from pathlib import Path +import subprocess +import sys +import time +import torch + +p = argparse.ArgumentParser(description=__doc__) +p.add_argument('study', type=Path) +a = p.parse_args() +plan = json.loads(a.study.read_text()) +assert plan.get('training_bytecache_root'), 'This amended runner requires an explicit training-only cache path' +assert plan['schema'] == 'seqfunc-matched-scale/v1' +assert plan['pretrained'] is False and plan['examples_per_epoch'] == 41020 +assert len(plan['runs']) == 2 and plan['shared']['batch_size'] == 4 +r = Path('/workspace/research') +here = Path(__file__).resolve().parent +source = Path(plan.get('source', r/'seqfunc-scale')) +sys.path.insert(0, str(source)) +from reference.alphagenome_50m.model import ModelConfig +configs = [ModelConfig.from_yaml(run['config']).to_dict() for run in plan['runs']] +differences = {key for key in configs[0] if configs[0][key] != configs[1][key]} +assert differences and differences <= {'encoder_width', 'channel_growth', 'hidden_dim', + 'num_sequence_blocks', 'num_attention_heads', 'head_hidden_size', + 'local_width'} +env = dict(os.environ, OPENBLAS_NUM_THREADS='1', OMP_NUM_THREADS='2', MKL_NUM_THREADS='2') +os.sched_setaffinity(0, sorted(os.sched_getaffinity(0))[:22]) +env.pop('SEQFUNC_TRAIN_BYTECACHE', None) +children, handles = {}, [] + +class AdoptedEvaluation: + def __init__(self, identity, directory): + self.identity, self.directory = identity, directory + self.pid = identity['pid'] + def poll(self): + try: + path = Path(f'/proc/{self.pid}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + active = (stat[0] not in 'ZX' and int(stat[19]) == self.identity['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == self.identity['cmdline_sha256']) + except OSError: + active = False + if active: + return None + try: + done = json.loads((self.directory/'pipeline-state.json').read_text())['phase'] == 'complete' + except (OSError, KeyError, json.JSONDecodeError): + done = False + return 0 if done else 1 + @property + def returncode(self): + return self.poll() + + +def launch(name, arguments, logfile, *, training=False): + handle = logfile.open('a') + handles.append(handle) + child_environment = dict(env) + if training: + child_environment['SEQFUNC_TRAIN_BYTECACHE'] = plan['training_bytecache_root'] + proc = subprocess.Popen([sys.executable, *map(str, arguments)], env=child_environment, + stdout=handle, stderr=subprocess.STDOUT) + children[name] = proc + print(json.dumps({'event': 'study_launch', 'name': name, 'pid': proc.pid}), flush=True) + +for run in plan['runs']: + directory = Path(run['output']) + directory.mkdir(parents=True, exist_ok=True) + checkpoints = [directory/name for name in ['progress.pt', 'last.pt']] + epochs = sorted(directory.glob('epoch-*.pt')) + if epochs: + checkpoints.append(epochs[-1]) + available = [(torch.load(path, map_location='cpu', weights_only=True)['global_step'], path) + for path in checkpoints if path.exists()] + step, checkpoint = max(available) if available else (0, None) + original = plan['cache_execution_amendment']['checkpoints'][run['name']] + assert step >= original['step'] + if step == original['step']: + checkpoint = Path(original['path']) + with checkpoint.open('rb') as stream: + assert hashlib.file_digest(stream, 'sha256').hexdigest() == original['sha256'] + shared = plan['shared'] + if step < shared['epochs']*plan['examples_per_epoch']: + args = [source/'train_batched.py', '--output', directory, '--config', run['config'], + '--workers', run['workers'], '--threads', 2, '--save-every', 1000, + '--keep-every', 0, '--log-every', 100] + for key in ['batch_size', 'epochs', 'seed', 'learning_rate', 'min_learning_rate', 'warmup', 'objective']: + args += ['--'+key.replace('_', '-'), shared[key]] + if shared.get('loss_scales'): + args += ['--loss-scales', shared['loss_scales']] + if checkpoint: + args += ['--resume', checkpoint] + launch(run['name']+'-train', args, directory/'train.log', training=True) + for epoch in plan['evaluation_epochs']: + evaluation = directory.parent/(run['name']+f'-epoch{epoch:02d}') + evaluation.mkdir(exist_ok=True) + name = run['name']+f'-eval{epoch:02d}' + state_path = evaluation/'pipeline-state.json' + if state_path.exists() and json.loads(state_path.read_text()).get('phase') == 'complete': + print(json.dumps({'event': 'completed_evaluation_preserved', 'name': name}), flush=True) + continue + if name in plan.get('adopted_evaluations', {}): + children[name] = AdoptedEvaluation(plan['adopted_evaluations'][name], evaluation) + print(json.dumps({'event': 'evaluation_adopted_without_interruption', 'name': name, 'pid': children[name].pid}), flush=True) + continue + args = [here/'run_candidate.py', directory/f'epoch-{epoch:02d}.pt', evaluation, + '--source', source, '--step', epoch*plan['examples_per_epoch'], + '--evaluation-lock', a.study.with_name('evaluation.lock'), + '--name', f"SeqFunc {run['name']} after {epoch} complete epochs", '--orientation', + '--lineage', plan['provenance']+' '+run['description']+ + ' Parameters are initialized locally with seed 17. The paired scale uses the same ' + 'batch-four interval order, objective, learning-rate schedule and full training permutations. ' + 'Only encoder/trunk/head/local-path width and transformer depth differ between scales.'] + launch(run['name']+f'-eval{epoch:02d}', args, evaluation/'pipeline.log') + +monitor_retired = False +while True: + status = {name: {'pid': proc.pid, 'returncode': proc.poll()} for name,proc in children.items()} + temporary = a.study.with_name('study-state-cached.tmp') + temporary.write_text(json.dumps(status, indent=2)+'\n') + temporary.replace(a.study.with_name('study-state-cached.json')) + if not monitor_retired and all(proc.poll() is not None for proc in children.values() + if isinstance(proc, AdoptedEvaluation)): + import signal + old = plan['retired_monitor_identity'] + try: + path = Path(f'/proc/{old["pid"]}') + stat = (path/'stat').read_text().split(') ', 1)[1].split() + if (int(stat[19]) == old['starttime_ticks'] + and hashlib.sha256((path/'cmdline').read_bytes()).hexdigest() == old['cmdline_sha256']): + os.kill(old['pid'], signal.SIGTERM) + os.kill(old['pid'], signal.SIGCONT) + except (FileNotFoundError, ProcessLookupError): + pass + monitor_retired = True + print(json.dumps({'event': 'old_monitor_retired_after_preserved_evaluation', 'pid': old['pid']}), flush=True) + if all(proc.poll() is not None for proc in children.values()): + break + time.sleep(5) +for handle in handles: + handle.close() +if any(proc.returncode for proc in children.values()): + raise RuntimeError('one or more study processes failed; inspect study-state.json and individual logs') diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_loss_scales.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_loss_scales.json new file mode 100644 index 0000000000000000000000000000000000000000..a8f484c060a0d4300b25557736abf741d6d67b15 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_loss_scales.json @@ -0,0 +1,32 @@ +{ + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_provenance.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..080bff0a2284d201557f4eee9e0c94fa443a741d --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/training_provenance.json @@ -0,0 +1,180 @@ +{ + "schema": "seqfunc-batched/v1", + "config": { + "input_length": 1048576, + "target_length": 196608, + "dense_bin_size": 128, + "trunk_bin_size": 1024, + "contact_bin_size": 2048, + "stem_stride": 8, + "encoder_width": 96, + "channel_growth": 96, + "num_downsampling_blocks": 7, + "conv_kernel_size": 5, + "dna_kernel_size": 15, + "hidden_dim": 1024, + "num_sequence_blocks": 16, + "num_attention_heads": 16, + "attention_head_dim": 64, + "mlp_expansion": 4.0, + "dropout": 0.0, + "head_hidden_size": 768, + "junction_rank": 32, + "splice_usage_tracks": 734, + "junction_tracks": 734, + "contact_tracks": 28, + "gradient_checkpointing": true, + "enable_pairwise": false, + "pair_width": 32, + "pair_downsample": 16, + "pair_update_interval": 2, + "local_width": 64, + "local_context_blocks": 6, + "splice_context_mode": "offset", + "local_to_trunk": true, + "residual_rms_cap": 0.1 + }, + "epoch": 8, + "offset": 0, + "global_step": 328160, + "examples": 41020, + "schedule_epochs": 14, + "objective": "variance_balanced", + "seed": 17, + "training_arguments": { + "input": "/workspace/data", + "output": "/workspace/research/experiments/scale13-large", + "config": "/workspace/research/studies/scale13/scale13-large.yaml", + "resume": "/workspace/research/experiments/scale13-large/progress.pt", + "batch_size": 4, + "epochs": 14, + "stop_at": 0, + "workers": 8, + "threads": 2, + "seed": 17, + "learning_rate": 0.0002, + "min_learning_rate": 2e-05, + "warmup": 4000, + "objective": "variance_balanced", + "loss_scales": "/workspace/research/studies/scale13/loss_scales.json", + "save_every": 1000, + "keep_every": 0, + "log_every": 100 + }, + "elapsed_seconds": 108193.25909642546, + "provenance": { + "data": "/workspace/data", + "split": "train", + "pretrained": false, + "world_size": 1, + "device": "NVIDIA H100 80GB HBM3", + "batch_reduction": "equal per example", + "train_manifest_sha256": "48133d27eedfb4b1d275cfca41f69dd075a2a0305ab2e8591e8e5e7529f36c64" + }, + "optimizer_steps": 82040, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + }, + "loss_scale_provenance": { + "split": "train", + "examples": 256, + "seed": 1729, + "source_moments_sha256": "7b7543c4d4d8d82ec9f9bc3f942ca65595671322f62c19391591df827d8fc588", + "training_indices_sha256": "b82df9ae11e95d9be5d6e5f85839cf23e25bbbe0a195eaee266ff52d366800dc", + "method": "reference MSE times inverse mean within-track training variance; floor0.01; splice BCE unchanged", + "mean_track_variance": { + "atac": 0.9602319499041136, + "dnase": 1.0768856864858252, + "procap": 0.3380534137610864, + "cage": 0.3114663911919963, + "rna_seq": 0.66617486157613, + "chip_tf": 2.7198812117366296, + "chip_histone": 3.405612880716664, + "contact_maps": 0.14704202117243276, + "splice_usage": 0.07913355074122039, + "junctions": 0.17919098248535248 + }, + "family_loss_scales": { + "atac": 1.041415045708339, + "dnase": 0.9286036694045732, + "procap": 2.9581124144681272, + "cage": 3.2106192779675315, + "rna_seq": 1.5011073783751905, + "chip_tf": 0.36766311546433506, + "chip_histone": 0.29363290398101966, + "contact_maps": 6.800777029767044, + "splice_usage": 12.636865029223356, + "junctions": 5.5806379658738825 + } + }, + "inference_ensemble": { + "method": "Equal mean of the same checkpoint evaluated on DNA and its reverse complement, mapped to original positions and public-metadata strand channels.", + "orientation_views": 2, + "fitted_ensemble_weights": false, + "weights": [ + 0.5, + 0.5 + ], + "dna_input": "Same provided 1,048,576bp, reversed and complemented; no extra flanking sequence.", + "site_scores": "Apply the existing training-prior probability link per orientation, then average probabilities.", + "sparse_outputs": "Each orientation chooses its own positions/pairs. Average the union with zero for unlisted coordinates; retain up to16,384 junctions by averaged predicted strength. Usage union has at most4,096 coordinates.", + "outputs": "All families and both entry points use the same two-view rule. No labels or targets used at inference.", + "created_utc": "2026-10-26T07:36:00.650275+00:00", + "parent_package": "/run-state/workspace/research/experiments/calib81-large-epoch08/package", + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "5f5507be4efce5af3167616de01c11da89fa4fe2b2c59dbdac0152e1a84d1348", + "perturb.py": "9defff26b93513ebe681c31485b75730aac911d9c0f9983f2c7aa615a01b67f8", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641" + }, + "unique_learned_parameters": 215626632, + "decoder_study": "/run-state/workspace/research/studies/rc17", + "decoder_assets_sha256": { + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + } + }, + "fixed_decoder117_intervention": { + "utc": "2026-10-26T10:57:56.711123+00:00", + "parent_inference_signature": { + "last.pt": "91f3382e440a1c61731b090dcf04d4de98d706f7c2f7ce956e1ade752e6c57eb", + "predict.py": "4f5dbddf5b6b701079aeaf38920fe414ce980f7161997f4d5c80952a6d834e51", + "perturb.py": "49065dc43816920ecac9de3ebcc4735673e4ff6594cf3f6bd34233b88c9bed73", + "prediction_contract.py": "b11af8f7cff82b35afc6282be387358a48464b78bd1dfa44452dace26a5e0a83", + "reference/alphagenome_50m/model.py": "c61f50a2a66a4f6950c6f6107a70710a5d0ab99a291b8bfaa1eb5565998d2fa3", + "reference/alphagenome_50m/layers.py": "c39b11287d3454627469fb9e196f23f218cf48c9856de9126f9e30e9e2662af8", + "reference/alphagenome_50m/dataset.py": "6ac7376529fc5b6725fdea8080de1109de4e3948d68607fc86ce38dc9c742919", + "splice_output.py": "e1b0627e036c60190e4d2a4653a915d9031b166ff3c1de7c23e5774009d70ef0", + "splice_calibration.json": "bd9368b32e739be65ca290d598ffcae5be6387c09f96048e183c57b7d570e641", + "rc_ensemble.py": "84a4d0e94ab86622e2c25e96f8835b2ef2bba77731ff7ff0bdc1c2de02cdcef4", + "single_prediction.py": "10f6216fdd72bf30b3a28ab6de9363e66b415526aa26a2bbfb6a84badbb67bec", + "strand_permutations.json": "afb0f0cf2e897fab3493de66601d7a66dac45a06316c2ac4a6cc828298820c60" + }, + "candidate_model_sha256": "9ddfecf4def3c0ce70a5ac8e831a22c6c5565c90719b5a36dc541acc12b7afb2", + "parameters": 215626632, + "inherited_examples": 328160, + "inherited_optimizer_steps": 82040, + "inherited_complete_passes": 8, + "additional_training_updates": 0, + "only_change": "Replace RC114 nearest global-feature decoding for dense, usage and junction heads with the exact selective linear-bin-center operation from completed111/96. Keep its original nearest splice classifier, contacts, learned tensors, four training-derived offsets and equal forward/RC views.", + "scope": "One fixed network and the same two orientation views serve every family and both entrypoints. No learned or fitted mixture, routing, new splice prior, motif mask, SWA, extra flanking DNA, variant-label training or additional SGD.", + "limitations": "Exploratory one-seed postselection. Interaction can worsen any changed readout. Full native replay, runtime and accuracy must be measured independently; unchanged paths require exact complete reports." + } +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_tracks.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_tracks.json new file mode 100644 index 0000000000000000000000000000000000000000..df8c6c512e329dd14d9464b206b67c07f9b4252a --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_tracks.json @@ -0,0 +1,17909 @@ +{ + "schema": "genome-sequence-to-function-report/v4", + "split": "valid", + "examples": 200, + "families": { + "atac": { + "mean_pearson": 0.5964424131030225, + "valid_tracks": 167, + "per_track_pearson": [ + 0.7150821317731466, + 0.7384597769957606, + 0.46765319269012345, + 0.6634561928586957, + 0.48002376644381833, + 0.47873044710711343, + 0.5000780274262977, + 0.5818374781252071, + 0.5937387251119809, + 0.6592758599118299, + 0.5972413451076276, + 0.5738951811351841, + 0.40827912828918145, + 0.5414086944210907, + 0.6975100975912346, + 0.5234225488426074, + 0.5718163277780991, + 0.5418736563911498, + 0.5774382856883261, + 0.5587379673791543, + 0.5282891542120608, + 0.49072347508564296, + 0.5688840878442294, + 0.5423149401973187, + 0.5419329169975334, + 0.5228890574808712, + 0.49976416687162206, + 0.5199291630154964, + 0.531367856368946, + 0.5924771715680335, + 0.5073433286908129, + 0.5474451953553782, + 0.5011908341921452, + 0.41961025953511233, + 0.38914320205826963, + 0.6049780096225882, + 0.5427052124423632, + 0.6507461966443554, + 0.5331860371295037, + 0.5266552588418986, + 0.6034821374225141, + 0.6066342112359779, + 0.5223068523937452, + 0.4974701917568982, + 0.5673011664082497, + 0.5845614954943211, + 0.45366956686112053, + 0.5073475304900273, + 0.5019092042454845, + 0.5447075630917196, + 0.5142226901602761, + 0.48222450009726453, + 0.5793987627015881, + 0.6147169003441324, + 0.47438381585378053, + 0.6993990619277798, + 0.815049329510579, + 0.5722688480657877, + 0.6086291205626224, + 0.6540543417068865, + 0.782025770824942, + 0.5178239364738694, + 0.704616053539596, + 0.6981061851591818, + 0.6523397887759269, + 0.5259325368046279, + 0.589467463868806, + 0.6517241824172603, + 0.8055206002886123, + 0.7255241983787389, + 0.5418654862206108, + 0.49475162844955606, + 0.470896755477366, + 0.5585178159573355, + 0.5189967038673632, + 0.5093690433692016, + 0.5139811995223124, + 0.5559908415066549, + 0.6007100578498618, + 0.5105566061374247, + 0.5942010051311947, + 0.6157475779248998, + 0.5309374619130488, + 0.5051619320140194, + 0.5392753026682978, + 0.5314536790151634, + 0.509106746549767, + 0.5009641736933537, + 0.46925165204502717, + 0.5239420663046582, + 0.5276771888865678, + 0.5236267607338618, + 0.5025400593300425, + 0.5214305878958012, + 0.5825834893306233, + 0.411634672383475, + 0.46167583582304456, + 0.45319450174295356, + 0.510117708248962, + 0.5267976403561887, + 0.4766076749043017, + 0.4790212861582686, + 0.523627817467579, + 0.48392919739351714, + 0.5636409129380842, + 0.5720504647865959, + 0.5583403983365309, + 0.539048383805137, + 0.5630586035701611, + 0.5732632818559876, + 0.6275369628400239, + 0.5465959736168934, + 0.5406167818365385, + 0.4590103243524842, + 0.48616339690463867, + 0.46020118503850704, + 0.49865187133968913, + 0.5017030337383163, + 0.5169313891083169, + 0.4946240412959147, + 0.6726162986287896, + 0.6915858071720518, + 0.7569266818414162, + 0.7426845797977709, + 0.7346201763667713, + 0.593239304013611, + 0.869489984072378, + 0.7434422186066915, + 0.7777653260508216, + 0.7004919296337121, + 0.7428113768878317, + 0.5809217208039111, + 0.6071979545621451, + 0.5804443717611316, + 0.7268726573757283, + 0.7633841033580422, + 0.7117634454416454, + 0.5558722500562302, + 0.4391276257540094, + 0.8294558454923145, + 0.8000110677167646, + 0.8216463640138951, + 0.8149793659333561, + 0.5891641347023024, + 0.8813265150920894, + 0.7928332369406469, + 0.7587057988013617, + 0.7493443164772491, + 0.8506755438100816, + 0.8293120236087066, + 0.6355831741922652, + 0.6407585915115865, + 0.7380334919869547, + 0.7663367367319491, + 0.7841868282294924, + 0.6229024587665701, + 0.7573661174124021, + 0.6158695652275915, + 0.6167235170223672, + 0.7039988611323696, + 0.619889992871274, + 0.7903731543157636, + 0.7930967078164967, + 0.6631876276672439, + 0.6094652248919854, + 0.6135132732521263, + 0.7559541166826195 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "dnase": { + "mean_pearson": 0.6732080577809814, + "valid_tracks": 305, + "per_track_pearson": [ + 0.5864997018445269, + 0.9340492902625601, + 0.8030322924462013, + 0.5466125745990831, + 0.5638470610268487, + 0.5688401987719857, + 0.9442455181100279, + 0.7454485403284399, + 0.4911971861794199, + 0.3731570240291048, + 0.8987938021167604, + 0.7321333108436722, + 0.45867482915202845, + 0.5498128753519808, + 0.5308464228146708, + 0.8267432407092722, + 0.8093410313991343, + 0.8591433238339458, + 0.8795278257617102, + 0.9063357320146809, + 0.7204507656581214, + 0.6490933649614813, + 0.9142676377064032, + 0.5610930787630047, + 0.5749710240948587, + 0.7950307877124876, + 0.4795549369219962, + 0.6980014098296339, + 0.5544334013796912, + 0.5811660693942311, + 0.6600785689830452, + 0.49560869614866987, + 0.8535223282185113, + 0.6398626796469575, + 0.8912896920562682, + 0.8994706012282888, + 0.4697119081282416, + 0.8404140833957913, + 0.8849532845092506, + 0.8726398996823197, + 0.8419450380029936, + 0.682978517572433, + 0.7178995671017859, + 0.6649094535945494, + 0.5603392961970893, + 0.6524147605431616, + 0.7472162245579199, + 0.5245453096383611, + 0.6241390928449893, + 0.4312604955595718, + 0.41498346935740144, + 0.5950066254073365, + 0.5370044838213904, + 0.8369620968168898, + 0.4713156089691144, + 0.43300150230387824, + 0.5006878739666986, + 0.552015599958025, + 0.5184034719801314, + 0.5483215560735726, + 0.4869101336918765, + 0.6406249624969913, + 0.5322993528197839, + 0.5232651941979506, + 0.5513434762635493, + 0.562453271758908, + 0.5877296808034366, + 0.5207181694620912, + 0.4134901924954665, + 0.4587400313280621, + 0.5326606989262288, + 0.6809152516680782, + 0.5459164730635483, + 0.6700494405991503, + 0.396192777762676, + 0.8232378879178432, + 0.9038202759060059, + 0.6942951811098146, + 0.4763880104634046, + 0.4935012573473328, + 0.5725150436639944, + 0.5453826752758182, + 0.4849121795517813, + 0.5247858087262782, + 0.6106157529483336, + 0.524294812583822, + 0.6487782646660887, + 0.5364277339988575, + 0.6664492689932028, + 0.5399812523518647, + 0.6638703737054901, + 0.437098047682797, + 0.7460236799949119, + 0.5053753383357559, + 0.7372873518214752, + 0.5368102732028536, + 0.7779513217796854, + 0.7470117268248767, + 0.40639544651426324, + 0.41685077349449057, + 0.5093474348393113, + 0.404568514655402, + 0.38570175621986136, + 0.7550895939701787, + 0.7282634205033892, + 0.43690769518255257, + 0.5241537378208243, + 0.617150746923064, + 0.859591210486151, + 0.5896041032795771, + 0.6009339574779284, + 0.6173404504452245, + 0.5469015639108379, + 0.4688594914411995, + 0.798474323000625, + 0.613139685059195, + 0.5297585688745231, + 0.5868592881783133, + 0.6735328068650487, + 0.5046087302380642, + 0.5640295047886092, + 0.5276071543237603, + 0.5970037151641163, + 0.5826196685418946, + 0.6499913491565755, + 0.5913095730955165, + 0.6822300062406921, + 0.6974066833041229, + 0.7182649731276725, + 0.6704024642807672, + 0.4933729828684185, + 0.6394567034123525, + 0.6715767998495532, + 0.8123452575258441, + 0.4772288778572822, + 0.36497532652149395, + 0.6347884092090543, + 0.5119591047465869, + 0.6727004353959682, + 0.6348860804949404, + 0.6454245785723771, + 0.5896618160944324, + 0.519350059379548, + 0.7972642808696264, + 0.657819013888953, + 0.7746974115113969, + 0.4336339923352708, + 0.4060279183432135, + 0.4348367590746078, + 0.3863055491565299, + 0.9079923718358317, + 0.3556700084311821, + 0.9127341577692218, + 0.7859966886560678, + 0.49463931684484846, + 0.3887420254987262, + 0.3885391186089819, + 0.39713909749051024, + 0.6873417757590479, + 0.7866889128791694, + 0.7768217817424142, + 0.6959918538555913, + 0.5953228708471492, + 0.5179245024519878, + 0.7079376362485099, + 0.6488469246725718, + 0.556167691222875, + 0.6707295829615396, + 0.49118017429852995, + 0.5293330006874964, + 0.4530787284945269, + 0.8442598860754815, + 0.5014909789458745, + 0.514012185855783, + 0.6931334543556669, + 0.7246553191673597, + 0.5590316801138585, + 0.3835068602388431, + 0.88651203430374, + 0.4119204235539944, + 0.8011092016826444, + 0.5413251658234652, + 0.6414970835791838, + 0.7600383302482082, + 0.48374974184809066, + 0.6116289923091587, + 0.6493205223180688, + 0.843461441860094, + 0.4480876501331013, + 0.6478465932417293, + 0.6123439399957707, + 0.747101235437298, + 0.9061311061765227, + 0.4852028800006109, + 0.6330477551138456, + 0.5301526157504299, + 0.68452190087205, + 0.7530398767629394, + 0.5977859122201787, + 0.49896878369713843, + 0.8826902355805194, + 0.8953464190943727, + 0.7658229396062103, + 0.8484661394598725, + 0.858508241995712, + 0.8500821203182808, + 0.8643714946996316, + 0.40233893641475965, + 0.42717504253250954, + 0.5526055547940429, + 0.5041461011348221, + 0.4989454748598664, + 0.6532063502104135, + 0.877821622406518, + 0.8600023018744193, + 0.7208861395406324, + 0.8439901567761559, + 0.8419710693086139, + 0.8713767782728222, + 0.9154342397710252, + 0.5894713391348505, + 0.6794532087560647, + 0.7316518238709956, + 0.703535903756414, + 0.8418953432923021, + 0.6382631104236873, + 0.9243058495302618, + 0.8963349755283754, + 0.8810790850303432, + 0.9121879464302259, + 0.9018910491197398, + 0.8920938830265575, + 0.8468162309617102, + 0.8712975497429791, + 0.9201150523106965, + 0.4981720450477982, + 0.9363685632749719, + 0.8902415778563922, + 0.8356431092902086, + 0.655219962749718, + 0.8819148803314284, + 0.5683593810913969, + 0.762942053903911, + 0.8848644581843212, + 0.6523053831248992, + 0.6430847301662237, + 0.7336279352280499, + 0.5659923073677534, + 0.6017602712247982, + 0.7619897044407641, + 0.8994852382469062, + 0.785935101888604, + 0.8348082466629582, + 0.6742606391790477, + 0.8794044760635443, + 0.8218587431454607, + 0.8527803928352069, + 0.91650369165109, + 0.8388847511577475, + 0.90825356398037, + 0.8817446758814206, + 0.5829160705910778, + 0.8860706451237298, + 0.6282560260910135, + 0.5917481420057245, + 0.8904525188351605, + 0.9138238696858723, + 0.8917091492509607, + 0.6705079530028971, + 0.6861995795720576, + 0.7014890411436674, + 0.8126221792977066, + 0.908634274080963, + 0.7608171427607986, + 0.9253483036277778, + 0.9297118442436183, + 0.6518945843893625, + 0.37278072183661376, + 0.5650904017086973, + 0.9085729890616302, + 0.5046653897696189, + 0.6656258562823424, + 0.8967616085487591, + 0.8423262240600722, + 0.6915286368548764, + 0.8760880215763326, + 0.90457057510031, + 0.8995799991880961, + 0.889914631227645, + 0.877714154600782, + 0.9030483127987253, + 0.8715640291239974, + 0.9401513062849293, + 0.7946791927965451, + 0.8481728248171543, + 0.9158580532200957, + 0.6136541091971438, + 0.6213901935522821, + 0.6217995831130632, + 0.5910973613046306, + 0.9171466214720327, + 0.8164770959274605, + 0.8591746168532641, + 0.8153035742161636, + 0.7832544295404551 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "procap": { + "mean_pearson": 0.5642503035362264, + "valid_tracks": 12, + "per_track_pearson": [ + 0.5287167005804667, + 0.5828511347566533, + 0.5396387124170955, + 0.577389833033833, + 0.5739711522719503, + 0.6092556871909295, + 0.5170728358676218, + 0.5718881948145695, + 0.5361185955807586, + 0.5680914616636868, + 0.5431284345955127, + 0.6228808996616401 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "cage": { + "mean_pearson": 0.3593844963442376, + "valid_tracks": 546, + "per_track_pearson": [ + 0.3659068502854044, + 0.4029546811740368, + 0.32287115476256073, + 0.2221512278123073, + 0.1288700190139293, + 0.34456312403995026, + 0.25419390658691243, + 0.23015613341824187, + 0.35780920435851227, + 0.3565888867710768, + 0.3874572570575114, + 0.2370838358498518, + 0.23362358550489262, + 0.19245684990353543, + 0.21845528499330888, + 0.255622424254333, + 0.3011109879308989, + 0.418185188197671, + 0.38659218424676073, + 0.40233739062779217, + 0.3201257524983503, + 0.3214351977670818, + 0.464367717247939, + 0.27512908869737024, + 0.3339057242934252, + 0.37076263068214205, + 0.3457991471924103, + 0.3821503625051293, + 0.3974986068152448, + 0.4187677613367676, + 0.25665147459933985, + 0.4380161209395739, + 0.27736644190339277, + 0.23790728436737418, + 0.4403359263404571, + 0.37076849194531597, + 0.39482587577297723, + 0.46130587255106037, + 0.4021141252461299, + 0.42505703744136875, + 0.47869546997067747, + 0.4045341533457929, + 0.3845749519005181, + 0.3313168362399955, + 0.2229014961466175, + 0.3176196889721813, + 0.33224620199122173, + 0.37268144418446775, + 0.21061380967698434, + 0.26032955145714404, + 0.40997977200096375, + 0.40996002316622543, + 0.4107182253249183, + 0.30099490281215807, + 0.47447293555394915, + 0.45633610986203943, + 0.2834974554602623, + 0.46854959901555565, + 0.3391321151538723, + 0.45085352408399604, + 0.43223450919419903, + 0.29760572486842046, + 0.37456189592692346, + 0.3194268070564683, + 0.3965857467108181, + 0.4557882500460493, + 0.44811643028550446, + 0.2661105547614245, + 0.2218878138091175, + 0.19133811097739312, + 0.356180389043651, + 0.4055506192911905, + 0.3788763754080054, + 0.3497093968938962, + 0.4033596568988871, + 0.3233849134339133, + 0.37908069599138944, + 0.3710849964143669, + 0.4409794120986503, + 0.41688123565719676, + 0.3720424501026914, + 0.2194365352562685, + 0.37464964531339495, + 0.4777807334438225, + 0.21835134240625623, + 0.38898920022860073, + 0.3572653173515152, + 0.4655433205747073, + 0.3148229670496258, + 0.4030431340521837, + 0.3313724218065232, + 0.37684978659990426, + 0.39170744070587044, + 0.20097386526154265, + 0.410104840905077, + 0.3375513130341697, + 0.4011575727291818, + 0.3959401782382203, + 0.36310408813400935, + 0.23802373561424112, + 0.379133522357547, + 0.38195444191142514, + 0.4167047158172883, + 0.46509548435755427, + 0.4316815349144307, + 0.4141292051736728, + 0.3120986611218186, + 0.4229304444761117, + 0.3394285699374958, + 0.3327153338139021, + 0.34323346145758715, + 0.3468120614488186, + 0.37189179945708895, + 0.3747736152359894, + 0.3137304322554592, + 0.40374079180463884, + 0.3689785474167855, + 0.38173153355809, + 0.37048061811104743, + 0.4058468142716336, + 0.421449179643466, + 0.39535568907932805, + 0.2956665284850668, + 0.3425951539474896, + 0.35401783113101404, + 0.4054021527866087, + 0.3843148909075494, + 0.41444638819360147, + 0.4744481229626826, + 0.41609156677139497, + 0.4337943068077655, + 0.3351958118442072, + 0.4365438974689006, + 0.4474671401269075, + 0.47958475362697306, + 0.3654079278994786, + 0.46165637964414774, + 0.3670064826555719, + 0.20246324003520572, + 0.42671729750489984, + 0.43176592789123336, + 0.37613808179049013, + 0.4264719240979081, + 0.3281907765602604, + 0.3560760307454857, + 0.37589250789384065, + 0.3625155635663115, + 0.4808947229881735, + 0.34484340324722484, + 0.35319631622391295, + 0.4234262239411468, + 0.3379152911880755, + 0.39433159700593245, + 0.3906423378860202, + 0.3702243813542546, + 0.3372653682176626, + 0.43967465619267626, + 0.47676986321853115, + 0.33594478475375644, + 0.30946197453535645, + 0.27934169533494807, + 0.3514897438141307, + 0.3550625820666356, + 0.40135800088523266, + 0.3447258022654254, + 0.3365873845238872, + 0.36954702024695835, + 0.29112052388718923, + 0.234340068777705, + 0.4044436149125074, + 0.37448123789507026, + 0.3299588882889645, + 0.376859417282474, + 0.46399075306790977, + 0.38721713443084405, + 0.2284312144624939, + 0.2650809840466341, + 0.35349946559959494, + 0.42264430290330496, + 0.4256159942828028, + 0.4129418220504466, + 0.4276819265075518, + 0.28498429073453657, + 0.4217022750516855, + 0.3761631909134284, + 0.302478819102815, + 0.35925840099433226, + 0.2640816203764239, + 0.46308522456594065, + 0.4009755897860191, + 0.3380818391300479, + 0.2672965641383211, + 0.2647397038122725, + 0.3651636764004503, + 0.3847460014061113, + 0.32509410434194624, + 0.37277651442457577, + 0.39818327637143786, + 0.2916758837770268, + 0.37501974211209466, + 0.19487853202455915, + 0.19761757446532915, + 0.18315960863022568, + 0.16996441817953045, + 0.24751415204839355, + 0.36105321670802826, + 0.3668229110902246, + 0.3155040729625298, + 0.17421651824522483, + 0.3423549479866653, + 0.4156020243233263, + 0.420294697130743, + 0.43056707786453174, + 0.3877287661456506, + 0.37742088842635646, + 0.3036350985730724, + 0.4187604069056086, + 0.42750963089531596, + 0.3888637430331472, + 0.4241076801330454, + 0.3831066782832672, + 0.305453823619629, + 0.3928514900035763, + 0.4006044485756443, + 0.46443997756120553, + 0.4360898561452745, + 0.3655889586563103, + 0.36880332777932734, + 0.41674239938709895, + 0.4623892291075823, + 0.3104683358680833, + 0.3222389291613105, + 0.44537652776594894, + 0.35940290001118186, + 0.3331360419446493, + 0.3508735716408694, + 0.46344068063974087, + 0.30910542238279537, + 0.411462671942941, + 0.40212651994326687, + 0.19382259042380537, + 0.3132775715945058, + 0.2551197324599976, + 0.39437742420178235, + 0.34842812654349087, + 0.38421538647981857, + 0.1864235081050826, + 0.42463871324189145, + 0.3825734620437602, + 0.41971356590387826, + 0.41171402159560966, + 0.2961800844119873, + 0.4568136269908182, + 0.2230869795951305, + 0.40575970611718454, + 0.3216416337175601, + 0.4580342597613145, + 0.41346979043210447, + 0.43485738662997886, + 0.48185340028491525, + 0.35516999159842255, + 0.39711135260100017, + 0.1877619906970264, + 0.41495858561299226, + 0.43650874959126756, + 0.3172321945737336, + 0.19845684910748124, + 0.20294746693097857, + 0.23454643066335265, + 0.3602982749676335, + 0.43462590598136847, + 0.22949139121402512, + 0.18191775733047377, + 0.3661958840908912, + 0.4082327647833924, + 0.35581534968348516, + 0.2606906646288595, + 0.12755581216723827, + 0.37222878189225883, + 0.2822866534717289, + 0.27236763413340337, + 0.36767615035286744, + 0.3479824902952963, + 0.40251334659617477, + 0.24403707677694797, + 0.25523301718452857, + 0.22833540320649123, + 0.24108303952749163, + 0.24674352842556535, + 0.3075934451819132, + 0.41212861509641135, + 0.3883984538631128, + 0.3818714479533804, + 0.3152560761743428, + 0.32115187901688697, + 0.4586295953879936, + 0.265151239985496, + 0.32408770793011166, + 0.40267167141098326, + 0.34448268765020323, + 0.3823351019057118, + 0.4010408238041987, + 0.4103913703494369, + 0.25014361252876344, + 0.4195856366953186, + 0.2707633373721356, + 0.24337741495504348, + 0.4245937488882283, + 0.3960658032023143, + 0.43383227626213244, + 0.45695553780337445, + 0.4668794788464033, + 0.42730561314483306, + 0.47779359367224866, + 0.3967894161066529, + 0.38969961462775216, + 0.3635438222864277, + 0.2795619058094343, + 0.34236797821488313, + 0.37574769075679754, + 0.4333838436804819, + 0.21546018502805683, + 0.28044796270742756, + 0.37464614413345004, + 0.4121265024344492, + 0.3985499802933538, + 0.29776457012763613, + 0.4698263125996336, + 0.44827594403002713, + 0.2949130534276596, + 0.4604984792630486, + 0.335206887266482, + 0.45131515129797756, + 0.42764448199442257, + 0.29542140686271, + 0.38154063214719264, + 0.33516927829848925, + 0.41381095144995356, + 0.44583070818631176, + 0.4353459162035526, + 0.2715148079571436, + 0.23517434451858923, + 0.19085661882116295, + 0.36105317414856763, + 0.4175583823082404, + 0.37470793519395634, + 0.3460483637608294, + 0.40811398546439553, + 0.3217217225379998, + 0.3675461670111855, + 0.36302678300562824, + 0.4255900794408324, + 0.4129885263731153, + 0.37377074632317747, + 0.2293218864002635, + 0.38211759562399505, + 0.47118321182852224, + 0.2314294585966316, + 0.3994463900627139, + 0.4059418860053654, + 0.46680616845138195, + 0.3035385170586375, + 0.42110778149297184, + 0.32936194603581354, + 0.37364416780740267, + 0.40028351940610374, + 0.20948512334929242, + 0.4014244813780954, + 0.3287161640040522, + 0.3804961742175125, + 0.40709435484506573, + 0.3569905384256439, + 0.23690894002289334, + 0.3932195385573974, + 0.39515598226372556, + 0.40384942830511006, + 0.4555754812351211, + 0.42720213425160714, + 0.4283044674440226, + 0.3019703568573579, + 0.3996846583259611, + 0.3411796400307811, + 0.33129079389813654, + 0.33439664594598784, + 0.34486386264948665, + 0.37968848275341166, + 0.3847763940569101, + 0.3023352296851272, + 0.38978464624409004, + 0.3592554923165045, + 0.3815887000691159, + 0.3587769003129858, + 0.4079751057713282, + 0.4019978282697572, + 0.3946517660570347, + 0.29731598510972235, + 0.3576160368810388, + 0.36199217361154323, + 0.4158067880546907, + 0.3694755096955216, + 0.4269663585029028, + 0.4737647321705784, + 0.4438410439246551, + 0.41331110116445163, + 0.3160053402132693, + 0.39638258117498354, + 0.4234276941937425, + 0.46499096184756844, + 0.37149588033220604, + 0.4538632293798107, + 0.3602935241074015, + 0.23468944993860272, + 0.4145750545918746, + 0.4138604932042305, + 0.3645457309380894, + 0.4303635346234888, + 0.32344955045059776, + 0.36430129194790617, + 0.37354581707602347, + 0.3596588134808929, + 0.47257617542644237, + 0.39573694300773166, + 0.3560220408165887, + 0.43638410354298396, + 0.341385631209211, + 0.3891909830279905, + 0.40038968315272916, + 0.37537787797627675, + 0.3652978774609345, + 0.43464637978383863, + 0.4797841755578225, + 0.34722463793901726, + 0.34429539845492074, + 0.290977800991222, + 0.33350655657436784, + 0.3674653067480494, + 0.4175323168200062, + 0.37514588263682463, + 0.3338465220803205, + 0.376271622272158, + 0.3246793346607501, + 0.22750687157815333, + 0.443219561153626, + 0.3800808807674182, + 0.33871386811466087, + 0.3775922384914403, + 0.46271681478889104, + 0.37662144132729686, + 0.21435669514743586, + 0.2710649357844821, + 0.36035689351982064, + 0.3923267826158592, + 0.43001617068906345, + 0.41235466165594686, + 0.4247198811717266, + 0.28680247333635495, + 0.42194534596026545, + 0.38247910305626653, + 0.34058555976579336, + 0.3700705566174192, + 0.2844281233378154, + 0.4610987551529721, + 0.40204449974778206, + 0.349967696245893, + 0.28982567306226886, + 0.2715527967075557, + 0.3205018074551948, + 0.405920065917935, + 0.34400109275362145, + 0.38984656346693713, + 0.40146183803316565, + 0.31023208097788446, + 0.3656768348131241, + 0.18731313165803637, + 0.18103515016996968, + 0.16472125565009751, + 0.15941028929642356, + 0.22189034784707126, + 0.37685343171225044, + 0.3432402792229782, + 0.3154497352022307, + 0.17290788329261583, + 0.36169955169448054, + 0.4324933050334896, + 0.4707083314603688, + 0.47425639363152033, + 0.41751420493079955, + 0.4089857981852178, + 0.3204778288974849, + 0.443457519355865, + 0.43533221691610063, + 0.3864744079385536, + 0.4284748174971231, + 0.40076665565131137, + 0.3434463091617187, + 0.4263506811775961, + 0.3972791987936314, + 0.5190604857608628, + 0.4389112411710613, + 0.409060167304029, + 0.37520898368410166, + 0.3953836020312061, + 0.42973141005530047, + 0.2668240956308426, + 0.2700414492502983, + 0.46467405021419866, + 0.3790019439420263, + 0.35007002431596046, + 0.3574198737391545, + 0.4694223966221002, + 0.338767481889429, + 0.39685126185257247, + 0.39093003792807157, + 0.187905738070875, + 0.30745132553708615, + 0.2532577051903625, + 0.41346574169084166, + 0.33134723774214236, + 0.38944672344967013, + 0.195875048460115, + 0.42262262898476377, + 0.3929417999870648, + 0.4164936967411523, + 0.42787137046258256, + 0.32624394931033907, + 0.49779444818808244, + 0.22007478768792052, + 0.39858644902112783, + 0.3101718554636888, + 0.44837520433471756, + 0.4395548184589524, + 0.4771201311184306, + 0.4694302300343529, + 0.37534239210662756, + 0.40042755080120823, + 0.17861647783852927, + 0.4041237289033895, + 0.4272792298884413, + 0.34211501291038043, + 0.1918842248376337, + 0.19204061359273525, + 0.22686579527986483, + 0.3667256025818978, + 0.4296664442294511, + 0.21641587208862073, + 0.16296085339865077 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "rna_seq": { + "mean_pearson": 0.6600583562190785, + "valid_tracks": 613, + "per_track_pearson": [ + 0.6327340703956594, + 0.717456660703047, + 0.6323240175476822, + 0.6761518401979278, + 0.6934746428684667, + 0.5096059221471781, + 0.6518541645157707, + 0.7057393225503579, + 0.6990931346963806, + 0.7117479105915211, + 0.6676481808153069, + 0.6450736818342759, + 0.5743074189767237, + 0.6971797404780969, + 0.6725830873538904, + 0.6833989328334179, + 0.717992876627545, + 0.6392137691696226, + 0.641867580187733, + 0.6447855756903058, + 0.6310063651383488, + 0.6990501266976892, + 0.6498111137441682, + 0.634768542831424, + 0.5298111611173385, + 0.7243569604990304, + 0.5965550657200838, + 0.6596855207482006, + 0.6346126315046747, + 0.6518473164214846, + 0.6720765325491613, + 0.6807095900775926, + 0.663382700892976, + 0.5879856661807603, + 0.6750800006433547, + 0.5593730837456117, + 0.6647026323310753, + 0.6695652736553351, + 0.621531267327006, + 0.6521795818435678, + 0.7104434580230593, + 0.6123946178081116, + 0.7188085939975974, + 0.7114996500589795, + 0.6571579637688464, + 0.5536498673616359, + 0.6409891870349302, + 0.7111023922191085, + 0.6693743005644675, + 0.7027214005543704, + 0.6893106145350281, + 0.6704089179901258, + 0.6079783650966839, + 0.6550002213476013, + 0.5355463641120594, + 0.6820353694320297, + 0.6456573497082931, + 0.721223977609925, + 0.7354025325656612, + 0.7545834433079655, + 0.7507191650043982, + 0.6876600291718912, + 0.7175219435242128, + 0.6319991386375124, + 0.7043348052158407, + 0.694492611278129, + 0.7177940327251304, + 0.7280509082945446, + 0.7087678424365247, + 0.7562728466812493, + 0.7252084862890422, + 0.72279791080147, + 0.6829882821692921, + 0.694449940511944, + 0.6928737982075723, + 0.6944576303273671, + 0.7118578437068991, + 0.7255917877462972, + 0.6790245012188661, + 0.7348690602073825, + 0.6925247746834635, + 0.6254658191241066, + 0.7471957682677219, + 0.7267342249385857, + 0.6338892097930047, + 0.680149651782065, + 0.7379140900002538, + 0.6598447662172294, + 0.719175253923445, + 0.6780837335556255, + 0.5829685097684645, + 0.7293265805434137, + 0.73624558561988, + 0.7197364943094051, + 0.7216395088935663, + 0.7303816583538723, + 0.7455410990738195, + 0.7258633762198422, + 0.6723876981819238, + 0.603207580324449, + 0.5887252681304187, + 0.6117187936230442, + 0.6431749111688672, + 0.7055545711095046, + 0.6280097734186143, + 0.6675743552052593, + 0.7230850672175356, + 0.6922281930740872, + 0.695085560698117, + 0.641939527808914, + 0.6870377895169533, + 0.689245377923845, + 0.6939324776244123, + 0.6769755815410164, + 0.6837636489661402, + 0.6946516745209033, + 0.6986306508090049, + 0.6858563128735251, + 0.6495225137250678, + 0.6761837943760279, + 0.6843673879833329, + 0.6341350747668489, + 0.6337052909245446, + 0.6628770179368563, + 0.7206693276284646, + 0.5742251764155656, + 0.5873584028713087, + 0.581288663995602, + 0.5930231414845429, + 0.6290419616019332, + 0.7201536228752953, + 0.6505266663616835, + 0.6484624656180286, + 0.6081932301194104, + 0.6431443979618455, + 0.7040818679475507, + 0.6744027272519901, + 0.6959103890785857, + 0.6415323467796036, + 0.6842360743702152, + 0.6176847141448357, + 0.6324313361051442, + 0.7055150925893827, + 0.6697983923466978, + 0.60495761969257, + 0.6169678270887601, + 0.6204817475254224, + 0.6466837134504859, + 0.5616349161338237, + 0.647451576765022, + 0.6093314425223488, + 0.6254978694997415, + 0.6546934278829143, + 0.5954382742041547, + 0.6296619231671411, + 0.6369971690072087, + 0.6273492847700405, + 0.6415569263700884, + 0.6314757957069675, + 0.7037425063688263, + 0.6937046947006235, + 0.686467943214255, + 0.6744345930964274, + 0.6026284945249367, + 0.5968514686859542, + 0.6016525685853431, + 0.6037867636705988, + 0.6231617751671722, + 0.6702658047284696, + 0.6657521872597565, + 0.7342631575052689, + 0.6207799966882155, + 0.7310297918473269, + 0.6745935582588637, + 0.6215756557519857, + 0.6349268646436769, + 0.602495945654889, + 0.6717167652722773, + 0.5707276822990579, + 0.5355432539504195, + 0.5547085085958352, + 0.7030608237384336, + 0.602975067213751, + 0.6932269685165499, + 0.5696446364348281, + 0.6499255166294581, + 0.6606397786770767, + 0.4883943836749828, + 0.6032912933568013, + 0.6360605268894883, + 0.7132265514008781, + 0.676717068903178, + 0.6410676835036601, + 0.5947475014528077, + 0.6846618030355546, + 0.6221349689450708, + 0.6578754983158166, + 0.43794058829911725, + 0.6573551806881975, + 0.6683792331852971, + 0.6528937898131351, + 0.7216031722965615, + 0.7013452023369833, + 0.6865297013848968, + 0.691549870175002, + 0.7010546049977724, + 0.6986332325785239, + 0.4237167919603562, + 0.6824320575844005, + 0.6673689572564823, + 0.6465406162617772, + 0.6227362692976958, + 0.6838112786037371, + 0.6916968991967062, + 0.6185713761986875, + 0.6980745221288256, + 0.6648066419157215, + 0.6150267910029354, + 0.7112966362613606, + 0.7158107365659607, + 0.7002509260202118, + 0.7713100891300939, + 0.6736972035932168, + 0.6771528727224686, + 0.6712013640892824, + 0.6528555687611114, + 0.6719114050030133, + 0.6522697975784586, + 0.5785704494982731, + 0.7452615095408226, + 0.6295964852407391, + 0.6847129129096031, + 0.6937351301573768, + 0.5713972372048163, + 0.7324138620366422, + 0.5373053238059481, + 0.5273891867947003, + 0.6927534692221404, + 0.7176473137404764, + 0.6858717521443088, + 0.6931152102662391, + 0.7403966859630755, + 0.6960084427878889, + 0.7483376399710869, + 0.3858896521226962, + 0.6255191751367314, + 0.6743566392426967, + 0.7027493490552696, + 0.6334018292837066, + 0.6197972637455691, + 0.62388577589392, + 0.6189154659336071, + 0.6257144497609449, + 0.631627257414446, + 0.6917838890901075, + 0.6509273121988225, + 0.5966826611092421, + 0.7043001066838862, + 0.6696600167082107, + 0.6911211022245735, + 0.615668104741598, + 0.6211057771018001, + 0.7356819435333064, + 0.7394793890845504, + 0.6802188902940164, + 0.7151203236145978, + 0.7572425540407328, + 0.5896344196069935, + 0.6879128714311745, + 0.6547976180777076, + 0.5259309697851468, + 0.6628737347550397, + 0.7447233547891209, + 0.7254202124869461, + 0.7292828354513547, + 0.726974160499434, + 0.4977319895026737, + 0.6287362095592577, + 0.7146795955643964, + 0.6872129902279797, + 0.7307463988535368, + 0.6488298617895601, + 0.6180515010602718, + 0.5849223888036363, + 0.6956606120802586, + 0.6663489766690763, + 0.6770434271831742, + 0.7191981990239381, + 0.7056304927180598, + 0.7274640606415415, + 0.6521276219517655, + 0.668064519358745, + 0.7274450435088997, + 0.6733626950335844, + 0.6163194450341808, + 0.5501210458299466, + 0.7243276820349208, + 0.7274634433905571, + 0.7370525060563768, + 0.6682881722150382, + 0.679442422360605, + 0.6499498966766434, + 0.6935405669404354, + 0.621261554078313, + 0.6545695451389232, + 0.7139698429071725, + 0.659624853903306, + 0.7217964828034403, + 0.7278610243907233, + 0.6919536467691066, + 0.7082230099729356, + 0.7153999404575025, + 0.6770949298629053, + 0.7062147798983951, + 0.6582013839492907, + 0.6249102313337918, + 0.6249392803940115, + 0.6633579551353663, + 0.73150008667215, + 0.6686437383308529, + 0.6756496948728002, + 0.7393634651008679, + 0.722137814123862, + 0.5795286856909031, + 0.6401629718415728, + 0.5122352577466471, + 0.6834277224102334, + 0.6594065589391509, + 0.7109158720249061, + 0.7474523323878199, + 0.695751227627671, + 0.7234982111601735, + 0.7363647227326158, + 0.7467803124453466, + 0.6342028238521843, + 0.737824748171023, + 0.7363529914726431, + 0.7538248953685742, + 0.7780026593731335, + 0.747362396643914, + 0.7784289703024706, + 0.7257000363783691, + 0.7386450056604291, + 0.6833826486057551, + 0.7517657511130522, + 0.7206498032352249, + 0.7126818385977337, + 0.7414825311390016, + 0.7652577872582851, + 0.627717109153483, + 0.7020606898260109, + 0.6714017130749429, + 0.6358412198071131, + 0.7583840861269651, + 0.7463186071892572, + 0.681644113173678, + 0.6657418349459027, + 0.6859914374114121, + 0.671446647480486, + 0.7319457739279661, + 0.65247204689899, + 0.6212241819464094, + 0.7303962991553281, + 0.6859912974352348, + 0.6925034180522289, + 0.6643456845107044, + 0.6910321617840285, + 0.6963746033936982, + 0.6848477508905831, + 0.6313489326248821, + 0.6167842332592571, + 0.5528601465908145, + 0.5680150944789872, + 0.5871614543015427, + 0.7554508364832827, + 0.6165478469223229, + 0.6397204968776029, + 0.7636783758034662, + 0.685139266807604, + 0.6891397967523245, + 0.6367749658047144, + 0.6880879423349177, + 0.6509505001693915, + 0.6713367262700511, + 0.7014167123646845, + 0.7671718092892481, + 0.7136821885268281, + 0.7183143635177571, + 0.6713523559382157, + 0.6311951322877948, + 0.6769227570460117, + 0.7551919146224225, + 0.6120324089701401, + 0.6311127246332511, + 0.6455912484627596, + 0.7472815672190658, + 0.5638621827310144, + 0.5709229921780117, + 0.5700697308334959, + 0.5661727770520724, + 0.65077818733704, + 0.7293161776292926, + 0.6328672033618679, + 0.6549793776653158, + 0.726652825414471, + 0.6979818543245746, + 0.7187384148439573, + 0.6651357792051685, + 0.7246454611647063, + 0.6617789990513916, + 0.6989007969344029, + 0.599780308101627, + 0.6520839684737438, + 0.7225074836310137, + 0.6711964174333721, + 0.5991889458339168, + 0.6013455599346105, + 0.6648289626988934, + 0.6972184116003154, + 0.6849003918863323, + 0.6150852916156635, + 0.593300224660758, + 0.6611191089309859, + 0.7174735007720907, + 0.5532266403610296, + 0.6138170472841566, + 0.6593296785916037, + 0.7128875639463919, + 0.6409135240574363, + 0.7052453403845216, + 0.6847237028585063, + 0.7020168673473184, + 0.7088304001932814, + 0.7076001503740914, + 0.5213476800022812, + 0.521890647012668, + 0.5248893396327701, + 0.525073994711083, + 0.6384771313974443, + 0.6664934733813962, + 0.662259766930497, + 0.7872395554179962, + 0.6638658144735061, + 0.7697220787768055, + 0.6652019805553213, + 0.640704163617709, + 0.6257951219727547, + 0.6064008036535017, + 0.7278524669010445, + 0.5443494024281531, + 0.5622973030233993, + 0.5357886869423243, + 0.7243473733178988, + 0.5780661339612084, + 0.6424435096381654, + 0.5780477838486163, + 0.6534301993174936, + 0.6915681902667471, + 0.498284872531514, + 0.6082426213375985, + 0.5875828760165589, + 0.6956124311539833, + 0.7062575573732959, + 0.60872282719353, + 0.5763472307913617, + 0.7127420698097211, + 0.6701614996443743, + 0.6680964637309096, + 0.43359510602666473, + 0.7007138278779326, + 0.6923214569980396, + 0.6353448328168303, + 0.7803074275015397, + 0.6853663118717735, + 0.7011701384774749, + 0.6992448294954027, + 0.6788482414958567, + 0.6999608373082202, + 0.36179140252583725, + 0.7172699082256594, + 0.6524587561701604, + 0.7699008536715424, + 0.7407829367025216, + 0.8076065805101489, + 0.7908543373451371, + 0.6484982257247541, + 0.7276749247751199, + 0.790643625168222, + 0.6695782356261251, + 0.8116035088581623, + 0.688264296802495, + 0.6524682988445526, + 0.75127326296255, + 0.6338764102109347, + 0.6245720813525415, + 0.6178977564550522, + 0.5881837803958457, + 0.5777519655587969, + 0.6016848552501144, + 0.5077879392692827, + 0.7733113866075702, + 0.5593133477044627, + 0.6686617832812648, + 0.7561369131355601, + 0.575047147888804, + 0.7394689701351361, + 0.517460391856353, + 0.5370928194750723, + 0.7104016395794498, + 0.6893507397118108, + 0.6412231300325885, + 0.6505169093772727, + 0.739880112474603, + 0.7445609143637011, + 0.8055164358427489, + 0.38316421309447896, + 0.6574093787903719, + 0.6715637356310707, + 0.7005476888912628, + 0.6802861795286779, + 0.6211744294086519, + 0.6349516597197845, + 0.629872600496039, + 0.6408628822382099, + 0.642724985290777, + 0.6948242718531413, + 0.6256843623347202, + 0.5473366552460753, + 0.7150034560296314, + 0.6739713205524483, + 0.7617063154861147, + 0.5731309725423988, + 0.5236640876641934, + 0.6872646166916192, + 0.698002322331914, + 0.6780313198472353, + 0.7500372334014447, + 0.7700860938462697, + 0.542546046072885, + 0.7027549641922163, + 0.6745506971474342, + 0.5196177102838438, + 0.6956048775400719, + 0.41120198727138857, + 0.5904783076937974, + 0.647524184812127, + 0.5995199501644525, + 0.6561542800493598, + 0.6342391779974293, + 0.6697990182066299, + 0.6800671618265838, + 0.5889142798146091, + 0.5646158631991772, + 0.5705759491287792, + 0.5658706622551032, + 0.5518238975726586, + 0.5576747584546784, + 0.5634549076925773, + 0.5606434465233432, + 0.5346818479062276, + 0.5243358606786495, + 0.6590279250105657, + 0.6493050791307271, + 0.6368681569247948, + 0.6448460253756828, + 0.6582489904919485, + 0.631625522375303, + null, + 0.6457270869778103, + 0.6974842055171508, + null, + 0.5469452386313942, + 0.6912050288485156, + 0.549154166604739, + 0.6748457413794468, + 0.6920600287053714, + 0.6910466641172932, + 0.689360826707657, + 0.6822798319466651, + 0.6754359936665877, + 0.6685987302025815, + 0.6387402458822493, + 0.6423917245637498, + 0.594099282610472, + 0.6471390409564163, + 0.5462215160329799, + 0.6654561304095032, + 0.6543059060497138, + null, + 0.654457617252566, + null, + null, + 0.6512006595449868, + null, + 0.6658065111158946, + 0.6467673107367511, + null, + 0.6804623323456556, + null, + null, + null, + null, + null, + null, + null, + 0.6642887155949443, + null, + null, + null, + null, + 0.5837554835730899, + null, + 0.6639827982905584, + null, + 0.6664855305634809, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6851808127631099, + 0.5279130087282468, + 0.6375054777189988, + 0.6670241081168513, + 0.586376732979916, + 0.6905589126685296, + null, + 0.6740815434660217, + null, + 0.6743748860290056, + null, + null, + 0.683099577902816, + 0.6925908070184187, + 0.6272700849845237, + 0.613493482200948, + null, + null, + 0.6312974887459014, + null, + null, + 0.6658109864227857, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + null, + 0.6748213511897735, + 0.6627281164794123, + 0.6748424484505142, + 0.6488392456186071, + null + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 0, + 307200, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 0, + 307200, + 0, + 307200, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0, + 0, + 307200, + 0, + 0, + 307200, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 0, + 307200, + 307200, + 307200, + 307200, + 0 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_tf": { + "mean_pearson": 0.47541400655136246, + "valid_tracks": 1617, + "per_track_pearson": [ + 0.25517189809856433, + 0.39916218278800264, + 0.32767340294109987, + 0.49678731625966616, + 0.4412684840711386, + 0.3314634550318222, + 0.4325219607462886, + 0.5046318103622641, + 0.4220852544940474, + 0.44842244415881743, + 0.41783576772500375, + 0.439621334615864, + 0.30876529881955883, + 0.5198623533263764, + 0.4314072879782416, + 0.4394716049319136, + 0.5152891983622996, + 0.39380958571522673, + 0.529695430717157, + 0.42882907205919074, + 0.4221713665036869, + 0.36996988190789615, + 0.3410461302586996, + 0.5552262105650587, + 0.40957866705381357, + 0.457225027174918, + 0.372219521328689, + 0.3993388938705427, + 0.4388361534760067, + 0.45898726822599595, + 0.4606759147606478, + 0.444729518539125, + 0.5131246902771047, + 0.3086529926430282, + 0.6876988032078614, + 0.30184623930744403, + 0.2686936287477081, + 0.19824806026054118, + 0.35536550698671165, + 0.24431780602312306, + 0.29972732990910267, + 0.4824794118596634, + 0.32988368832418513, + 0.4327718940137604, + 0.31813294173009454, + 0.4181521630855069, + 0.32267202286103236, + 0.2424714249392864, + 0.4880673595122921, + 0.4602610821711112, + 0.4100448639797275, + 0.38330463283821053, + 0.4049769694225413, + 0.44804256921811864, + 0.4420057034202289, + 0.45206711749051204, + 0.37748151310624395, + 0.5441416297154155, + 0.5524152146764014, + 0.5532994232589007, + 0.3461451231280925, + 0.22488854840923023, + 0.41808205178323304, + 0.44871458524178404, + 0.292682674952511, + 0.3755533319389753, + 0.37221426773547217, + 0.49304589994305026, + 0.5144767413834275, + 0.4640459244678194, + 0.38394430342659713, + 0.7279548038894711, + 0.37337652799090415, + 0.399027013967689, + 0.39016613030203134, + 0.4512855874343059, + 0.3238652384709625, + 0.675442832522403, + 0.43550777563038173, + 0.4532781795323565, + 0.41633934165870207, + 0.4336264028548437, + 0.5018625727128713, + 0.4637525571438754, + 0.3802011259128012, + 0.33720424146493433, + 0.4681345266154454, + 0.4849390264711645, + 0.8133381612213478, + 0.5040731636765178, + 0.4044408891037079, + 0.3638616532665235, + 0.4841469271427451, + 0.39753080041150846, + 0.44214083699540907, + 0.5719922858114176, + 0.3572699260126672, + 0.3986545958380482, + 0.40674197608414103, + 0.34553594994087017, + 0.2657843452678443, + 0.39743289747962185, + 0.35399544234355473, + 0.3383025393828596, + 0.38615648122158347, + 0.27184734426898877, + 0.34097355115431893, + 0.36165045788244615, + 0.3936671365209007, + 0.44831693228003183, + 0.2765289219814156, + 0.3820427870621779, + 0.358488787304867, + 0.4086877181176399, + 0.4552129827018812, + 0.4598500051876293, + 0.4230565604153857, + 0.4366822622138477, + 0.4092493635389054, + 0.4627611156364588, + 0.4182301053894794, + 0.4521604655501028, + 0.4332205620301543, + 0.35277831291637607, + 0.41697063858502303, + 0.36967258059856106, + 0.40765622332001633, + 0.37862255975637915, + 0.40721484674008057, + 0.5492289112534137, + 0.4677267389675732, + 0.39187307137116256, + 0.4451745779326081, + 0.6426009893495884, + 0.38307575800857296, + 0.4440590727566473, + 0.45516831141710606, + 0.34157984965321336, + 0.3974641275307591, + 0.3755344242960552, + 0.4150520532760907, + 0.4561020136988889, + 0.5824428945808588, + 0.3929179441739022, + 0.44903320107950284, + 0.5254246551157735, + 0.4708021978716365, + 0.33367646971662496, + 0.445231672636753, + 0.48902446586256787, + 0.4629829947646218, + 0.3881348105784964, + 0.4995204975888346, + 0.5058559728555464, + 0.4481245136813896, + 0.44071213374148666, + 0.36415974783205163, + 0.413128271991793, + 0.41910280670499356, + 0.4678018310178997, + 0.47827919779414024, + 0.4057227849885777, + 0.432753557204315, + 0.4386177730618453, + 0.4640401524714724, + 0.4116264945754788, + 0.4121908625590561, + 0.4954007088196769, + 0.34074544180081734, + 0.46485253607015525, + 0.44540409644820467, + 0.39073595250717125, + 0.5127553996153206, + 0.39818628848135573, + 0.36911681828487986, + 0.4551197839189689, + 0.32526151918536284, + 0.4856427665199551, + 0.509629497581123, + 0.3735524243395724, + 0.4766293011095528, + 0.455549090588774, + 0.37103237224864, + 0.4467900515830107, + 0.38317959705631893, + 0.4814532206109307, + 0.4693784192371023, + 0.34270606282425964, + 0.3759807125523115, + 0.4398165326844583, + 0.38072691666993336, + 0.39856145146632604, + 0.45974197199760275, + 0.5173336810492177, + 0.39083831732017005, + 0.38837048590901324, + 0.3989515153368238, + 0.3964874339662448, + 0.3088415195392782, + 0.49643521601083745, + 0.41002627510129924, + 0.3813549569300328, + 0.42413773405688976, + 0.3723696686637515, + 0.47400612819578625, + 0.41739708205617176, + 0.4973797339314973, + 0.4069203582533536, + 0.3670756824927891, + 0.5217787213149291, + 0.4036507308252393, + 0.4766381839849881, + 0.45146394932291306, + 0.4614880046410558, + 0.8768136153080776, + 0.5002337641753958, + 0.42896832336428947, + 0.47901573395505015, + 0.4838196320865539, + 0.5634609671549922, + 0.6254206672158915, + 0.6453850895104143, + 0.461855201241965, + 0.4176301341113457, + 0.5107811103591778, + 0.742854274043237, + 0.3894528725238051, + 0.42087060816389643, + 0.5618753959118208, + 0.5372805864445646, + 0.6022519371064045, + 0.4820812168256878, + 0.3668944069093116, + 0.4344972741441817, + 0.6151215909139097, + 0.47888170933543156, + 0.39780967507417214, + 0.6268529706360138, + 0.5680072300622431, + 0.47330590300518, + 0.5137315850596492, + 0.7215607521567908, + 0.5095947717339087, + 0.5373925586473116, + 0.46686272525219275, + 0.771659918003735, + 0.36278952159713557, + 0.6516499777973723, + 0.43181886329322855, + 0.5633529149790444, + 0.46271039913015943, + 0.4015116653773119, + 0.7108120365511954, + 0.5329619214768662, + 0.37491943557484714, + 0.3988869764148729, + 0.6578552234527377, + 0.418540526420485, + 0.4629848288343555, + 0.518241262670061, + 0.4223888708029445, + 0.516889382849098, + 0.5195225185549421, + 0.4539242988601468, + 0.4136931400707951, + 0.4900095918985254, + 0.39416450175240964, + 0.37481861609784944, + 0.6414536140227235, + 0.4087744477675542, + 0.47178711047051747, + 0.5923013061833939, + 0.4274222453393411, + 0.6701953245963654, + 0.3936234360054639, + 0.3943452147938546, + 0.5732663100361413, + 0.5152715622017913, + 0.5942561441677539, + 0.7387654455055739, + 0.6819097843446761, + 0.4413293518167157, + 0.48980155919185137, + 0.5840941906542859, + 0.3909067715460234, + 0.374945693347733, + 0.5798249022426897, + 0.5552828166635061, + 0.7503799358939564, + 0.5131342701468561, + 0.39628437449273046, + 0.4867758469694823, + 0.39740139545114, + 0.49002494595500407, + 0.7158877406014439, + 0.6402172810480932, + 0.5679410797836472, + 0.5524851226933031, + 0.5187382922512238, + 0.5306566810976253, + 0.7495493597238212, + 0.33487103687044084, + 0.4687716216336358, + 0.36259435671252593, + 0.6242385409098247, + 0.41749283275262394, + 0.7461401488948981, + 0.45737737777766446, + 0.4595592460158783, + 0.7081895837614149, + 0.5339891823624818, + 0.3678786856589323, + 0.5815179434291021, + 0.4315257331054949, + 0.52858786646811, + 0.5122783117613815, + 0.37380685357639487, + 0.521336653110463, + 0.4307109401586678, + 0.47538896495283656, + 0.44390818340772226, + 0.3911762972815761, + 0.44409421589265335, + 0.548006831317015, + 0.6044404297077282, + 0.6761233111703719, + 0.4546397489763626, + 0.33983677279687974, + 0.4834988734353272, + 0.49160139472940295, + 0.45052051440474616, + 0.5017853240951741, + 0.5138986526028395, + 0.40918869877794867, + 0.3631834182466041, + 0.5357245407842417, + 0.7709872502465114, + 0.5453550844837197, + 0.6191469606795164, + 0.6035732176735834, + 0.7797747972818995, + 0.5889038349096211, + 0.39193785780812135, + 0.6046654452926667, + 0.4188981967998742, + 0.5618336030823792, + 0.5085991663747294, + 0.41865851331488213, + 0.5176686109399959, + 0.7918613728395373, + 0.2642261152798627, + 0.49077586950094265, + 0.49536663869431513, + 0.6365311155608785, + 0.40658616125622943, + 0.3544540580869931, + 0.4587870357512967, + 0.7305460013410948, + 0.42168608842436084, + 0.46706861129320554, + 0.5242061536292337, + 0.4789496129704301, + 0.45467860416605216, + 0.5217746947885208, + 0.6284891217476675, + 0.4814883708730034, + 0.43880047833450625, + 0.5707031139426326, + 0.4768123504911135, + 0.587156444680966, + 0.5768944291196295, + 0.5530246881401817, + 0.5708386722533894, + 0.370621867655479, + 0.39675857388922425, + 0.589987120909385, + 0.3528008213306249, + 0.7069807181495336, + 0.3546657699220182, + 0.5794502698137348, + 0.6125497051426056, + 0.6972278075454814, + 0.3387944817298646, + 0.4067952805305334, + 0.3756337726653962, + 0.4288924895462446, + 0.3581564275254731, + 0.33984500312126215, + 0.6353929135083463, + 0.42386959539283503, + 0.6899082256940501, + 0.5989977972966889, + 0.48491681293431166, + 0.6085857047359378, + 0.40463278693480775, + 0.5738852250669886, + 0.5574706557210038, + 0.4936631909984911, + 0.41236021815706697, + 0.4856397481420601, + 0.5322206844676568, + 0.5506661548145252, + 0.41879312255741646, + 0.38713806214187974, + 0.6452696328290423, + 0.49719098748871315, + 0.6668565013828771, + 0.6620665013629297, + 0.5430560498054505, + 0.3652149754829327, + 0.40181938774910436, + 0.3961453136389008, + 0.5349991879556611, + 0.7545752712291427, + 0.4932337953726467, + 0.40610928589270123, + 0.75067951987729, + 0.47188210868115554, + 0.6359359460785128, + 0.5500878385257406, + 0.4623632066694581, + 0.5185990074734295, + 0.4836567226786427, + 0.49739674387683896, + 0.5899501104180132, + 0.5623933150065976, + 0.5500670714648893, + 0.4735379582395007, + 0.4928303426236974, + 0.5538684079029208, + 0.5137921404489137, + 0.7742210938166169, + 0.510778887368875, + 0.5121235337429753, + 0.46429757851783543, + 0.338769494422384, + 0.6735143756159871, + 0.437431945449673, + 0.3812562877900248, + 0.5358495647371896, + 0.4891806869579115, + 0.37718081647960117, + 0.49907866745669843, + 0.46306514282954603, + 0.785025535459062, + 0.3340817824439505, + 0.42517289940620717, + 0.39391124594155263, + 0.4526385911747247, + 0.42530827207109145, + 0.3461092033533207, + 0.45110288629431244, + 0.3031677588335894, + 0.446764339500585, + 0.7541419410279312, + 0.3574742701207001, + 0.42857161338150973, + 0.5211624218395611, + 0.43814323247239273, + 0.4273591381288014, + 0.5002318616519444, + 0.3809520911879348, + 0.5380785538889395, + 0.5504581990049453, + 0.2872949182796575, + 0.650364206152602, + 0.7010691574204384, + 0.524647502659465, + 0.5046480277137909, + 0.5554522457481016, + 0.4539802903246513, + 0.477036649521922, + 0.5852392384175821, + 0.7878178344034051, + 0.43071980013561184, + 0.6194392468854716, + 0.37683509556604006, + 0.4773740493468751, + 0.5060204276723062, + 0.5111893229787511, + 0.4234535586401973, + 0.3764483704468121, + 0.3724890568339119, + 0.4183911654973245, + 0.5955844463763376, + 0.4894404667509297, + 0.3665645254628225, + 0.42618306823467056, + 0.4807580343389195, + 0.3346588624432554, + 0.5674000219476197, + 0.5145005967825209, + 0.4508978216271237, + 0.47185280199270696, + 0.3588605816654816, + 0.44870293343768825, + 0.5226048887279169, + 0.6008768827157677, + 0.44525045049700196, + 0.7884029754954495, + 0.3986474310014413, + 0.5682716263491302, + 0.7485411758953985, + 0.6450959684945814, + 0.3417282626902125, + 0.5270782257368948, + 0.4356534233787934, + 0.3814920446124535, + 0.40226659617531557, + 0.4165779425694296, + 0.4877954503293614, + 0.511942637731364, + 0.39996814289093574, + 0.6289424257416629, + 0.7763296643555164, + 0.45742876971559726, + 0.47259135682280407, + 0.4570493089035027, + 0.5248565532426018, + 0.5393749200201192, + 0.5046512162830091, + 0.5203339566965783, + 0.39271042429832, + 0.6751348077288102, + 0.5038303803723676, + 0.5682608744075456, + 0.7235022974435753, + 0.35346033078851763, + 0.6033661064204439, + 0.407133090833548, + 0.48089816066294777, + 0.3688492933740436, + 0.5339634016821316, + 0.5184986743328345, + 0.36313285708195947, + 0.4629090040226758, + 0.4294351490378409, + 0.5502478655582171, + 0.607768712143772, + 0.4852674167578981, + 0.486491190310534, + 0.43445749987600396, + 0.7490462757403351, + 0.5341849821520002, + 0.5582880323006173, + 0.4856691006799095, + 0.4807394002709835, + 0.6265655387659237, + 0.383495811994432, + 0.583653619961859, + 0.4045219337273224, + 0.5652616791439118, + 0.44773768135422115, + 0.3774884855410819, + 0.39512880385445165, + 0.4958343433956355, + 0.39949102099211026, + 0.5227167630374692, + 0.7433174238420002, + 0.7363799811481576, + 0.4307056488947158, + 0.39228929095564835, + 0.7098905641583817, + 0.46452296629575524, + 0.4132922983631986, + 0.546620153265604, + 0.3825501635780235, + 0.610617296502785, + 0.4488348369065547, + 0.5824091090022726, + 0.6006670614236338, + 0.4760557959131643, + 0.5293931108121189, + 0.6654501467325857, + 0.5406258658676971, + 0.3918689866964547, + 0.45034321575383934, + 0.33633867660109207, + 0.5553874395884683, + 0.3712269344282328, + 0.5353939043606966, + 0.5185347832915451, + 0.4214576486966215, + 0.7305308296188578, + 0.3553610123549158, + 0.7358389860193116, + 0.4704274502978063, + 0.5537639437888396, + 0.7521974116550488, + 0.49970445902125277, + 0.5677706560178234, + 0.44530847253884426, + 0.7142472679751543, + 0.43930613942875846, + 0.3623063902926628, + 0.580883057806174, + 0.5390412373804858, + 0.651841485623282, + 0.5636813582954388, + 0.4268048056584494, + 0.5014199718977538, + 0.3329576593239576, + 0.48490772337812604, + 0.5791064011108248, + 0.5325247088393851, + 0.3828817173244606, + 0.5641042832653194, + 0.6027084838755897, + 0.5828283671587401, + 0.459139005304309, + 0.4336497605337561, + 0.5234813727588326, + 0.48078240027157765, + 0.5700084307395836, + 0.5128020038942829, + 0.3932428037949111, + 0.4320996804297902, + 0.5288339449404785, + 0.42345183011363197, + 0.7449452853038668, + 0.5392170292958036, + 0.4744308312766131, + 0.5150439697576057, + 0.42718042180077, + 0.46549715638074896, + 0.5535906057474324, + 0.5658177522065259, + 0.7182464310667495, + 0.5577106270505687, + 0.5882616389058175, + 0.5451771143000109, + 0.5182097041566861, + 0.5896368592134524, + 0.4312752353320203, + 0.6505157076523354, + 0.6073931347657331, + 0.733379868932147, + 0.539601581936267, + 0.5604408205973547, + 0.5525298844483825, + 0.46874606147001285, + 0.550557711875903, + 0.3778996785173649, + 0.6930675099234036, + 0.5087613937386654, + 0.7438130413069552, + 0.7728196389271218, + 0.5119747122985797, + 0.4753188025993672, + 0.5189989904480088, + 0.3916229612709524, + 0.52145874885578, + 0.7664627249151323, + 0.5034154231639913, + 0.5363203584087937, + 0.598847465928144, + 0.5433415127368526, + 0.376834058325216, + 0.6243754780345857, + 0.4534406559206625, + 0.41625202864534355, + 0.7430193403655088, + 0.4374224500822086, + 0.4631968292538281, + 0.4266199737113363, + 0.5002020234006884, + 0.5750573537515526, + 0.6433515357096515, + 0.687420712143845, + 0.6330701112129984, + 0.6010172526996901, + 0.4822974090629012, + 0.6201631368199363, + 0.5772666521620821, + 0.7871212686580076, + 0.7552456803429185, + 0.39548874652185356, + 0.40945137459874636, + 0.3909196312472344, + 0.5379647295720845, + 0.4354172501857094, + 0.7963115785089339, + 0.40099134667726494, + 0.44073104336895597, + 0.5920381716998697, + 0.504007541956734, + 0.5837378983337428, + 0.6375846862984107, + 0.4906115776913026, + 0.6299415598379537, + 0.4041671525675464, + 0.6981809488567968, + 0.7903478750393221, + 0.7721448341740458, + 0.6000843057280426, + 0.4557280426399309, + 0.5779652272978436, + 0.7535652714409674, + 0.7016810954384903, + 0.6382744190750815, + 0.5117489306225382, + 0.49019524810445086, + 0.5098714464268984, + 0.7457762023616109, + 0.45014854685035977, + 0.46164820559087705, + 0.3514921865074817, + 0.614747524750115, + 0.3490657413861565, + 0.442814781642025, + 0.691900353888037, + 0.5548710274610866, + 0.6950657452234329, + 0.5638833992614729, + 0.5398072603422688, + 0.6777811634468061, + 0.645724739006067, + 0.5806054069007279, + 0.4069127901774884, + 0.6599649576122398, + 0.5395639823216163, + 0.4284390525883295, + 0.38176224615736104, + 0.502634176263635, + 0.6522955898454196, + 0.5372448051800213, + 0.6276557358892889, + 0.591029965992852, + 0.4481273987566586, + 0.4342416554670125, + 0.6995745763904817, + 0.5224999692160881, + 0.6272797982312517, + 0.7322513703827876, + 0.4365280162531752, + 0.41316235298856496, + 0.411480246325842, + 0.4058672226467809, + 0.7235240927098306, + 0.7438090653711322, + 0.5788038649011165, + 0.4438511921173623, + 0.49696571104331155, + 0.41400052762877426, + 0.7914994052722347, + 0.3790214106283004, + 0.5052665609901401, + 0.36537850889073054, + 0.5301759609266055, + 0.4038242736369891, + 0.5819739505129623, + 0.5110022472157042, + 0.38551704190813874, + 0.709238009253548, + 0.5300918930901705, + 0.4490964438357113, + 0.3682951227000881, + 0.3798703519951169, + 0.43074234577404447, + 0.44561145030039795, + 0.39816723322636793, + 0.39060487786344367, + 0.4265389382758286, + 0.42064206653794495, + 0.43899911713605927, + 0.36445539620638834, + 0.46863245402922415, + 0.22760058805428923, + 0.30255843357227996, + 0.381448497932984, + 0.4019741667211752, + 0.464264293569427, + 0.46084414559464104, + 0.3950916803263349, + 0.4171349661596651, + 0.38243006258083845, + 0.6924571065851346, + 0.3711774054977634, + 0.35181502007805976, + 0.7968932001113228, + 0.38189378036016636, + 0.4586069654942354, + 0.37689234314093795, + 0.4588453830289401, + 0.31722783507566443, + 0.4382504082267958, + 0.3708231966731226, + 0.5063508137240823, + 0.2894830542123254, + 0.42724275603876954, + 0.4129020952666821, + 0.5272924219350683, + 0.42599321916498994, + 0.4828306940640187, + 0.3207927375336392, + 0.4959429413554076, + 0.4331726116303406, + 0.3857614951043371, + 0.483958983149917, + 0.4138620342261387, + 0.4854022421154248, + 0.42247932640778835, + 0.3992459100473911, + 0.8142044328043008, + 0.34603869820556304, + 0.49595850426853166, + 0.37854845594991793, + 0.436257948687444, + 0.35384434492049893, + 0.42455513216257834, + 0.4684802424877158, + 0.5750068222837331, + 0.3544751261043594, + 0.4014108137370117, + 0.31488629196828727, + 0.42297887916066174, + 0.4818578916654296, + 0.46075776329558377, + 0.4887464791493528, + 0.3863731756128858, + 0.3856327613791608, + 0.4773756526147698, + 0.44639165773596406, + 0.46086616619153525, + 0.33924913339322277, + 0.3327005790845971, + 0.5012645903343254, + 0.4659039867093154, + 0.43376098384092815, + 0.4033500008664893, + 0.4255681635036544, + 0.8479548146595285, + 0.3428488727240656, + 0.3773462496022085, + 0.3823398965118167, + 0.3617272936670979, + 0.3974245935170822, + 0.3964195321353808, + 0.3537698268843945, + 0.37925772824753806, + 0.3648998664228977, + 0.4972179762947121, + 0.4816240148958224, + 0.46925963140357857, + 0.49614147681208565, + 0.5075456703325814, + 0.41157009560348284, + 0.5813265973620763, + 0.46792759187428834, + 0.5025316045126648, + 0.5186298605206494, + 0.4762691039119908, + 0.4485310337460181, + 0.3501868135038463, + 0.4480596271453226, + 0.5557479841909291, + 0.5113477647693527, + 0.3824527600603605, + 0.4148033656266776, + 0.445964769665007, + 0.497077406316989, + 0.38230733907679865, + 0.43488598704529613, + 0.5159653991987961, + 0.5680767841916053, + 0.3859510872974381, + 0.4468790326352887, + 0.39692179296692043, + 0.3941730802773613, + 0.38774939450843765, + 0.7996281493411679, + 0.4335637360048733, + 0.4732932428078523, + 0.39250571904385145, + 0.4026862936309002, + 0.43206939660984134, + 0.7152619993294493, + 0.47219147589331634, + 0.4658596644867838, + 0.4692705322572408, + 0.4425617457097352, + 0.4874605720354865, + 0.48642297061154743, + 0.5193382885446353, + 0.31944775270791, + 0.39557516276696136, + 0.471535518904533, + 0.479498899530194, + 0.43962717667346146, + 0.4435419235412677, + 0.44043099982597544, + 0.3816178116171249, + 0.42266670018559027, + 0.4291660351751576, + 0.43792745160646707, + 0.6262973856206229, + 0.40684216138080376, + 0.44510732340008513, + 0.3705560981458082, + 0.607603335726931, + 0.4280591628419729, + 0.4034452460243151, + 0.46971008223682015, + 0.48582428908075465, + 0.4734549795548216, + 0.3980186151108597, + 0.5752520784673281, + 0.5043724486786906, + 0.3800882480268147, + 0.4756044218805502, + 0.47339774334849805, + 0.5262960435059879, + 0.42534653319071447, + 0.48550214538833564, + 0.3781814899787478, + 0.6713369528843885, + 0.4150850840583401, + 0.4730159892657658, + 0.5015766270240678, + 0.4497397332598294, + 0.4427876265478627, + 0.4773198710735365, + 0.36562216742634424, + 0.4229509664907851, + 0.6164572505716774, + 0.42163140103121377, + 0.4092178494578861, + 0.5371865009231458, + 0.44564888366516897, + 0.43660549440112273, + 0.6436231627054598, + 0.44395777189869023, + 0.5131108700301913, + 0.46878016219683355, + 0.4350818203340851, + 0.5306780744660741, + 0.43225934730336396, + 0.5159842399858178, + 0.44170371664484687, + 0.5086390659310164, + 0.40016638939347954, + 0.5125791447212282, + 0.4599198308837946, + 0.3781629806634326, + 0.3552335328977885, + 0.42802759170036114, + 0.45643551242844077, + 0.5739863922565778, + 0.4844176226734684, + 0.47405123079418077, + 0.517322299864679, + 0.42750087852194685, + 0.42266669389206174, + 0.50649098657794, + 0.3911833684749409, + 0.43125896697585303, + 0.5146838268402875, + 0.42817743275358644, + 0.5418748773151711, + 0.5194666089654337, + 0.4275214291557029, + 0.4329949094948624, + 0.5380344617921046, + 0.4530997094697295, + 0.46528356649334074, + 0.37546770087023484, + 0.5240334365863364, + 0.6153152866422424, + 0.8024726014597254, + 0.4529056517711986, + 0.4466437913058295, + 0.45012187672093895, + 0.4182341085581914, + 0.3881025526832581, + 0.40335190859562764, + 0.49178787008113223, + 0.501407523828506, + 0.45365452599797146, + 0.46938115499047434, + 0.5237824299368538, + 0.549855391873032, + 0.40064687966101564, + 0.39323744352259715, + 0.4880551029920812, + 0.4375187731149091, + 0.41330138391746657, + 0.45571226092410017, + 0.4214660203576321, + 0.43812957567337685, + 0.3430670291476891, + 0.5386699252817871, + 0.4624374321292006, + 0.4267460185842454, + 0.4273301314399481, + 0.3426384275740828, + 0.49509367642085966, + 0.3268392885464573, + 0.5584207996141556, + 0.44594028065065383, + 0.37520445616626097, + 0.4947778302855973, + 0.4633636408012427, + 0.38195250836755507, + 0.623134151848035, + 0.5544900004674828, + 0.3729517000162521, + 0.5099403994516827, + 0.4427717987684391, + 0.4689992500820653, + 0.4879752775950651, + 0.3359377369752912, + 0.4265996236348812, + 0.5299452818864786, + 0.4114988081106328, + 0.46956897594198493, + 0.452260398046674, + 0.4841779322135153, + 0.5017125634438497, + 0.4279336645376493, + 0.5311633657795536, + 0.495079343774524, + 0.4312177077237565, + 0.3671912415985768, + 0.48971523715423887, + 0.39166514923798834, + 0.42805499503260264, + 0.44212936847889134, + 0.49962589526812523, + 0.5589119774117783, + 0.48709835764099024, + 0.4444306327345199, + 0.5139740229216013, + 0.4679510914966262, + 0.4625955007793538, + 0.5444092337216627, + 0.4253030747400077, + 0.35702674754085556, + 0.5135034505783995, + 0.4557117378130025, + 0.47404555865784165, + 0.46365909498372365, + 0.4695983107566557, + 0.5196674859333097, + 0.3987273282793022, + 0.47299665013774256, + 0.3818724328764742, + 0.3926398303013503, + 0.3717344938579728, + 0.4533475004134967, + 0.39197752058565455, + 0.46688630579246426, + 0.5427606881160028, + 0.47043243605480334, + 0.5056145525478682, + 0.7833530775156119, + 0.579058496834913, + 0.4748820105515134, + 0.43271569663539583, + 0.4846367632607445, + 0.5046239072869697, + 0.37604248505078897, + 0.5903395076700283, + 0.4985863277719141, + 0.49428674386862625, + 0.4921282152294305, + 0.43142892091260604, + 0.5235193764992534, + 0.529967091260593, + 0.43890200531670404, + 0.5577149902782087, + 0.5354499087938418, + 0.4153123039228824, + 0.46497052214941886, + 0.5404484033308008, + 0.45753936319869454, + 0.48979411849518406, + 0.3801619367295917, + 0.5744461863707876, + 0.539349966550873, + 0.8289068678642247, + 0.5533534410864562, + 0.47657002222459616, + 0.29951052865354716, + 0.49127776173168136, + 0.44527768491192465, + 0.6223389360311824, + 0.37132795820799713, + 0.5718811124927585, + 0.3676515020606737, + 0.42025496129462553, + 0.40869395109380546, + 0.39165939233178293, + 0.4754502538012313, + 0.3975179664710075, + 0.4584573554946102, + 0.4686503865521, + 0.44811709897149316, + 0.4027515421449576, + 0.39898383864394554, + 0.4395384185583313, + 0.3920986651645083, + 0.576329027128131, + 0.44740253853969025, + 0.538275165825377, + 0.4456196117439836, + 0.4076032669471631, + 0.5770820358988182, + 0.5006347244456812, + 0.4051652454371679, + 0.6094378656295086, + 0.3433335233928691, + 0.4820693355722714, + 0.3905681807610278, + 0.4225801835670555, + 0.3565054581307753, + 0.42581893612367266, + 0.5185591521943499, + 0.38003516098066664, + 0.7025900832348976, + 0.4735476034493045, + 0.4069996149047876, + 0.4397916704870429, + 0.41627422058776187, + 0.6154073967252478, + 0.3524378753225843, + 0.41901128455905085, + 0.4121395423926951, + 0.4673772995009259, + 0.39959595645812257, + 0.44141393202831253, + 0.34180875971637675, + 0.48559377727181413, + 0.7738157747433675, + 0.5271639677144021, + 0.4083168093630273, + 0.4304444836367993, + 0.46114726652015975, + 0.3768423641090839, + 0.40320891139581566, + 0.47823373205098907, + 0.7580897487854575, + 0.5718063920423146, + 0.4835663587198462, + 0.4216622558384132, + 0.4007704035093626, + 0.43967147818421515, + 0.3688869635637064, + 0.4861553136939609, + 0.5456402154426251, + 0.4125872074557216, + 0.41021309432090264, + 0.407395626165695, + 0.5215151790097178, + 0.378691130317752, + 0.5706807791451055, + 0.4928873144386815, + 0.3216031738295728, + 0.4281441059408825, + 0.4455068015726055, + 0.5377354593807491, + 0.5452647727479661, + 0.25739449724754115, + 0.7306575734954333, + 0.7186272029611425, + 0.21998477796359295, + 0.352916560814285, + 0.44910500596264685, + 0.485470949151874, + 0.33711964379415466, + 0.4353144859169425, + 0.4798616923548491, + 0.4882252936473866, + 0.5023188429274773, + 0.3987287074148959, + 0.4571553087898194, + 0.3791897441155696, + 0.34008125379066384, + 0.4545419735140978, + 0.4117513113416769, + 0.40570234142657763, + 0.5265594038894816, + 0.4583404644381437, + 0.5529693678764421, + 0.4599779945420625, + 0.6168767300216846, + 0.44796308644314026, + 0.4916054470638858, + 0.5137299581113735, + 0.47125189753997365, + 0.3569697126312482, + 0.6389986302109899, + 0.647421970310341, + 0.47168664778427616, + 0.39405123281161314, + 0.4205318032269852, + 0.5504939827871732, + 0.4934084435656238, + 0.532761890331753, + 0.5368481292085062, + 0.4938425532115834, + 0.4605326470002213, + 0.3952083614738619, + 0.42500075422279515, + 0.5842426643337987, + 0.5080575571647061, + 0.4857387681340101, + 0.5071440173869023, + 0.4206203717193756, + 0.2828557687506647, + 0.38313847579759824, + 0.41297848727521824, + 0.5512078490989805, + 0.48369705255002443, + 0.39581883930715356, + 0.49685074143258895, + 0.3945437916467317, + 0.3562017933803087, + 0.4194749064276871, + 0.43686474160201166, + 0.40838171813269225, + 0.4582502888484388, + 0.6388672138116432, + 0.41068412326940584, + 0.4716233248147812, + 0.5854250538441306, + 0.4172285240088615, + 0.5302585054194604, + 0.6134590883546379, + 0.47541840326469725, + 0.5170772627176927, + 0.5049034006101784, + 0.4575767976352815, + 0.3613674382086631, + 0.3823316114696767, + 0.5074395407931822, + 0.5293580155475254, + 0.49868750968966563, + 0.3993606793876188, + 0.4887908888997975, + 0.5111630316379412, + 0.531031751282995, + 0.4204153182014013, + 0.48194776175190235, + 0.3994511524920801, + 0.49816471484232555, + 0.459453691904308, + 0.43668674977027266, + 0.41309637832632295, + 0.4785728289435948, + 0.48881229511540997, + 0.33187024024526046, + 0.4086189729187545, + 0.4529593215450842, + 0.4280233127623302, + 0.49641580951874925, + 0.5099405343282963, + 0.5169744099874805, + 0.4782208291913399, + 0.5151572241581396, + 0.41919792383517385, + 0.48570565003312266, + 0.4888835466931222, + 0.45703496566620383, + 0.45901075087519644, + 0.4717285546959255, + 0.4649559638716137, + 0.48667177486583857, + 0.4973691087165521, + 0.7311852092348295, + 0.5572451042314022, + 0.5190971198620145, + 0.5506157911350917, + 0.5333042993906913, + 0.5140105838998996, + 0.5500793769399056, + 0.38299581267565713, + 0.4204081205906331, + 0.48312393168792056, + 0.3629618514025534, + 0.46441459076873265, + 0.5130238729180584, + 0.4549982406832638, + 0.4600324492233829, + 0.48472880069616897, + 0.3464423062423888, + 0.5598784280697078, + 0.5032151835455457, + 0.40366018039625884, + 0.4541168813382418, + 0.40294574807955413, + 0.3739347867347133, + 0.47763809935534324, + 0.4671408936087044, + 0.4445164146201834, + 0.4794068444240509, + 0.413023618580343, + 0.5008752262716161, + 0.38948303444092364, + 0.41093548095637206, + 0.45080757349094774, + 0.517540289193683, + 0.5078764976229512, + 0.48375718464619766, + 0.42423802126735394, + 0.36787794670887664, + 0.3719656384103847, + 0.460381974070888, + 0.43032589094269247, + 0.4741694402030351, + 0.49828630621712333, + 0.4469596251590487, + 0.42327214395146406, + 0.37591723396873683, + 0.42918257221534467, + 0.546466695539327, + 0.4253907766001863, + 0.21952665517698114, + 0.3807759286159421, + 0.4629444716851509, + 0.3286796631667118, + 0.3171497363407549, + 0.479510049148735, + 0.47754805052246374, + 0.3495249714064556, + 0.3283897569583573, + 0.4264071545121244, + 0.4162190966684461, + 0.5532727220432125, + 0.34610860048573394, + 0.3493820636127407, + 0.3733236175826735, + 0.39412741687329766, + 0.30052721878235866, + 0.3668673383835531, + 0.34368238728813233, + 0.3792693698146332, + 0.3514715628736753, + 0.470470268219261, + 0.3798630069261143, + 0.4211635140117398, + 0.42326252638555406, + 0.5111231177993367, + 0.3676175671392462, + 0.4260768949261558, + 0.42120273573799755, + 0.4336545557067683, + 0.42615199504403367, + 0.4298109764216283, + 0.41938513063208244, + 0.6520612242946915, + 0.3584848451504071, + 0.46987854795143985, + 0.5952093489131619, + 0.5102152867034591, + 0.3817502035560507, + 0.4851913962494076, + 0.4360934996010857, + 0.48034391338842347, + 0.4903062268092963, + 0.37594718634748076, + 0.5928061077004576, + 0.44842152990417344, + 0.44601306195081397, + 0.37959577781062526, + 0.4527306688154457, + 0.4149773235747712, + 0.47896600078370993, + 0.5135908861170908, + 0.4684367838547899, + 0.48545743672609154, + 0.582793388654999, + 0.46147311617782943, + 0.42704669760278113, + 0.5253429577766533, + 0.5939072347634793, + 0.5458681770953281, + 0.4612781406373504, + 0.5938298558064944, + 0.5811625407069271, + 0.40080131513703476, + 0.4126918267013044, + 0.4628775154082412, + 0.3643233673289582, + 0.6906863663687391, + 0.4998020819655397, + 0.4469463797598688, + 0.3709981252788305, + 0.4816433481023389, + 0.4168604891402605, + 0.5370296647209141, + 0.4547786838527207, + 0.4860571095968118, + 0.546613100066763, + 0.3511088213656298, + 0.43264477347793806, + 0.5648418837574842, + 0.36616064613118443, + 0.4447145215818824, + 0.3424228287120547, + 0.3914768886971452, + 0.4157194737517175, + 0.4200426532496155, + 0.4904537732269943, + 0.3634887305400374, + 0.34311392436377547, + 0.4517979799255188, + 0.4173038547380358, + 0.3896420583218137, + 0.46938891936265364, + 0.48896400166630094, + 0.4424921356560818, + 0.542432373801725, + 0.5732278903667812, + 0.47962337698852153, + 0.3742245386072086, + 0.49531967805814037, + 0.45548454379008635, + 0.41215690175649, + 0.4502838374372877, + 0.34418367807401273, + 0.3427750567348821, + 0.3268830319433727, + 0.3826385508415103, + 0.45516012009704887, + 0.6589015028625319, + 0.45032643764824565, + 0.44284164339381465, + 0.31154446736551356, + 0.37279170132331274, + 0.3333069507140069, + 0.4252832664434511, + 0.3844066549114276, + 0.4077075199066679, + 0.37057925649203427, + 0.43478136708869025, + 0.40797940232951696, + 0.43480176555703137, + 0.4322317094114822, + 0.39492083612825635, + 0.5154772182889427, + 0.5120912438063973, + 0.4058722663778492, + 0.45112602240444133, + 0.4657688096623175, + 0.4103680339065867, + 0.3831074169281228, + 0.5262551274461874, + 0.46626613653590054, + 0.5933930154751677, + 0.4721026713693227, + 0.4631031900736302, + 0.4621609380673937, + 0.46772264036509614, + 0.36804259737104933, + 0.6293487095954646, + 0.4193234271088276, + 0.2435444360819687, + 0.49626073464591175, + 0.6181042657817143, + 0.387977623207849, + 0.5060519737295508, + 0.7715555338745806, + 0.46502800132876293, + 0.5434579241934793, + 0.32672752654341136, + 0.5070200516402011, + 0.37186508582573446, + 0.38626358468318306, + 0.3735493491158374, + 0.32568356674959276, + 0.4612984976579279, + 0.27090917074514737, + 0.4626689599732403, + 0.4696447039026832, + 0.45145808858957515, + 0.4160204858208402, + 0.3999961529340729, + 0.37955447186389357, + 0.4830867914012728, + 0.5328442186433865, + 0.556847617672407, + 0.3810393971643459, + 0.46684164325359817, + 0.432284137360158, + 0.7413546894573938, + 0.42906979343839974, + 0.5695846541787846, + 0.7437226950149121, + 0.7217416123890286, + 0.4891265527340496, + 0.40627142692547186, + 0.3811793781305298, + 0.4739048092282944, + 0.5160596206945809, + 0.5039493314824431, + 0.5915141942949289, + 0.43797867398628537, + 0.5616492407243036, + 0.4794758549985476, + 0.41000755188677246, + 0.7095791062997006, + 0.6880899986478334, + 0.5069206405758011, + 0.4309718037271196, + 0.4858728197082675, + 0.4338888206891026, + 0.471913778343655, + 0.6527583802143364, + 0.6768204840341993, + 0.4406847385532403, + 0.45225449214560093, + 0.45271758723062877, + 0.39775281073748864, + 0.5063352849539846, + 0.3847899306250175, + 0.5479437261726045, + 0.5476755815631774, + 0.38035522469928545, + 0.4706297693310899, + 0.4833837097139182, + 0.40670265459332466, + 0.5489025016666123, + 0.5395632359944845, + 0.43906810169131383, + 0.39486555532261364, + 0.5024193101660008, + 0.4621008418612733, + 0.378323226559824, + 0.3606071730460154, + 0.3124072667176785, + 0.2676435623082674, + 0.31686549511480905, + 0.5082586861817062, + 0.5625897702120105, + 0.4242067608756903, + 0.33377946012573684, + 0.42835822877452207, + 0.5097845391941083, + 0.5038163539234514, + 0.5313113674849312, + 0.4240934925075919, + 0.5174786088060529, + 0.5633735861673552, + 0.43438262845121867, + 0.5977215359640948, + 0.5512308054135873, + 0.4690319963695742, + 0.44728664088334835, + 0.3946878503822812, + 0.26208702037477366, + 0.31829872928324404, + 0.30244396135579404, + 0.5714270313241784, + 0.4776266828573822, + 0.44195069645118185, + 0.5838729365213924, + 0.6048237982105561, + 0.3425582838416714, + 0.5024617633201957, + 0.3992812853484627, + 0.7332701285544082, + 0.4623952784796322, + 0.46969593070192883, + 0.3850582444907196, + 0.4731938171298525, + 0.5170778103463682, + 0.4633258257533098, + 0.4357500662454436, + 0.5129233123954627, + 0.4757506261235228, + 0.5907169111898926, + 0.4439775447914708, + 0.5203773703022101, + 0.4709092274594151, + 0.45602997102370435, + 0.43905554479812003, + 0.49657357715876954, + 0.464166870165079, + 0.5158778509831168, + 0.39939126055121055, + 0.37397814441115196, + 0.3702960203427841, + 0.44490816173705056, + 0.48800096792856434, + 0.4264962435739665, + 0.4929053290733338, + 0.5153018758997103, + 0.4727463505542488, + 0.4407659892578683, + 0.45942363650329104, + 0.2833477956489939, + 0.4534141932193771, + 0.3377867362462778, + 0.4693847050886874, + 0.2464473369309439, + 0.6797602727183362, + 0.3486910176156507, + 0.527285860201009, + 0.6039437600444807, + 0.5225501501390306, + 0.2902121497206078, + 0.48176221197006913, + 0.38308081475151196, + 0.7854490988453199, + 0.30363933912964286, + 0.5713380777022768, + 0.5323096109923984, + 0.5628993010319326, + 0.6445425353359675, + 0.5112012512224623, + 0.6719172260835824, + 0.48372153750696945, + 0.44193238043684674, + 0.3744292017064261, + 0.48098415851655313, + 0.33573925173470437, + 0.7290656622933384 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "chip_histone": { + "mean_pearson": 0.5122565353843086, + "valid_tracks": 1116, + "per_track_pearson": [ + 0.5921869826115648, + 0.7688963370828424, + 0.7418195374747736, + 0.8242344998981957, + 0.751489913530931, + 0.7009137224883474, + 0.7499944048382453, + 0.6888019377778218, + 0.44290038668223314, + 0.7291216242256491, + 0.3865082900111449, + 0.5301034303377403, + 0.4479562407734585, + 0.41603365095378675, + 0.4516936978479742, + 0.37287203341707903, + 0.4190329780210601, + 0.6008947730055166, + 0.6126126031676663, + 0.49211575316681394, + 0.37082665204735216, + 0.6989559943187833, + 0.4910308531419823, + 0.6168643049556991, + 0.4000475805320547, + 0.36483651966765296, + 0.4740458080556961, + 0.5804137294212026, + 0.6794425925349329, + 0.525039507949303, + 0.35461756576095105, + 0.4515651182287488, + 0.45560762156122125, + 0.40180145033545306, + 0.5114644409796452, + 0.38978526261553326, + 0.5198802900551676, + 0.575828961758147, + 0.6432554353827753, + 0.6358265436545986, + 0.798263558022923, + 0.7895560004496525, + 0.783716864608028, + 0.6761759921194966, + 0.5286928649264608, + 0.5803974396119441, + 0.7889550331400241, + 0.45934978374309215, + 0.42644404670993097, + 0.5773929793279797, + 0.4728574000523302, + 0.784976228668105, + 0.8011224266086933, + 0.5315937647662508, + 0.5161506922555922, + 0.43617198043147376, + 0.4155261477455483, + 0.474376986587205, + 0.4580942155257366, + 0.5950896039773524, + 0.5643280851443696, + 0.5227107007513481, + 0.5312939419398394, + 0.5650627013988906, + 0.6875563384543846, + 0.4720445822629859, + 0.42712236109953916, + 0.6447072166282256, + 0.4786395873861623, + 0.5575227288282237, + 0.5666684665506903, + 0.5506373295439487, + 0.5121055332777378, + 0.5899390479476968, + 0.4899180556406368, + 0.601230658024171, + 0.37733600327244343, + 0.41600156777346387, + 0.460491287483768, + 0.3732600004740867, + 0.489845285952258, + 0.6201698107469789, + 0.4485930118456821, + 0.4344697640380928, + 0.4310044036543346, + 0.5231972962214194, + 0.5163585371268609, + 0.497608734289399, + 0.40635810412465667, + 0.4452128052106053, + 0.3986185755026707, + 0.394370167818699, + 0.5667500513760627, + 0.5861528877741454, + 0.469254113204238, + 0.6280604273728712, + 0.41536685135147955, + 0.5975252023026261, + 0.45428181724480293, + 0.5064707562704756, + 0.49200757861766636, + 0.5749083650063523, + 0.3574700698627379, + 0.5157441849614349, + 0.38608792936498365, + 0.3996924287095473, + 0.524829683327051, + 0.5760252561700464, + 0.38540859359793667, + 0.8182572821272557, + 0.8128947050487727, + 0.8176706937369491, + 0.8364543225428549, + 0.5941349948915366, + 0.7180109398320464, + 0.7288513427787371, + 0.7669192161976695, + 0.7184090379038901, + 0.5984542295563376, + 0.671847065045887, + 0.5845670655535354, + 0.7815223290466059, + 0.6784155561295101, + 0.4636853144007975, + 0.3709610596507471, + 0.35235917684996987, + 0.380490790230872, + 0.5931460818113455, + 0.477002631367974, + 0.3305142564001711, + 0.512955192878037, + 0.4856681583344955, + 0.5328657811388227, + 0.424680052447472, + 0.45030194612669194, + 0.5298345608395226, + 0.515553555984689, + 0.4093162238563122, + 0.4342376582619976, + 0.43345098220039335, + 0.4071412853254296, + 0.3488583209823241, + 0.47499971088287435, + 0.6217501728364886, + 0.34097046892429506, + 0.5843702471679055, + 0.3950854645133713, + 0.5207031152662914, + 0.4775739207741828, + 0.5891266727013857, + 0.3560614600210914, + 0.5327451996594251, + 0.3702948094496434, + 0.39887077178090596, + 0.45767185553050493, + 0.3927978664551235, + 0.30772851281207875, + 0.23036025297866117, + 0.4009920207923848, + 0.39946915678689265, + 0.5750775343149029, + 0.570154563267956, + 0.6418776902170664, + 0.5740388100568362, + 0.41332200457509677, + 0.5399464269099894, + 0.5961378487785627, + 0.5865419929448019, + 0.4869277289726596, + 0.5202275383929869, + 0.4545552851708587, + 0.3457057252747495, + 0.5247056311183019, + 0.49986547239686674, + 0.31553904735441113, + 0.3405132638650325, + 0.5351863669435728, + 0.6136973187114149, + 0.5967202172414209, + 0.5239668926383574, + 0.3640116219360679, + 0.599271887606814, + 0.541975609399082, + 0.5510632735306005, + 0.3439956302732937, + 0.5985083141455616, + 0.5067726288171303, + 0.5539396968210153, + 0.29044505126795905, + 0.4813823384409814, + 0.3732532475278743, + 0.30910865060094367, + 0.5737337741779678, + 0.48329356484944075, + 0.5927951162227915, + 0.4287460242850463, + 0.6052650627056071, + 0.40520648672193466, + 0.5031651341848998, + 0.43276754502737563, + 0.37000464479105627, + 0.30721673504248115, + 0.4139806840310998, + 0.41332123744563126, + 0.6515446741936649, + 0.3997502989540913, + 0.33207701905831255, + 0.509239701203914, + 0.607147988920436, + 0.44828782451275123, + 0.4936062981215718, + 0.4674338390155963, + 0.4008348977203351, + 0.4258777233458159, + 0.41269834592034405, + 0.5663776414503372, + 0.6482827495746867, + 0.6469219896908933, + 0.6073975086432889, + 0.3935691375248388, + 0.5592880572693607, + 0.4968682712243643, + 0.31745825702453284, + 0.38318070306443014, + 0.40115089248299757, + 0.2782640171669879, + 0.5788323072085974, + 0.37057971668846706, + 0.3528155422467211, + 0.5702826025468604, + 0.3294046916488355, + 0.3272191765443733, + 0.5723895559632828, + 0.44639840093842253, + 0.36029819921629125, + 0.6502826074352512, + 0.5911646813186932, + 0.3385816420759909, + 0.34090254931810804, + 0.5144221349981518, + 0.35697197406029896, + 0.4860846420639344, + 0.3613219379016558, + 0.5521370983396204, + 0.492090574403048, + 0.432041381250748, + 0.3950081774793751, + 0.3668069897703602, + 0.500365330344957, + 0.3864585009989784, + 0.4777806123618982, + 0.3663008668698484, + 0.43434143462536645, + 0.396432374446545, + 0.4089804242290201, + 0.48764190077630437, + 0.4556217332685926, + 0.41663366406575114, + 0.46677506697348503, + 0.4900530648928121, + 0.36631457886325336, + 0.5793767349582525, + 0.37206314478083263, + 0.40398758032022436, + 0.43762800046818656, + 0.3647593416791792, + 0.3981928796575374, + 0.5674532771235105, + 0.3097406856626074, + 0.388471920927364, + 0.3777735773730937, + 0.3870611091899191, + 0.40421664716753397, + 0.4413067100560933, + 0.41111074060427233, + 0.5280625053758653, + 0.3341713679344554, + 0.5010473911839067, + 0.5393140392200687, + 0.3519546320781491, + 0.3447168162060546, + 0.6485871930053416, + 0.5312009954740307, + 0.3188348656980348, + 0.5940356067301962, + 0.5557177012790527, + 0.5202072954519278, + 0.5285603842887197, + 0.6070161673291318, + 0.5735829425138812, + 0.5891762582231845, + 0.5830191288719138, + 0.5864693658853747, + 0.46606899784732, + 0.5265685394980213, + 0.5903275430868327, + 0.5305580130814668, + 0.49636801753344395, + 0.6765024533247298, + 0.3657306756834814, + 0.6339520253640784, + 0.6857281621598393, + 0.7471278142363921, + 0.7232947222006433, + 0.5546831652585812, + 0.5913084547101084, + 0.4532782363876991, + 0.7121518308134509, + 0.5205341337655361, + 0.531435335958934, + 0.3668715262380913, + 0.5559836350433913, + 0.4668147107987174, + 0.4815890729894311, + 0.5965911371392476, + 0.3842760384939395, + 0.46112599648459357, + 0.36911909591698644, + 0.48991988381206303, + 0.41940145069994716, + 0.3750126757648006, + 0.42749238618665786, + 0.307829575511803, + 0.3825496271091226, + 0.4169422013548806, + 0.42528886412003264, + 0.4110218172502212, + 0.4224372392833432, + 0.6451542932231571, + 0.5385440001826146, + 0.49363962943802475, + 0.550279657849438, + 0.5116092048967483, + 0.36412645921448106, + 0.3407174302298272, + 0.6275707927548184, + 0.4348991344568205, + 0.3766284943478653, + 0.49081143347079415, + 0.6177763994470743, + 0.6446851185213227, + 0.49262270639974004, + 0.39584484025138156, + 0.5208730276738728, + 0.4157393801643673, + 0.3845991365812573, + 0.5486449685304028, + 0.3875224596846625, + 0.8302030294800758, + 0.5706228545368336, + 0.4745339775597618, + 0.47883882185143567, + 0.5034479892076259, + 0.49498294912669943, + 0.4685702184525457, + 0.5783092885501631, + 0.45849630411342895, + 0.5995398666544054, + 0.5065293882472671, + 0.5687873036552799, + 0.6129887882393815, + 0.5237079168996209, + 0.5391355110429524, + 0.4471530061431993, + 0.461516583702071, + 0.6553675395139537, + 0.6259110535139442, + 0.47341676038647273, + 0.63979228045754, + 0.5030245776045954, + 0.6020318748173681, + 0.5043000759052051, + 0.5437614749655193, + 0.482406933523076, + 0.5940608396701916, + 0.6110129065764025, + 0.6608210951820112, + 0.5816579021306204, + 0.5755197328634108, + 0.5605144043116611, + 0.5289469822840422, + 0.438947809161213, + 0.4445262937747718, + 0.37934070858803404, + 0.43498355374157965, + 0.3664959359229471, + 0.3964911075196969, + 0.42043102531788673, + 0.5827725148933202, + 0.4078155746568012, + 0.3737432257983108, + 0.4384921579963536, + 0.6068162390855304, + 0.6809978443277744, + 0.6292905457356729, + 0.6211259538132429, + 0.45087886810600475, + 0.630217016173349, + 0.49890245132182764, + 0.661503849557665, + 0.5124858737403335, + 0.6204685714933372, + 0.4895343743444697, + 0.6115970449685334, + 0.6304373929514079, + 0.5374335018610408, + 0.5292472089153978, + 0.5646194922930025, + 0.6395039491135356, + 0.5851752601471145, + 0.4668160755601572, + 0.44038845880474153, + 0.3869356230424923, + 0.4117703259915272, + 0.5746084150271726, + 0.6154304146853093, + 0.36411910177942947, + 0.6642189623601711, + 0.6424385202382052, + 0.531473915653883, + 0.44879928683618775, + 0.5176447427197134, + 0.6175200903479378, + 0.5752194616599111, + 0.5238506013251923, + 0.8759177015859511, + 0.5092713838219334, + 0.6218215205680957, + 0.5667930971436702, + 0.5995805222159074, + 0.5060430163484961, + 0.43560151785250256, + 0.6366121965890982, + 0.4971476920317772, + 0.4618026310315203, + 0.49036785851079423, + 0.4796428044396848, + 0.49710303438633796, + 0.45815479825919225, + 0.49003623381001843, + 0.48192343676067223, + 0.5083537758143387, + 0.351485764807968, + 0.5188245233908928, + 0.36059252533049435, + 0.43300981754049717, + 0.32030196328863025, + 0.5222064735503904, + 0.4169476450067555, + 0.3630488605398548, + 0.5932076219800142, + 0.4402872001538435, + 0.32005554699204697, + 0.5782691669361302, + 0.5034803765619176, + 0.5053622365756919, + 0.5657428336917854, + 0.44655974965958234, + 0.45171921710466006, + 0.6692429670519209, + 0.46369077545758675, + 0.5689975802411622, + 0.46796827746992603, + 0.33817821230938105, + 0.3180674056065659, + 0.43886477786788075, + 0.5346963762449294, + 0.6439162505832585, + 0.623564490725078, + 0.5242272546094745, + 0.5848386875419241, + 0.5247910176073258, + 0.5474156219572147, + 0.477344845110418, + 0.46237854880398965, + 0.4754621735529148, + 0.45427330202671967, + 0.514652792263484, + 0.44513013993574274, + 0.47980914198634933, + 0.5719602393928983, + 0.502437249562487, + 0.4763886318511817, + 0.39991572233261935, + 0.4443964375673643, + 0.547658365796887, + 0.6024777197394371, + 0.5820692629343245, + 0.5934161406173792, + 0.6167712795550906, + 0.646894094326921, + 0.6824222915490533, + 0.6657721945411931, + 0.5044901325273272, + 0.548901485213364, + 0.532316761999919, + 0.5688895844402471, + 0.43977677947827026, + 0.5423643853188478, + 0.5152383712958843, + 0.5987605786425523, + 0.4285419916057562, + 0.6126283515111417, + 0.6693575182977214, + 0.6347648278671323, + 0.5870651889097215, + 0.6594676850669715, + 0.561345763881255, + 0.4896027222474861, + 0.41111455325724844, + 0.4690429554812922, + 0.38114816194218126, + 0.4436134936042398, + 0.5942228911451913, + 0.636319153098068, + 0.514761162635222, + 0.4124301730543806, + 0.44242065319859714, + 0.5393245463790072, + 0.5430164802972332, + 0.4374185222630422, + 0.5581203583436961, + 0.5266013110520735, + 0.40634931110909955, + 0.4591881897942392, + 0.4243127323092314, + 0.501841005730061, + 0.4864942100161382, + 0.38216435353701406, + 0.4499691133998827, + 0.5152892687258613, + 0.4415740624967194, + 0.4281943192325914, + 0.47535979877965306, + 0.4238698330881213, + 0.41759471044459134, + 0.46943288792595267, + 0.5829665131855998, + 0.4546376319802808, + 0.3319355692215063, + 0.6087638645307927, + 0.43319017695820095, + 0.5901765470303489, + 0.5827849721405381, + 0.4171053954617608, + 0.5474974161656427, + 0.6295868343133927, + 0.5370630025509672, + 0.6783682286082372, + 0.505955194300694, + 0.5654623088879577, + 0.5339374973856864, + 0.5256101351131657, + 0.5842967533362559, + 0.5514785085824712, + 0.4110300037568421, + 0.5335088137230911, + 0.573281313066662, + 0.4737205799592835, + 0.35866414208762654, + 0.5037478236803212, + 0.5420894758701345, + 0.5099256549487177, + 0.6723815412326736, + 0.5819673951588908, + 0.5834731129915411, + 0.600743337955113, + 0.6643429646204924, + 0.585848141377689, + 0.6414369246878291, + 0.6628023327744953, + 0.6454471529698257, + 0.4761804301925156, + 0.5075569613152007, + 0.46032435180967163, + 0.37263897162188925, + 0.33892908521965026, + 0.4132427373459092, + 0.4382822856428163, + 0.381414182943088, + 0.3446104618210453, + 0.5323137280225707, + 0.34268873005157074, + 0.5008607553783108, + 0.38444553368642936, + 0.5694580187262122, + 0.6332322646697657, + 0.5224775749908208, + 0.38765352670636116, + 0.4262783762329231, + 0.5918607243852599, + 0.3854774085652164, + 0.5378219482697838, + 0.35671790030469236, + 0.8485251156797906, + 0.5664415801830326, + 0.4642476275289174, + 0.6492301174981375, + 0.608767571053208, + 0.5365376368140498, + 0.47001596145027125, + 0.5303046566957424, + 0.6169320031912875, + 0.4843882192084512, + 0.6225230977950904, + 0.5805655828235643, + 0.34033812002098907, + 0.5395296894862862, + 0.35022755788434246, + 0.42024280978120276, + 0.3583670921805951, + 0.5608137843129232, + 0.46440951690940974, + 0.4255090479717622, + 0.5906211614934533, + 0.6117494974514831, + 0.4325661941202864, + 0.5264876750121746, + 0.4841320473036275, + 0.5020458388571851, + 0.5333776191677113, + 0.5147838306388712, + 0.38923700426830493, + 0.3825146696010611, + 0.6229684689966742, + 0.46744105943113196, + 0.4158247591691869, + 0.5555601955214127, + 0.4422675358246797, + 0.48538300231391723, + 0.3634934406147697, + 0.4118406018197582, + 0.4682310881326346, + 0.5200853517704271, + 0.523535043906913, + 0.3927648316760735, + 0.5177060962454311, + 0.3389060587704852, + 0.545591076520364, + 0.5061821276194022, + 0.5266515405368238, + 0.5385283825810014, + 0.4426354892868367, + 0.498884429904436, + 0.4123155339024162, + 0.5446169031149024, + 0.4600787422254953, + 0.504145970238745, + 0.5118369748474523, + 0.4023850860589988, + 0.5438866772069899, + 0.3120873384286495, + 0.6775619965604656, + 0.6799317949195061, + 0.5781413254507746, + 0.5680485843485987, + 0.6218867324485393, + 0.6184391028260476, + 0.31074809776193085, + 0.43928539124649885, + 0.3921860358628425, + 0.44255708813008127, + 0.37535437212042627, + 0.3820984720722041, + 0.4418544913348675, + 0.5095147411322638, + 0.4934152260094849, + 0.39396414697176857, + 0.4188384734542528, + 0.3384874498702523, + 0.4818813932781005, + 0.27067470220279527, + 0.4627814442046489, + 0.5206010356265436, + 0.5309758577684783, + 0.5417558739382593, + 0.46582180694661646, + 0.5745133204062415, + 0.32635124799539, + 0.559055262025023, + 0.49868823902626447, + 0.4424056325601389, + 0.4265760479122714, + 0.4583302063289684, + 0.4676022875883451, + 0.47681741843913544, + 0.5057403337232095, + 0.5325952814336908, + 0.5724396156983844, + 0.38422623944475126, + 0.41795220427889557, + 0.4988518182882349, + 0.5585308875676849, + 0.541003310020869, + 0.5537861869185555, + 0.3958181940120525, + 0.6247285218192441, + 0.5073293960972156, + 0.49497798088951067, + 0.6312696538803704, + 0.46323006750185586, + 0.4003057510474914, + 0.3634372621236077, + 0.4416529705759801, + 0.48848168357317406, + 0.44479414377674564, + 0.4565053461675089, + 0.4660656884500887, + 0.5851268895425665, + 0.6263992791929847, + 0.6834590896856462, + 0.5779803197237109, + 0.5910147221738201, + 0.5405752563628844, + 0.617632919601924, + 0.5497201710412143, + 0.6524322146543527, + 0.39914121669466773, + 0.537585896531129, + 0.6406524693724965, + 0.6884793008902731, + 0.23280500507281338, + 0.4225986078216824, + 0.4673576038913514, + 0.5850561131811525, + 0.5679964936585956, + 0.5806688379094082, + 0.43299898075218535, + 0.5417064791309834, + 0.7566989531143035, + 0.627502166532113, + 0.7942035616538892, + 0.8077447402886585, + 0.6734268064868639, + 0.7728717807685638, + 0.7779714505905204, + 0.44246715564181344, + 0.45275546815701556, + 0.46052894323596166, + 0.5574816656804117, + 0.3807743077414122, + 0.5217788106323674, + 0.415578042959657, + 0.5223400822214922, + 0.4570327089491819, + 0.43630481220514183, + 0.35990414973979495, + 0.4672730764019588, + 0.4614763410634726, + 0.4910481732640052, + 0.47016891465723887, + 0.3577201031483419, + 0.3839629569358683, + 0.3699267282864378, + 0.390902418742734, + 0.2987162234850567, + 0.3828618625058428, + 0.3492012137519018, + 0.5641093402561536, + 0.6113460753878525, + 0.4096179236918889, + 0.3506315510953588, + 0.303973435697994, + 0.5549606575901328, + 0.6578128938451635, + 0.656663957931792, + 0.41503719169714515, + 0.39795754558005975, + 0.2957422445628348, + 0.5385851820791846, + 0.36577848132602203, + 0.4301818899720592, + 0.5127205264405126, + 0.5796990864094613, + 0.4983006455758523, + 0.34884708972579925, + 0.5026323153296238, + 0.5656944943045589, + 0.5489581172277292, + 0.490141219097475, + 0.5446870090311379, + 0.4404503109360333, + 0.5667384686857049, + 0.5324702925500153, + 0.6103861581369218, + 0.39917145856727404, + 0.533264451601868, + 0.7249453454795912, + 0.4835054691806759, + 0.7022364418792668, + 0.40219840636635756, + 0.5694317542743801, + 0.72583589206858, + 0.5889409718663209, + 0.4358904031607321, + 0.41903242582729666, + 0.37894671198883906, + 0.35840796196370817, + 0.5180572442424667, + 0.6370877276108775, + 0.510338041976563, + 0.44487863251970783, + 0.3655851503371145, + 0.5641779476518575, + 0.5500177030991923, + 0.5826694157944513, + 0.6136835836743662, + 0.5676646019197192, + 0.5229399866687219, + 0.46357814960749477, + 0.6117399699380353, + 0.34230403503623597, + 0.45615532157884137, + 0.5147549548326238, + 0.4506652044899465, + 0.6877892164152376, + 0.6834473627236419, + 0.6264527475464405, + 0.6879606586658195, + 0.7623602760632548, + 0.6150828400108755, + 0.770969436405044, + 0.5667407068963641, + 0.5022680843434674, + 0.479247378478047, + 0.6254180820593945, + 0.41514004596756543, + 0.5613624047746306, + 0.5722264385032051, + 0.5587761798078588, + 0.5140310488433875, + 0.4825822331650763, + 0.4992950759019655, + 0.5461957858951411, + 0.45805120284703277, + 0.6090941271659402, + 0.5857292066968706, + 0.6002977607957685, + 0.6365275813461, + 0.6824727368631199, + 0.619763916499039, + 0.49881153498122194, + 0.6528008857805474, + 0.6236513568196906, + 0.394882898837756, + 0.48210767817229266, + 0.4590524936120807, + 0.4522717623974754, + 0.43917738293062586, + 0.47516862220084144, + 0.545572468261449, + 0.5250236081333894, + 0.5294751861304551, + 0.520418895779682, + 0.33635878724003243, + 0.48543112512132963, + 0.5883392098610059, + 0.5340644872523359, + 0.6338139307462327, + 0.6633742188287545, + 0.753136492262791, + 0.3830996432413198, + 0.4220032931402407, + 0.5656676788204835, + 0.44396111064577015, + 0.3192447283744811, + 0.38945515939355296, + 0.37450931457883935, + 0.697353986737371, + 0.6544350988593204, + 0.6734254722105883, + 0.38771682857276674, + 0.4317836069202647, + 0.3765093972173455, + 0.6293579428338485, + 0.5365087483658324, + 0.3021554793792751, + 0.3397999991771498, + 0.5315878104060603, + 0.47611328201175845, + 0.5520668509244067, + 0.5580378404984823, + 0.47532659439891045, + 0.4542435955938092, + 0.3377067986517484, + 0.7055788821323333, + 0.5923293800778837, + 0.7145747355661456, + 0.5963571315625087, + 0.32652898292115984, + 0.5574088949153648, + 0.5762465115525578, + 0.5703600352111469, + 0.5194931504690705, + 0.43098171271817953, + 0.5387896175633329, + 0.5941662800783217, + 0.5810393914519156, + 0.519586728647774, + 0.40347143440995387, + 0.5380626626451918, + 0.44888070202224045, + 0.510557029557894, + 0.6545807742072149, + 0.613663452734363, + 0.5986450225785296, + 0.49788492627326, + 0.5140814878810807, + 0.5142639342394564, + 0.46588456612270246, + 0.5327600918009234, + 0.4548629350155762, + 0.32658455079193927, + 0.5102441548564169, + 0.5984253493392511, + 0.3064905486854994, + 0.7728057538728187, + 0.4849551107381108, + 0.6248143763320311, + 0.7459557398049508, + 0.6886631198830201, + 0.41930255769936664, + 0.8165311698737925, + 0.5177068541526486, + 0.6880925436943238, + 0.42424234831748664, + 0.6377610858199261, + 0.6572694913385472, + 0.7295439098810623, + 0.6719289567970007, + 0.7239982264444953, + 0.6328395107905339, + 0.4716807345468434, + 0.6555355624706177, + 0.5905778653874364, + 0.5029781115076521, + 0.730946101615684, + 0.38002569745417286, + 0.6427222726057533, + 0.6629918070275594, + 0.7347150540878636, + 0.6712603529111265, + 0.7336689854894705, + 0.5931308623104471, + 0.44784383963245655, + 0.5734236396199565, + 0.5732169117772843, + 0.5445332600766313, + 0.4494477121673678, + 0.4118859180433204, + 0.48803661350401645, + 0.5683593900348912, + 0.5831712144266405, + 0.46261088901144765, + 0.5939906433547463, + 0.6040411798796859, + 0.4710861980130047, + 0.601169393739015, + 0.585932357350305, + 0.559552418191698, + 0.49607038207109844, + 0.4515847351751013, + 0.49676806749333485, + 0.5603836756771714, + 0.4730340799980238, + 0.39844500838958763, + 0.4911286049593202, + 0.41608227130138276, + 0.551569380356136, + 0.46949264004448477, + 0.39635384277600566, + 0.4055993080484504, + 0.43055900893757737, + 0.5401655800605007, + 0.5112754486570487, + 0.40962086858420027, + 0.36525618872111904, + 0.4617938905011587, + 0.6083026229539449, + 0.6703984379101068, + 0.35579659154205384, + 0.5746374053985663, + 0.6275361619393537, + 0.6072238006259912, + 0.6070061429757762, + 0.4598493118510498, + 0.6063934219281467, + 0.47298555397787334, + 0.6419002578912997, + 0.5088392313029613, + 0.6684945321157699, + 0.6225516022648087, + 0.6232197342868014, + 0.3507628815161532, + 0.3954667194806083, + 0.4422106375711023, + 0.5422027382326645, + 0.5156300406874108, + 0.594260542027958, + 0.37868028305852297, + 0.6153067539683176, + 0.5496394508556033, + 0.5920931062701092, + 0.5817866671729361, + 0.5485281332984363, + 0.4652107266886435, + 0.5812528079855238, + 0.6093568730287517, + 0.5879519947133928, + 0.5670339919508008, + 0.4769428719135155, + 0.3937403937546855, + 0.42206284094524155, + 0.41943319343069985, + 0.2877385403839629, + 0.4350947244584532, + 0.5661918892934878, + 0.4740775536904006, + 0.45584678372177684, + 0.48808970633227844, + 0.44655089471604775, + 0.5647164459018705, + 0.5896057162747356, + 0.40559472005946645, + 0.5189717152423197, + 0.6070997671581349, + 0.42388139887617526, + 0.48757273099451676, + 0.5606307391642235, + 0.5287955055256498, + 0.6408253306353593, + 0.3949259079968387, + 0.5957441789315066, + 0.4412112557328695, + 0.6978144784141809, + 0.6338289414766493, + 0.6836063085439822, + 0.35285522767339905, + 0.3818297715022048, + 0.35773235133061715, + 0.3313095465944491, + 0.37535179621725645, + 0.5254530107224314, + 0.5627267012004742, + 0.324308234699998, + 0.4328624667472921, + 0.48282994027951776, + 0.5302368101150932, + 0.5961649931406029, + 0.5591231854191848, + 0.6928758134325619, + 0.3543046929116518, + 0.48818488814625177, + 0.5272444194071941, + 0.4997876283101827, + 0.4869636353628515, + 0.6019431189235466, + 0.5500073575839771, + 0.3157389215937792, + 0.596956580967815, + 0.5357313760755437, + 0.48146677483221817, + 0.6261725924777913, + 0.4650291441397679, + 0.41280523375568945, + 0.44674451479650756, + 0.4999322545072452, + 0.4908479308793024, + 0.44414289212237135, + 0.49045346465876044, + 0.6584178493886802, + 0.5022268065976965, + 0.6617646576854229, + 0.5676935382968917, + 0.5759991616273998, + 0.6339243529975602, + 0.5764406383631832, + 0.5586056835425389, + 0.5285525427586246, + 0.6849456612914181, + 0.6974215082338687, + 0.5956389209234589, + 0.6119003332813093, + 0.7717709425417018, + 0.4944257899343735, + 0.5279203354779562, + 0.40906218762514085, + 0.5801378500461101, + 0.6188496834796355, + 0.5709650628884181, + 0.49437844341054055, + 0.5887292189653099, + 0.41271728977601957, + 0.518831127809663, + 0.4435480796982697, + 0.33417120463007977, + 0.507546780781074, + 0.4896563829100894, + 0.49014774686243867, + 0.49870241458149567, + 0.3849998655301121, + 0.6723476637575563, + 0.7015766802432525, + 0.5631351286854869, + 0.5915484534971546, + 0.6705560873750855, + 0.5337869477820013, + 0.3583770154516228, + 0.4637852628550625, + 0.43508853550172016, + 0.4077658713449124, + 0.46244466860566086, + 0.5045127833944953, + 0.5515408598863326, + 0.5009087348203979, + 0.3672917652066263 + ], + "values_per_track": [ + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200, + 307200 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_sites": { + "mean_pearson": 0.43115650430280533, + "valid_tracks": 4, + "per_track_pearson": [ + 0.4640417506982821, + 0.40767757311550207, + 0.4601687009774838, + 0.39273799241995344 + ], + "values_per_track": [ + 39321600, + 39321600, + 39321600, + 39321600 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0 + ] + }, + "splice_usage": { + "mean_pearson": 0.6419120847985565, + "valid_tracks": 626, + "per_track_pearson": [ + 0.6558698528334993, + 0.6448013852794453, + 0.6554623633358972, + 0.6503559973279724, + 0.6743455438512582, + 0.6486302173094404, + 0.55715830721227, + 0.6058765165517342, + 0.6478085424878276, + 0.6491334651968667, + 0.6548374171428536, + 0.6606397717807047, + 0.6277878136902499, + 0.6370785356819134, + 0.6439132529905585, + 0.6495704504463892, + 0.6404581926633154, + 0.6338382761084662, + 0.6724038287920662, + 0.6566002344238304, + 0.6763686928992438, + 0.662180430602436, + 0.6688126233882126, + 0.6339699666967443, + 0.647642877313021, + 0.6454474981668412, + 0.6499284983973987, + 0.6423421024804085, + 0.6426985448894312, + 0.6413591575246563, + 0.6491356867376128, + 0.6651502567497518, + 0.6243058863282163, + 0.6766877716414613, + 0.6395550464301609, + 0.6771984791904111, + 0.623753990369039, + 0.6493713904376361, + 0.6397320126644281, + 0.661974079378428, + 0.6700166382131036, + 0.6390653355720425, + 0.6299865289244435, + 0.6342694231949084, + 0.6579627614046905, + 0.6298928551707254, + 0.6483170380046235, + 0.6602016857040471, + 0.6489266776090112, + 0.6510830294575142, + 0.6269231708623189, + 0.6340337122250473, + 0.6448172025909868, + 0.652721511110325, + 0.6277120429834429, + 0.6530962078107937, + 0.6514273068101663, + 0.6587261533930744, + 0.6573680295056945, + 0.6528455329047561, + 0.6482966787262213, + 0.6569248830976916, + 0.654849374663801, + 0.6367678407475305, + 0.6418533836674638, + 0.6359732433061359, + 0.653542170374947, + 0.6508731267227832, + 0.6320370439075716, + 0.6357391835261166, + 0.6623734414913283, + 0.6527210956801226, + 0.6576044533857294, + 0.6437340280230982, + 0.6599221090506413, + 0.658210196564094, + 0.6485085887634461, + 0.6559148639494957, + 0.6421922918110278, + 0.6465291668348098, + 0.6700108766918628, + 0.6526004966046661, + 0.6532043364803694, + 0.6546627699689535, + 0.6510021047410124, + 0.6484153810847135, + 0.6569922370311071, + 0.6681654224439607, + 0.6569223438079576, + 0.6477060613380312, + 0.6419939827401754, + 0.6462098152424903, + 0.6617505811397969, + 0.6463069209994353, + 0.6539934276332097, + 0.6674466697224817, + 0.6489955023369356, + 0.659056335350487, + 0.6660401718771733, + 0.6570215896958301, + 0.6345025705484657, + 0.6394590728974059, + 0.656536541664844, + 0.6384482457006512, + 0.6497768721762118, + 0.6448475251793379, + 0.6338250268699216, + 0.661302504657845, + 0.663695525054844, + 0.6521743532844555, + 0.6400106446569716, + 0.6524815534426374, + 0.6578965171194199, + 0.6648114066637723, + 0.6591573160078671, + 0.6455970659754184, + 0.6475582809168333, + 0.6452298847393638, + 0.6395492281614931, + 0.6569149683005588, + 0.6534919585576957, + 0.642443109924591, + 0.651906843641477, + 0.6485587351872162, + 0.624971759309125, + null, + 0.6684363285162003, + 0.6506914993227663, + 0.6343284817929159, + 0.6529520486270618, + 0.6459216572314797, + 0.6449814334175129, + 0.6558179454680395, + 0.6402618633133731, + 0.6439137606943857, + 0.6482515352635783, + null, + 0.6355767348711591, + 0.6445278156946436, + 0.6187521189087961, + 0.6333446899490447, + 0.6392842518641066, + 0.6352387501606725, + 0.6527310882558047, + 0.6662487037924878, + 0.6347237331762369, + 0.6431639875323667, + 0.6464860127490824, + 0.6362503518223172, + 0.6476374930701739, + 0.6452001574441641, + 0.64803107123521, + 0.6397120112623009, + 0.6444349829183293, + 0.6462910127567785, + 0.6489493138196724, + 0.6412320824940456, + 0.6395536112326363, + 0.644120336533067, + 0.6545771571279054, + 0.6549169385813773, + 0.6470584066998878, + 0.6709172179645861, + 0.6495424281847737, + 0.6401111526628294, + 0.6545449904379873, + 0.6504289432447793, + 0.6227224374984701, + 0.6210043148606696, + 0.6212973444246747, + 0.6488559818771125, + 0.6509747276782146, + 0.6322082441891261, + 0.6481772071286707, + 0.6461653873591742, + 0.6183547898674872, + 0.6336533434884315, + 0.6448837425430374, + 0.6320204729659018, + 0.6528624224090744, + 0.6520311936634928, + 0.6464185358397806, + 0.6087054051410233, + 0.64659286294001, + 0.6455441565834573, + 0.6149390535486017, + 0.6619678979833415, + 0.6470075476928254, + 0.653607000046757, + 0.6352680195761202, + 0.6591972480235273, + 0.6477992098139592, + 0.6575056709242312, + 0.6665983580433023, + 0.656555575619606, + 0.6634099495919773, + 0.6613070717641588, + 0.6302414731433146, + 0.6561836265815499, + 0.669020219913712, + null, + 0.6708606328274582, + 0.675802786361637, + 0.666152482614306, + 0.6672137219855152, + 0.6702704740452506, + 0.6707988234005947, + 0.6519958089192829, + null, + null, + 0.6680603618834243, + 0.6762570412184444, + null, + 0.6431561147790403, + 0.6155929463897403, + 0.6498009087408518, + 0.6781570743964094, + 0.6709732319944445, + 0.6745981701983635, + 0.6702023429771067, + null, + 0.6779928875010289, + 0.6640882936125813, + null, + 0.6097459129951374, + null, + 0.6385880249570505, + 0.6637076378401233, + 0.6739344515839487, + 0.6821579295682408, + null, + 0.659814426201574, + 0.6557380411907554, + 0.6629418266756039, + null, + 0.6551355423398592, + null, + 0.6580517575621813, + null, + 0.631569935688785, + 0.6611240514257364, + null, + 0.6524681029788965, + 0.6553656215557173, + null, + null, + 0.6671229649752668, + 0.6687852093081992, + 0.6586397626604413, + 0.6654355146477706, + null, + 0.668740106495289, + null, + 0.6664579884715648, + 0.6844964086853462, + null, + 0.5648744976189077, + 0.6607705394489751, + 0.6708200131827314, + null, + 0.6590031432516614, + 0.6800120336691097, + null, + 0.6798479215768691, + 0.6792641723486632, + 0.6797432622426404, + null, + null, + null, + null, + 0.6774200802768505, + null, + null, + 0.6626770105126734, + 0.6642517158632992, + 0.6736222050514986, + null, + 0.6807018263537206, + 0.6760265332007054, + null, + null, + 0.6754196864057928, + 0.6617123414828628, + 0.6669341126831868, + 0.6699608956655717, + 0.6791834800184094, + 0.674382228879525, + 0.6499360646369654, + 0.6712324712755118, + 0.660227062861095, + 0.6607306748835541, + 0.6708237675439686, + 0.6771474939121283, + null, + 0.6498596988276653, + 0.663722537633844, + 0.6658665859841891, + 0.6595398769223884, + 0.6401110563723496, + 0.6198336509666681, + 0.668875364757296, + 0.6582630479624509, + 0.6562692753908239, + 0.6657835281781631, + 0.6695800063667647, + 0.6643396143123216, + null, + 0.6613492768526764, + 0.6868822652055344, + 0.6816712839874574, + null, + 0.6648089781865951, + 0.5771505765982756, + null, + 0.6743586027217606, + null, + 0.6608311135472575, + 0.6669781090221657, + 0.6734023606530949, + 0.6381983390013631, + 0.6694385062758612, + 0.6522499775601317, + 0.6496214456868762, + null, + null, + 0.6615798065578559, + 0.6374549435185572, + null, + 0.668636199749438, + null, + 0.6728906472949323, + 0.6646826950765793, + 0.6561046603574578, + 0.6553790462778263, + null, + null, + null, + null, + 0.6756717206421438, + null, + 0.6648289855118155, + 0.6706396329077959, + null, + null, + 0.666920902910314, + 0.6736024964362034, + 0.6670285717501878, + null, + 0.6555101313203242, + 0.6555323213423642, + 0.6571611105854849, + null, + 0.6781157654512229, + null, + null, + 0.6731740994751824, + null, + 0.6646509353402417, + null, + null, + 0.6576230396456583, + 0.6630279154497357, + 0.6542937531431532, + 0.6572352045155735, + 0.6467741305510794, + null, + 0.649840966516971, + 0.6516617541823636, + 0.6326554580796611, + 0.6031880427754828, + 0.6040711914357961, + 0.6519732872475809, + 0.6324097992753015, + 0.5635199170118707, + 0.6367297265006605, + 0.6280651910519469, + 0.6308375944990062, + 0.6259560488722393, + 0.6484222586895972, + 0.6365481148911539, + 0.6570058983399654, + 0.6190476883852065, + 0.6105826705979023, + 0.6261196295050758, + 0.6417960199073726, + 0.6497482255878936, + 0.6488597684328754, + 0.6684591195130961, + 0.6705651877648502, + 0.6186088537116906, + 0.5471142320142162, + 0.6462638660043747, + 0.6404906222609347, + 0.6216759671995398, + 0.6568790782840763, + 0.624214639512657, + 0.6170156493757387, + 0.6312718239696256, + 0.5935928754168783, + 0.564927272567095, + 0.6192818493204867, + 0.545974949342638, + 0.634434370845195, + 0.5294375198658107, + 0.6461476972292953, + 0.6308260262750621, + 0.6491336292115523, + 0.6681845329908763, + 0.5795944833608252, + 0.5925286661739508, + 0.6192308699414314, + 0.6364472776262159, + 0.5678773285210218, + 0.6039866716918647, + 0.5872533737291091, + 0.6149837558526225, + 0.588429578364095, + 0.6046615600054024, + 0.5706306290512568, + 0.6261953170139433, + 0.6119892067674669, + 0.5677964744459906, + 0.6194074765348856, + 0.6438349132816242, + 0.6280585424194255, + 0.6239597837479268, + 0.6427388761221088, + 0.6279308759029756, + 0.6521667271004816, + 0.6540760824501878, + 0.643922350642931, + 0.6350771032483081, + 0.6456836159492457, + 0.6489408607115226, + 0.6647141640753896, + 0.6599850791631365, + 0.6263484900371968, + 0.6434686820786335, + 0.63156402901981, + 0.6378106084637871, + 0.6417509076211443, + 0.6231607256114767, + 0.6296821000657126, + 0.6302664737293336, + 0.6210849948802479, + 0.6115827547593438, + 0.6326758093346225, + 0.631227539059237, + 0.6218835649563247, + 0.6274325741139086, + 0.6590193341847778, + 0.6371062841901134, + 0.6297418817249616, + 0.6161703617203247, + 0.6190038210064213, + 0.6217141557800949, + 0.6315954782315543, + 0.6291689961329254, + 0.6312590113620598, + 0.6394433142153859, + 0.6313902439641088, + 0.6360453634103461, + 0.6352310305198651, + 0.634647652883406, + 0.6546549932687704, + 0.6445765045363474, + 0.6620914574127024, + 0.6298230403877034, + 0.649967971724533, + 0.6293815394098617, + 0.6292590018059141, + 0.6468938083503073, + 0.6316280911971195, + 0.6197095499226167, + 0.6255780431898564, + 0.6299364097533694, + 0.6340427536846452, + 0.6221207229693079, + 0.6170721641870327, + 0.6147029287917396, + 0.6251951917247818, + 0.6360180776553007, + 0.6483211319627511, + 0.6309316466485861, + 0.6413390082374169, + 0.6121550269073868, + 0.6382013733871643, + 0.6344544224913955, + 0.6567764007648668, + 0.6309111097938613, + 0.6269668315626061, + 0.6521236619668295, + null, + 0.6427287746886328, + 0.6101409340253074, + 0.649530816138653, + 0.6583377085085051, + 0.6591835392898922, + 0.635769882935621, + 0.6373581907063824, + 0.6650142845107587, + 0.6579025536439086, + 0.6617635668314411, + null, + 0.6042928218035944, + 0.6535532000497027, + 0.6721606216989782, + 0.666003509506064, + 0.6540696033915177, + 0.6502039133934311, + 0.6305727914465314, + 0.6508282056093867, + 0.672802094133943, + 0.6645034679217526, + 0.6328173374036397, + 0.6832212415036453, + 0.6552538404435746, + 0.6636272474851646, + 0.6755407031002727, + 0.6357065456325944, + 0.6204635824184782, + 0.5958052498758765, + 0.5916052789086629, + 0.589898373995385, + 0.659723794318824, + 0.64093053997684, + 0.6565318035927316, + 0.655950164965362, + 0.656211632223534, + 0.6547281336203532, + 0.6607129817362074, + 0.651509298347972, + 0.6412943390909733, + 0.6462628499962384, + 0.6191157544337145, + 0.6431298515412324, + 0.6390527108928273, + 0.6364906984324128, + 0.6606655122221536, + 0.6518051220324497, + 0.6461605051748376, + 0.6335859873465512, + 0.6555455269556074, + 0.644650774917267, + 0.6519399263163472, + 0.6322993640899183, + 0.6332060170046088, + 0.6365421819253705, + 0.6463607735778599, + 0.6223502711774126, + 0.6433748184125654, + 0.6328968955083903, + 0.6640187961869904, + 0.6529289167238577, + 0.6158658280902369, + 0.6626569191327051, + 0.6282723775257238, + 0.6547855919471743, + 0.651389292376593, + 0.62214779082778, + 0.6462008550390579, + 0.6358034530891983, + 0.6392837714170125, + 0.6517414101894925, + 0.612688220148154, + 0.6360271476473996, + 0.6423404686487151, + null, + 0.5979697737990189, + 0.6678111574675444, + 0.6082834042516877, + 0.6671156854861322, + 0.6399684155633822, + 0.6358307820198688, + 0.6159824307474536, + null, + null, + 0.6237963922803752, + 0.6669621417918513, + null, + 0.6201413856715434, + 0.5850571379096434, + 0.6098314976737832, + 0.6225959898128672, + 0.6345447276065557, + 0.635447764031139, + 0.639398742043099, + null, + 0.6200022743717812, + 0.6374692949404268, + null, + 0.5875707168869384, + null, + 0.5989704990818197, + 0.6185954902086882, + 0.6077892355224818, + 0.6181833399616846, + null, + 0.6096631090397037, + 0.6056191727632624, + 0.6488589343741099, + null, + 0.5904723900315384, + null, + 0.6249225817786922, + null, + 0.6228045665309416, + 0.6159014760893183, + null, + 0.6105539148137851, + 0.6208484488913326, + null, + null, + 0.6009749804913235, + 0.6598279254799481, + 0.6083078150121715, + 0.613906949926106, + null, + 0.618506838748766, + null, + 0.6160363929831918, + 0.6389498961474225, + null, + 0.5108972863497822, + 0.6255821228997235, + 0.6069930195943182, + null, + 0.6335539330584384, + 0.6433513910200377, + null, + 0.6338301541014943, + 0.6162425279762527, + 0.6448567352615108, + null, + null, + null, + null, + 0.6488564760472287, + null, + null, + 0.6485147222593544, + 0.64124559239735, + 0.6137123358463493, + null, + 0.6362161172256259, + 0.6382241076273617, + null, + null, + 0.6489524697811027, + 0.6400543338379507, + 0.6514965003673018, + 0.6317497194087134, + 0.6529679410431815, + 0.6659931333784918, + 0.6127912857422269, + 0.6403268053347245, + 0.6121664263954802, + 0.6429816816313246, + 0.6318859602255993, + 0.6346720872453203, + null, + 0.611012331852167, + 0.6154643523962626, + 0.6194405439331804, + 0.6215698159740507, + 0.6348784706023101, + 0.6185778896699725, + 0.6366947035235015, + 0.6084208629587691, + 0.6127537390048464, + 0.6456726135414932, + 0.641248563053531, + 0.6388681836565949, + null, + 0.6265885986800579, + 0.6401646335297403, + 0.6549694465576551, + null, + 0.6305586977583405, + 0.5770051324906998, + null, + 0.6547043625010232, + null, + 0.5923502451929754, + 0.6477758714218897, + 0.6361084682706302, + 0.6154501760153378, + 0.6299657371197203, + 0.6246852495981429, + 0.620524686626688, + null, + null, + 0.614272285067319, + 0.6222332619790099, + null, + 0.6170875675367736, + null, + 0.6325715649567608, + 0.6228559688104449, + 0.6484553825081856, + 0.6194217160677762, + null, + null, + null, + null, + 0.6687733275774748, + null, + 0.6290519995168946, + 0.6408798657413743, + null, + null, + 0.6575163707204442, + 0.6263978589438512, + 0.6246977682054056, + null, + 0.6504103051266334, + 0.6545951012902358, + 0.6125134799206723, + null, + 0.6565218888321714, + null, + null, + 0.6623191943347264, + null, + 0.6173603080398964, + null, + null, + 0.6185656101859991, + 0.611626403075068, + 0.6190189818345214, + 0.6138253607403196, + 0.6176561640013173, + null, + 0.6072683329239446 + ], + "values_per_track": [ + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245, + 13245 + ], + "missing_prediction_ratio": [ + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702, + 0.3552283880709702 + ] + }, + "junctions": { + "mean_pearson": 0.45595872192240394, + "valid_tracks": 626, + "per_track_pearson": [ + 0.5829796350862045, + 0.522878873013264, + 0.6090519442331265, + 0.5585256827235451, + 0.6176758531571057, + 0.6509875680396954, + 0.0910809303065105, + 0.33156008439522644, + 0.5888145199520465, + 0.5223906526901066, + 0.5345666608589995, + 0.4532466718261355, + 0.34746841303301773, + 0.425980959322774, + 0.4861276500811827, + 0.42810310923970796, + 0.4599471263983987, + 0.3942610548137951, + 0.46929499390631274, + 0.4750969903724612, + 0.48479960380368514, + 0.5880424652559705, + 0.4823213141374942, + 0.5350858746231445, + 0.39860420558554394, + 0.41809664226637144, + 0.42259183505734427, + 0.5775341823847634, + 0.5803691137871233, + 0.5175436553613716, + 0.5229389171483587, + 0.634724403923335, + 0.5668061356114411, + 0.5929402950273145, + 0.638087423184158, + 0.5638571423421442, + 0.6049792171676677, + 0.45998904250258993, + 0.5583337836573311, + 0.4298436266663869, + 0.4622431943562, + 0.422383870002197, + 0.4538140149327794, + 0.5118500617073217, + 0.4809364261476672, + 0.4387452646420174, + 0.4170385383491526, + 0.5149321169271301, + 0.4368284890127719, + 0.5378618187576033, + 0.5024179918817308, + 0.5518495606243321, + 0.4087157600528622, + 0.4971259660210802, + 0.39798668661867914, + 0.41454945223219225, + 0.4341085015108582, + 0.38266666734312366, + 0.3787708501831279, + 0.4648454575093267, + 0.3069143682892139, + 0.3788582088242721, + 0.40745123387389426, + 0.4415223337592682, + 0.43355692651041206, + 0.4983045065266572, + 0.5102974986555519, + 0.44665437137107067, + 0.43973477355606194, + 0.43750672369067506, + 0.4129528981369091, + 0.451270465121032, + 0.4588004432682569, + 0.39954521941851767, + 0.5332212232433501, + 0.5296743353597728, + 0.5187181334794596, + 0.3923839024219054, + 0.5168926692010717, + 0.5111354942501525, + 0.47716769684125054, + 0.4165485986317431, + 0.4685948232705938, + 0.5161910148047256, + 0.5419855085024025, + 0.39292959979565106, + 0.45828284311472317, + 0.45314139702276357, + 0.4073277178664135, + 0.4432221414388154, + 0.36427159669542686, + 0.4578935247854477, + 0.46080641485269974, + 0.5262981484330149, + 0.4624376218219849, + 0.37322149811832284, + 0.4302914262836872, + 0.6289385025451526, + 0.5312703574609017, + 0.48224652749600105, + 0.43432382780224854, + 0.5116850343328513, + 0.5428408470419448, + 0.5280976271299183, + 0.5274052869529923, + 0.509357865233163, + 0.27427748571451394, + 0.5756429346092139, + 0.4820641972806721, + 0.5392375932003779, + 0.47067469488790437, + 0.5362634029110004, + 0.4871922181534327, + 0.5213512586582881, + 0.5293654638222465, + 0.4723242343049274, + 0.43195898076945644, + 0.4139765413669662, + 0.41610670713911946, + 0.4804957646903476, + 0.4320293144342665, + 0.44419872508376945, + 0.45269335691615037, + 0.4608948240137707, + 0.3980669375285576, + null, + 0.516279003708158, + 0.47891166293210147, + 0.2514605864520385, + 0.3588422654218835, + 0.53386510944069, + 0.5677914232721444, + 0.33930783906213113, + 0.46574748730237486, + 0.6033435417827702, + 0.5149721992446699, + null, + 0.5606315967259724, + 0.5025049787914986, + 0.5348051791656322, + 0.5068671594588964, + 0.49150126984409037, + 0.39321467441480495, + 0.43833718485148204, + 0.5034434660576224, + 0.28818753528806823, + 0.38041336108851725, + 0.42527182054961676, + 0.34555086394173284, + 0.38736150701916333, + 0.46597157699761677, + 0.5715633535778123, + 0.47227750839265525, + 0.7119296043617743, + 0.5579275995231416, + 0.5124096587122938, + 0.5596870262324395, + 0.4852267053305354, + 0.4167232038370843, + 0.5412272993149216, + 0.49117742557611677, + 0.41687561363572745, + 0.5992470969384703, + 0.5628075193092311, + 0.2759611551954417, + 0.573163746925008, + 0.5863823322240942, + 0.5819139631175468, + 0.30793332302703624, + 0.38277362516076513, + 0.37150525803576384, + 0.36368918842395775, + 0.5167637824308543, + 0.3405925481442404, + 0.42688566945011397, + 0.5049264476668628, + 0.4222436657846142, + 0.3285222254338497, + 0.34815314168013495, + 0.5220812916760249, + 0.43021718144673216, + 0.3426195830361499, + 0.3830580095412623, + 0.5655464103146753, + 0.510233168552129, + 0.5250042919637002, + 0.5450417560776284, + 0.4884228371454086, + 0.5185556744557205, + 0.5417020589910689, + 0.5086859987459622, + 0.5950171371485375, + 0.47897962948607514, + 0.5204985005681777, + 0.5356982114614526, + 0.4800225326404947, + 0.45445431014043397, + 0.4529891284131769, + 0.6127012618895262, + 0.529062838050919, + null, + 0.5197736660704555, + 0.5285048317652109, + 0.5689742522575783, + 0.5286488523387884, + 0.361180278784202, + 0.3939631155837578, + 0.4706815954535513, + null, + null, + 0.5471960180631898, + 0.5204213535013477, + null, + 0.40173705632926254, + 0.45423203466158063, + 0.35202227882448645, + 0.41631898353950836, + 0.48302221020776, + 0.47845867218892474, + 0.24907794055400162, + null, + 0.6041715378856226, + 0.5102104806222482, + null, + 0.4367327230405094, + null, + 0.5193408124719586, + 0.5701956012866627, + 0.4697202632247244, + 0.4537361707622219, + null, + 0.2716770619253155, + 0.3251458716371432, + 0.47075679920656976, + null, + 0.37097805074614115, + null, + 0.5877336854864998, + null, + 0.3133659486256113, + 0.46529650351019997, + null, + 0.5113689473480509, + 0.4658889585015142, + null, + null, + 0.46768662848265863, + 0.526874144070576, + 0.38270502090248604, + 0.392198628207151, + null, + 0.5674638332681925, + null, + 0.5198876518260341, + 0.39866984101437364, + null, + 0.37843463709996844, + 0.5019164988804095, + 0.5418430129268362, + null, + 0.4130689148769948, + 0.49024989974068667, + null, + 0.6008634066611218, + 0.6229668774713306, + 0.6112573254308251, + null, + null, + null, + null, + 0.503472245891851, + null, + null, + 0.3919010932388751, + 0.47286969286855074, + 0.6448243089404204, + null, + 0.5298892994780717, + 0.5609313388158614, + null, + null, + 0.5286936214052892, + 0.4419856322886822, + 0.49629897095994574, + 0.4950304957577388, + 0.5090971427341265, + 0.5232139473701523, + 0.40239860452406956, + 0.47328563798224615, + 0.4469420405205849, + 0.3564449063023464, + 0.47835090152614335, + 0.5769231137803161, + null, + 0.521781767857844, + 0.5411127830884476, + 0.25276802348825683, + 0.37858216577191106, + 0.42413950651561133, + 0.38478739846051846, + 0.3842854764624937, + 0.45174560300937144, + 0.47557027330519974, + 0.3695331310642391, + 0.5754190627193699, + 0.4447531689232436, + null, + 0.4377980462912458, + 0.44268949936058494, + 0.4408995896210428, + null, + 0.4177332875182021, + 0.30491506048842265, + null, + 0.5552334821732361, + null, + 0.45559412520337245, + 0.46022818385830316, + 0.49409581454942403, + 0.42651177889568503, + 0.3828899148459971, + 0.41301413725099606, + 0.38953377969728936, + null, + null, + 0.4929553690603887, + 0.4459949643184746, + null, + 0.5570236840727336, + null, + 0.5686009953393678, + 0.5142032908355875, + 0.3877402771280272, + 0.49776810231799135, + null, + null, + null, + null, + 0.3771571953851194, + null, + 0.4355827518467667, + 0.6231561350873683, + null, + null, + 0.5379537820899234, + 0.463266574578411, + 0.40923216055547795, + null, + 0.5723181346719852, + 0.5215509865136733, + 0.4412779486525474, + null, + 0.6479603377587628, + null, + null, + 0.5298665692939234, + null, + 0.30873430895596454, + null, + null, + 0.388254982093832, + 0.4637539773020985, + 0.5111820233741637, + 0.47244623309318384, + 0.4881782061309685, + null, + 0.47166789992117136, + 0.5254648195500637, + 0.43329390761866277, + 0.597782523169899, + 0.34313972810872023, + 0.6399559005247203, + 0.5119730256436591, + 0.24678128725386259, + 0.41702504381585137, + 0.5116403062196264, + 0.49675161403589385, + 0.33492455895764045, + 0.5071811484172157, + 0.34715598776405154, + 0.343362624808428, + 0.36273764240253326, + 0.37192963260365686, + 0.42229522405380604, + 0.5463897663520798, + 0.5995680940655396, + 0.5679843771716977, + 0.6200058564302545, + 0.591373166704556, + 0.457538133363376, + 0.6208034274577089, + 0.43013755605003123, + 0.3994287207425693, + 0.473863167180068, + 0.5155431917974821, + 0.45259769135039657, + 0.26476883268256163, + 0.4037593549991137, + 0.47622958193464393, + 0.37318744321689873, + 0.5145355538363654, + 0.26650817595839427, + 0.48649985300232773, + 0.22301919934635375, + 0.40164530790469155, + 0.5002473216851719, + 0.43759165131833405, + 0.5460334241744281, + 0.33331058930001345, + 0.2932987117154483, + 0.28477707402986346, + 0.39595345219239225, + 0.34157010244977654, + 0.34745723648730914, + 0.5389006623696218, + 0.3586564313432552, + 0.38674124335036947, + 0.35658681989150814, + 0.3476078665206038, + 0.3142854040339184, + 0.321634889193734, + 0.2568504993985113, + 0.44844803217490303, + 0.46016857127899985, + 0.43331339414685843, + 0.4320820049601899, + 0.43963941318116967, + 0.5676824290571999, + 0.39294978740334835, + 0.44081019714157355, + 0.4504996413241976, + 0.30460574833676546, + 0.3604782293754512, + 0.4401069439287724, + 0.39569796758350223, + 0.33729530747449843, + 0.3418663970085914, + 0.38607873134185017, + 0.4888612159538937, + 0.47528578734700117, + 0.39505069034033324, + 0.40547801683307716, + 0.37694617762257726, + 0.41212624794590536, + 0.42259078435565944, + 0.4449529135264547, + 0.46814256499299556, + 0.42018886431758146, + 0.41066514484855976, + 0.506744898288832, + 0.4727886825660496, + 0.3760470983856128, + 0.44990169111035133, + 0.5005373671926435, + 0.4467247326927478, + 0.40515203177896425, + 0.3971908680754706, + 0.4988017860770559, + 0.3327931700878283, + 0.4214254779346095, + 0.316520554913997, + 0.36813437849203823, + 0.3631984499629168, + 0.41838878905783317, + 0.39919818944190655, + 0.5409497928820112, + 0.5684666503333393, + 0.3861463755864776, + 0.4787441820685211, + 0.3630399373969669, + 0.4661953797270206, + 0.5283331729751332, + 0.4287863767276537, + 0.3223338693384473, + 0.45153985766230514, + 0.4059230992963505, + 0.39466244516149085, + 0.42229587320096235, + 0.40504873508250383, + 0.3428383155433209, + 0.44787688471102055, + 0.3936469969358155, + 0.42349415287479986, + 0.465917772586279, + 0.3947906923018497, + 0.3639270814485644, + 0.28972476314707113, + 0.4765772887221912, + 0.4361090783502004, + 0.382738689908506, + 0.45368313953841305, + 0.48937419163598184, + null, + 0.4944332639925523, + 0.5369501854031774, + 0.31496517539265173, + 0.5976025156130628, + 0.42294109693429555, + 0.2472927050584182, + 0.3542556982824414, + 0.43221360669744524, + 0.5307477005783041, + 0.34828793362830535, + null, + 0.4236976963157171, + 0.37665720438211875, + 0.35823554657974604, + 0.37523377456629753, + 0.4962261796951588, + 0.39533231672337216, + 0.4205705179927455, + 0.5520537566700789, + 0.2513091135505231, + 0.31720681664014994, + 0.4622351925475378, + 0.39072182148310075, + 0.36164483461639085, + 0.44518024911241805, + 0.5151302584720676, + 0.34011431282663973, + 0.6150350808717236, + 0.6489978469500572, + 0.5177825253024861, + 0.5047392978697576, + 0.40764338451062904, + 0.4555022130751738, + 0.3098140175919731, + 0.39259569685964163, + 0.34738678488488745, + 0.5353679540397354, + 0.5601969927332225, + 0.41603154472814835, + 0.3775241242323835, + 0.4943070721480346, + 0.4414337145093571, + 0.3860608323998358, + 0.4116959305617438, + 0.4334927467199279, + 0.43181031432576195, + 0.437719681967947, + 0.3072890503851338, + 0.3719572384700637, + 0.35907966296088617, + 0.5037235576290726, + 0.4535445458977782, + 0.44504382890738087, + 0.4432003563569925, + 0.3387156249359401, + 0.39854352542230914, + 0.5228667575814725, + 0.532575364568703, + 0.4876614680073244, + 0.6369258599077446, + 0.5288903329037844, + 0.4303661989496556, + 0.5429678785998779, + 0.4886128343677261, + 0.5524263386613946, + 0.585705877155895, + 0.42560866699560523, + 0.3299241669682995, + 0.3877835810386359, + 0.3998088151606081, + 0.3922831685579287, + 0.47112431672961236, + 0.557512065096056, + 0.5160034302858596, + null, + 0.5382277606473475, + 0.5339646792397517, + 0.4558403020484914, + 0.565609104453494, + 0.5773635363217705, + 0.37230485655035045, + 0.47090693608256046, + null, + null, + 0.4243367242880784, + 0.5548367473687523, + null, + 0.2868997979536225, + 0.4495853234526233, + 0.34315712247850766, + 0.3858729629468626, + 0.3958400369737296, + 0.36965649313411664, + 0.4181187574561683, + null, + 0.44562405114890186, + 0.3814419763441712, + null, + 0.38324661807958355, + null, + 0.5354277044942399, + 0.4502396649460089, + 0.4636466683723402, + 0.3505833478832496, + null, + 0.49575153697037844, + 0.33713562205839054, + 0.4816800509229433, + null, + 0.4587125228708586, + null, + 0.4250328181503992, + null, + 0.4489278269307745, + 0.4347637941758413, + null, + 0.4269067876639673, + 0.5622702985806327, + null, + null, + 0.4080768940513764, + 0.363578339502293, + 0.27952658207663045, + 0.3686991010166015, + null, + 0.4446140102018488, + null, + 0.3157438538234258, + 0.4647269976360178, + null, + 0.28944497948994447, + 0.5012348309476543, + 0.43418507486096863, + null, + 0.42889182542765586, + 0.5135396839619539, + null, + 0.5569685152494259, + 0.5265786312267883, + 0.5770277606447356, + null, + null, + null, + null, + 0.531077639480194, + null, + null, + 0.4812283880171429, + 0.4520446372507718, + 0.5914425851778013, + null, + 0.46839623362233274, + 0.5462886972418151, + null, + null, + 0.4566849425409472, + 0.2376898300030128, + 0.4684112559812901, + 0.36988683678028633, + 0.3727753515430256, + 0.4512517691532004, + 0.3214399818666256, + 0.3380443511659531, + 0.25088994533152564, + 0.4386711308517819, + 0.537630585914447, + 0.355136147317688, + null, + 0.4026146706156469, + 0.5244733916889236, + 0.5200541986432373, + 0.5409284772228734, + 0.3255232338371857, + 0.46118699104700445, + 0.44514390437244794, + 0.46587525507248095, + 0.5151784654100817, + 0.4628955748894557, + 0.4350014751136207, + 0.4517657898933005, + null, + 0.3932780042606678, + 0.4935886804591874, + 0.4372399110052767, + null, + 0.5417434766541763, + 0.36960331780165323, + null, + 0.5032653154656944, + null, + 0.6010948157549056, + 0.5047801134559491, + 0.47125022716249737, + 0.5516152782761519, + 0.48559930072872964, + 0.5620203402626088, + 0.5160895460577375, + null, + null, + 0.44983804787672516, + 0.6487337565396216, + null, + 0.34327248463455384, + null, + 0.461600914429043, + 0.5080740238816012, + 0.45403791604605986, + 0.5227197590889655, + null, + null, + null, + null, + 0.40852674924066495, + null, + 0.5037288930688419, + 0.5492082034046827, + null, + null, + 0.4450031764761906, + 0.361550711145546, + 0.3886443727507118, + null, + 0.4175308077358484, + 0.43689705669076756, + 0.4543284316505589, + null, + 0.5383452192947791, + null, + null, + 0.42202964176148894, + null, + 0.38268413850159494, + null, + null, + 0.432388409995948, + 0.5665419765922144, + 0.5768668743109666, + 0.6373337963893642, + 0.5716939141594652, + null, + 0.41903498984680226 + ], + "values_per_track": [ + 302, + 304, + 281, + 316, + 302, + 295, + 284, + 288, + 312, + 282, + 297, + 295, + 308, + 320, + 300, + 304, + 286, + 312, + 314, + 306, + 278, + 310, + 308, + 269, + 281, + 304, + 312, + 295, + 296, + 290, + 292, + 309, + 268, + 329, + 273, + 317, + 262, + 319, + 286, + 298, + 293, + 291, + 262, + 270, + 305, + 296, + 271, + 279, + 282, + 294, + 274, + 272, + 264, + 294, + 268, + 294, + 289, + 302, + 292, + 284, + 290, + 305, + 302, + 293, + 298, + 316, + 296, + 298, + 293, + 306, + 281, + 310, + 297, + 302, + 279, + 295, + 311, + 299, + 290, + 294, + 302, + 316, + 292, + 309, + 285, + 323, + 292, + 306, + 280, + 316, + 287, + 291, + 285, + 294, + 296, + 288, + 295, + 289, + 302, + 316, + 300, + 305, + 281, + 313, + 293, + 300, + 267, + 299, + 301, + 296, + 280, + 298, + 308, + 295, + 294, + 296, + 290, + 296, + 277, + 293, + 280, + 295, + 272, + 304, + 247, + 124, + 289, + 287, + 264, + 268, + 276, + 290, + 287, + 266, + 285, + 299, + 124, + 295, + 268, + 304, + 274, + 268, + 272, + 290, + 285, + 262, + 254, + 288, + 272, + 283, + 270, + 282, + 273, + 308, + 290, + 291, + 264, + 277, + 283, + 274, + 264, + 288, + 319, + 287, + 265, + 278, + 289, + 302, + 265, + 285, + 273, + 293, + 266, + 292, + 260, + 257, + 276, + 287, + 272, + 286, + 287, + 268, + 271, + 279, + 284, + 271, + 292, + 297, + 279, + 305, + 288, + 272, + 280, + 284, + 287, + 283, + 272, + 288, + 306, + 300, + 129, + 304, + 295, + 301, + 296, + 311, + 283, + 304, + 119, + 132, + 303, + 308, + 130, + 290, + 273, + 274, + 293, + 302, + 290, + 298, + 123, + 322, + 291, + 133, + 256, + 142, + 275, + 285, + 284, + 297, + 138, + 276, + 284, + 285, + 127, + 297, + 130, + 308, + 118, + 300, + 298, + 137, + 269, + 305, + 129, + 128, + 282, + 306, + 306, + 297, + 125, + 283, + 129, + 308, + 291, + 144, + 260, + 311, + 275, + 138, + 280, + 299, + 121, + 305, + 285, + 295, + 134, + 129, + 129, + 120, + 301, + 141, + 127, + 294, + 282, + 312, + 120, + 293, + 292, + 125, + 130, + 287, + 276, + 299, + 296, + 293, + 294, + 293, + 289, + 295, + 281, + 314, + 271, + 134, + 285, + 296, + 280, + 297, + 294, + 288, + 285, + 290, + 305, + 296, + 276, + 293, + 123, + 311, + 296, + 312, + 107, + 313, + 279, + 121, + 291, + 134, + 312, + 288, + 303, + 276, + 294, + 292, + 286, + 141, + 117, + 306, + 294, + 134, + 277, + 130, + 301, + 298, + 275, + 301, + 140, + 128, + 125, + 126, + 292, + 137, + 289, + 315, + 115, + 141, + 284, + 301, + 270, + 125, + 292, + 297, + 282, + 124, + 314, + 134, + 129, + 293, + 122, + 309, + 127, + 129, + 259, + 308, + 291, + 291, + 284, + 129, + 302, + 266, + 285, + 276, + 280, + 316, + 281, + 287, + 265, + 293, + 278, + 289, + 277, + 292, + 292, + 269, + 273, + 284, + 270, + 280, + 293, + 286, + 283, + 299, + 266, + 280, + 275, + 302, + 287, + 279, + 253, + 285, + 272, + 248, + 282, + 273, + 282, + 251, + 282, + 286, + 278, + 284, + 274, + 283, + 255, + 291, + 278, + 266, + 262, + 265, + 260, + 249, + 269, + 264, + 281, + 266, + 269, + 278, + 279, + 281, + 276, + 287, + 267, + 293, + 273, + 277, + 272, + 280, + 259, + 263, + 279, + 274, + 276, + 295, + 279, + 279, + 276, + 290, + 283, + 278, + 265, + 289, + 292, + 263, + 270, + 282, + 272, + 262, + 288, + 274, + 273, + 287, + 281, + 290, + 269, + 285, + 279, + 289, + 269, + 292, + 279, + 261, + 274, + 276, + 280, + 276, + 288, + 265, + 270, + 285, + 281, + 284, + 267, + 302, + 279, + 279, + 260, + 283, + 260, + 253, + 271, + 285, + 279, + 273, + 283, + 269, + 120, + 282, + 280, + 282, + 253, + 277, + 286, + 283, + 266, + 287, + 262, + 107, + 287, + 272, + 286, + 287, + 269, + 282, + 286, + 282, + 269, + 268, + 267, + 277, + 273, + 260, + 271, + 279, + 271, + 265, + 290, + 273, + 266, + 296, + 270, + 276, + 273, + 309, + 286, + 267, + 254, + 292, + 279, + 245, + 269, + 276, + 270, + 258, + 283, + 271, + 246, + 282, + 280, + 286, + 274, + 287, + 269, + 272, + 273, + 284, + 243, + 269, + 278, + 284, + 280, + 291, + 271, + 283, + 271, + 279, + 278, + 284, + 260, + 310, + 293, + 117, + 282, + 297, + 268, + 273, + 292, + 294, + 283, + 126, + 120, + 319, + 307, + 125, + 285, + 289, + 260, + 293, + 287, + 281, + 283, + 128, + 279, + 270, + 130, + 268, + 125, + 295, + 281, + 296, + 285, + 135, + 273, + 285, + 272, + 129, + 281, + 109, + 283, + 126, + 267, + 275, + 133, + 275, + 279, + 135, + 125, + 290, + 283, + 304, + 297, + 118, + 257, + 126, + 282, + 268, + 126, + 271, + 286, + 269, + 129, + 276, + 284, + 119, + 293, + 299, + 279, + 136, + 124, + 126, + 111, + 289, + 109, + 109, + 282, + 294, + 296, + 129, + 286, + 302, + 119, + 125, + 289, + 291, + 275, + 303, + 285, + 274, + 272, + 294, + 265, + 265, + 295, + 282, + 121, + 288, + 279, + 294, + 285, + 286, + 286, + 292, + 273, + 298, + 283, + 270, + 275, + 127, + 280, + 280, + 300, + 123, + 289, + 299, + 122, + 305, + 122, + 301, + 281, + 296, + 255, + 292, + 268, + 263, + 122, + 123, + 284, + 279, + 129, + 281, + 115, + 293, + 289, + 286, + 272, + 137, + 127, + 118, + 107, + 284, + 115, + 260, + 297, + 123, + 130, + 279, + 285, + 280, + 119, + 279, + 287, + 293, + 107, + 308, + 123, + 121, + 275, + 120, + 265, + 109, + 125, + 271, + 289, + 304, + 286, + 295, + 119, + 295 + ], + "missing_prediction_ratio": [ + 0.7384105960264901, + 0.7763157894736842, + 0.7188612099644128, + 0.7436708860759493, + 0.7649006622516556, + 0.7220338983050848, + 0.7077464788732394, + 0.7395833333333334, + 0.7724358974358975, + 0.7375886524822695, + 0.7306397306397306, + 0.7152542372881356, + 0.724025974025974, + 0.746875, + 0.7666666666666667, + 0.75, + 0.7132867132867133, + 0.7371794871794872, + 0.7579617834394905, + 0.738562091503268, + 0.7014388489208633, + 0.7612903225806451, + 0.7077922077922078, + 0.7620817843866171, + 0.7188612099644128, + 0.7828947368421053, + 0.7564102564102564, + 0.7728813559322034, + 0.7128378378378378, + 0.7482758620689656, + 0.7328767123287672, + 0.7443365695792881, + 0.75, + 0.7477203647416414, + 0.7362637362637363, + 0.7381703470031545, + 0.7137404580152672, + 0.7335423197492164, + 0.6678321678321678, + 0.738255033557047, + 0.7474402730375427, + 0.7869415807560137, + 0.767175572519084, + 0.7444444444444445, + 0.760655737704918, + 0.7533783783783784, + 0.7416974169741697, + 0.7311827956989247, + 0.7092198581560284, + 0.7312925170068028, + 0.7481751824817519, + 0.7573529411764706, + 0.7196969696969697, + 0.7687074829931972, + 0.7313432835820896, + 0.6938775510204082, + 0.7197231833910035, + 0.7715231788079471, + 0.7054794520547946, + 0.7077464788732394, + 0.7172413793103448, + 0.780327868852459, + 0.7649006622516556, + 0.7303754266211604, + 0.738255033557047, + 0.7689873417721519, + 0.7094594594594594, + 0.7483221476510067, + 0.726962457337884, + 0.7843137254901961, + 0.7330960854092526, + 0.7516129032258064, + 0.734006734006734, + 0.7682119205298014, + 0.7383512544802867, + 0.7593220338983051, + 0.7459807073954984, + 0.7290969899665551, + 0.7413793103448276, + 0.7619047619047619, + 0.7417218543046358, + 0.7436708860759493, + 0.7363013698630136, + 0.7475728155339806, + 0.7403508771929824, + 0.7739938080495357, + 0.773972602739726, + 0.7222222222222222, + 0.7321428571428571, + 0.7563291139240507, + 0.7421602787456446, + 0.738831615120275, + 0.7403508771929824, + 0.7585034013605442, + 0.7263513513513513, + 0.75, + 0.7322033898305085, + 0.7058823529411765, + 0.7715231788079471, + 0.7563291139240507, + 0.73, + 0.7475409836065574, + 0.7580071174377224, + 0.7763578274760383, + 0.757679180887372, + 0.7633333333333333, + 0.7153558052434457, + 0.7625418060200669, + 0.7574750830564784, + 0.75, + 0.7428571428571429, + 0.7550335570469798, + 0.7597402597402597, + 0.7220338983050848, + 0.7414965986394558, + 0.7668918918918919, + 0.7241379310344828, + 0.7027027027027027, + 0.703971119133574, + 0.7542662116040956, + 0.7035714285714286, + 0.7389830508474576, + 0.7316176470588235, + 0.7664473684210527, + 0.6882591093117408, + 0.9838709677419355, + 0.7647058823529411, + 0.7317073170731707, + 0.7651515151515151, + 0.75, + 0.75, + 0.7586206896551724, + 0.7456445993031359, + 0.7293233082706767, + 0.7578947368421053, + 0.7324414715719063, + 0.9758064516129032, + 0.7627118644067796, + 0.75, + 0.7763157894736842, + 0.7518248175182481, + 0.7052238805970149, + 0.7169117647058824, + 0.7103448275862069, + 0.7368421052631579, + 0.7213740458015268, + 0.7086614173228346, + 0.7118055555555556, + 0.6507352941176471, + 0.7173144876325088, + 0.7111111111111111, + 0.74822695035461, + 0.717948717948718, + 0.7694805194805194, + 0.7551724137931034, + 0.7560137457044673, + 0.7159090909090909, + 0.8014440433212996, + 0.7137809187279152, + 0.7518248175182481, + 0.696969696969697, + 0.71875, + 0.7460815047021944, + 0.7073170731707317, + 0.6943396226415094, + 0.7589928057553957, + 0.7474048442906575, + 0.7847682119205298, + 0.6943396226415094, + 0.7298245614035088, + 0.7032967032967034, + 0.726962457337884, + 0.7518796992481203, + 0.75, + 0.7269230769230769, + 0.7354085603112841, + 0.7028985507246377, + 0.7386759581881533, + 0.7022058823529411, + 0.7377622377622378, + 0.7282229965156795, + 0.7201492537313433, + 0.6789667896678967, + 0.7275985663082437, + 0.7640845070422535, + 0.7749077490774908, + 0.7226027397260274, + 0.7205387205387206, + 0.7060931899641577, + 0.7967213114754098, + 0.7361111111111112, + 0.7683823529411765, + 0.725, + 0.6866197183098591, + 0.7003484320557491, + 0.6784452296819788, + 0.6580882352941176, + 0.7256944444444444, + 0.7549019607843137, + 0.74, + 0.9689922480620154, + 0.7631578947368421, + 0.7050847457627119, + 0.6843853820598007, + 0.7398648648648649, + 0.7588424437299035, + 0.7279151943462897, + 0.7171052631578947, + 0.9915966386554622, + 0.9848484848484849, + 0.7425742574257426, + 0.7435064935064936, + 0.9769230769230769, + 0.7206896551724138, + 0.6373626373626373, + 0.6824817518248175, + 0.6962457337883959, + 0.7019867549668874, + 0.7137931034482758, + 0.6879194630872483, + 0.967479674796748, + 0.7204968944099379, + 0.7285223367697594, + 0.9849624060150376, + 0.64453125, + 0.9929577464788732, + 0.6763636363636364, + 0.7298245614035088, + 0.7112676056338029, + 0.7171717171717171, + 0.9710144927536232, + 0.6920289855072463, + 0.7253521126760564, + 0.7052631578947368, + 0.9763779527559056, + 0.696969696969697, + 0.9923076923076923, + 0.711038961038961, + 0.9745762711864406, + 0.7, + 0.7080536912751678, + 0.9708029197080292, + 0.6840148698884758, + 0.7147540983606557, + 0.9689922480620154, + 0.9765625, + 0.6914893617021277, + 0.7287581699346405, + 0.7058823529411765, + 0.7138047138047138, + 0.952, + 0.7067137809187279, + 1.0, + 0.6883116883116883, + 0.7079037800687286, + 0.9791666666666666, + 0.7038461538461539, + 0.7009646302250804, + 0.7018181818181818, + 0.9710144927536232, + 0.6642857142857143, + 0.7123745819397993, + 1.0, + 0.7442622950819672, + 0.6842105263157895, + 0.7186440677966102, + 0.9776119402985075, + 0.9922480620155039, + 0.9612403100775194, + 0.9666666666666667, + 0.7142857142857143, + 0.950354609929078, + 0.9921259842519685, + 0.7278911564625851, + 0.7375886524822695, + 0.7211538461538461, + 0.9916666666666667, + 0.7133105802047781, + 0.7191780821917808, + 0.992, + 0.9769230769230769, + 0.7247386759581882, + 0.6739130434782609, + 0.7224080267558528, + 0.7027027027027027, + 0.7098976109215017, + 0.6972789115646258, + 0.6996587030716723, + 0.698961937716263, + 0.6711864406779661, + 0.6868327402135231, + 0.7261146496815286, + 0.6752767527675276, + 0.9925373134328358, + 0.7192982456140351, + 0.7128378378378378, + 0.7214285714285714, + 0.7104377104377104, + 0.7448979591836735, + 0.7256944444444444, + 0.7157894736842105, + 0.6896551724137931, + 0.6950819672131148, + 0.7094594594594594, + 0.7137681159420289, + 0.7133105802047781, + 0.967479674796748, + 0.729903536977492, + 0.7432432432432432, + 0.7371794871794872, + 1.0, + 0.7188498402555911, + 0.6774193548387096, + 0.9834710743801653, + 0.718213058419244, + 0.9925373134328358, + 0.7243589743589743, + 0.6979166666666666, + 0.7095709570957096, + 0.7028985507246377, + 0.7346938775510204, + 0.6952054794520548, + 0.6958041958041958, + 0.9787234042553191, + 0.9914529914529915, + 0.696078431372549, + 0.7040816326530612, + 0.9701492537313433, + 0.6787003610108303, + 0.9846153846153847, + 0.7209302325581395, + 0.7013422818791947, + 0.6509090909090909, + 0.7209302325581395, + 0.9857142857142858, + 0.984375, + 0.944, + 0.9682539682539683, + 0.6815068493150684, + 0.9635036496350365, + 0.71280276816609, + 0.7587301587301587, + 0.991304347826087, + 0.9929078014184397, + 0.7077464788732394, + 0.7142857142857143, + 0.7148148148148148, + 0.984, + 0.726027397260274, + 0.7272727272727273, + 0.6702127659574468, + 0.9919354838709677, + 0.7738853503184714, + 0.9850746268656716, + 0.9844961240310077, + 0.7303754266211604, + 0.9836065573770492, + 0.7249190938511327, + 0.9763779527559056, + 1.0, + 0.667953667953668, + 0.7305194805194806, + 0.7216494845360825, + 0.7079037800687286, + 0.6866197183098591, + 0.9922480620155039, + 0.695364238410596, + 0.6954887218045113, + 0.7368421052631579, + 0.7318840579710145, + 0.7035714285714286, + 0.8164556962025317, + 0.6725978647686833, + 0.7142857142857143, + 0.6641509433962264, + 0.7133105802047781, + 0.7014388489208633, + 0.7058823529411765, + 0.7075812274368231, + 0.7328767123287672, + 0.7397260273972602, + 0.7174721189591078, + 0.6886446886446886, + 0.6654929577464789, + 0.674074074074074, + 0.7178571428571429, + 0.7064846416382252, + 0.6958041958041958, + 0.7102473498233216, + 0.7658862876254181, + 0.7443609022556391, + 0.6928571428571428, + 0.730909090909091, + 0.7119205298013245, + 0.7317073170731707, + 0.7204301075268817, + 0.6521739130434783, + 0.7403508771929824, + 0.7352941176470589, + 0.7096774193548387, + 0.7375886524822695, + 0.73992673992674, + 0.75177304964539, + 0.6972111553784861, + 0.6737588652482269, + 0.6713286713286714, + 0.6690647482014388, + 0.6690140845070423, + 0.7992700729927007, + 0.7491166077738516, + 0.6862745098039216, + 0.718213058419244, + 0.7517985611510791, + 0.7105263157894737, + 0.7175572519083969, + 0.7320754716981132, + 0.6923076923076923, + 0.7188755020080321, + 0.7100371747211895, + 0.7083333333333334, + 0.7437722419928826, + 0.6992481203007519, + 0.724907063197026, + 0.7517985611510791, + 0.7455197132616488, + 0.697508896797153, + 0.677536231884058, + 0.7212543554006968, + 0.7228464419475655, + 0.7610921501706485, + 0.7106227106227107, + 0.7111913357400722, + 0.7058823529411765, + 0.6892857142857143, + 0.6254826254826255, + 0.6730038022813688, + 0.6881720430107527, + 0.7372262773722628, + 0.6884057971014492, + 0.7593220338983051, + 0.6953405017921147, + 0.6989247311827957, + 0.6992753623188406, + 0.6896551724137931, + 0.6925795053003534, + 0.7230215827338129, + 0.7245283018867924, + 0.740484429065744, + 0.7123287671232876, + 0.6844106463878327, + 0.6888888888888889, + 0.7163120567375887, + 0.7389705882352942, + 0.7213740458015268, + 0.7534722222222222, + 0.6897810218978102, + 0.6923076923076923, + 0.7282229965156795, + 0.697508896797153, + 0.7241379310344828, + 0.7100371747211895, + 0.656140350877193, + 0.6953405017921147, + 0.7093425605536332, + 0.6617100371747212, + 0.7157534246575342, + 0.6845878136200717, + 0.6896551724137931, + 0.6861313868613139, + 0.7210144927536232, + 0.75, + 0.7427536231884058, + 0.71875, + 0.7056603773584905, + 0.737037037037037, + 0.7017543859649122, + 0.7295373665480427, + 0.7147887323943662, + 0.6966292134831461, + 0.6920529801324503, + 0.7096774193548387, + 0.7204301075268817, + 0.6884615384615385, + 0.7031802120141343, + 0.6884615384615385, + 0.6798418972332015, + 0.6752767527675276, + 0.7333333333333333, + 0.6810035842293907, + 0.663003663003663, + 0.734982332155477, + 0.6765799256505576, + 0.975, + 0.7304964539007093, + 0.7214285714285714, + 0.7163120567375887, + 0.6996047430830039, + 0.7292418772563177, + 0.7447552447552448, + 0.7208480565371025, + 0.6766917293233082, + 0.7247386759581882, + 0.6717557251908397, + 0.9626168224299065, + 0.7386759581881533, + 0.6691176470588235, + 0.7482517482517482, + 0.7700348432055749, + 0.6765799256505576, + 0.6879432624113475, + 0.6643356643356644, + 0.6843971631205674, + 0.6579925650557621, + 0.6865671641791045, + 0.6554307116104869, + 0.6787003610108303, + 0.684981684981685, + 0.6423076923076924, + 0.6605166051660517, + 0.7168458781362007, + 0.7490774907749077, + 0.7169811320754716, + 0.7586206896551724, + 0.7619047619047619, + 0.7406015037593985, + 0.706081081081081, + 0.7222222222222222, + 0.6702898550724637, + 0.673992673992674, + 0.7313915857605178, + 0.6853146853146853, + 0.6853932584269663, + 0.6771653543307087, + 0.702054794520548, + 0.7670250896057348, + 0.6326530612244898, + 0.6802973977695167, + 0.7355072463768116, + 0.6703703703703704, + 0.686046511627907, + 0.696113074204947, + 0.6715867158671587, + 0.7113821138211383, + 0.7163120567375887, + 0.7428571428571429, + 0.7237762237762237, + 0.7226277372262774, + 0.7073170731707317, + 0.7174721189591078, + 0.6654411764705882, + 0.6593406593406593, + 0.7570422535211268, + 0.7325102880658436, + 0.6877323420074349, + 0.6906474820143885, + 0.6725352112676056, + 0.7857142857142857, + 0.711340206185567, + 0.7195571955719557, + 0.7420494699646644, + 0.6420664206642066, + 0.6738351254480287, + 0.6762589928057554, + 0.6584507042253521, + 0.6576923076923077, + 0.7677419354838709, + 0.7201365187713311, + 0.9743589743589743, + 0.7375886524822695, + 0.7441077441077442, + 0.6567164179104478, + 0.717948717948718, + 0.7568493150684932, + 0.7517006802721088, + 0.6749116607773852, + 0.9682539682539683, + 0.9583333333333334, + 0.799373040752351, + 0.7785016286644951, + 0.96, + 0.6982456140350877, + 0.6505190311418685, + 0.6653846153846154, + 0.7303754266211604, + 0.7212543554006968, + 0.7046263345195729, + 0.7279151943462897, + 0.9609375, + 0.7347670250896058, + 0.6555555555555556, + 0.9923076923076923, + 0.6492537313432836, + 0.984, + 0.6644067796610169, + 0.7330960854092526, + 0.7432432432432432, + 0.7052631578947368, + 0.9259259259259259, + 0.6886446886446886, + 0.6842105263157895, + 0.6838235294117647, + 0.9922480620155039, + 0.6725978647686833, + 0.981651376146789, + 0.6819787985865724, + 0.9523809523809523, + 0.704119850187266, + 0.7054545454545454, + 0.9774436090225563, + 0.6872727272727273, + 0.7168458781362007, + 1.0, + 0.984, + 0.7586206896551724, + 0.7279151943462897, + 0.75, + 0.734006734006734, + 0.9745762711864406, + 0.6653696498054474, + 1.0, + 0.7092198581560284, + 0.7164179104477612, + 0.9761904761904762, + 0.6420664206642066, + 0.6993006993006993, + 0.6877323420074349, + 0.9689922480620154, + 0.677536231884058, + 0.7253521126760564, + 0.9747899159663865, + 0.7337883959044369, + 0.725752508361204, + 0.7383512544802867, + 0.9485294117647058, + 0.967741935483871, + 0.9761904761904762, + 0.954954954954955, + 0.7301038062283737, + 0.981651376146789, + 0.9908256880733946, + 0.7092198581560284, + 0.7448979591836735, + 0.75, + 0.9767441860465116, + 0.7202797202797203, + 0.7350993377483444, + 0.9747899159663865, + 0.96, + 0.71280276816609, + 0.7079037800687286, + 0.7090909090909091, + 0.7392739273927392, + 0.712280701754386, + 0.7043795620437956, + 0.6580882352941176, + 0.7142857142857143, + 0.6566037735849056, + 0.6679245283018868, + 0.7457627118644068, + 0.6808510638297872, + 0.9834710743801653, + 0.7465277777777778, + 0.6953405017921147, + 0.7414965986394558, + 0.7192982456140351, + 0.6818181818181818, + 0.7062937062937062, + 0.702054794520548, + 0.7032967032967034, + 0.7449664429530202, + 0.6925795053003534, + 0.6777777777777778, + 0.6872727272727273, + 0.968503937007874, + 0.7392857142857143, + 0.725, + 0.7833333333333333, + 0.983739837398374, + 0.7439446366782007, + 0.6889632107023411, + 0.9836065573770492, + 0.7049180327868853, + 0.9918032786885246, + 0.7342192691029901, + 0.7188612099644128, + 0.7466216216216216, + 0.6705882352941176, + 0.7328767123287672, + 0.6567164179104478, + 0.6387832699619772, + 0.9754098360655737, + 0.959349593495935, + 0.7323943661971831, + 0.6845878136200717, + 0.9844961240310077, + 0.708185053380783, + 0.991304347826087, + 0.726962457337884, + 0.7024221453287197, + 0.6958041958041958, + 0.7169117647058824, + 0.9343065693430657, + 0.9763779527559056, + 0.9830508474576272, + 0.9626168224299065, + 0.6936619718309859, + 0.9826086956521739, + 0.6846153846153846, + 0.7609427609427609, + 0.983739837398374, + 0.9923076923076923, + 0.7204301075268817, + 0.712280701754386, + 0.6785714285714286, + 0.9747899159663865, + 0.6953405017921147, + 0.7177700348432056, + 0.7098976109215017, + 0.9906542056074766, + 0.7857142857142857, + 0.967479674796748, + 0.9834710743801653, + 0.7381818181818182, + 0.9666666666666667, + 0.6754716981132075, + 0.9541284403669725, + 0.992, + 0.6826568265682657, + 0.7197231833910035, + 0.7401315789473685, + 0.7167832167832168, + 0.7186440677966102, + 0.9831932773109243, + 0.7050847457627119 + ] + }, + "contact_maps": { + "mean_pearson": 0.38043744553598574, + "valid_tracks": 28, + "per_track_pearson": [ + 0.3716412354800516, + 0.40206579283017835, + 0.42919586671173826, + 0.3595774146783457, + 0.2929615546279432, + 0.3377174360117163, + 0.3920844515299356, + 0.2947852341115412, + 0.4523259281883103, + 0.42310087432507704, + 0.3556537529682108, + 0.4140659210570622, + 0.40668220479825185, + 0.3479079735526697, + 0.3754637951729096, + 0.4715003672967436, + 0.10507814290987041, + 0.4052199554479566, + 0.40125603890473416, + 0.4048529584351213, + 0.36818951461695965, + 0.366632357346287, + 0.35145563177047984, + 0.4184340890063714, + 0.42157304298370374, + 0.4355254223441207, + 0.4039339452609476, + 0.4433675726403638 + ], + "values_per_track": [ + 25021379, + 25091190, + 20440544, + 21266921, + 22404268, + 20385006, + 20430091, + 22398973, + 19466103, + 19706533, + 20370661, + 18894329, + 19455687, + 20487691, + 20662439, + 19880454, + 12936503, + 19837607, + 24811009, + 20250504, + 20249859, + 20468632, + 19757763, + 20485378, + 19944390, + 19750895, + 19989917, + 19835794 + ], + "missing_prediction_ratio": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ] + } + }, + "overall_mean_family_pearson": 0.5227708114071791, + "invalid_reason": null +} diff --git a/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_variants.json b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_variants.json new file mode 100644 index 0000000000000000000000000000000000000000..a4d3529179a31983c6d0b4e342c93089ba3c1265 --- /dev/null +++ b/tasks/genome-sequence-to-function/genome-sequence-to-function-astra-xhigh-1h100-48h-20261022a/submissions/S041/validation_variants.json @@ -0,0 +1,125 @@ +{ + "eqtl_sign": { + "rows": 17052, + "perturbations": 165, + "metric": "auroc", + "model": 0.5706520239936081, + "anchors": { + "AlphaGenome": 0.8263569969634328 + } + }, + "eqtl_causality": { + "rows": 17934, + "perturbations": 174, + "metric": "auroc", + "model": 0.5433474290617147, + "anchors": { + "AlphaGenome": 0.8325427468284611 + } + }, + "eqtl_coefficient": { + "rows": 8575, + "perturbations": 165, + "metric": "spearman", + "model": 0.24705074296141663, + "anchors": { + "AlphaGenome": 0.4133539365891599 + } + }, + "sqtl_causality": { + "rows": 21340, + "perturbations": 194, + "metric": "auprc", + "model": 0.7536302682576997, + "anchors": { + "AlphaGenome": 0.8447114201904594 + } + }, + "clinvar_noncoding": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.6457456805917281, + "anchors": { + "AlphaGenome": 0.9409857958177288 + } + }, + "clinvar_splice_region": { + "rows": 194, + "perturbations": 194, + "metric": "auprc", + "model": 0.7048370898552894, + "anchors": { + "AlphaGenome": 0.7224791623996103 + } + }, + "clinvar_missense": { + "rows": 199, + "perturbations": 199, + "metric": "auprc", + "model": 0.32204879054464763, + "anchors": { + "AlphaGenome": 0.373826292978903 + } + }, + "mfass": { + "rows": 195, + "perturbations": 195, + "metric": "auprc", + "model": 0.49504606852726185, + "anchors": { + "AlphaGenome": 0.7342557141920668 + } + }, + "caqtl": { + "rows": 198, + "perturbations": 198, + "metric": "auprc", + "model": 0.13781464670763982, + "anchors": { + "AlphaGenome": 0.8075760167741196 + } + }, + "bqtl_spi1": { + "rows": 160, + "perturbations": 160, + "metric": "auprc", + "model": 0.2214636709105544, + "anchors": { + "AlphaGenome": 0.8167450781363793 + } + }, + "enhancer_gene": { + "rows": 200, + "perturbations": 200, + "metric": "auprc", + "model": 0.40797520459422204, + "anchors": { + "AlphaGenome": 0.9594307235541584 + } + }, + "traitgym_mendelian": { + "rows": 292, + "perturbations": 292, + "metric": "auprc", + "model": 0.21983696912091938, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.5026323772738239, + "Enformer_L2_L2.plus.all": 0.5452938267693526, + "CADD.plus.RawScore": 0.7189111372856957, + "GPN-MSA_absLLR.plus.score": 0.6729149139043132 + } + }, + "traitgym_complex": { + "rows": 299, + "perturbations": 299, + "metric": "auprc", + "model": 0.2709726043637668, + "anchors": { + "Borzoi_L2_L2.plus.all": 0.294501577162467, + "Enformer_L2_L2.plus.all": 0.269227301087889, + "CADD.plus.RawScore": 0.41341570280137435, + "GPN-MSA_absLLR.plus.score": 0.4213971415058371 + } + } +} diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S009/model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S009/model.py new file mode 100644 index 0000000000000000000000000000000000000000..dd8029dd09fabebdafa2cb65a14e9a4b8b65c2a9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S009/model.py @@ -0,0 +1,77 @@ +"""Shared metadata-conditioned convolutional operator, trained from scratch.""" +import math +import numpy as np +import torch +from torch import nn +import torch.nn.functional as F + +DATASETS=['ape2d/burgers','ape2d/decay_turb','ape2d/diff','ape2d/fisher','ape2d/gs_alpha','ape2d/gs_beta','ape2d/gs_delta','ape2d/gs_epsilon','ape2d/gs_gamma','ape2d/gs_iota','ape2d/gs_kappa','ape2d/gs_theta','ape2d/kdv','ape2d/kolm_flow','ape2d/ks','ape2d/sh','pdearena/navier_stokes_2d','well/active_matter','well/turbulent_radiative_layer_2D'] +CONSTANTS=['Viscosity','Viscosity X','Viscosity Y','Diffusivity','Reactivity','Feed Rate','Kill Rate','Critical Number','Domain Extent','tcool','L','zeta','alpha','buoyancy_y'] +MAXC=11 + +def metadata(meta,dataset): + vals=[math.log(max(float(meta.get('dt',1)),1e-12))/5] + vals += [math.log(max(float(v),1e-12))/5 for v in meta.get('domain_extent',[1,1])] + bc=meta.get('boundary_conditions',['periodic']*4) + vals += [float(b=='periodic') for b in bc]+[float(b=='wall') for b in bc] + const=meta.get('constants',{}) + for k in CONSTANTS: + v=float(const.get(k,0)); vals += [math.copysign(math.log1p(abs(v)*100),v)/8,float(k in const)] + return np.array(vals,np.float32),DATASETS.index(dataset)+1 if dataset in DATASETS else 0,(bc[0]=='periodic',bc[2]=='periodic') + +METADIM=3+8+2*len(CONSTANTS) + +def pad(x,n,bc): + x=F.pad(x,(n,n,0,0),mode='circular' if bc[1] else 'replicate') + return F.pad(x,(0,0,n,n),mode='circular' if bc[0] else 'replicate') + +class Block(nn.Module): + def __init__(self,c,cond): + super().__init__(); self.dw=nn.Conv2d(c,c,5,groups=c); self.norm=nn.LayerNorm(c); self.fc1=nn.Linear(c,2*c);self.fc2=nn.Linear(2*c,c) + self.film=nn.Linear(cond,2*c); nn.init.zeros_(self.film.weight);nn.init.zeros_(self.film.bias) + self.scale=nn.Parameter(torch.ones(c)*0.1) + def forward(self,x,e,bc): + h=self.dw(pad(x,2,bc)).permute(0,2,3,1);h=self.norm(h) + g,b=self.film(e).chunk(2,-1);h=h*(1+g[:,None,None,:])+b[:,None,None,:] + h=self.fc2(F.gelu(self.fc1(h)))*self.scale + return x+h.permute(0,3,1,2) + +class Operator(nn.Module): + def __init__(self,width=48,depth=2): + super().__init__(); self.width=width;self.depth=depth;d=192;ws=[width,width*2,width*4,width*6] + self.ds=nn.Embedding(len(DATASETS)+1,32) + self.meta=nn.Sequential(nn.Linear(METADIM+32+MAXC*3,d),nn.SiLU(),nn.Linear(d,d),nn.SiLU()) + self.stem=nn.Conv2d(4*MAXC+2,ws[0],3) + self.enc=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws]) + self.down=nn.ModuleList([nn.Conv2d(ws[i],ws[i+1],2,stride=2) for i in range(3)]) + self.up=nn.ModuleList([nn.Conv2d(ws[i+1]+ws[i],ws[i],1) for i in range(3)]) + self.dec=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws[:-1]]) + self.head=nn.Conv2d(ws[0],MAXC,1);nn.init.zeros_(self.head.weight);nn.init.zeros_(self.head.bias) + def forward(self,x,md,ids,bc=(True,True)): + # Reversible feature normalization is computed from supplied states, with fixed learned parameters. + b,t,c,h,w=x.shape + mean=x[:,-1].mean((-2,-1),keepdim=True) + std=x.std((1,3,4),unbiased=False).clamp_min(1e-3)[:,:,None,None] + z=(x-mean[:,None])/std[:,None] + mask=torch.ones(b,c,device=x.device) + sm=torch.cat([torch.asinh(mean.flatten(1)),torch.log(std.flatten(1))/5,mask],1) + if c0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/infer.sh b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..bbf5ac5eb2acfbbad07b871df5e8d072b2304862 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/infer.sh @@ -0,0 +1,3 @@ +#!/bin/bash +set -euo pipefail +exec python "$(dirname "$0")/launch.py" "$@" diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/kinetic.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/kinetic.py new file mode 100644 index 0000000000000000000000000000000000000000..eb920d27341683a42347392c7a7a52cad881cf9a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/kinetic.py @@ -0,0 +1,137 @@ +"""Portable, fixed-parameter angular moment operator; PyTorch JIT accelerates inference.""" +import math +from pathlib import Path +from typing import Tuple +import numpy as np +import torch +torch._C._jit_set_texpr_fuser_enabled(False) +torch._C._jit_override_can_fuse_on_cpu(False) +torch._C._jit_override_can_fuse_on_gpu(False) +torch._C._jit_set_profiling_executor(False) + +@torch.jit.script +def flow(s:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor]: + c=s[0].real;q=s[1].real*.5;r=s[1].imag*.5;sr=s[2].real;si=s[2].imag + base0=2*alpha*q-.8*zeta*((c-sr)*q-si*r);base1=alpha*r-.4*zeta*((c+sr)*r-si*q);pc=.1*c.mean();ps=psi + for _ in range(iterations): + e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]) + sig0=base0+.4*((c+sr)*e0+si*e1);sig1=base1+.2*(si*e0+(c-sr)*e1) + sf=torch.fft.rfft2(torch.stack([sig0,sig1]));raw=(kx*ky*sf[0]+(ky*ky-kx*kx)*sf[1])*inv*inv*mask + ps=(raw+pc*ps)/(1+pc) + vel=torch.fft.irfft2(torch.stack([1j*ky*ps,-1j*kx*ps]),s=[nx,ny]);e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]);om=torch.fft.irfft2((kx*kx+ky*ky)*ps,s=[nx,ny]) + return vel,e0,e1,om,ps + +@torch.jit.script +def rhs(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + s=torch.fft.ifft2(v);vel,e0,e1,om,ps=flow(s,psi,khx,khy,inv,maskh,alpha,zeta,iterations,nx,ny) + dx=torch.fft.ifft2(1j*kx*v);dy=torch.fft.ifft2(1j*ky*v);out=-vel[0]*dx-vel[1]*dy + A=e1+zeta*s[1].imag;B=-e0-zeta*s[1].real + out=out+1j*an*.5*om*s + out[:-1]=out[:-1]+.5j*an[:-1]*(A-1j*B)*s[1:] + out[1:]=out[1:]+.5j*an[1:]*(A+1j*B)*s[:-1] + out[0]=out[0].real.to(out.dtype) + nf=torch.fft.fft2(out)*mask;nf[0,0,0].zero_() + return nf,ps + +@torch.jit.script +def integrate(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,e:torch.Tensor,e2:torch.Tensor,q:torch.Tensor,f1:torch.Tensor,f2:torch.Tensor,f3:torch.Tensor,alpha:float,zeta:float,iterations:int,substeps:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + ps=psi + for _ in range(substeps): + nv,ps=rhs(v,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + a=e2*v+q*nv;na,ps=rhs(a,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + b=e2*v+q*na;nb,ps=rhs(b,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + cc=e2*a+q*(2*nb-nv);nc,ps=rhs(cc,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + v=e*v+f1*nv+2*f2*(na+nb)+f3*nc + return v,ps + +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if n0).all();moment=((y[3]-y[6]).square()+4*y[4].square()<=1.21*y[0].square()).all() + if not (finite and positive and moment):break + out.append(y) + return torch.stack(out) if out else self.context.new_empty((0,11,*self.original_shape)) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/make_bingham_table.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/make_bingham_table.py new file mode 100644 index 0000000000000000000000000000000000000000..3cba92fb5f0503306c910a27c10b6e42c7099aa6 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/make_bingham_table.py @@ -0,0 +1,9 @@ +import numpy as np +from scipy.special import ive +from pathlib import Path +s=np.linspace(0,12,8193);mu=-np.expm1(-s);r=mu[1:];k=np.where(r<.53,2*r+r**3+5*r**5/6,np.where(r<.85,-.4+1.39*r+.43/(1-r),1/(r**3-4*r**2+3*r))) +for _ in range(6): + z=ive(1,k)/ive(0,k);k=np.maximum(k+(r-z)/np.maximum(1-z*z-z/k,1e-14),1e-10) +t=np.zeros((65,len(s)),np.float32);t[0]=1 +for j in range(1,65):t[j,1:]=ive(j,k)/ive(0,k) +np.save(Path(__file__).with_name('bingham_table.npy'),t) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/model.py new file mode 100644 index 0000000000000000000000000000000000000000..dd8029dd09fabebdafa2cb65a14e9a4b8b65c2a9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/model.py @@ -0,0 +1,77 @@ +"""Shared metadata-conditioned convolutional operator, trained from scratch.""" +import math +import numpy as np +import torch +from torch import nn +import torch.nn.functional as F + +DATASETS=['ape2d/burgers','ape2d/decay_turb','ape2d/diff','ape2d/fisher','ape2d/gs_alpha','ape2d/gs_beta','ape2d/gs_delta','ape2d/gs_epsilon','ape2d/gs_gamma','ape2d/gs_iota','ape2d/gs_kappa','ape2d/gs_theta','ape2d/kdv','ape2d/kolm_flow','ape2d/ks','ape2d/sh','pdearena/navier_stokes_2d','well/active_matter','well/turbulent_radiative_layer_2D'] +CONSTANTS=['Viscosity','Viscosity X','Viscosity Y','Diffusivity','Reactivity','Feed Rate','Kill Rate','Critical Number','Domain Extent','tcool','L','zeta','alpha','buoyancy_y'] +MAXC=11 + +def metadata(meta,dataset): + vals=[math.log(max(float(meta.get('dt',1)),1e-12))/5] + vals += [math.log(max(float(v),1e-12))/5 for v in meta.get('domain_extent',[1,1])] + bc=meta.get('boundary_conditions',['periodic']*4) + vals += [float(b=='periodic') for b in bc]+[float(b=='wall') for b in bc] + const=meta.get('constants',{}) + for k in CONSTANTS: + v=float(const.get(k,0)); vals += [math.copysign(math.log1p(abs(v)*100),v)/8,float(k in const)] + return np.array(vals,np.float32),DATASETS.index(dataset)+1 if dataset in DATASETS else 0,(bc[0]=='periodic',bc[2]=='periodic') + +METADIM=3+8+2*len(CONSTANTS) + +def pad(x,n,bc): + x=F.pad(x,(n,n,0,0),mode='circular' if bc[1] else 'replicate') + return F.pad(x,(0,0,n,n),mode='circular' if bc[0] else 'replicate') + +class Block(nn.Module): + def __init__(self,c,cond): + super().__init__(); self.dw=nn.Conv2d(c,c,5,groups=c); self.norm=nn.LayerNorm(c); self.fc1=nn.Linear(c,2*c);self.fc2=nn.Linear(2*c,c) + self.film=nn.Linear(cond,2*c); nn.init.zeros_(self.film.weight);nn.init.zeros_(self.film.bias) + self.scale=nn.Parameter(torch.ones(c)*0.1) + def forward(self,x,e,bc): + h=self.dw(pad(x,2,bc)).permute(0,2,3,1);h=self.norm(h) + g,b=self.film(e).chunk(2,-1);h=h*(1+g[:,None,None,:])+b[:,None,None,:] + h=self.fc2(F.gelu(self.fc1(h)))*self.scale + return x+h.permute(0,3,1,2) + +class Operator(nn.Module): + def __init__(self,width=48,depth=2): + super().__init__(); self.width=width;self.depth=depth;d=192;ws=[width,width*2,width*4,width*6] + self.ds=nn.Embedding(len(DATASETS)+1,32) + self.meta=nn.Sequential(nn.Linear(METADIM+32+MAXC*3,d),nn.SiLU(),nn.Linear(d,d),nn.SiLU()) + self.stem=nn.Conv2d(4*MAXC+2,ws[0],3) + self.enc=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws]) + self.down=nn.ModuleList([nn.Conv2d(ws[i],ws[i+1],2,stride=2) for i in range(3)]) + self.up=nn.ModuleList([nn.Conv2d(ws[i+1]+ws[i],ws[i],1) for i in range(3)]) + self.dec=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws[:-1]]) + self.head=nn.Conv2d(ws[0],MAXC,1);nn.init.zeros_(self.head.weight);nn.init.zeros_(self.head.bias) + def forward(self,x,md,ids,bc=(True,True)): + # Reversible feature normalization is computed from supplied states, with fixed learned parameters. + b,t,c,h,w=x.shape + mean=x[:,-1].mean((-2,-1),keepdim=True) + std=x.std((1,3,4),unbiased=False).clamp_min(1e-3)[:,:,None,None] + z=(x-mean[:,None])/std[:,None] + mask=torch.ones(b,c,device=x.device) + sm=torch.cat([torch.asinh(mean.flatten(1)),torch.log(std.flatten(1))/5,mask],1) + if cMAXC: + return np.concatenate([predict(model,x[:,i:i+MAXC].cpu().numpy(),meta,ds,steps,device,detail_decay,physics,active_tau,kinetic_cfg,arena) for i in range(0,c,MAXC)],1) + factor=2 if min(h,w)>=192 or min(h,w)>=128 and max(h,w)/min(h,w)>2 else 1 + z=F.avg_pool2d(x,factor) if factor>1 else x + md,di,bc=metadata(meta,ds);md=torch.tensor(md,device=device)[None];ids=torch.tensor([di],device=device) + detail=x[-1]-F.interpolate(z[-1,None],size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else None + ctx=z[None];out=[] + solver=SpectralUpdate(meta,(h,w),device) if physics and eligible(meta,c) else None + if arena and arena_eligible(meta,c) and min(h,w)>1:solver=ArenaOperator(meta,x[-1]) + fullstate=x[-1] + for k in range(steps): + with torch.autocast(device_type=device,dtype=torch.bfloat16,enabled=device=='cuda'):y=model(ctx,md,ids,bc) + if not torch.isfinite(y).all():y=ctx[:,-1].clone() + if solver is not None: + candidate=solver.step(fullstate) + if torch.isfinite(candidate).all(): + fullstate=candidate;y=(F.avg_pool2d(candidate,factor) if factor>1 else candidate)[None] + else:solver=None + ctx=torch.cat([ctx[:,1:],y[:,None]],1) + yh=F.interpolate(y,size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else y[0] + if detail is not None:yh=yh+detail*(detail_decay**(k+1)) + if solver is not None:yh=fullstate + yh=torch.where(torch.isfinite(yh),yh,x[-1]) + out.append(yh.cpu().numpy()) + pred=np.stack(out).astype(np.float32) + if active_tau is not None: + raw=torch.as_tensor(pred,device=device);projected=project_active(x,raw,meta,active_tau) + pred=torch.where(torch.isfinite(projected),projected,raw).cpu().numpy() + if kinetic_cfg and str(meta.get('pde',''))=='active_matter' and c==11 and list(meta.get('fields',[]))==['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3'] and all(k in meta.get('constants',{}) for k in ['L','alpha','zeta']): + from kinetic import KineticOperator + cfg=dict(kinetic_cfg) + try: + op=KineticOperator(meta,x,**cfg) + except RuntimeError: + if not cfg.get('cuda_graph',False):raise + cfg['cuda_graph']=False;op=KineticOperator(meta,x,**cfg) + physical=op.rollout(steps,guard=True);used=len(physical) + if used:pred[:used]=physical.cpu().numpy() + print(f'kinetic frames: {used}/{steps}',flush=True) + return pred + +def main(): + torch.set_num_threads(4);torch.manual_seed(0);torch.backends.cudnn.benchmark=False;torch.backends.cudnn.deterministic=True + inp,out=map(Path,sys.argv[1:]);here=Path(os.environ['PDE_SURROGATE_PACKAGE_DIR']) if 'PDE_SURROGATE_PACKAGE_DIR' in os.environ else Path(__file__).resolve().parent + device='cuda' if torch.cuda.is_available() else 'cpu';cfg=json.loads((here/'config.json').read_text()) + model=Operator(cfg['width'],cfg['depth']).to(device).eval() + model.load_state_dict(torch.load(here/'weights.pt',map_location=device,weights_only=True));(out/'cases').mkdir(parents=True,exist_ok=True) + index=json.loads((inp/'index.json').read_text());t0=time.time() + for i,case in enumerate(index['cases']): + with h5py.File(inp/'cases'/(case['case_id']+'.h5'),'r') as f: + x=f['input'][()];meta=dict(f.attrs);meta['constants']=json.loads(meta.get('constants','{}')) + y=predict(model,x,meta,case['dataset'],int(case['n_steps_out']),device,cfg.get('detail_decay',0.8),cfg.get('physics',False),cfg.get('active_tau',None),cfg.get('kinetic',None),cfg.get('arena',False)) + with h5py.File(out/'cases'/(case['case_id']+'.h5'),'w') as f:f.create_dataset('prediction',data=y) + print(i+1,len(index['cases']),case['dataset'],'seconds',round(time.time()-t0,1),flush=True) +if __name__=='__main__':main() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/spectral.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/spectral.py new file mode 100644 index 0000000000000000000000000000000000000000..9bccb91a9cf5e34afe7cf7b9a254c13e8b07d0a0 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S016/spectral.py @@ -0,0 +1,77 @@ +"""Metadata-parameterized Fourier update written independently in this session. + +The fixed differential operators supply a physics branch of the shared neural +surrogate. No trajectories or solver parameter fitting are accessed at inference. +""" +import math +import torch + +SUPPORTED={'Diffusion','Fisher-KPP','Gray Scott','Swift-Hohenberg','Kuramoto-Sivashinsky','Korteweg-de-Vries','Burgers','Navier-Stokes: Decaying Turbulence','Navier-Stokes: Kolmogorov Flow'} + +def eligible(meta,c): + name=str(meta.get('pde','')) + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + needed={'Diffusion':['Viscosity X','Viscosity Y'],'Fisher-KPP':['Diffusivity','Reactivity'],'Gray Scott':['Feed Rate','Kill Rate'],'Swift-Hohenberg':['Reactivity','Critical Number'],'Kuramoto-Sivashinsky':[],'Korteweg-de-Vries':['Viscosity'],'Burgers':['Viscosity'],'Navier-Stokes: Decaying Turbulence':['Viscosity'],'Navier-Stokes: Kolmogorov Flow':['Viscosity']} + return name in SUPPORTED and all(k in meta.get('constants',{}) for k in needed.get(name,[])) and c==nc and all(str(b)=='periodic' for b in meta.get('boundary_conditions',[])) + +def phi(z,n): + near=z.abs()<0.5;safe=torch.where(near,torch.ones_like(z),z) + v=torch.expm1(safe) + for j in range(1,n):v=v-safe**j/math.factorial(j) + v=v/safe**n;zs=torch.where(near,z,torch.zeros_like(z));series=torch.zeros_like(z) + for j in range(16):series=series+zs**j/math.factorial(n+j) + return torch.where(near,series,v) + +class SpectralUpdate: + def __init__(self,meta,shape,device): + self.order=4 + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.float();self.ky=ky.float();k2=kx*kx+ky*ky;self.k2=k2.float() + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).float() + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=2. + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=(torch.arange(ny,device=device)+7/16)*ly/ny;self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(torch.complex64 if v.is_complex() else torch.float32) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + v=self.fft(u)/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + for _ in range(self.substeps): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;v=stage+self.etd2*(self.nonlinear(stage)-nv);continue + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + v=self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c) + return self.ifft(v*self.observation) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/README.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/README.md new file mode 100644 index 0000000000000000000000000000000000000000..4f14c85549559220db8ce4eb907eac35e29fb9af --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/README.md @@ -0,0 +1,25 @@ +# Shared neural and spectral surrogate + +A single shared neural operator is evaluated for every case and every autoregressive step. A deterministic spectral branch conditions on PDE metadata and uses independently implemented differential operators where their form was verified on training data. Unsupported equations use the shared neural update. There are no separately trained specialist checkpoints, no pretrained weights and no inference-time fitting. All runtime assets are included. + +Selection uses validation rollouts only. The neural model was trained from random initialization on all train pools, with sampling probabilities documented in EXPERIMENTS.md. See EXPERIMENTS.md for per-dataset results and negative experiments. Run `bash infer.sh INPUT_DIR OUTPUT_DIR`. + +Active matter also uses a fixed Fourier angular-moment Smoluchowski/Stokes operator. A recurrent context encoder inserts the observed low moments into its physical state; no parameter optimization or training occurs at inference. The Bingham initialization table contains mathematical Bessel ratios generated from scratch. Neural forecasts remain available as a fallback if a physical update is non-finite or violates positivity/moment bounds. The entry point pins the package directory through an open file descriptor so atomic publication cannot mix runtime assets. + +The active context encoder transports hidden angular moments with the supplied concentration and orientation, then applies a fixed linear Stokes observation layer and a necessary moment-positivity projection. It consumes only the four supplied frames. This is a fixed state computation; model and solver parameters do not change at inference. + +The active hidden state uses three fixed uncertainty quadrature nodes: its encoded center and symmetric small angular perturbations. Each node uses the same physical parameters and shared neural model; identical computations are shared. Outputs average the common valid physical prefix. A fixed local random generator makes the quadrature reproducible for every input, independent of case IDs and ordering. No separate specialist checkpoints or inference-time training are used. + +Buoyant smoke uses a fixed staggered-grid finite-difference fluid operator with semi-Lagrangian advection, explicit viscosity, and a pressure projection. Numerical spacing and scheme were identified using training trajectories only, with no generator code or initial-condition seeds. The implementation uses only PyTorch. + +Neural forecasts use a fixed, horizon-dependent shrinkage toward profiles averaged over periodic directions, with nonnegative density and concentration. The coefficient was selected on nine training rollouts before validation. The physical operators are unaffected. + +Pressure forecasts have an additional fixed uncertainty decoder: after ten steps, spatial pressure patterns gradually shrink toward their own predicted spatial mean. The schedule was selected on training trajectories. No target statistics are used during inference. + +The shared neural operator evaluates long rectangular grids at native resolution. The checkpoint manifest and experiment ledger record training grid sizes, rollout horizon, sampling probabilities and checkpoint lineage. + +Reaction-diffusion updates retain a finer internal Fourier grid and use double-precision second-order exponential integration. The fixed observation operator maps cell averages to and from that state. Numerical choices were selected on training trajectories and tested on validation rollouts. CUDA graphs accelerate the fixed updates without changing their arithmetic. + +The Burgers and forced-vorticity updates also retain a finer internal Fourier grid in double precision. A fixed cell-averaging observation map returns the native output grid. Training-only resolution studies selected a twofold grid increase; smaller time steps and fourfold grids added little. + +The shared neural network includes a metadata-conditioned global Fourier feature layer. It was added with a zero output projection, preserving the parent model at initialization, then trained jointly across all four groups. New parameters were initialized from scratch in this session. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/infer.sh b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..bbf5ac5eb2acfbbad07b871df5e8d072b2304862 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/infer.sh @@ -0,0 +1,3 @@ +#!/bin/bash +set -euo pipefail +exec python "$(dirname "$0")/launch.py" "$@" diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/regularize.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/regularize.py new file mode 100644 index 0000000000000000000000000000000000000000..884d0f0848f06eea3c3a7150048b4447467d8210 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S044/regularize.py @@ -0,0 +1,18 @@ +"""Fixed spatial uncertainty shrinkage and nonnegative material scalars.""" +import torch + +def regularize(pred,meta,profile_mix=0.75,horizon=30,pressure_mix=0.,pressure_start=10,pressure_end=30): + bc=meta.get('boundary_conditions',['periodic']*4) + axes=tuple(axis for axis,edge in [(-2,0),(-1,2)] if bc[edge]=='periodic') + y=pred + if axes and profile_mix: + weight=(torch.arange(1,len(pred)+1,device=pred.device,dtype=pred.dtype)/horizon).clamp_max(1)*profile_mix + y=pred*(1-weight[:,None,None,None])+pred.mean(axes,keepdim=True)*weight[:,None,None,None] + y=y.clone() + for i,name in enumerate(meta.get('fields',[])): + if i1;self.initialized=False + self.double=gs_double if meta['pde']=='Gray Scott' else advection_double if advective else False + self.real_dtype=torch.float64 if self.double else torch.float32 + self.complex_dtype=torch.complex128 if self.double else torch.complex64 + if self.fine:shape=tuple(int(scale*s) for s in shape) + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.to(self.real_dtype);self.ky=ky.to(self.real_dtype);k2=kx*kx+ky*ky;self.k2=k2.to(self.real_dtype) + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if self.fine: + ox,oy=self.original_shape;fx=torch.fft.fftfreq(ox,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(oy,device=device,dtype=torch.float64)[None,:];self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=gs_dt;self.order=gs_order + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=torch.arange(ny,device=device,dtype=self.real_dtype)*ly/ny+7/16*ly/self.original_shape[1] if self.fine else (torch.arange(ny,device=device)+7/16)*ly/ny + self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(self.complex_dtype if v.is_complex() else self.real_dtype) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + self.graph=None + if cuda_graph and torch.device(device).type=='cuda' and name!='Diffusion': + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + self.buffer=torch.zeros((nc,nx,ny//2+1),device=device,dtype=self.complex_dtype) + try: + def body():self.buffer.copy_(self.substep(self.buffer)) + stream=torch.cuda.Stream();stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3):body() + torch.cuda.current_stream().wait_stream(stream);torch.cuda.synchronize();graph=torch.cuda.CUDAGraph() + with torch.cuda.graph(graph):body() + self.graph=graph + except RuntimeError: + self.graph=None + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + if self.fine: + if not self.initialized: + latent=torch.fft.irfft2(torch.fft.rfft2(u.to(self.real_dtype))/self.observation,s=self.original_shape);v=self.fft(resample(latent,self.shape));self.initialized=True + else:v=self.internal + else:v=self.fft(u.to(self.real_dtype))/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + if self.graph is not None: + self.buffer.copy_(v) + for _ in range(self.substeps):self.graph.replay() + v=self.buffer + else: + for _ in range(self.substeps):v=self.substep(v) + if self.fine: + self.internal=v + low=resample(self.ifft(v),self.original_shape) + return torch.fft.irfft2(torch.fft.rfft2(low)*self.observation,s=self.original_shape).float() + return self.ifft(v*self.observation).float() + def substep(self,v): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;return stage+self.etd2*(self.nonlinear(stage)-nv) + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + return self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/EXPERIMENTS.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/EXPERIMENTS.md new file mode 100644 index 0000000000000000000000000000000000000000..af5bed54de57cd533f00cd3b906456d7c2b84737 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/EXPERIMENTS.md @@ -0,0 +1,2104 @@ +# Experiment ledger + +All development uses train/valid pools of `/workspace/data/split.json`. No held-out targets or case definitions are used. One H100 GPU is available. No pretrained assets are used. + +## E000 — package initialization +- Hypothesis: a spatial-mean prior provides a finite, portable starting package while training begins. +- Changed factor: parameter-free spatial-mean prediction. +- Seed: not applicable (deterministic). +- Validation: fixed 81-case subset (seed 71, all start at frame 0), reward 0.313512; 0 failed cases. +- Checkpoint hash: no checkpoint; predictor SHA256 307c11c9f4f3f186b96a34597822e1e40e95d47d6159b3de3fe0934ebae13d7d. +- Outcome: initial package published; not a trained candidate or a reproduced baseline run. + +## E001 — coupled, metadata-conditioned multiscale model +- Hypothesis: joint channel modeling, physical metadata, scale-aware normalization and periodic boundaries improve cross-equation prediction relative to the documented starter weaknesses. +- Changed factors: new shared convolutional architecture; all 19 dataset identifiers represented in one parameter set; group-balanced training. +- Initialization seed: 20261021. +- Validation: pending. +- Checkpoint hash: pending. +- Outcome: preparing train-only reduced-resolution cache and validation data. + +### E000 validation VRMSE by dataset + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 1.000634 | +| ape2d/decay_turb | 0.999991 | +| ape2d/diff | 0.999973 | +| ape2d/fisher | 1.581309 | +| ape2d/gs_alpha | 1.005787 | +| ape2d/gs_beta | 1.023021 | +| ape2d/gs_delta | 1.591578 | +| ape2d/gs_epsilon | 1.004624 | +| ape2d/gs_gamma | 1.000439 | +| ape2d/gs_iota | 1.633860 | +| ape2d/gs_kappa | 1.079744 | +| ape2d/gs_theta | 1.574410 | +| ape2d/kdv | 1.000191 | +| ape2d/kolm_flow | 0.999992 | +| ape2d/ks | 0.999992 | +| ape2d/sh | 1.041517 | +| pdearena/navier_stokes_2d | 1.000318 | +| well/active_matter | 0.992894 | +| well/turbulent_radiative_layer_2D | 1.502768 | + +## E001-smoke — 100 updates +- Hypothesis / changed factor: Short pipeline check of new coupled operator with 40 base channels, 12 samples per batch, one-step objective, float16 cache; no baseline retraining. Float16 cache will be replaced to reduce quantization error. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Validation: fixed 81-case subset; reward **0.322278**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.779417 | +| ape2d/decay_turb | 0.919250 | +| ape2d/diff | 0.573262 | +| ape2d/fisher | 1.128033 | +| ape2d/gs_alpha | 0.999278 | +| ape2d/gs_beta | 1.093050 | +| ape2d/gs_delta | 1.567459 | +| ape2d/gs_epsilon | 0.717784 | +| ape2d/gs_gamma | 0.761942 | +| ape2d/gs_iota | 1.560647 | +| ape2d/gs_kappa | 0.956638 | +| ape2d/gs_theta | 1.511911 | +| ape2d/kdv | 0.764748 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.939004 | +| ape2d/sh | 0.919993 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — intermediate 2,000 updates +- Hypothesis / changed factor: Intermediate checkpoint of the same 8,000-update one-step run; all other factors unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `cc58e45b55743bae890d3c3f212be97b831186431591d2501186f69378ce6f37`. +- Validation: fixed 81-case subset; reward **0.263620**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.817172 | +| ape2d/decay_turb | 2.389786 | +| ape2d/diff | 0.837767 | +| ape2d/fisher | 1.272807 | +| ape2d/gs_alpha | 2.153748 | +| ape2d/gs_beta | 8.179170 | +| ape2d/gs_delta | 1.369924 | +| ape2d/gs_epsilon | 1.198089 | +| ape2d/gs_gamma | 1.805734 | +| ape2d/gs_iota | 1.964417 | +| ape2d/gs_kappa | 0.904690 | +| ape2d/gs_theta | 1.609477 | +| ape2d/kdv | 1.151686 | +| ape2d/kolm_flow | 2.257709 | +| ape2d/ks | 2.152634 | +| ape2d/sh | 0.836066 | +| pdearena/navier_stokes_2d | 1.003965 | +| well/active_matter | 2.135917 | +| well/turbulent_radiative_layer_2D | 1.657301 | + +Outcome: did not improve validation reward; retained current package. + +## E003 — metadata-conditioned spectral branch +- Hypothesis / changed factor: Add independently implemented ETDRK4 differential operators to the shared model. All neural weights unchanged from E001-smoke; neural branch still applied every step on every case. Equation parameters come from metadata; solver forms selected using train-only diagnostics. Kolmogorov remains with the neural update pending forcing/drag diagnosis. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Runtime code SHA256: `9e848b7d80d919c0b32a15d07bf4643a1b56b5235004827d0950660afa2e4952`. +- Validation: fixed 81-case subset; reward **0.401067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — one-step operator, 8,000 updates +- Hypothesis / changed factor: Joint channel ConvNeXt U-Net, metadata conditioning, group-balanced batches, float32 reduced-grid cache, one-step log variance-normalized MSE. Random initialization; 48 base channels and 16 samples per batch. Development inference wrapper relative-path error was corrected before scoring. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `9a645f05c3a17872093f7f461376b1fedd663ceefbbd879ef8f1cd4b8d8e58fb`. +- Validation: fixed 81-case subset; reward **0.350858**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: did not improve validation reward; retained current package. + +## E004 — trained neural branch and forced turbulence operator +- Hypothesis / changed factor: Replace smoke neural weights with E001 step 8,000, still one shared network across all cases. Add Kolmogorov forcing -4 cos(4y) and drag 0.2, identified only in training diagnostics. Both effects are recorded together; full group scores isolate transfer changes from corpus changes. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `d8dd1c803a436415851c119241900367c593b0acfb904f6f5b3f3b17b7444478`. +- Validation: fixed 81-case subset; reward **0.437086**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## E005A — physical output constraints and horizon decay (validation ablations) + +Hypothesis: exact tensor identities and reduced confidence at longer horizons stabilize active matter. All experiments reuse E001 8,000-update predictions; seed 20261021. No new weights or training data. Checkpoint SHA256: df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d. + +The diffusion coefficient 0.05, trace(D)=concentration and E=sym(gradient velocity) were checked on training pools. Horizon decay candidates 10, 20 and 40 steps are hyperparameters selected on validation, not parameters fitted to validation targets. These ablations are not published until integrated into a portable package and replayed. + +### active_diffusion: Analytic concentration diffusion only + +Reward 0.351796; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.620833 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_constraints: Concentration diffusion and algebraic tensor projection + +Reward 0.352944; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.570211 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_strain: Recompute strain from predicted velocity + +Reward 0.338342; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 2.350641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: negative; strain derivatives amplify velocity errors. + +### active_projection: Weighted Fourier projection of velocity and strain + +Reward 0.352917; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.571374 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_10: Weighted projection and exponential shrinkage, tau=10 + +Reward 0.375912; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832421 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_20: Weighted projection and exponential shrinkage, tau=20 + +Reward 0.372622; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.911641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_40: Weighted projection and exponential shrinkage, tau=40 + +Reward 0.365053; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.123714 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +## Train-only solver diagnostics (no validation selection) + +Initial Navier–Stokes sign and step-size hypotheses failed (non-finite or high training error); smaller steps and the opposite vorticity convention succeeded. KS requires separate fourth derivatives, not the squared Laplacian. Kolmogorov forcing and drag were identified using training trajectories only. Doubling the latent solver grid from 256 to 512 barely changed Gray–Scott/KS training errors and was rejected as a general fix. Grid averaging, integrator order and timestep are under investigation. These are diagnostic tests, not held-out experiments; validation VRMSE is not applicable. No checkpoint was created. Source, configurations and train results are retained under research/physics_diagnostics. + +## E005B — integrated physical projection and horizon regularization +- Hypothesis / changed factor: E004 model plus concentration diffusion, weighted Fourier projection of velocity and strain, tensor identities and exponential relaxation toward context means for active-matter fluctuations, tau=10 steps. All weights unchanged; tau selected from fixed validation ablations. The neural model is still applied to every case. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `83bec07505f714951064c32e8cb4cc20720921d9b62a1e4c1ae3bba353048798`. +- Validation: fixed 81-case subset; reward **0.462140**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Full validation of physics_smoke + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.400925**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.060324 | +| ape2d/decay_turb | 0.000387 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.002465 | +| ape2d/gs_alpha | 0.022938 | +| ape2d/gs_beta | 0.012380 | +| ape2d/gs_delta | 0.146653 | +| ape2d/gs_epsilon | 0.026477 | +| ape2d/gs_gamma | 0.222599 | +| ape2d/gs_iota | 0.036049 | +| ape2d/gs_kappa | 0.109056 | +| ape2d/gs_theta | 0.135853 | +| ape2d/kdv | 0.025460 | +| ape2d/kolm_flow | 0.979667 | +| ape2d/ks | 0.080605 | +| ape2d/sh | 0.002222 | +| pdearena/navier_stokes_2d | 0.923463 | +| well/active_matter | 1.120708 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E002 — intermediate 2,000 multi-step updates +- Hypothesis / changed factor: E005B inference architecture and physical operators; replace only shared neural weights with the 2,000-update checkpoint of four-step fine-tuning with 0.01 input noise. +- Seed: 20261021. +- Checkpoint SHA256: `d720a4dbeee0c16a125aca0cee22ee060e1e80c6b863efbd169fd69b5f2c9efc`. +- Runtime code SHA256: `f4b3be2c3e227fcfa3e8cb9774970780928d6635210a425fa2c6b07e8f08e8a5`. +- Validation: fixed 81-case subset; reward **0.473422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.559579 | +| well/active_matter | 0.827134 | +| well/turbulent_radiative_layer_2D | 0.628780 | + +Outcome: improvement; published atomically. + +## E006 — observation and integration operators +- Hypothesis / changed factor: Same E005B neural weights, projection and horizon regularization. Spectral branch now models finite 8-cell averaging, the corresponding Kolmogorov forcing location, and ETDRK2 substeps for Fisher/Swift–Hohenberg/KS/KdV confirmed on training trajectories. Gray–Scott uses ETDRK4 max substep 2. No reference generator code or seeds used. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.490722**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Contract and determinism verification + +Seed 0 inference checks on three validation inputs spanning scalar periodic, 11-channel periodic and rectangular mixed-boundary data, plus independently constructed unseen 2-channel 7×11 and 13-channel 129×177 inputs. All shapes/dtypes were correct, every output finite, maximum rerun difference exactly 0. Small-grid padding and non-finite-step fallback added without changing any predictions on the three validation cases (bitwise parity). No synthetic training trajectories were generated. Evidence: research/universal_v1/contract_checks.json. + +## E007 — combine improved physics and multi-step training +- Hypothesis / changed factor: E006 spectral observation/integration model and E005B output projection, with shared neural weights after 4,000 four-step fine-tuning updates. Corpus and transfer gains combined; all weights originate from E001 random initialization. +- Seed: 20261021. +- Checkpoint SHA256: `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.506925**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.531697 | +| well/active_matter | 0.821690 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: improvement; published atomically. + +## E002 — four-step training with input perturbations +- Hypothesis / changed factor: Fine-tune E001 for 8,000 updates with four autoregressive steps in the loss, 0.01 normalized input noise, lr 0.0003, batch 12. Test whether recovery from model errors improves long rollouts. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `1d2b51ab44d7b8212b22c4c1894b02b03697b7ed1d30f756f216e6076006e48d`. +- Validation: fixed 81-case subset; reward **0.398333**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.411539 | +| ape2d/decay_turb | 0.688351 | +| ape2d/diff | 0.194736 | +| ape2d/fisher | 0.729932 | +| ape2d/gs_alpha | 0.660633 | +| ape2d/gs_beta | 0.856659 | +| ape2d/gs_delta | 0.790159 | +| ape2d/gs_epsilon | 0.457965 | +| ape2d/gs_gamma | 0.625919 | +| ape2d/gs_iota | 0.527543 | +| ape2d/gs_kappa | 0.587907 | +| ape2d/gs_theta | 0.768902 | +| ape2d/kdv | 0.263418 | +| ape2d/kolm_flow | 0.973194 | +| ape2d/ks | 0.959533 | +| ape2d/sh | 0.611700 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.929633 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: did not improve validation reward; retained current package. + +## E002 — final multi-step checkpoint with validated inference operators +- Hypothesis / changed factor: Replace only the E007 shared neural weights with E002 after all 8,000 four-step fine-tuning updates. Physical operators and active-matter horizon regularization unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.511096**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.801125 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## Full validation of phase2_4000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.500153**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045446 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.527867 | +| well/active_matter | 0.839802 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E009 — active-fluid moment physics, train-only diagnostics + +Hypothesis: the transfer error is partly due to missing orientation-distribution moments. Implemented independent Bingham closure and a Fourier angular-moment Smoluchowski/Stokes system, based on public equation descriptions (the local The Well active_matter README) and training-only checks. No generator code, seeds, pretrained weights or normalization statistics used. Translation and rotation diffusivities 0.05 and beta 0.8 were verified against train fields; alpha and zeta come from supplied metadata. + +Changed factors: Bingham closure vs retained angular moments; spatial grids 128/256; angular resolution 12/24/40; zero vs Bingham initialization; fixed recurrent context encoder that inserts observed low moments. The encoder updates physical state only; no parameters or gradients are fitted at inference. + +Seed: not applicable (deterministic analytical operators). Checkpoint: none; these are not published candidates. Validation VRMSE: not yet run for any dataset. Four selected TRAIN trajectories only. Bingham 30-step train mean VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5): 0.0261, 0.5173, 0.8618, 1.0525. Doubling spatial resolution did not resolve closure drift. At (3,-2), 12 angular modes with context encoding reduced four-step train VRMSE to about 0.0044, versus 0.153 for Bingham. Twelve modes were inadequate at high activity; increasing angular resolution is being tested. A GPU prototype initially failed due to TorchScript complex-number optimization; this is a portability failure, not a scored model. + +Outcome: promising mechanism, still under development; retained the validated shared neural/spectral model. Detailed train results and source variants remain in research/physics_diagnostics. + + +## E009b — active kinetic resolution and implementation diagnostics + +Hypothesis: angular and spatial discretization limit the kinetic closure. Changed factors: M=32/40/64 even angular modes, 128 versus native256 spatial grid, spatial2/3 filtering enabled/disabled, eager versus CUDA graph execution. Seed: none (deterministic operators); neural checkpoint unchanged. Validation: not evaluated for these train-only numerical diagnostics. Runtime source and all numerical results are in physics_diagnostics. + +Native256/M32 thirty-step TRAIN VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5) was 0.025519, 0.014763, 0.100166, 0.880100. M40 and M64 gave nearly equal short-rollout errors, whereas 128 spatial resolution remained inaccurate or unstable at high activity. Removing the spatial filter did not help. Four-step errors at native256/M64 were approximately 0.0024/0.0031/0.0041/0.0055 for (9,-3), and 0.0233/0.0483/0.0937/0.1448 for (17,-5). Outcome: select native256/M32 for validation, retaining a finite physical-prefix fallback. TorchScript fusion failures and CUDA graph scalar-copy failures were fixed before scoring. CUDA graphs reduce launch overhead; all state storage addresses remain fixed. + +## E011a — buoyant smoke discretization, train-only identification + +Hypothesis: a fixed incompressible-fluid operator can reduce PDEArena rollout error once the observed grid layout is represented correctly. Independent MAC solver with semi-Lagrangian advection, explicit viscosity and discrete Neumann pressure projection. Training data reveal that saved velocities are lower faces with the final zero normal face omitted. Correct discrete divergence is about 0.00016 RMS. Physical domain32 and stored interval6 account for endpoint metadata conventions. + +Changed factors on three TRAIN trajectories: stored interval6 versus6.461538, internal time steps0.25/0.5/1/1.5, viscosity0/0.001/0.005/0.008/0.01/0.012, and linear versus MacCormack scalar advection. Seed: none; checkpoint: none (analytic prototype). Validation per dataset: not yet evaluated. At interval6, linear scalar advection and viscosity0.01, one-step mean train VRMSE was 0.0000243. MacCormack with the same viscosity gave0.0981; reducing the internal step to0.25 with MacCormack gave0.3391. Thus closer continuum integration did not reproduce the observed numerical dynamics. Eight additional complete TRAIN rollouts with the selected scheme had mean VRMSE below0.00011. A separate torch-only double-precision implementation reproduced the SciPy prototype. Outcome: proceed to validation; no held-out data or generator code used. Detailed tested values and errors are retained in arena_initial_results.json, arena_scheme_results.json, arena_long_results.json and arena_torch_results.json. + +## E010 — fixed angular-moment active-fluid operator +- Hypothesis / changed factor: Add the independently implemented Smoluchowski/Stokes angular-moment operator to the same phase2-final shared neural model. M32 even angular modes, native256 grid, six Stokes iterations, ETDRK4, deterministic CUDA graph; fixed Bingham initialization and observed low-moment context encoding. No inference parameter fitting. All twelve active validation cases used all thirty physical steps. Exact repeat determinism verified on low/high-activity validation inputs and a diffusion input; two novel grid/channel schemas also passed. Runtime pins the complete package directory during atomic updates. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `d11ea64e646593878f730df43933d3f1e64f481a34083bb7687be973e51e2d33`. +- Validation: fixed 81-case subset; reward **0.572674**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E011 — fixed MAC-grid buoyant-fluid operator +- Hypothesis / changed factor: Keep the E010 shared neural checkpoint and all other inference components; add a metadata-conditioned MAC fluid update for the buoyant-smoke schema. The selected numerical scheme was identified on train pools only. Quick validation reran all twelve PDEArena cases and reused E010 predictions for unchanged operators with identical weights. PDEArena mean VRMSE is 0.00008349. Exact repeat determinism passed for two smoke cases, diffusion, and two novel grid/channel schemas. Expanded end-to-end validation is next. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `397cac1ec1b4314c9e5832071d0cc8a3451a85a8d891ba469a64722ac4d69f65`. +- Validation: fixed 81-case subset; reward **0.713334**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E012a — neural forecast uncertainty, train-only ablation + +Hypothesis: spatial averaging reduces uncertain late-time turbulent fluctuations while retaining vertical mean profiles. Same phase2-final checkpoint (SHA256 d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420), seed20261021. Nine TRAIN trajectories, one for each cooling time; no validation used to select the coefficient. Changed factors: nonnegative material scalars; Gaussian smoothing with final width1/2/4/8/16 pixels; linear horizon-dependent shrinkage toward the mean over periodic directions with final weight0.25/0.5/0.75/1.0 at step30. + +| Variant | Train VRMSE | +|---|---:| +| raw | 0.538547 | +| positive | 0.538100 | +| blur_1 | 0.537242 | +| blur_2 | 0.536386 | +| blur_4 | 0.534177 | +| blur_8 | 0.529554 | +| blur_16 | 0.526478 | +| profile_0.25 | 0.530485 | +| profile_0.5 | 0.525987 | +| profile_0.75 | 0.524946 | +| profile_1.0 | 0.527495 | + +Validation per dataset: not evaluated in this ablation. Outcome: coefficient0.75 selected for subsequent validation. Positivity alone helped very little; smoothing cannot resolve most pressure error. Roughly one-fifth of pressure MSE comes from spatial-mean error, so a correction to the mean alone is insufficient. + +## E012 — fixed spatial uncertainty shrinkage +- Hypothesis / changed factor: The same shared neural weights and physical operators as E011. Add horizon-dependent shrinkage toward means over periodic directions (weight 0.75 by step30), plus nonnegative density/concentration. The coefficient was selected on nine training trajectories before validation. All nine radiative-layer validation rollouts were rerun; unchanged physical outputs were reused for the other72 cases. Exact determinism passed on three radiative-layer inputs and two novel grid/channel schemas. Validation radiative-layer VRMSE improved from0.579723 to0.546835. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715448**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546835 | + +Outcome: improvement; published atomically. + +## E013 — native rectangular inference before fine-tuning +- Hypothesis / changed factor: Change only the shared neural inference grid for rectangular inputs from64x192 to native128x384, keeping the E012 checkpoint and all other operators/regularization fixed. This tests resolution transfer before native-grid fine-tuning. Radiative-layer VRMSE worsened from0.546835 to0.769883; retain E012. All other validation predictions are unchanged physical outputs. The result motivates retraining the shared operator on native grids rather than changing inference resolution alone. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `cc9804d49168dff3c2a3868118de9b3b3bccbd2b0982814f94617a3115b7b6fa`. +- Validation: fixed 81-case subset; reward **0.703067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.769883 | + +Outcome: did not improve validation reward; retained current package. + +## E008 — larger-batch transfer-focused training, update 5000 +- Hypothesis / changed factor: Continue the same shared network from E002 final weights: batch 48, four-step loss, 0.01 input noise, corpus probability 0.1 and 0.3 for each transfer group, lr 0.0003 cosine, sampling seed 20261022; initialization seed remains 20261021. Inference operators frozen to E012 (angular-moment active matter, MAC buoyant smoke and fixed neural uncertainty shrinkage). Quick validation reruns all nine radiative-layer trajectories and reuses fixed physical predictions for the other seventy-two cases, which are independent of the neural weights; expanded validation runs end-to-end. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715555**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + + +## E015a — reaction-diffusion spatial/time convergence, training only + +Hypothesis: unresolved reaction harmonics cause remaining Gray–Scott drift. Changed factors: grid256/512/1024 after finite8-cell observation deconvolution; ETDRK4 maxdt2/0.5 vs ETDRK2 maxdt0.2/0.25/0.3/0.5; optional physical context encoding preserving unresolved modes; finite-averaging ratio2/4/8/16 versus its continuum limit. Deterministic numerical operators, no random seed or learned checkpoint. Validation per dataset: not yet evaluated. + +On selected TRAIN gamma/delta trajectories, grid256/RK4dm2 three-step VRMSE was gamma .000764/.000799/.001016 and delta .000817/.001631/.002567. Grid512/RK2dm.25 gave gamma .000197/.000268/.000394 and delta .000105/.000222/.000335. Grid1024 and the recurrent physical-state encoder gave no additional improvement. Ratio8 remained best among tested observation filters. Thus prior negative grid-only tests did not rule out improvement when resolution and observation inversion were changed together. A portable double-precision GPU implementation is undergoing full train rollouts. Numerical results and all variants are retained in gs_resolution_results.json, gs_resolution_time_results.json, gs_context_results.json and gs_observation_results.json. + +A separate spectral CUDA-graph implementation reproduced ten different corpus training cases exactly over three forecast steps (max absolute difference0 for each). Some runtime reduction was measured while shared training was active, e.g. three Navier–Stokes steps .70s to .47s. This changes execution only; no numerical parameters are fitted at inference. + +## E016 design — shared global spectral feature mixing, not yet trained + +Hypothesis: acoustic pressure patterns need communication across the native rectangular grid. Prepared a shared Fourier feature-mixing residual at the first neural stage, with even extensions in nonperiodic directions and metadata-conditioned complex gains. All weights will be trained jointly on all pools; this is not a dataset specialist. Adds306552 trainable parameters to the2.759547M shared net, total3066099. New output projection starts at zero. Seed20261024 for the new parameters. Initial model matches the existing parent exactly, and a training-only backward pass gives finite nonzero gradients. No optimizer updates or validation results yet; not a publishable candidate. Source and initialization checks in research/global_v1. Continue native-grid training first to establish a comparison. + +## E015b — offline numerical coefficient calibration, negative train result + +Hypothesis: small global corrections to Gray–Scott diffusivities and elapsed time reduce residual error. Fitted three fixed coefficients offline to one-step errors on TRAIN gamma/delta trajectories, with the fine512/RK2dm.25 operator. No inference-time fitting. Seed: none; checkpoint: none (analytic calibration); validation per dataset: not run because full TRAIN rollouts worsened. + +Fitted options: `{"gs_scale": 2, "gs_dt": 0.25, "gs_order": 2, "gs_double": true, "gs_diffusion_u": 1.9999732723556725e-05, "gs_diffusion_v": 1.0000209495791168e-05, "gs_time_scale": 0.9999851457782819}`. One-step normalized MSE decreased from 2.3365036901186555e-08 to 1.9611423235582817e-08. + +| Train dataset | Variant | 26-step VRMSE | +|---|---|---:| +| gs_gamma | uncalibrated | 0.029923 | +| gs_gamma | calibrated | 0.031195 | +| gs_delta | uncalibrated | 0.013152 | +| gs_delta | calibrated | 0.014885 | +| gs_alpha | uncalibrated | 0.002711 | +| gs_alpha | calibrated | 0.002466 | +| gs_theta | uncalibrated | 0.010527 | +| gs_theta | calibrated | 0.049919 | + +Outcome: rejected. The tiny one-step improvement did not carry over to long rollouts, particularly on the unused theta training trajectory. Retain nominal physical coefficients. + +## E015 — finer reaction-diffusion state with matched integration +- Hypothesis / changed factor: Keep the same phase3/update5000 shared neural checkpoint. Gray–Scott now retains a2x internal Fourier grid, uses double-precision ETDRK2 with max internal dt0.25, and observes it through the finite8-cell averaging filter. Nominal physical coefficients retained after calibration failed full train rollouts. Enable deterministic CUDA graphs for spectral updates. Reran all48 corpus validation cases; corpus mean improved from0.032509 to0.010893. Other33 cases reuse identical parent outputs. Exact repeat checks passed on two Gray–Scott cases, Kolmogorov flow and novel schemas; all24 unchanged corpus cases matched parent outputs bit for bit. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `3962bd91a5cf0d9b11b901a10dda99623d4d995b034f57f75c81e62bc41c8aff`. +- Validation: fixed 81-case subset; reward **0.755127**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + +## Full validation of arena_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.696465**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045295 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E014 — native-grid eight-step training, update 1000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.756053**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 1000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755034**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546620 | + +Outcome: did not improve validation reward; retained current package. + +## E017a — active-state observation diagnostics on training trajectories + +Hypotheses: remaining active-fluid drift arises either from numerical resolution or inconsistent hidden angular moments after observed low moments are inserted. Changed factors: time step, spatial grid, Stokes iterations, finite-difference moment inversion, fixed velocity-constrained moment observation, positivity projection, rigid angular transport, and Bingham amplitude transport. All diagnostics use the first trajectory in the corresponding TRAIN pool, starting at frame zero; no validation or held-out targets. The context encoder uses only four observed frames and updates physical state; solver and encoder parameters are fixed. No parameter optimization at inference. + +Seed: deterministic physical operators, no new random parameters. Shared checkpoint unchanged; current selected SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Validation VRMSE per dataset: not evaluated for these train-only diagnostics. + + +kinetic_accuracy_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"name": "baseline", "params": [17, -5]} | 0.07398549 | 0.13735820 | 4 | +| {"name": "smaller_time_step", "params": [17, -5]} | 0.07397478 | 0.13734564 | 4 | +| {"name": "stokes_iterations", "params": [17, -5]} | 0.07398360 | 0.13735427 | 4 | +| {"name": "upsample_384", "params": [17, -5]} | 0.07628622 | 0.14166540 | 4 | +| {"name": "upsample_512", "params": [17, -5]} | 0.07628392 | 0.14165913 | 4 | + +kinetic_encoder_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"last": false, "params": [3, -2], "weight": 0} | 0.00370120 | 0.00440354 | 4 | +| {"last": false, "params": [3, -2], "weight": 1} | 0.01032324 | 0.01005470 | 4 | +| {"last": true, "params": [3, -2], "weight": 0.25} | 0.00641056 | 0.00683838 | 4 | +| {"last": true, "params": [3, -2], "weight": 1} | 0.02690719 | 0.02838808 | 4 | +| {"last": false, "params": [9, -3], "weight": 0} | 0.00375097 | 0.00547713 | 4 | +| {"last": false, "params": [9, -3], "weight": 1} | 0.14332058 | 0.16554202 | 4 | +| {"last": true, "params": [9, -3], "weight": 0.25} | 0.04367409 | 0.04571036 | 4 | +| {"last": true, "params": [9, -3], "weight": 1} | 0.12102679 | 0.12214902 | 4 | +| {"last": false, "params": [17, -5], "weight": 0} | 0.07398282 | 0.13735300 | 4 | +| {"last": false, "params": [17, -5], "weight": 1} | 0.35820127 | 0.58563930 | 4 | +| {"last": true, "params": [17, -5], "weight": 0.25} | 0.05379023 | 0.08525104 | 4 | +| {"last": true, "params": [17, -5], "weight": 1} | 0.20885072 | 0.28744033 | 4 | + +kinetic_flow_encoder_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "weight": 0.0} | 0.02551891 | 0.05217000 | 30 | +| {"params": [1, -1], "weight": 0.25} | 0.01215404 | 0.02445865 | 30 | +| {"params": [1, -1], "weight": 1.0} | 0.00162011 | 0.00254364 | 30 | +| {"params": [3, -2], "weight": 0.0} | 0.01476343 | 0.04135585 | 30 | +| {"params": [3, -2], "weight": 0.25} | 0.00774478 | 0.02111043 | 30 | +| {"params": [3, -2], "weight": 1.0} | 0.00182436 | 0.00253463 | 30 | +| {"params": [9, -3], "weight": 0.0} | 0.10013896 | 0.45469648 | 30 | +| {"params": [9, -3], "weight": 0.25} | 0.03468125 | 0.16266733 | 30 | +| {"params": [9, -3], "weight": 1.0} | 0.04499391 | 0.18140420 | 30 | +| {"params": [17, -5], "weight": 0.0} | 0.54988378 | 1.29260397 | 30 | +| {"params": [17, -5], "weight": 0.25} | 0.48679241 | 1.30647564 | 30 | +| {"params": [17, -5], "weight": 1.0} | 0.62024931 | 1.32728279 | 30 | + +kinetic_transport_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.00372676 | 0.00443898 | 4 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187841 | 0.00173137 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 0.0} | 0.00375795 | 0.00446493 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 1.0} | 0.00187924 | 0.00173272 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00041423 | 0.00039727 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043372 | 0.00043870 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 0.0} | 0.00041337 | 0.00038921 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 1.0} | 0.00039989 | 0.00040295 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.00621011 | 0.01090280 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00566252 | 0.01021467 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 0.0} | 0.00627246 | 0.01102048 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 1.0} | 0.00555848 | 0.01004915 | 4 | + +kinetic_transport_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 0.0} | 0.02551866 | 0.05216906 | 30 | +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00162010 | 0.00254361 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.01480152 | 0.04140588 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183204 | 0.00254354 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00303750 | 0.01372149 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00258502 | 0.00757161 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.45993094 | 1.23510826 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.46775045 | 1.23808050 | 30 | + +Outcome: finer numerics did not improve forecasts. Temporal-derivative inversion was mostly harmful. Fixed velocity observation improves low-activity cases; rigid transport additionally improves moderate/high activity. Bingham amplitude transport adds little in short tests. Select phase transport plus observation weight1 for validation based on complete 30-step training evidence. No candidate from this entry alone was published. + +- Diagnostic source kinetic_torch.py: SHA256 `1082bbf868d579e4c101b86d5fccffc9e43ba1c45e569e1c1dcbc4a19c5ddd6a`. +- Diagnostic source kinetic_flow_encoder.py: SHA256 `e0d1e5022cc8e0595c1c9fd43dc15ba4e77c7998e6fef139c5b58e2df30dde22`. +- Diagnostic source kinetic_transport_encoder.py: SHA256 `4bbb953264d06a28cd3832509172ae68240a24ec779116a3e0377346818e26e9`. + +## E014 — native-grid eight-step training, update 2000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.757118**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 2000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755962**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.532789 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of fine_gs_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.730834**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261022; checkpoint SHA256 `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.049812 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Expanded-validation protocol: all561 non-active cases were rerun with this complete package. The24 active-matter outputs were reused from the earlier end-to-end E011 run after verifying identical active solver assets/configuration and thirty physical steps for every case, so those outputs are independent of the changed shared neural weights. The official scorer evaluates all585 trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E017 — geometric and velocity-constrained context state +- Hypothesis / changed factor: Keep the same shared neural checkpoint and all other operators from parent.json. Change only the active-matter four-frame physical-state encoder: rotate higher angular moments with observed second-moment orientation, scale by observed concentration, use twelve fixed regularized Stokes observation steps for F4, and project F4 into its necessary moment-positivity disk. Fixed observer weight1, ridge0.01 and phase confidence threshold0.1 were selected using train-only full30 diagnostics. No learned or solver parameter changes at inference. All twelve active validation cases rerun; sixty-nine identical parent outputs reused. Exact physical-prefix coverage is in physical_coverage.json. No new random parameters; inherited initialization seed20261021 and sampling seed recorded in the checkpoint manifest. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.802217**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E017b — combine context encoder and native checkpoint +- Hypothesis / changed factor: Combine the E017 fixed physical context encoder with the better shared native-grid checkpoint selected at phase4 update2000. No training, new random parameters or numerical changes. All runtime modules/configuration outside kinetic.py and the shared weights were checked identical between parents. Twelve active outputs are independent of shared weights after verifying all30 steps use the physical operator. Official scoring reuses those12 outputs and69 matching neural-parent outputs. Replayed three neural cases and two novel schemas with exact determinism. Active determinism was already checked for the identical physical encoder. Inherited initialization seed20261021, native-stage sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.803282**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E018 — finer internal advection grids +- Hypothesis / changed factor: Keep the same shared neural checkpoint and active-state encoder. Change only Burgers and Kolmogorov physical updates to a twofold internal Fourier grid in double precision, retaining unresolved state between outputs and using the same finite-cell observation map. Preserve the physical Kolmogorov forcing phase when refining its grid. Three full26 training trajectories per equation supported the change; doubling substeps, using a fourfold grid, or encoding unresolved modes through all four context frames added little. No random parameters or new training. Rerun the six affected validation cases; reuse seventy-five identical parent outputs. All six used26/26 physical updates. Exact repeat checks include Burgers, Kolmogorov, an unchanged Gray-Scott case and two novel schemas. Inherited initialization seed20261021 and sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `970b00933280c87d32c3d322412f89d56ad1969897bbdd0b83dbeccc51231cf8`. +- Validation: fixed 81-case subset; reward **0.806376**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E017 expanded active-matter validation + +Changed factor: expand from twelve to all24 active-matter validation trajectories, all start0. Twelve existing predictions were reused only after exact context equality checks; the other twelve were replayed with the same complete E017 package. Official scorer active-matter VRMSE **0.08948507**, versus 0.19604967 for the earlier physical encoder on the same24 cases. Failed cases 0. Other dataset VRMSE: not reevaluated in this experiment. Deterministic encoder; inherited neural initialization seed20261021, sampling seed 20261023. Checkpoint SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Outcome: expanded group-level validation, no comparison between different-size aggregate rewards. + +## E018a — further fixed-operator train diagnostics + +Hypotheses and changed factors: (1) interpolated velocity/strain can drive hidden moments between context frames; (2) an entropic angular-moment projection can repair the full distribution; (3) moment and velocity filters should differ; (4) unresolved Fourier modes cause Burgers/Kolmogorov error; (5) a finer grid, retained context state, time-step refinement or alternative exponential integration repairs KdV. All diagnostics use TRAIN trajectories only, with exact source indices recorded by scripts. No new random parameters or learned checkpoint. Validation VRMSE per dataset: not evaluated in these diagnostics; E018 separately evaluates the selected grid change. + + +kinetic_forced_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"cubic": false, "observe_initial": false, "params": [3, -2]} | 0.00187798 | 0.00173005 | 4 | +| {"cubic": true, "observe_initial": false, "params": [3, -2]} | 0.00187867 | 0.00173198 | 4 | +| {"cubic": true, "observe_initial": true, "params": [3, -2]} | 0.00413512 | 0.00506158 | 4 | +| {"cubic": false, "observe_initial": false, "params": [9, -3]} | 0.01194177 | 0.01268802 | 4 | +| {"cubic": true, "observe_initial": false, "params": [9, -3]} | 0.01874249 | 0.02191580 | 4 | +| {"cubic": true, "observe_initial": true, "params": [9, -3]} | 0.01873903 | 0.02190757 | 4 | +| {"cubic": false, "observe_initial": false, "params": [17, -5]} | 0.02985521 | 0.04619750 | 4 | +| {"cubic": true, "observe_initial": false, "params": [17, -5]} | 0.04395452 | 0.07376709 | 4 | +| {"cubic": true, "observe_initial": true, "params": [17, -5]} | 0.04395483 | 0.07376773 | 4 | + +kinetic_moment_projection_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187845 | 0.00173138 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043149 | 0.00043489 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00560499 | 0.01007434 | 4 | + +kinetic_moment_projection_long_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00161964 | 0.00254240 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183213 | 0.00254286 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00240461 | 0.00671084 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.48482975 | 1.27747452 | 30 | + +kinetic_unmasked_flow_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187846 | 0.00173107 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00041315 | 0.00041952 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00551630 | 0.01016449 | 4 | + +nonlinear_resolution_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963583 | 0.05441318 | 26 | +| {"double": true, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963618 | 0.05441412 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1} | 0.00572061 | 0.01058393 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 2} | 0.00572059 | 0.01058392 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01216232 | 0.02993135 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01185469 | 0.02870498 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1} | 0.00759149 | 0.02465414 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 2} | 0.00759146 | 0.02465408 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001462 | 0.00006308 | 26 | +| {"double": true, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 2} | 0.00001457 | 0.00006339 | 26 | + +nonlinear_resolution_expanded_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00301647 | 0.00326871 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037459 | 0.00050441 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037497 | 0.00050395 | 26 | +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00124680 | 0.00155330 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012381 | 0.00019031 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012380 | 0.00019021 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.08121000 | 0.15702826 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.02042678 | 0.05910929 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.01992439 | 0.05732524 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.04068160 | 0.08604752 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00718886 | 0.02328656 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00716170 | 0.02330063 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035132 | 0.00209063 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035139 | 0.00208930 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00034709 | 0.00206470 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203683 | 0.01849953 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203362 | 0.01849631 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01204391 | 0.01851219 | 26 | + +nonlinear_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 0} | 0.00572061 | 0.01058393 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 17} | 0.00037297 | 0.00050443 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 93} | 0.00012373 | 0.00019031 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 0} | 0.00759136 | 0.02465325 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 17} | 0.02068291 | 0.05987191 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 93} | 0.00716521 | 0.02321832 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 17} | 0.00035139 | 0.00208930 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 93} | 0.01203363 | 0.01849633 | 26 | + +kdv_numerics_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 0} | 0.00001493 | 0.00006064 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001457 | 0.00006350 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 0} | 0.00001457 | 0.00006347 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00035151 | 0.00209107 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 17} | 0.00019026 | 0.00094255 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00034583 | 0.00205838 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 17} | 0.00034618 | 0.00206044 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01203692 | 0.01849972 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 93} | 0.01360971 | 0.02156036 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01205018 | 0.01851917 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 93} | 0.01204985 | 0.01851889 | 26 | + +Outcomes: interpolated active forcing was harmful at moderate/high activity. Entropic moment projection modestly improved moderate activity but worsened the complete high-activity rollout (0.467750 to0.484830); not selected. Retaining full velocity modes has only a small short-rollout effect; full30 test remains ongoing. Twofold grids robustly improved Burgers and Kolmogorov across three training trajectories. Fourfold grids and more substeps added little; four-frame unresolved-state encoding added little or hurt. KdV grid/integrator variants were not consistently better. A proposed small-domain KdV test found no training cases below domain extent15 and ran no rollouts; replaced it with tests on the existing three training trajectories. Low-frequency differential fits recover nonlinear coefficients near6; fitted dissipative corrections were inconsistent, so no coefficient change was selected. + +The phase5 global-mixing trainer initially stopped before optimizer creation because channels_last was applied to a five-dimensional Fourier parameter. Applying the layout only to Conv2d modules fixed it; restart used the identical checkpoint and seed20261024, with no optimizer updates lost. Exactly one GPU trainer remains. + +Diagnostic source kinetic_forced_context.py: SHA256 `cebf38a2f07fbf2ea664c5b00ac17bd5dd1cef2a128b75a2d7e978fbc102e3a0`. + +Diagnostic source kinetic_moment_projection.py: SHA256 `fc796f88bda7f7802bab82a8e41d76dc1e08e1fc3a98fb9eed20cd45e26fc69a`. + +Diagnostic source kinetic_unmasked_flow.py: SHA256 `4a40d463091ef1dd6a2432e2ce79f0a417c61af1943369711062c6e0113a059b`. + +Diagnostic source spectral_resolution.py: SHA256 `907c567c87c46bb3adf9e11b1282d1a3e1400751d5e5c0bfdd835350d3062420`. + +Diagnostic source spectral_kdv_numerics.py: SHA256 `e3fcb35b99158c156691e324dae512d17f9fca287fb072270e1adbc6437b1ffe`. + +## E018b — distinct active flow and moment filters, negative full-rollout result + +Hypothesis: the stored active velocity/strain retain modes above the moment dealiasing cutoff, so the Stokes solve should retain those modes. Changed only the velocity/strain spectral mask, preserving the moment mask. Four complete30-step TRAIN rollouts, deterministic physical operator, no new seed or checkpoint; validation per dataset not evaluated. + +| Training parameters | Mean VRMSE | +|---|---:| +| [1, -1] | 0.00161508 | +| [3, -2] | 0.00182706 | +| [9, -3] | 0.00258145 | +| [17, -5] | 0.46823457 | + +Outcome: negligible improvement at low/moderate activity and slight worsening at high activity (0.467750 to0.468235); retained the validated operator. The direct training spectral audit found concentration/director power above the2/3 cutoff at float32 roundoff, while high-activity strain retains about0.0022 of its spectral norm there. This numerical distinction is real but does not explain the dominant rollout error. + +## E016 — shared global spectral mixing, update 1000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `3cd1d10c39cf842ae764c3a6390d8af9c784fc4cc2141213e7647256be66ba8c`. +- Validation: fixed 81-case subset; reward **0.803408**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.560153 | + +Outcome: did not improve validation reward; retained current package. + +## E016 — shared global spectral mixing, update 1000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `a0019d25f7e2d44c69f90d0ada4f0459c943fbffa3503dc585227648c836f516`. +- Validation: fixed 81-case subset; reward **0.802480**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.574692 | + +Outcome: did not improve validation reward; retained current package. + +## E020 — pressure uncertainty shrinkage +- Hypothesis / changed factor: Keep the shared neural weights and fixed physical operators. Add pressure-only output shrinkage toward each predicted spatial mean: zero through step10, linear ramp to0.75 at step30. The existing periodic-profile shrinkage remains unchanged. Select this fixed schedule from nine train rollouts for both native and global1000 checkpoints; native training mean VRMSE0.519754 to0.518445, global0.536804 to0.534698. No new random parameters or inference-time fitting; inherited initialization seed20261021 and sampling seed20261023. Rerun all nine neural validation cases, reuse seventy-two identical physical outputs. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `1b8f91e26862343f0238655ab2b43f24ad67da270779efcfa7ef02a841f95ab6`. +- Validation: fixed 81-case subset; reward **0.806391**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Outcome: improvement; published atomically. + +## E020a — pressure schedule training sweep + +Changed factor: pressure-only mixing toward its predicted spatial mean, with start0/5/10, end15/30, strength0/0.25/0.5/0.75/1. Nine training trajectories, one per cooling parameter; train membership was audited again. Native checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4 and global1000 checkpoint from phase5/eval001000_shrink were both evaluated. No random coefficients; inherited initialization20261021, native sampling20261023 and global sampling/new-parameter20261024. Validation per dataset: not evaluated in this training sweep; selected native variant validated separately in E020. + +| Model | Start | End | Strength | Training VRMSE | Pressure VRMSE | +|---|---:|---:|---:|---:|---:| +| native | 0 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 15 | 0.25 | 0.51936945 | 1.08422407 | +| native | 0 | 15 | 0.5 | 0.52051218 | 1.08879500 | +| native | 0 | 15 | 0.75 | 0.52311825 | 1.09921928 | +| native | 0 | 15 | 1.0 | 0.52708753 | 1.11509640 | +| native | 0 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 30 | 0.25 | 0.51915815 | 1.08337887 | +| native | 0 | 30 | 0.5 | 0.51916075 | 1.08338927 | +| native | 0 | 30 | 0.75 | 0.51976061 | 1.08578873 | +| native | 0 | 30 | 1.0 | 0.52095018 | 1.09054701 | +| native | 5 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 15 | 0.25 | 0.51899641 | 1.08273192 | +| native | 5 | 15 | 0.5 | 0.51948753 | 1.08469640 | +| native | 5 | 15 | 0.75 | 0.52120696 | 1.09157413 | +| native | 5 | 15 | 1.0 | 0.52409830 | 1.10313946 | +| native | 5 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 30 | 0.25 | 0.51903559 | 1.08288863 | +| native | 5 | 30 | 0.5 | 0.51874419 | 1.08172304 | +| native | 5 | 30 | 0.75 | 0.51888335 | 1.08227969 | +| native | 5 | 30 | 1.0 | 0.51945286 | 1.08455773 | +| native | 10 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 15 | 0.25 | 0.51861991 | 1.08122590 | +| native | 10 | 15 | 0.5 | 0.51848526 | 1.08068733 | +| native | 10 | 15 | 0.75 | 0.51935090 | 1.08414988 | +| native | 10 | 15 | 1.0 | 0.52118939 | 1.09150382 | +| native | 10 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 30 | 0.25 | 0.51902077 | 1.08282937 | +| native | 10 | 30 | 0.5 | 0.51858337 | 1.08107977 | +| native | 10 | 30 | 0.75 | 0.51844492 | 1.08052597 | +| native | 10 | 30 | 1.0 | 0.51860631 | 1.08117150 | +| global1000 | 0 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 15 | 0.25 | 0.53561169 | 1.15099039 | +| global1000 | 0 | 15 | 0.5 | 0.53658240 | 1.15487321 | +| global1000 | 0 | 15 | 0.75 | 0.53962210 | 1.16703201 | +| global1000 | 0 | 15 | 1.0 | 0.54454053 | 1.18670573 | +| global1000 | 0 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 30 | 0.25 | 0.53562902 | 1.15105970 | +| global1000 | 0 | 30 | 0.5 | 0.53524687 | 1.14953112 | +| global1000 | 0 | 30 | 0.75 | 0.53566622 | 1.15120852 | +| global1000 | 0 | 30 | 1.0 | 0.53687798 | 1.15605554 | +| global1000 | 5 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 15 | 0.25 | 0.53535586 | 1.14996706 | +| global1000 | 5 | 15 | 0.5 | 0.53561634 | 1.15100898 | +| global1000 | 5 | 15 | 0.75 | 0.53755790 | 1.15877521 | +| global1000 | 5 | 15 | 1.0 | 0.54106119 | 1.17278837 | +| global1000 | 5 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 30 | 0.25 | 0.53563652 | 1.15108971 | +| global1000 | 5 | 30 | 0.5 | 0.53500332 | 1.14855691 | +| global1000 | 5 | 30 | 0.75 | 0.53491749 | 1.14821358 | +| global1000 | 5 | 30 | 1.0 | 0.53538269 | 1.15007438 | +| global1000 | 10 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 15 | 0.25 | 0.53513905 | 1.14909983 | +| global1000 | 10 | 15 | 0.5 | 0.53481845 | 1.14781741 | +| global1000 | 10 | 15 | 0.75 | 0.53584572 | 1.15192652 | +| global1000 | 10 | 15 | 1.0 | 0.53814365 | 1.16111824 | +| global1000 | 10 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 30 | 0.25 | 0.53575180 | 1.15155082 | +| global1000 | 10 | 30 | 0.5 | 0.53504651 | 1.14872968 | +| global1000 | 10 | 30 | 0.75 | 0.53469765 | 1.14733423 | +| global1000 | 10 | 30 | 1.0 | 0.53471068 | 1.14738634 | + +Outcome: both models selected start10/end30/strength0.75, but the improvement was small. Earlier/stronger shrinkage is not selected. The global model remains worse than the native model after this correction. + +Global1000 checkpoint SHA256 `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. Study source SHA256 `2d9ddca63b268fbc9b52f5ac1bf800e7400ae7458fbd09f80a21cecaca84ffe9`. + +## E021 preparation — shared global-statistics residual head + +Hypothesis: predicting spatially uniform channel changes explicitly will help the pressure/global-mean dynamics that dominate radiative-layer error. Add one shared MLP head using the existing metadata embedding plus normalized means and standard deviations of each of the four context frames. It emits a uniform residual for every channel on every dataset; there are no dataset-specific specialist weights. The spatial network remains the selected native ConvNeXt body for this prototype. Added56075 parameters, total2815622, seed20261027. Output projection initialized to zero. + +Initialization check on one native TRAIN window: exact parent prediction parity (maximum difference0), finite nonzero output-layer gradient norm 0.043989721685647964. No optimizer updates, no new trained checkpoint, no validation per dataset and no publication. Parent checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4. Consider only after the longer-horizon experiment; regenerate against the actually selected parent architecture if it changes. Prototype source SHA256 `8de97bfff13028b7d33adfd16564d8b7b4f00d2ac50e67d2f051549480cefbe8`. + +## E022a — fixed hidden-state uncertainty quadrature, training diagnostics + +Hypothesis: averaging fixed hidden-state uncertainty may reduce late chaotic active-fluid error. After the four-frame state encoder, evolve center and symmetric smooth higher-moment phase perturbations with the same fixed physical parameters. Perturbations leave observedF0/F2 unchanged and enforce F4 positivity. Fixed latent-noise seed20261026, phase standard deviations0.01 and0.0001. Compare equal quadrature weights and half-center/quarter-endpoint weights. No neural training, no new learned checkpoint; validation per dataset not evaluated. These are physical-state quadrature nodes, not separately trained dataset models. + +kinetic_state_quadrature_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.01 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.01 | 0.00162322 | +| [1, -1] | half_center | 0.01 | 0.00162206 | +| [3, -2] | center | 0.01 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.01 | 0.00183122 | +| [3, -2] | half_center | 0.01 | 0.00183157 | +| [9, -3] | center | 0.01 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.01 | 0.16464736 | +| [9, -3] | half_center | 0.01 | 0.12350380 | +| [17, -5] | center | 0.01 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.01 | 0.44187382 | +| [17, -5] | half_center | 0.01 | 0.42681998 | +kinetic_state_quadrature_small_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.0001 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.0001 | 0.00162011 | +| [1, -1] | half_center | 0.0001 | 0.00161957 | +| [3, -2] | center | 0.0001 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.0001 | 0.00183261 | +| [3, -2] | half_center | 0.0001 | 0.00183262 | +| [9, -3] | center | 0.0001 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.0001 | 0.00258179 | +| [9, -3] | half_center | 0.0001 | 0.00257420 | +| [17, -5] | center | 0.0001 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.0001 | 0.45577160 | +| [17, -5] | half_center | 0.0001 | 0.45775375 | + +Outcome:0.01 is rejected because severe moderate-activity degradation outweighs the high-activity gain. The0.0001 quadrature leaves low/moderate errors nearly unchanged and reduces the hardest training mean from0.467763 to0.455772, but requires three physical integrations. Expand train tests to four additional parameter combinations before considering validation. Center includes an extra FFT roundtrip and positivity projection, so tiny roundoff differences from E017 are expected. + +Additional KdV observation-filter test on train0/17/93, powers0/0.5/1/2: removing deconvolution worsens all three; doubling it helps17 slightly but barely changes93 and hurts0. Keep nominal finite8-cell observation model. No KdV runtime change. + +## Full validation of advection_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775038**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Expanded-validation protocol: all561 non-active cases were replayed with the complete E018 package. All24 active outputs were reused from the expanded E017 validation after checking identical physical assets/configuration, exact input equality for reused quick cases, and thirty physical steps in every case. Hence those outputs are independent of the changed shared neural checkpoint. All585 validation trajectories are scored together by the official scorer. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of pressure_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775053**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Exact reuse protocol: model weights and all runtime operators except pressure regularization are identical to E018. Input metadata confirms only the nine radiative-layer cases contain a pressure field; their inputs match the already replayed E020 quick cases exactly. Reuse those nine new forecasts and576 unchanged E018 full forecasts. The official scorer evaluates all585 validation trajectories together. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E016 — shared global spectral mixing, update 2000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `96a49ac4734af31edd0aa5345f51515a899ef067e36abeb9e2347781c28fc055`. +- Validation: fixed 81-case subset; reward **0.807185**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E022b — expanded state-quadrature training study + +Changed factor: expand the fixed1e-4 hidden-state quadrature test from four to eight TRAIN parameter combinations, same latent seed20261026 and same fixed physical operator. No new learned weights. Validation per dataset not yet evaluated. + +| Additional training parameters | Quadrature | Mean VRMSE | +|---|---|---:| +| [5, -4] | center | 0.20956509 | +| [5, -4] | symmetric_mean | 0.20925570 | +| [5, -4] | half_center | 0.20933180 | +| [7, -3] | center | 0.00780154 | +| [7, -3] | symmetric_mean | 0.00781992 | +| [7, -3] | half_center | 0.00781529 | +| [11, -4] | center | 0.08376674 | +| [11, -4] | symmetric_mean | 0.08358850 | +| [11, -4] | half_center | 0.08358082 | +| [15, -4] | center | 0.26489031 | +| [15, -4] | symmetric_mean | 0.26059937 | +| [15, -4] | half_center | 0.26147711 | + +Across all eight cases, center mean 0.12997754, equal-weight quadrature 0.12788370. Outcome: select equal weights for validation; three physical integrations per active rollout. No publication from training results alone. + +## E016 — shared global spectral mixing, update 2000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `f06079b546194d4243540a3578b4df4e547efc8b64bb67210a4d5fe41648b2bd`. +- Validation: fixed 81-case subset; reward **0.806418**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515442 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775847**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match E018. Parsed per-case logs verify552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E023 — shared spatial-mean forecasting branch + +Hypothesis: remaining radiative-layer mean-pressure drift can be predicted from four-frame global statistics and metadata while the shared autoregressive operator handles spatial structure. Train one common MLP across all19 datasets; every branch will be evaluated on every inference case. No dataset-specific checkpoint selection. Random initialization seed20261028; CPU-only, two threads, batch512, 12000 AdamW updates, cosine learning rate0.0005, half windows start0 and half random, horizons1–30 capped by available train frames, group probabilities0.1/0.1/0.1/0.7. Loss is smooth absolute mean error normalized by target spatial variance; targets never enter model inputs. Uses only the10115 audited training records. Spatial moments derived with float64 accumulation. No pretrained assets. + +Changed factor: add a shared horizon-conditioned mean predictor to the fixed autoregressive spatial forecast. CPU training is concurrent with the one-GPU 16-step experiment. This is distinct from the earlier untrained E021 residual-mean prototype. Validation and checkpoint selection pending. No package change at preparation. + +## E023a — validation mean-branch checkpoint and blend comparison + +Changed factors: shared mean-branch checkpoint at updates2000/6000/12000 and fixed blend0/.25/.5/.75/1. Exact spatial-MSE decomposition on the nine radiative-layer VALID trajectories, keeping the same global2000 spatial prediction. These are validation selection results, not training. Other dataset VRMSE unchanged by the composite physical branch, not reevaluated here. All mean-model weights were randomly initialized here with seed20261028, shared across all19 datasets. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.50464012 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.25 | 0.48902082 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.5 | 0.48371851 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.75 | 0.48967327 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 1.0 | 0.50368029 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 6000 | 0.0 | 0.50464012 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.25 | 0.49150598 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.5 | 0.48589594 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.75 | 0.48943812 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 1.0 | 0.50039246 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 12000 | 0.0 | 0.50464012 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.25 | 0.48902526 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.5 | 0.48202704 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.75 | 0.48522405 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 1.0 | 0.49632276 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | + +Outcome: select update12000, blend0.5; all other tested choices are inferior. Full mean replacement is consistently worse than a partial blend. Actual float32 composite-package replay and determinism checks are required before publication; analytical score alone is not published. + +## E022 — fixed hidden-state quadrature +- Hypothesis / changed factor: Integrate three fixed hidden-angular-state quadrature nodes after the four-frame physical encoder, using one common neural model and identical fixed physical parameters. Nodes share all common computations; no separately trained specialists. Seed20261026, phase scale0.0001, equal weights, observedF0/F2 unchanged, F4 positivity preserved. Eight train trajectories supported this choice; larger perturbations were rejected. Rerun all twelve active validation cases, all30 physical steps; reuse sixty-nine matching non-active outputs. Any validated shared-neural progress during the replay is retained as documented in neural_rebase.json. No inference-time fitting. Inherited neural seeds/checkpoint in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `def27fe183c189d25508bcde6a013f2f6560b6b7d68526b782530bd4b1d7c730`. +- Validation: fixed 81-case subset; reward **0.807374**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E023 — shared mean forecast, composite replay +- Hypothesis / changed factor: Add one shared horizon-conditioned spatial-mean forecasting branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261028; two CPU threads, batch512, 12000 total updates, selected update12000, blend0.5. Inputs are four context means and variances plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in mean_branch_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809033**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809649**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `e7dd6ac71780c453f0a23e995e48d0c94ad68af36df21c2d8993fb7ba914c8a7`. +- Validation: fixed 81-case subset; reward **0.809341**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477935 | + +Outcome: did not improve validation reward; retained current package. + +## E024 — spatial-profile conditioned shared mean branch + +Hypothesis: four-frame channel means and variances omit thermal-layer geometry; coarse profiles along each axis may predict global mean drift more accurately. Changed factor: add eight-bin spatial marginal profiles for both axes, all channels and all four context frames, to the same shared horizon-conditioned MLP. Profiles are centered by context means and normalized by context standard deviations. Random initialization seed20261029; otherwise same shared19-dataset training distribution and12000-update schedule as E023. CPU-only, two threads, no extra GPU. Train data membership audited before deriving profiles; float64 averaging followed by float32 storage. Feature parity between training-cache input and inference from raw TRAIN context passed (feature_checks.json). All parameters initialized from scratch; no pretrained assets. + +Validation pending. Compare against the raw spatial forecast from the currently selected phase6 update1000, obtained by setting only the old mean blend to zero and replaying nine valid contexts. The final candidate, if selected, will replace the existing shared mean branch rather than adding a dataset-specific specialist. No publication from training loss. + +## E024a — profile-branch validation selection + +Changed factors: profile-branch checkpoint2000/6000/12000 and fixed mean blend0/.25/.5/.75/1. Same9 valid radiative-layer cases and phase6 update1000 spatial forecast; the old mean branch was disabled for this ablation replay. Other dataset VRMSE not reevaluated at this selection stage. Shared random initialization seed20261029. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.47627939 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.25 | 0.47072551 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.5 | 0.47161600 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.75 | 0.47837804 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 1.0 | 0.48981324 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 6000 | 0.0 | 0.47627939 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.25 | 0.47073488 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.5 | 0.46949443 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.75 | 0.47282608 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 1.0 | 0.48034141 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 12000 | 0.0 | 0.47627939 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.25 | 0.47050116 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.5 | 0.46970411 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.75 | 0.47404605 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 1.0 | 0.48293819 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | + +Outcome: update6000/blend0.5 selected at0.46949443, versus0.47388509 with E023 and0.47627894 without any mean adjustment. Additional training to12000 improved training loss but slightly worsened validation. Actual composite replay and exact contract checks are running before any publication. + +## E024 — profile-conditioned shared mean forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a profile-conditioned branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261029; two CPU threads, batch512, 12000 total updates, selected update6000, blend0.5. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_mean_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `8dc8554c96b10cffe81e8707063febceb81c18933dd345ac0dfaa329237377d6`. +- Validation: fixed 81-case subset; reward **0.809986**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Outcome: improvement; published atomically. + +## E025 — jointly forecast spatial means and coarse profiles + +Hypothesis: remaining broad thermal-layer structure errors require a forecast of spatial profiles, not only a corrected global mean. Changed factor: shared MLP predicts both global channel means and centered eight-bin marginal profiles on both axes. Enforce agreement between the predicted profile means and global means. Train across all19 datasets from random initialization seed20261030 with the same audited train statistics/profiles,12000 CPU-only updates, batch512, two threads; loss0.25 normalized mean error +0.75 normalized coarse-profile error, with the same smooth absolute form. No new data or pretrained assets. Every branch remains shared and evaluated on every case. + +Inference candidate will interpolate centered coarse-profile residuals with periodic or clamped edges according to metadata, and blend them into the common autoregressive forecast. Physical outputs remain authoritative. Uniform and spatial correction coefficients will be selected on validation, followed by actual package replay and exact determinism checks. Initialization shape, finite-output, zero-correction identity, and mean/profile consistency checks passed on an audited TRAIN context. Validation pending; no publication from training loss. + +## E025a — joint mean/profile validation selection + +Changed factors: joint statistic checkpoint2000/6000/12000, mean correction0/.5/1 and profile correction0/.25/.5/1. Same9 VALID radiative-layer trajectories and phase6 update1000 raw spatial forecasts. MSE decomposition includes the uniform correction, centered profile correction, and all cross terms; actual float32 replay is required next. Shared random seed20261030. Other dataset VRMSE not reevaluated in this selection stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.25 | 0.47635292 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.5 | 0.47958239 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 1.0 | 0.49471505 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.0 | 0.47005735 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.25 | 0.47010562 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.5 | 0.47337008 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 1.0 | 0.48871343 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.0 | 0.48430118 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.25 | 0.48426968 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.5 | 0.48736599 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 1.0 | 0.50220535 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.25 | 0.47518010 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.5 | 0.47612881 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 1.0 | 0.48395044 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.0 | 0.46912443 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.25 | 0.46794480 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.5 | 0.46886089 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 1.0 | 0.47675092 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.0 | 0.47781927 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.25 | 0.47659139 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.5 | 0.47738680 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 1.0 | 0.48486302 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.25 | 0.47462923 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.5 | 0.47500679 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 1.0 | 0.48171365 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.0 | 0.46724263 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.25 | 0.46555366 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.5 | 0.46594474 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 1.0 | 0.47282504 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.0 | 0.47431715 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.25 | 0.47263591 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.5 | 0.47296605 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 1.0 | 0.47954663 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | + +Outcome: select update12000, mean blend0.5, profile blend0.25, VRMSE0.46555366 versus E0240.46949440. At update2000 any profile correction worsened the best mean-only result; profile prediction benefited from further training. Larger profile blends were inferior. Actual replay and contract checks pending; no analytical-only publication. + +## E025 — joint shared mean/profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261030; two CPU threads, batch512, 12000 total updates, selected update12000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `67752d3d72ef3e1cf10485e57abe77cdbc9e99eb48ccb69ce72c8d12d6ab62bd`. +- Validation: fixed 81-case subset; reward **0.810291**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Outcome: improvement; published atomically. + +## E026 — sixteen-bin shared statistical forecast + +Hypothesis: eight-bin profiles smooth thin thermal interfaces; sixteen bins may improve the shared coarse forecast. Changed factor: double the input and output profile resolution on both axes from8 to16 bins, retaining one shared MLP across all19 datasets. Random initialization seed20261031. Same12000-update CPU-only training schedule, two threads, group sampling, loss weighting and horizon distribution as E025; the wider input/output layers increase parameter count. Train-only profile cache membership audited before reading, with float64 averaging. Source snapshot and train/inference feature-parity checks retained in profile16_forecast_v1. No pretrained assets. + +Validation pending; compare on the same9 valid thermal trajectories and the same phase6 update1000 raw spatial forecast. No package change before actual replay and exact contract checks. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/config.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9d795a26dcdbd86daed999f26a0ccbe93d3c87f0 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/config.json @@ -0,0 +1,43 @@ +{ + "width": 48, + "depth": 2, + "detail_decay": 0.8, + "physics": true, + "active_tau": 10.0, + "kinetic": { + "modes": 32, + "resolution": 256, + "iterations": 6, + "cuda_graph": true, + "transport": "phase", + "observer_weight": 1.0, + "observer_ridge": 0.01, + "observer_iterations": 12, + "quadrature_scale": 0.0001, + "quadrature_seed": 20261026 + }, + "arena": true, + "neural_regularization": { + "profile_mix": 0.75, + "horizon": 30, + "pressure_mix": 0.75, + "pressure_start": 10, + "pressure_end": 30 + }, + "spectral_options": { + "gs_scale": 2, + "gs_dt": 0.25, + "gs_order": 2, + "gs_double": true, + "cuda_graph": true, + "advection_scale": 2, + "advection_double": true + }, + "native_rectangular": true, + "global_mixing": true, + "mean_forecast": { + "weight": 0.5, + "profile_weight": 0.25, + "width": 192 + } +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/infer.sh b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..bbf5ac5eb2acfbbad07b871df5e8d072b2304862 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/infer.sh @@ -0,0 +1,3 @@ +#!/bin/bash +set -euo pipefail +exec python "$(dirname "$0")/launch.py" "$@" diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/make_bingham_table.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/make_bingham_table.py new file mode 100644 index 0000000000000000000000000000000000000000..3cba92fb5f0503306c910a27c10b6e42c7099aa6 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/make_bingham_table.py @@ -0,0 +1,9 @@ +import numpy as np +from scipy.special import ive +from pathlib import Path +s=np.linspace(0,12,8193);mu=-np.expm1(-s);r=mu[1:];k=np.where(r<.53,2*r+r**3+5*r**5/6,np.where(r<.85,-.4+1.39*r+.43/(1-r),1/(r**3-4*r**2+3*r))) +for _ in range(6): + z=ive(1,k)/ive(0,k);k=np.maximum(k+(r-z)/np.maximum(1-z*z-z/k,1e-14),1e-10) +t=np.zeros((65,len(s)),np.float32);t[0]=1 +for j in range(1,65):t[j,1:]=ive(j,k)/ive(0,k) +np.save(Path(__file__).with_name('bingham_table.npy'),t) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/model_manifest.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/model_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..053e5f07bb591db8cd50147c958ac87dac332114 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/model_manifest.json @@ -0,0 +1,128 @@ +{ + "contract": "pde-surrogate-rollout/v1", + "name": "universal-convnext-fourier-physical-operator", + "description": "One shared metadata-conditioned neural operator with fixed independently implemented spectral, angular-moment and staggered-grid physical updates; all neural weights trained from scratch.", + "parameters": 3348622, + "context_frames": 4, + "sources_used": [ + "train pools of /workspace/data/split.json; all 19 dataset identifiers", + "Independent equation and observation operators checked against training pools; fixed hyperparameters selected on validation.", + "Public active-fluid equation description in The Well; independently implemented kinetic moment equations and mathematical Bingham initialization table, verified on training pools.", + "Training-pool identification of MAC discretization, viscosity, advection scheme and time/grid metadata conversion for buoyant smoke; independent solver implementation." + ], + "pretrained_assets": [], + "checkpoint_sha256": "63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6", + "training_step": 1000, + "seed": 20261025, + "selection": "Validation rollouts only; see EXPERIMENTS.md.", + "runtime_code_sha256": "67752d3d72ef3e1cf10485e57abe77cdbc9e99eb48ccb69ce72c8d12d6ab62bd", + "initialization_seed": 20261021, + "training_sampling_seed": 20261025, + "training_lineage": "/run-state/workspace/research/universal_v1/phase6/initial.pt", + "training_config": { + "out": "/run-state/workspace/research/universal_v1/phase6", + "index": "/run-state/workspace/research/universal_v1/train_native_trl_index.json", + "steps": 12000, + "width": 48, + "depth": 2, + "batch": 6, + "lr": 5e-05, + "seed": 20261025, + "unroll": 16, + "save_every": 1000, + "workers": 6, + "resume": "/run-state/workspace/research/universal_v1/phase6/initial.pt", + "compile": false, + "noise": 0.01, + "corpus_prob": 0.25, + "group_weights": "0.05,0.075,0.075,0.8" + }, + "added_parameters_seed": 20261024, + "active_context_encoder": "fixed angular-state transport, Stokes observation layer, and moment positivity", + "active_encoder_source_sha256": "cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f", + "physical_state_quadrature": { + "nodes": 3, + "weights": [ + 0.3333333333333333, + 0.3333333333333333, + 0.3333333333333333 + ], + "phase_scale": 0.0001, + "seed": 20261026 + }, + "auxiliary_parameters": 282523, + "mean_forecast": { + "parameters": 282523, + "seed": 20261030, + "step": 12000, + "checkpoint_sha256": "56b7067dd85f4ba3aa13b34aafc87e1267cf7c14e7a898de9c50044b1f559c04", + "training_checkpoint_sha256": "c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916", + "source_sha256": "75b96fd03a17e93bbf4bde60e3e983537debf55df3dc325326748eeb2caa3c8d", + "training_config": { + "steps": 12000, + "batch": 512, + "seed": 20261030, + "width": 192, + "probabilities": [ + 0.1, + 0.1, + 0.1, + 0.7 + ], + "initial_context_probability": 0.5, + "learning_rate": 0.0005, + "mean_loss_weight": 0.25, + "profile_loss_weight": 0.75, + "statistic_cache_manifest": { + "source": "audited train_native_trl_index.json only", + "index_sha256": "27427edbddcce02804ef3bb098649ab058eb64a9564430b08e309d10cba4834c", + "records": 10115, + "shape": [ + 10115, + 101, + 11 + ], + "variance": "population spatial variance; float64 accumulation, float32 storage", + "workers": 2 + }, + "sources_used": [ + "ape2d/burgers", + "ape2d/decay_turb", + "ape2d/diff", + "ape2d/fisher", + "ape2d/gs_alpha", + "ape2d/gs_beta", + "ape2d/gs_delta", + "ape2d/gs_epsilon", + "ape2d/gs_gamma", + "ape2d/gs_iota", + "ape2d/gs_kappa", + "ape2d/gs_theta", + "ape2d/kdv", + "ape2d/kolm_flow", + "ape2d/ks", + "ape2d/sh", + "pdearena/navier_stokes_2d", + "well/active_matter", + "well/turbulent_radiative_layer_2D" + ], + "pretrained_assets": [], + "profile_cache_manifest": { + "source": "audited train_native_trl_index.json only", + "index_sha256": "27427edbddcce02804ef3bb098649ab058eb64a9564430b08e309d10cba4834c", + "shape": [ + 10115, + 101, + 11, + 2, + 8 + ], + "features": "eight equal bins of each spatial marginal per channel; float64 averaging, float32 storage", + "workers": 2 + } + }, + "weight": 0.5, + "profile_weight": 0.25, + "application": "same model evaluated on every case; no inference-time fitting" + } +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/predict.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/predict.py new file mode 100644 index 0000000000000000000000000000000000000000..30f6262b40f7d2753c5c2b38b68a4b211505a8fe --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/predict.py @@ -0,0 +1,87 @@ +import os +os.environ.setdefault('OMP_NUM_THREADS','4') +import json,sys,time +from pathlib import Path +import numpy as np +import h5py +import torch +import torch.nn.functional as F +from model import Operator,metadata,MAXC +from spectral import SpectralUpdate,eligible +from constraints import project_active +from arena import ArenaOperator,eligible as arena_eligible +from regularize import regularize +from mean_model import MeanForecast, forecast_statistics, correct_statistics + +@torch.no_grad() +def predict(model,x,meta,ds,steps,device='cuda',detail_decay=0.8,physics=False,active_tau=None,kinetic_cfg=None,arena=False,neural_regularization=None,native_rectangular=False,spectral_options=None): + if steps==0:return np.empty((0,*x.shape[1:]),dtype=np.float32) + x=torch.as_tensor(x,device=device,dtype=torch.float32) + t,c,h,w=x.shape + if c>MAXC: + return np.concatenate([predict(model,x[:,i:i+MAXC].cpu().numpy(),meta,ds,steps,device,detail_decay,physics,active_tau,kinetic_cfg,arena,neural_regularization,native_rectangular,spectral_options) for i in range(0,c,MAXC)],1) + factor=2 if min(h,w)>=192 or min(h,w)>=128 and max(h,w)/min(h,w)>2 and not native_rectangular else 1 + z=F.avg_pool2d(x,factor) if factor>1 else x + md,di,bc=metadata(meta,ds);md=torch.tensor(md,device=device)[None];ids=torch.tensor([di],device=device) + future_means,future_profiles=forecast_statistics(model.mean_forecast,x,md,ids,steps) if hasattr(model,'mean_forecast') else (None,None) + detail=x[-1]-F.interpolate(z[-1,None],size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else None + ctx=z[None];out=[] + solver=SpectralUpdate(meta,(h,w),device,**(spectral_options or {})) if physics and eligible(meta,c) else None + if arena and arena_eligible(meta,c) and min(h,w)>1:solver=ArenaOperator(meta,x[-1]) + fullstate=x[-1];physical_steps=0;physical_name=type(solver).__name__ if solver is not None else None + for k in range(steps): + with torch.autocast(device_type=device,dtype=torch.bfloat16,enabled=device=='cuda'):y=model(ctx,md,ids,bc) + if not torch.isfinite(y).all():y=ctx[:,-1].clone() + if solver is not None: + candidate=solver.step(fullstate) + if torch.isfinite(candidate).all(): + physical_steps+=1 + fullstate=candidate;y=(F.avg_pool2d(candidate,factor) if factor>1 else candidate)[None] + else:solver=None + ctx=torch.cat([ctx[:,1:],y[:,None]],1) + yh=F.interpolate(y,size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else y[0] + if detail is not None:yh=yh+detail*(detail_decay**(k+1)) + if solver is not None:yh=fullstate + yh=torch.where(torch.isfinite(yh),yh,x[-1]) + out.append(yh.cpu().numpy()) + pred=np.stack(out).astype(np.float32) + if physical_name:print(f'physical steps: {physical_name} {physical_steps}/{steps}',flush=True) + if neural_regularization is not None and solver is None: + raw=torch.as_tensor(pred,device=device);pred=regularize(raw,meta,**neural_regularization).cpu().numpy() + if future_means is not None and solver is None: + pred=correct_statistics(torch.as_tensor(pred,device=device),future_means,future_profiles,model.mean_weight,model.profile_weight,bc).cpu().numpy() + if active_tau is not None: + raw=torch.as_tensor(pred,device=device);projected=project_active(x,raw,meta,active_tau) + pred=torch.where(torch.isfinite(projected),projected,raw).cpu().numpy() + if kinetic_cfg and str(meta.get('pde',''))=='active_matter' and c==11 and list(meta.get('fields',[]))==['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3'] and all(k in meta.get('constants',{}) for k in ['L','alpha','zeta']): + from kinetic import KineticOperator + cfg=dict(kinetic_cfg) + try: + op=KineticOperator(meta,x,**cfg) + except RuntimeError: + if not cfg.get('cuda_graph',False):raise + cfg['cuda_graph']=False;op=KineticOperator(meta,x,**cfg) + physical=op.rollout(steps,guard=True);used=len(physical) + if used:pred[:used]=physical.cpu().numpy() + print(f'kinetic frames: {used}/{steps}',flush=True) + return pred + +def main(): + torch.set_num_threads(4);torch.manual_seed(0);torch.backends.cudnn.benchmark=False;torch.backends.cudnn.deterministic=True + inp,out=map(Path,sys.argv[1:]);here=Path(os.environ['PDE_SURROGATE_PACKAGE_DIR']) if 'PDE_SURROGATE_PACKAGE_DIR' in os.environ else Path(__file__).resolve().parent + device='cuda' if torch.cuda.is_available() else 'cpu';cfg=json.loads((here/'config.json').read_text()) + model=Operator(cfg['width'],cfg['depth']).to(device).eval() + model.load_state_dict(torch.load(here/'weights.pt',map_location=device,weights_only=True));(out/'cases').mkdir(parents=True,exist_ok=True) + if cfg.get('mean_forecast'): + model.mean_forecast=MeanForecast(cfg['mean_forecast']['width']).to(device).eval() + model.mean_forecast.load_state_dict(torch.load(here/'mean_weights.pt',map_location=device,weights_only=True)) + model.mean_weight=float(cfg['mean_forecast']['weight']) + model.profile_weight=float(cfg['mean_forecast']['profile_weight']) + index=json.loads((inp/'index.json').read_text());t0=time.time() + for i,case in enumerate(index['cases']): + with h5py.File(inp/'cases'/(case['case_id']+'.h5'),'r') as f: + x=f['input'][()];meta=dict(f.attrs);meta['constants']=json.loads(meta.get('constants','{}')) + y=predict(model,x,meta,case['dataset'],int(case['n_steps_out']),device,cfg.get('detail_decay',0.8),cfg.get('physics',False),cfg.get('active_tau',None),cfg.get('kinetic',None),cfg.get('arena',False),cfg.get('neural_regularization',None),cfg.get('native_rectangular',False),cfg.get('spectral_options',None)) + with h5py.File(out/'cases'/(case['case_id']+'.h5'),'w') as f:f.create_dataset('prediction',data=y) + print(i+1,len(index['cases']),case['dataset'],'seconds',round(time.time()-t0,1),flush=True) +if __name__=='__main__':main() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/spectral.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/spectral.py new file mode 100644 index 0000000000000000000000000000000000000000..5b278afcdaa70ba7a8a40a0d3a1c40ab6aa94669 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S053/spectral.py @@ -0,0 +1,115 @@ +"""Metadata-parameterized Fourier update written independently in this session. + +The fixed differential operators supply a physics branch of the shared neural +surrogate. No trajectories or solver parameter fitting are accessed at inference. +""" +import math +import torch +from numerics import resample + +SUPPORTED={'Diffusion','Fisher-KPP','Gray Scott','Swift-Hohenberg','Kuramoto-Sivashinsky','Korteweg-de-Vries','Burgers','Navier-Stokes: Decaying Turbulence','Navier-Stokes: Kolmogorov Flow'} + +def eligible(meta,c): + name=str(meta.get('pde','')) + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + needed={'Diffusion':['Viscosity X','Viscosity Y'],'Fisher-KPP':['Diffusivity','Reactivity'],'Gray Scott':['Feed Rate','Kill Rate'],'Swift-Hohenberg':['Reactivity','Critical Number'],'Kuramoto-Sivashinsky':[],'Korteweg-de-Vries':['Viscosity'],'Burgers':['Viscosity'],'Navier-Stokes: Decaying Turbulence':['Viscosity'],'Navier-Stokes: Kolmogorov Flow':['Viscosity']} + return name in SUPPORTED and all(k in meta.get('constants',{}) for k in needed.get(name,[])) and c==nc and all(str(b)=='periodic' for b in meta.get('boundary_conditions',[])) + +def phi(z,n): + near=z.abs()<0.5;safe=torch.where(near,torch.ones_like(z),z) + v=torch.expm1(safe) + for j in range(1,n):v=v-safe**j/math.factorial(j) + v=v/safe**n;zs=torch.where(near,z,torch.zeros_like(z));series=torch.zeros_like(z) + for j in range(16):series=series+zs**j/math.factorial(n+j) + return torch.where(near,series,v) + +class SpectralUpdate: + def __init__(self,meta,shape,device,cuda_graph=True,gs_scale=1,gs_dt=2.,gs_order=4,gs_double=False,advection_scale=1,advection_double=False): + advective=meta['pde'] in {'Burgers','Navier-Stokes: Kolmogorov Flow'} + scale=gs_scale if meta['pde']=='Gray Scott' else advection_scale if advective else 1 + self.order=4;self.original_shape=shape;self.fine=scale>1;self.initialized=False + self.double=gs_double if meta['pde']=='Gray Scott' else advection_double if advective else False + self.real_dtype=torch.float64 if self.double else torch.float32 + self.complex_dtype=torch.complex128 if self.double else torch.complex64 + if self.fine:shape=tuple(int(scale*s) for s in shape) + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.to(self.real_dtype);self.ky=ky.to(self.real_dtype);k2=kx*kx+ky*ky;self.k2=k2.to(self.real_dtype) + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if self.fine: + ox,oy=self.original_shape;fx=torch.fft.fftfreq(ox,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(oy,device=device,dtype=torch.float64)[None,:];self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=gs_dt;self.order=gs_order + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=torch.arange(ny,device=device,dtype=self.real_dtype)*ly/ny+7/16*ly/self.original_shape[1] if self.fine else (torch.arange(ny,device=device)+7/16)*ly/ny + self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(self.complex_dtype if v.is_complex() else self.real_dtype) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + self.graph=None + if cuda_graph and torch.device(device).type=='cuda' and name!='Diffusion': + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + self.buffer=torch.zeros((nc,nx,ny//2+1),device=device,dtype=self.complex_dtype) + try: + def body():self.buffer.copy_(self.substep(self.buffer)) + stream=torch.cuda.Stream();stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3):body() + torch.cuda.current_stream().wait_stream(stream);torch.cuda.synchronize();graph=torch.cuda.CUDAGraph() + with torch.cuda.graph(graph):body() + self.graph=graph + except RuntimeError: + self.graph=None + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + if self.fine: + if not self.initialized: + latent=torch.fft.irfft2(torch.fft.rfft2(u.to(self.real_dtype))/self.observation,s=self.original_shape);v=self.fft(resample(latent,self.shape));self.initialized=True + else:v=self.internal + else:v=self.fft(u.to(self.real_dtype))/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + if self.graph is not None: + self.buffer.copy_(v) + for _ in range(self.substeps):self.graph.replay() + v=self.buffer + else: + for _ in range(self.substeps):v=self.substep(v) + if self.fine: + self.internal=v + low=resample(self.ifft(v),self.original_shape) + return torch.fft.irfft2(torch.fft.rfft2(low)*self.observation,s=self.original_shape).float() + return self.ifft(v*self.observation).float() + def substep(self,v): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;return stage+self.etd2*(self.nonlinear(stage)-nv) + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + return self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/config.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/config.json new file mode 100644 index 0000000000000000000000000000000000000000..a215d5ae47e0e27e2106ae5fd7fd9e614de7a43a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/config.json @@ -0,0 +1,50 @@ +{ + "width": 48, + "depth": 2, + "detail_decay": 0.8, + "physics": true, + "active_tau": 10.0, + "kinetic": { + "modes": 32, + "resolution": 256, + "iterations": 6, + "cuda_graph": true, + "transport": "phase", + "observer_weight": 1.0, + "observer_ridge": 0.01, + "observer_iterations": 12, + "quadrature_scale": 0.0003, + "quadrature_seed": 20261026 + }, + "arena": true, + "neural_regularization": { + "profile_mix": 0.75, + "horizon": 30, + "pressure_mix": 0.0, + "pressure_start": 10, + "pressure_end": 30 + }, + "spectral_options": { + "gs_scale": 2, + "gs_dt": 0.25, + "gs_order": 2, + "gs_double": true, + "cuda_graph": true, + "advection_scale": 2, + "advection_double": true + }, + "native_rectangular": true, + "global_mixing": true, + "mean_forecast": { + "weight": 0.5, + "profile_weight": 0.25, + "width": 192 + }, + "momentum_pressure": { + "weight": 0.05 + }, + "coarse_forecast": { + "width": 192, + "weight": 0.25 + } +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/mean_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/mean_model.py new file mode 100644 index 0000000000000000000000000000000000000000..b4eaad429a2f439a5a6074d0e7ad97dd55fdce9a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/mean_model.py @@ -0,0 +1,96 @@ +"""Shared context-statistic forecast; all parameters learned from random initialization.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + + +class MeanForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4*2*16) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), nn.Linear(width, MAXC*33), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, profiles=None): + # One independently specified horizon per row. All supplied statistics + # describe only the four context frames; no inference-time fitting. + b, t, c = means.shape + assert t == 4 and c <= MAXC + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = (means - last[:, None]) / scale[:, None] + relative = relative.clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + assert profiles is not None + last_profiles=profiles[:,-1] + profiles=(profiles-means[:,:,:,None,None])/scale[:,None,:,None,None] + profiles=F.pad(profiles.clamp(-100,100),(0,0,0,0,0,MAXC-c)).flatten(1) + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale)*mask/5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + profiles, + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + delta=self.net(torch.cat([features,md,self.ds(ids),time],1)).reshape(b,MAXC,33)[:,:c] + future_mean=last+scale*delta[:,:,0]*h + future_profiles=last_profiles+scale[:,:,None,None]*delta[:,:,1:].reshape(b,c,2,16)*h[:,:,None,None] + future_profiles=future_profiles-future_profiles.mean(-1,keepdim=True)+future_mean[:,:,None,None] + return future_mean,future_profiles + + +@torch.no_grad() +def forecast_statistics(model, context, md, ids, steps): + mean = context.double().mean((-2, -1)).float() + variance = context.double().var((-2, -1), unbiased=False).float() + profx=F.adaptive_avg_pool2d(context.double(),(16,1)).squeeze(-1) + profy=F.adaptive_avg_pool2d(context.double(),(1,16)).squeeze(-2) + profiles=torch.stack([profx,profy],-2).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + profiles=profiles[None].expand(steps,-1,-1,-1,-1)) + + +@torch.no_grad() +def forecast_means(model,context,md,ids,steps): + return forecast_statistics(model,context,md,ids,steps)[0] + +def interpolate_profile(profile,size,periodic): + position=(torch.arange(size,device=profile.device,dtype=profile.dtype)+.5)*16/size-.5 + left=position.floor().long();weight=position-left + if periodic: + i=left.remainder(16);j=(left+1).remainder(16) + else: + i=left.clamp(0,15);j=(left+1).clamp(0,15) + output=profile[...,i]*(1-weight)+profile[...,j]*weight + return output-output.mean(-1,keepdim=True) + +@torch.no_grad() +def correct_statistics(prediction,means,profiles,mean_weight,profile_weight,bc): + raw=prediction.double();center=raw.mean((-2,-1)) + means=torch.where(torch.isfinite(means),means.double(),center) + output=raw+mean_weight*(means-center)[:,:,None,None] + if profile_weight: + current_x=F.adaptive_avg_pool2d(raw,(16,1)).squeeze(-1) + current_y=F.adaptive_avg_pool2d(raw,(1,16)).squeeze(-2) + current=torch.stack([current_x,current_y],-2) + forecast=torch.where(torch.isfinite(profiles),profiles.double(),current) + difference=(forecast-forecast.mean(-1,keepdim=True))-(current-current.mean(-1,keepdim=True)) + dx=interpolate_profile(difference[:,:,0],raw.shape[-2],bc[0]) + dy=interpolate_profile(difference[:,:,1],raw.shape[-1],bc[1]) + output=output+profile_weight*(dx[:,:,:,None]+dy[:,:,None,:]) + return output.float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/momentum.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/momentum.py new file mode 100644 index 0000000000000000000000000000000000000000..b8148c38e71e00a0b24a0a11ff8494ebc95c96d7 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S079/momentum.py @@ -0,0 +1,49 @@ +"""Fixed pressure-gradient projection from forecast momentum; PyTorch only.""" +import math +import torch + + +def gradient(values,spacing,axis): + x=values.movedim(axis,0);out=torch.empty_like(x) + out[1:-1]=(x[2:]-x[:-2])/(2*spacing) + out[0]=(-3*x[0]+4*x[1]-x[2])/(2*spacing) + out[-1]=(3*x[-1]-4*x[-2]+x[-3])/(2*spacing) + return out.movedim(0,axis) + + +def project_faces(fx,fy,dx,dy): + nx,ny=fx.shape[-2:];device=fx.device;dtype=fx.dtype + rhs=(torch.roll(fx,1,-2)-fx)/dx + rhs[...,:-1]-=fy/dy;rhs[...,1:]+=fy/dy + kx=torch.fft.fftfreq(nx,device=device,dtype=dtype) + ky=torch.arange(ny,device=device,dtype=dtype) + eigen=4*torch.sin(math.pi*kx[:,None]).square()/dx**2+4*torch.sin(math.pi*ky[None,:]/(2*ny)).square()/dy**2 + eigen[0,0]=1 + cosine=(2/ny)**.5*torch.cos(math.pi/ny*ky[:,None]*(ky[None,:]+.5));cosine[0]*=2**-.5 + coefficients=torch.fft.fft(rhs@cosine.T,dim=-2,norm='ortho')/eigen + coefficients[...,0,0]=0 + return torch.fft.ifft(coefficients,dim=-2,norm='ortho').real@cosine + + +@torch.no_grad() +def pressure_projection(context,prediction,meta,weight=.05): + fields=list(meta.get('fields',[]));bc=list(meta.get('boundary_conditions',[])) + names=['density','pressure','velocity_0','velocity_1'] + if len(fields)!=prediction.shape[1] or not all(name in fields for name in names):return prediction + if len(bc)!=4 or bc[:2]!=['periodic','periodic'] or 'periodic' in bc[2:]:return prediction + if min(prediction.shape[-2:])<3 or not len(prediction) or not weight:return prediction + try: + dt=float(meta.get('dt',0));lx,ly=map(float,meta.get('domain_extent',[0,0])) + except (ValueError,TypeError):return prediction + if not all(math.isfinite(v) and v>0 for v in [dt,lx,ly]):return prediction + state=torch.cat([context,prediction]).double();density,pressure,vx,vy=[fields.index(name) for name in names] + rho=state[:,density];u=state[:,vx];v=state[:,vy];nx,ny=rho.shape[-2:];dx=lx/nx;dy=ly/ny + ddx=lambda value:(torch.roll(value,-1,-2)-torch.roll(value,1,-2))/(2*dx) + force_x=-gradient(rho*u,dt,0)-ddx(rho*u*u)-gradient(rho*u*v,dy,-1) + force_y=-gradient(rho*v,dt,0)-ddx(rho*u*v)-gradient(rho*v*v,dy,-1) + faces_x=.5*(force_x+torch.roll(force_x,-1,-2));faces_y=.5*(force_y[...,:-1]+force_y[...,1:]) + estimate=project_faces(faces_x,faces_y,dx,dy)[len(context):] + original=prediction[:,pressure].double();center=original.mean((-2,-1),keepdim=True) + updated=(1-weight)*original+weight*(estimate+center) + result=prediction.clone();result[:,pressure]=torch.where(torch.isfinite(updated),updated,original).float() + return result diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/EXPERIMENTS.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/EXPERIMENTS.md new file mode 100644 index 0000000000000000000000000000000000000000..26ae02e88b19e22a9cdd6bb8de718ddc5601a6f3 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/EXPERIMENTS.md @@ -0,0 +1,3375 @@ +# Experiment ledger + +All development uses train/valid pools of `/workspace/data/split.json`. No held-out targets or case definitions are used. One H100 GPU is available. No pretrained assets are used. + +## E000 — package initialization +- Hypothesis: a spatial-mean prior provides a finite, portable starting package while training begins. +- Changed factor: parameter-free spatial-mean prediction. +- Seed: not applicable (deterministic). +- Validation: fixed 81-case subset (seed 71, all start at frame 0), reward 0.313512; 0 failed cases. +- Checkpoint hash: no checkpoint; predictor SHA256 307c11c9f4f3f186b96a34597822e1e40e95d47d6159b3de3fe0934ebae13d7d. +- Outcome: initial package published; not a trained candidate or a reproduced baseline run. + +## E001 — coupled, metadata-conditioned multiscale model +- Hypothesis: joint channel modeling, physical metadata, scale-aware normalization and periodic boundaries improve cross-equation prediction relative to the documented starter weaknesses. +- Changed factors: new shared convolutional architecture; all 19 dataset identifiers represented in one parameter set; group-balanced training. +- Initialization seed: 20261021. +- Validation: pending. +- Checkpoint hash: pending. +- Outcome: preparing train-only reduced-resolution cache and validation data. + +### E000 validation VRMSE by dataset + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 1.000634 | +| ape2d/decay_turb | 0.999991 | +| ape2d/diff | 0.999973 | +| ape2d/fisher | 1.581309 | +| ape2d/gs_alpha | 1.005787 | +| ape2d/gs_beta | 1.023021 | +| ape2d/gs_delta | 1.591578 | +| ape2d/gs_epsilon | 1.004624 | +| ape2d/gs_gamma | 1.000439 | +| ape2d/gs_iota | 1.633860 | +| ape2d/gs_kappa | 1.079744 | +| ape2d/gs_theta | 1.574410 | +| ape2d/kdv | 1.000191 | +| ape2d/kolm_flow | 0.999992 | +| ape2d/ks | 0.999992 | +| ape2d/sh | 1.041517 | +| pdearena/navier_stokes_2d | 1.000318 | +| well/active_matter | 0.992894 | +| well/turbulent_radiative_layer_2D | 1.502768 | + +## E001-smoke — 100 updates +- Hypothesis / changed factor: Short pipeline check of new coupled operator with 40 base channels, 12 samples per batch, one-step objective, float16 cache; no baseline retraining. Float16 cache will be replaced to reduce quantization error. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Validation: fixed 81-case subset; reward **0.322278**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.779417 | +| ape2d/decay_turb | 0.919250 | +| ape2d/diff | 0.573262 | +| ape2d/fisher | 1.128033 | +| ape2d/gs_alpha | 0.999278 | +| ape2d/gs_beta | 1.093050 | +| ape2d/gs_delta | 1.567459 | +| ape2d/gs_epsilon | 0.717784 | +| ape2d/gs_gamma | 0.761942 | +| ape2d/gs_iota | 1.560647 | +| ape2d/gs_kappa | 0.956638 | +| ape2d/gs_theta | 1.511911 | +| ape2d/kdv | 0.764748 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.939004 | +| ape2d/sh | 0.919993 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — intermediate 2,000 updates +- Hypothesis / changed factor: Intermediate checkpoint of the same 8,000-update one-step run; all other factors unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `cc58e45b55743bae890d3c3f212be97b831186431591d2501186f69378ce6f37`. +- Validation: fixed 81-case subset; reward **0.263620**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.817172 | +| ape2d/decay_turb | 2.389786 | +| ape2d/diff | 0.837767 | +| ape2d/fisher | 1.272807 | +| ape2d/gs_alpha | 2.153748 | +| ape2d/gs_beta | 8.179170 | +| ape2d/gs_delta | 1.369924 | +| ape2d/gs_epsilon | 1.198089 | +| ape2d/gs_gamma | 1.805734 | +| ape2d/gs_iota | 1.964417 | +| ape2d/gs_kappa | 0.904690 | +| ape2d/gs_theta | 1.609477 | +| ape2d/kdv | 1.151686 | +| ape2d/kolm_flow | 2.257709 | +| ape2d/ks | 2.152634 | +| ape2d/sh | 0.836066 | +| pdearena/navier_stokes_2d | 1.003965 | +| well/active_matter | 2.135917 | +| well/turbulent_radiative_layer_2D | 1.657301 | + +Outcome: did not improve validation reward; retained current package. + +## E003 — metadata-conditioned spectral branch +- Hypothesis / changed factor: Add independently implemented ETDRK4 differential operators to the shared model. All neural weights unchanged from E001-smoke; neural branch still applied every step on every case. Equation parameters come from metadata; solver forms selected using train-only diagnostics. Kolmogorov remains with the neural update pending forcing/drag diagnosis. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Runtime code SHA256: `9e848b7d80d919c0b32a15d07bf4643a1b56b5235004827d0950660afa2e4952`. +- Validation: fixed 81-case subset; reward **0.401067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — one-step operator, 8,000 updates +- Hypothesis / changed factor: Joint channel ConvNeXt U-Net, metadata conditioning, group-balanced batches, float32 reduced-grid cache, one-step log variance-normalized MSE. Random initialization; 48 base channels and 16 samples per batch. Development inference wrapper relative-path error was corrected before scoring. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `9a645f05c3a17872093f7f461376b1fedd663ceefbbd879ef8f1cd4b8d8e58fb`. +- Validation: fixed 81-case subset; reward **0.350858**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: did not improve validation reward; retained current package. + +## E004 — trained neural branch and forced turbulence operator +- Hypothesis / changed factor: Replace smoke neural weights with E001 step 8,000, still one shared network across all cases. Add Kolmogorov forcing -4 cos(4y) and drag 0.2, identified only in training diagnostics. Both effects are recorded together; full group scores isolate transfer changes from corpus changes. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `d8dd1c803a436415851c119241900367c593b0acfb904f6f5b3f3b17b7444478`. +- Validation: fixed 81-case subset; reward **0.437086**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## E005A — physical output constraints and horizon decay (validation ablations) + +Hypothesis: exact tensor identities and reduced confidence at longer horizons stabilize active matter. All experiments reuse E001 8,000-update predictions; seed 20261021. No new weights or training data. Checkpoint SHA256: df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d. + +The diffusion coefficient 0.05, trace(D)=concentration and E=sym(gradient velocity) were checked on training pools. Horizon decay candidates 10, 20 and 40 steps are hyperparameters selected on validation, not parameters fitted to validation targets. These ablations are not published until integrated into a portable package and replayed. + +### active_diffusion: Analytic concentration diffusion only + +Reward 0.351796; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.620833 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_constraints: Concentration diffusion and algebraic tensor projection + +Reward 0.352944; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.570211 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_strain: Recompute strain from predicted velocity + +Reward 0.338342; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 2.350641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: negative; strain derivatives amplify velocity errors. + +### active_projection: Weighted Fourier projection of velocity and strain + +Reward 0.352917; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.571374 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_10: Weighted projection and exponential shrinkage, tau=10 + +Reward 0.375912; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832421 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_20: Weighted projection and exponential shrinkage, tau=20 + +Reward 0.372622; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.911641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_40: Weighted projection and exponential shrinkage, tau=40 + +Reward 0.365053; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.123714 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +## Train-only solver diagnostics (no validation selection) + +Initial Navier–Stokes sign and step-size hypotheses failed (non-finite or high training error); smaller steps and the opposite vorticity convention succeeded. KS requires separate fourth derivatives, not the squared Laplacian. Kolmogorov forcing and drag were identified using training trajectories only. Doubling the latent solver grid from 256 to 512 barely changed Gray–Scott/KS training errors and was rejected as a general fix. Grid averaging, integrator order and timestep are under investigation. These are diagnostic tests, not held-out experiments; validation VRMSE is not applicable. No checkpoint was created. Source, configurations and train results are retained under research/physics_diagnostics. + +## E005B — integrated physical projection and horizon regularization +- Hypothesis / changed factor: E004 model plus concentration diffusion, weighted Fourier projection of velocity and strain, tensor identities and exponential relaxation toward context means for active-matter fluctuations, tau=10 steps. All weights unchanged; tau selected from fixed validation ablations. The neural model is still applied to every case. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `83bec07505f714951064c32e8cb4cc20720921d9b62a1e4c1ae3bba353048798`. +- Validation: fixed 81-case subset; reward **0.462140**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Full validation of physics_smoke + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.400925**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.060324 | +| ape2d/decay_turb | 0.000387 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.002465 | +| ape2d/gs_alpha | 0.022938 | +| ape2d/gs_beta | 0.012380 | +| ape2d/gs_delta | 0.146653 | +| ape2d/gs_epsilon | 0.026477 | +| ape2d/gs_gamma | 0.222599 | +| ape2d/gs_iota | 0.036049 | +| ape2d/gs_kappa | 0.109056 | +| ape2d/gs_theta | 0.135853 | +| ape2d/kdv | 0.025460 | +| ape2d/kolm_flow | 0.979667 | +| ape2d/ks | 0.080605 | +| ape2d/sh | 0.002222 | +| pdearena/navier_stokes_2d | 0.923463 | +| well/active_matter | 1.120708 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E002 — intermediate 2,000 multi-step updates +- Hypothesis / changed factor: E005B inference architecture and physical operators; replace only shared neural weights with the 2,000-update checkpoint of four-step fine-tuning with 0.01 input noise. +- Seed: 20261021. +- Checkpoint SHA256: `d720a4dbeee0c16a125aca0cee22ee060e1e80c6b863efbd169fd69b5f2c9efc`. +- Runtime code SHA256: `f4b3be2c3e227fcfa3e8cb9774970780928d6635210a425fa2c6b07e8f08e8a5`. +- Validation: fixed 81-case subset; reward **0.473422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.559579 | +| well/active_matter | 0.827134 | +| well/turbulent_radiative_layer_2D | 0.628780 | + +Outcome: improvement; published atomically. + +## E006 — observation and integration operators +- Hypothesis / changed factor: Same E005B neural weights, projection and horizon regularization. Spectral branch now models finite 8-cell averaging, the corresponding Kolmogorov forcing location, and ETDRK2 substeps for Fisher/Swift–Hohenberg/KS/KdV confirmed on training trajectories. Gray–Scott uses ETDRK4 max substep 2. No reference generator code or seeds used. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.490722**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Contract and determinism verification + +Seed 0 inference checks on three validation inputs spanning scalar periodic, 11-channel periodic and rectangular mixed-boundary data, plus independently constructed unseen 2-channel 7×11 and 13-channel 129×177 inputs. All shapes/dtypes were correct, every output finite, maximum rerun difference exactly 0. Small-grid padding and non-finite-step fallback added without changing any predictions on the three validation cases (bitwise parity). No synthetic training trajectories were generated. Evidence: research/universal_v1/contract_checks.json. + +## E007 — combine improved physics and multi-step training +- Hypothesis / changed factor: E006 spectral observation/integration model and E005B output projection, with shared neural weights after 4,000 four-step fine-tuning updates. Corpus and transfer gains combined; all weights originate from E001 random initialization. +- Seed: 20261021. +- Checkpoint SHA256: `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.506925**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.531697 | +| well/active_matter | 0.821690 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: improvement; published atomically. + +## E002 — four-step training with input perturbations +- Hypothesis / changed factor: Fine-tune E001 for 8,000 updates with four autoregressive steps in the loss, 0.01 normalized input noise, lr 0.0003, batch 12. Test whether recovery from model errors improves long rollouts. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `1d2b51ab44d7b8212b22c4c1894b02b03697b7ed1d30f756f216e6076006e48d`. +- Validation: fixed 81-case subset; reward **0.398333**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.411539 | +| ape2d/decay_turb | 0.688351 | +| ape2d/diff | 0.194736 | +| ape2d/fisher | 0.729932 | +| ape2d/gs_alpha | 0.660633 | +| ape2d/gs_beta | 0.856659 | +| ape2d/gs_delta | 0.790159 | +| ape2d/gs_epsilon | 0.457965 | +| ape2d/gs_gamma | 0.625919 | +| ape2d/gs_iota | 0.527543 | +| ape2d/gs_kappa | 0.587907 | +| ape2d/gs_theta | 0.768902 | +| ape2d/kdv | 0.263418 | +| ape2d/kolm_flow | 0.973194 | +| ape2d/ks | 0.959533 | +| ape2d/sh | 0.611700 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.929633 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: did not improve validation reward; retained current package. + +## E002 — final multi-step checkpoint with validated inference operators +- Hypothesis / changed factor: Replace only the E007 shared neural weights with E002 after all 8,000 four-step fine-tuning updates. Physical operators and active-matter horizon regularization unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.511096**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.801125 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## Full validation of phase2_4000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.500153**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045446 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.527867 | +| well/active_matter | 0.839802 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E009 — active-fluid moment physics, train-only diagnostics + +Hypothesis: the transfer error is partly due to missing orientation-distribution moments. Implemented independent Bingham closure and a Fourier angular-moment Smoluchowski/Stokes system, based on public equation descriptions (the local The Well active_matter README) and training-only checks. No generator code, seeds, pretrained weights or normalization statistics used. Translation and rotation diffusivities 0.05 and beta 0.8 were verified against train fields; alpha and zeta come from supplied metadata. + +Changed factors: Bingham closure vs retained angular moments; spatial grids 128/256; angular resolution 12/24/40; zero vs Bingham initialization; fixed recurrent context encoder that inserts observed low moments. The encoder updates physical state only; no parameters or gradients are fitted at inference. + +Seed: not applicable (deterministic analytical operators). Checkpoint: none; these are not published candidates. Validation VRMSE: not yet run for any dataset. Four selected TRAIN trajectories only. Bingham 30-step train mean VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5): 0.0261, 0.5173, 0.8618, 1.0525. Doubling spatial resolution did not resolve closure drift. At (3,-2), 12 angular modes with context encoding reduced four-step train VRMSE to about 0.0044, versus 0.153 for Bingham. Twelve modes were inadequate at high activity; increasing angular resolution is being tested. A GPU prototype initially failed due to TorchScript complex-number optimization; this is a portability failure, not a scored model. + +Outcome: promising mechanism, still under development; retained the validated shared neural/spectral model. Detailed train results and source variants remain in research/physics_diagnostics. + + +## E009b — active kinetic resolution and implementation diagnostics + +Hypothesis: angular and spatial discretization limit the kinetic closure. Changed factors: M=32/40/64 even angular modes, 128 versus native256 spatial grid, spatial2/3 filtering enabled/disabled, eager versus CUDA graph execution. Seed: none (deterministic operators); neural checkpoint unchanged. Validation: not evaluated for these train-only numerical diagnostics. Runtime source and all numerical results are in physics_diagnostics. + +Native256/M32 thirty-step TRAIN VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5) was 0.025519, 0.014763, 0.100166, 0.880100. M40 and M64 gave nearly equal short-rollout errors, whereas 128 spatial resolution remained inaccurate or unstable at high activity. Removing the spatial filter did not help. Four-step errors at native256/M64 were approximately 0.0024/0.0031/0.0041/0.0055 for (9,-3), and 0.0233/0.0483/0.0937/0.1448 for (17,-5). Outcome: select native256/M32 for validation, retaining a finite physical-prefix fallback. TorchScript fusion failures and CUDA graph scalar-copy failures were fixed before scoring. CUDA graphs reduce launch overhead; all state storage addresses remain fixed. + +## E011a — buoyant smoke discretization, train-only identification + +Hypothesis: a fixed incompressible-fluid operator can reduce PDEArena rollout error once the observed grid layout is represented correctly. Independent MAC solver with semi-Lagrangian advection, explicit viscosity and discrete Neumann pressure projection. Training data reveal that saved velocities are lower faces with the final zero normal face omitted. Correct discrete divergence is about 0.00016 RMS. Physical domain32 and stored interval6 account for endpoint metadata conventions. + +Changed factors on three TRAIN trajectories: stored interval6 versus6.461538, internal time steps0.25/0.5/1/1.5, viscosity0/0.001/0.005/0.008/0.01/0.012, and linear versus MacCormack scalar advection. Seed: none; checkpoint: none (analytic prototype). Validation per dataset: not yet evaluated. At interval6, linear scalar advection and viscosity0.01, one-step mean train VRMSE was 0.0000243. MacCormack with the same viscosity gave0.0981; reducing the internal step to0.25 with MacCormack gave0.3391. Thus closer continuum integration did not reproduce the observed numerical dynamics. Eight additional complete TRAIN rollouts with the selected scheme had mean VRMSE below0.00011. A separate torch-only double-precision implementation reproduced the SciPy prototype. Outcome: proceed to validation; no held-out data or generator code used. Detailed tested values and errors are retained in arena_initial_results.json, arena_scheme_results.json, arena_long_results.json and arena_torch_results.json. + +## E010 — fixed angular-moment active-fluid operator +- Hypothesis / changed factor: Add the independently implemented Smoluchowski/Stokes angular-moment operator to the same phase2-final shared neural model. M32 even angular modes, native256 grid, six Stokes iterations, ETDRK4, deterministic CUDA graph; fixed Bingham initialization and observed low-moment context encoding. No inference parameter fitting. All twelve active validation cases used all thirty physical steps. Exact repeat determinism verified on low/high-activity validation inputs and a diffusion input; two novel grid/channel schemas also passed. Runtime pins the complete package directory during atomic updates. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `d11ea64e646593878f730df43933d3f1e64f481a34083bb7687be973e51e2d33`. +- Validation: fixed 81-case subset; reward **0.572674**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E011 — fixed MAC-grid buoyant-fluid operator +- Hypothesis / changed factor: Keep the E010 shared neural checkpoint and all other inference components; add a metadata-conditioned MAC fluid update for the buoyant-smoke schema. The selected numerical scheme was identified on train pools only. Quick validation reran all twelve PDEArena cases and reused E010 predictions for unchanged operators with identical weights. PDEArena mean VRMSE is 0.00008349. Exact repeat determinism passed for two smoke cases, diffusion, and two novel grid/channel schemas. Expanded end-to-end validation is next. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `397cac1ec1b4314c9e5832071d0cc8a3451a85a8d891ba469a64722ac4d69f65`. +- Validation: fixed 81-case subset; reward **0.713334**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E012a — neural forecast uncertainty, train-only ablation + +Hypothesis: spatial averaging reduces uncertain late-time turbulent fluctuations while retaining vertical mean profiles. Same phase2-final checkpoint (SHA256 d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420), seed20261021. Nine TRAIN trajectories, one for each cooling time; no validation used to select the coefficient. Changed factors: nonnegative material scalars; Gaussian smoothing with final width1/2/4/8/16 pixels; linear horizon-dependent shrinkage toward the mean over periodic directions with final weight0.25/0.5/0.75/1.0 at step30. + +| Variant | Train VRMSE | +|---|---:| +| raw | 0.538547 | +| positive | 0.538100 | +| blur_1 | 0.537242 | +| blur_2 | 0.536386 | +| blur_4 | 0.534177 | +| blur_8 | 0.529554 | +| blur_16 | 0.526478 | +| profile_0.25 | 0.530485 | +| profile_0.5 | 0.525987 | +| profile_0.75 | 0.524946 | +| profile_1.0 | 0.527495 | + +Validation per dataset: not evaluated in this ablation. Outcome: coefficient0.75 selected for subsequent validation. Positivity alone helped very little; smoothing cannot resolve most pressure error. Roughly one-fifth of pressure MSE comes from spatial-mean error, so a correction to the mean alone is insufficient. + +## E012 — fixed spatial uncertainty shrinkage +- Hypothesis / changed factor: The same shared neural weights and physical operators as E011. Add horizon-dependent shrinkage toward means over periodic directions (weight 0.75 by step30), plus nonnegative density/concentration. The coefficient was selected on nine training trajectories before validation. All nine radiative-layer validation rollouts were rerun; unchanged physical outputs were reused for the other72 cases. Exact determinism passed on three radiative-layer inputs and two novel grid/channel schemas. Validation radiative-layer VRMSE improved from0.579723 to0.546835. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715448**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546835 | + +Outcome: improvement; published atomically. + +## E013 — native rectangular inference before fine-tuning +- Hypothesis / changed factor: Change only the shared neural inference grid for rectangular inputs from64x192 to native128x384, keeping the E012 checkpoint and all other operators/regularization fixed. This tests resolution transfer before native-grid fine-tuning. Radiative-layer VRMSE worsened from0.546835 to0.769883; retain E012. All other validation predictions are unchanged physical outputs. The result motivates retraining the shared operator on native grids rather than changing inference resolution alone. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `cc9804d49168dff3c2a3868118de9b3b3bccbd2b0982814f94617a3115b7b6fa`. +- Validation: fixed 81-case subset; reward **0.703067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.769883 | + +Outcome: did not improve validation reward; retained current package. + +## E008 — larger-batch transfer-focused training, update 5000 +- Hypothesis / changed factor: Continue the same shared network from E002 final weights: batch 48, four-step loss, 0.01 input noise, corpus probability 0.1 and 0.3 for each transfer group, lr 0.0003 cosine, sampling seed 20261022; initialization seed remains 20261021. Inference operators frozen to E012 (angular-moment active matter, MAC buoyant smoke and fixed neural uncertainty shrinkage). Quick validation reruns all nine radiative-layer trajectories and reuses fixed physical predictions for the other seventy-two cases, which are independent of the neural weights; expanded validation runs end-to-end. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715555**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + + +## E015a — reaction-diffusion spatial/time convergence, training only + +Hypothesis: unresolved reaction harmonics cause remaining Gray–Scott drift. Changed factors: grid256/512/1024 after finite8-cell observation deconvolution; ETDRK4 maxdt2/0.5 vs ETDRK2 maxdt0.2/0.25/0.3/0.5; optional physical context encoding preserving unresolved modes; finite-averaging ratio2/4/8/16 versus its continuum limit. Deterministic numerical operators, no random seed or learned checkpoint. Validation per dataset: not yet evaluated. + +On selected TRAIN gamma/delta trajectories, grid256/RK4dm2 three-step VRMSE was gamma .000764/.000799/.001016 and delta .000817/.001631/.002567. Grid512/RK2dm.25 gave gamma .000197/.000268/.000394 and delta .000105/.000222/.000335. Grid1024 and the recurrent physical-state encoder gave no additional improvement. Ratio8 remained best among tested observation filters. Thus prior negative grid-only tests did not rule out improvement when resolution and observation inversion were changed together. A portable double-precision GPU implementation is undergoing full train rollouts. Numerical results and all variants are retained in gs_resolution_results.json, gs_resolution_time_results.json, gs_context_results.json and gs_observation_results.json. + +A separate spectral CUDA-graph implementation reproduced ten different corpus training cases exactly over three forecast steps (max absolute difference0 for each). Some runtime reduction was measured while shared training was active, e.g. three Navier–Stokes steps .70s to .47s. This changes execution only; no numerical parameters are fitted at inference. + +## E016 design — shared global spectral feature mixing, not yet trained + +Hypothesis: acoustic pressure patterns need communication across the native rectangular grid. Prepared a shared Fourier feature-mixing residual at the first neural stage, with even extensions in nonperiodic directions and metadata-conditioned complex gains. All weights will be trained jointly on all pools; this is not a dataset specialist. Adds306552 trainable parameters to the2.759547M shared net, total3066099. New output projection starts at zero. Seed20261024 for the new parameters. Initial model matches the existing parent exactly, and a training-only backward pass gives finite nonzero gradients. No optimizer updates or validation results yet; not a publishable candidate. Source and initialization checks in research/global_v1. Continue native-grid training first to establish a comparison. + +## E015b — offline numerical coefficient calibration, negative train result + +Hypothesis: small global corrections to Gray–Scott diffusivities and elapsed time reduce residual error. Fitted three fixed coefficients offline to one-step errors on TRAIN gamma/delta trajectories, with the fine512/RK2dm.25 operator. No inference-time fitting. Seed: none; checkpoint: none (analytic calibration); validation per dataset: not run because full TRAIN rollouts worsened. + +Fitted options: `{"gs_scale": 2, "gs_dt": 0.25, "gs_order": 2, "gs_double": true, "gs_diffusion_u": 1.9999732723556725e-05, "gs_diffusion_v": 1.0000209495791168e-05, "gs_time_scale": 0.9999851457782819}`. One-step normalized MSE decreased from 2.3365036901186555e-08 to 1.9611423235582817e-08. + +| Train dataset | Variant | 26-step VRMSE | +|---|---|---:| +| gs_gamma | uncalibrated | 0.029923 | +| gs_gamma | calibrated | 0.031195 | +| gs_delta | uncalibrated | 0.013152 | +| gs_delta | calibrated | 0.014885 | +| gs_alpha | uncalibrated | 0.002711 | +| gs_alpha | calibrated | 0.002466 | +| gs_theta | uncalibrated | 0.010527 | +| gs_theta | calibrated | 0.049919 | + +Outcome: rejected. The tiny one-step improvement did not carry over to long rollouts, particularly on the unused theta training trajectory. Retain nominal physical coefficients. + +## E015 — finer reaction-diffusion state with matched integration +- Hypothesis / changed factor: Keep the same phase3/update5000 shared neural checkpoint. Gray–Scott now retains a2x internal Fourier grid, uses double-precision ETDRK2 with max internal dt0.25, and observes it through the finite8-cell averaging filter. Nominal physical coefficients retained after calibration failed full train rollouts. Enable deterministic CUDA graphs for spectral updates. Reran all48 corpus validation cases; corpus mean improved from0.032509 to0.010893. Other33 cases reuse identical parent outputs. Exact repeat checks passed on two Gray–Scott cases, Kolmogorov flow and novel schemas; all24 unchanged corpus cases matched parent outputs bit for bit. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `3962bd91a5cf0d9b11b901a10dda99623d4d995b034f57f75c81e62bc41c8aff`. +- Validation: fixed 81-case subset; reward **0.755127**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + +## Full validation of arena_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.696465**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045295 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E014 — native-grid eight-step training, update 1000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.756053**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 1000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755034**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546620 | + +Outcome: did not improve validation reward; retained current package. + +## E017a — active-state observation diagnostics on training trajectories + +Hypotheses: remaining active-fluid drift arises either from numerical resolution or inconsistent hidden angular moments after observed low moments are inserted. Changed factors: time step, spatial grid, Stokes iterations, finite-difference moment inversion, fixed velocity-constrained moment observation, positivity projection, rigid angular transport, and Bingham amplitude transport. All diagnostics use the first trajectory in the corresponding TRAIN pool, starting at frame zero; no validation or held-out targets. The context encoder uses only four observed frames and updates physical state; solver and encoder parameters are fixed. No parameter optimization at inference. + +Seed: deterministic physical operators, no new random parameters. Shared checkpoint unchanged; current selected SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Validation VRMSE per dataset: not evaluated for these train-only diagnostics. + + +kinetic_accuracy_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"name": "baseline", "params": [17, -5]} | 0.07398549 | 0.13735820 | 4 | +| {"name": "smaller_time_step", "params": [17, -5]} | 0.07397478 | 0.13734564 | 4 | +| {"name": "stokes_iterations", "params": [17, -5]} | 0.07398360 | 0.13735427 | 4 | +| {"name": "upsample_384", "params": [17, -5]} | 0.07628622 | 0.14166540 | 4 | +| {"name": "upsample_512", "params": [17, -5]} | 0.07628392 | 0.14165913 | 4 | + +kinetic_encoder_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"last": false, "params": [3, -2], "weight": 0} | 0.00370120 | 0.00440354 | 4 | +| {"last": false, "params": [3, -2], "weight": 1} | 0.01032324 | 0.01005470 | 4 | +| {"last": true, "params": [3, -2], "weight": 0.25} | 0.00641056 | 0.00683838 | 4 | +| {"last": true, "params": [3, -2], "weight": 1} | 0.02690719 | 0.02838808 | 4 | +| {"last": false, "params": [9, -3], "weight": 0} | 0.00375097 | 0.00547713 | 4 | +| {"last": false, "params": [9, -3], "weight": 1} | 0.14332058 | 0.16554202 | 4 | +| {"last": true, "params": [9, -3], "weight": 0.25} | 0.04367409 | 0.04571036 | 4 | +| {"last": true, "params": [9, -3], "weight": 1} | 0.12102679 | 0.12214902 | 4 | +| {"last": false, "params": [17, -5], "weight": 0} | 0.07398282 | 0.13735300 | 4 | +| {"last": false, "params": [17, -5], "weight": 1} | 0.35820127 | 0.58563930 | 4 | +| {"last": true, "params": [17, -5], "weight": 0.25} | 0.05379023 | 0.08525104 | 4 | +| {"last": true, "params": [17, -5], "weight": 1} | 0.20885072 | 0.28744033 | 4 | + +kinetic_flow_encoder_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "weight": 0.0} | 0.02551891 | 0.05217000 | 30 | +| {"params": [1, -1], "weight": 0.25} | 0.01215404 | 0.02445865 | 30 | +| {"params": [1, -1], "weight": 1.0} | 0.00162011 | 0.00254364 | 30 | +| {"params": [3, -2], "weight": 0.0} | 0.01476343 | 0.04135585 | 30 | +| {"params": [3, -2], "weight": 0.25} | 0.00774478 | 0.02111043 | 30 | +| {"params": [3, -2], "weight": 1.0} | 0.00182436 | 0.00253463 | 30 | +| {"params": [9, -3], "weight": 0.0} | 0.10013896 | 0.45469648 | 30 | +| {"params": [9, -3], "weight": 0.25} | 0.03468125 | 0.16266733 | 30 | +| {"params": [9, -3], "weight": 1.0} | 0.04499391 | 0.18140420 | 30 | +| {"params": [17, -5], "weight": 0.0} | 0.54988378 | 1.29260397 | 30 | +| {"params": [17, -5], "weight": 0.25} | 0.48679241 | 1.30647564 | 30 | +| {"params": [17, -5], "weight": 1.0} | 0.62024931 | 1.32728279 | 30 | + +kinetic_transport_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.00372676 | 0.00443898 | 4 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187841 | 0.00173137 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 0.0} | 0.00375795 | 0.00446493 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 1.0} | 0.00187924 | 0.00173272 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00041423 | 0.00039727 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043372 | 0.00043870 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 0.0} | 0.00041337 | 0.00038921 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 1.0} | 0.00039989 | 0.00040295 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.00621011 | 0.01090280 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00566252 | 0.01021467 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 0.0} | 0.00627246 | 0.01102048 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 1.0} | 0.00555848 | 0.01004915 | 4 | + +kinetic_transport_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 0.0} | 0.02551866 | 0.05216906 | 30 | +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00162010 | 0.00254361 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.01480152 | 0.04140588 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183204 | 0.00254354 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00303750 | 0.01372149 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00258502 | 0.00757161 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.45993094 | 1.23510826 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.46775045 | 1.23808050 | 30 | + +Outcome: finer numerics did not improve forecasts. Temporal-derivative inversion was mostly harmful. Fixed velocity observation improves low-activity cases; rigid transport additionally improves moderate/high activity. Bingham amplitude transport adds little in short tests. Select phase transport plus observation weight1 for validation based on complete 30-step training evidence. No candidate from this entry alone was published. + +- Diagnostic source kinetic_torch.py: SHA256 `1082bbf868d579e4c101b86d5fccffc9e43ba1c45e569e1c1dcbc4a19c5ddd6a`. +- Diagnostic source kinetic_flow_encoder.py: SHA256 `e0d1e5022cc8e0595c1c9fd43dc15ba4e77c7998e6fef139c5b58e2df30dde22`. +- Diagnostic source kinetic_transport_encoder.py: SHA256 `4bbb953264d06a28cd3832509172ae68240a24ec779116a3e0377346818e26e9`. + +## E014 — native-grid eight-step training, update 2000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.757118**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 2000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755962**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.532789 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of fine_gs_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.730834**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261022; checkpoint SHA256 `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.049812 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Expanded-validation protocol: all561 non-active cases were rerun with this complete package. The24 active-matter outputs were reused from the earlier end-to-end E011 run after verifying identical active solver assets/configuration and thirty physical steps for every case, so those outputs are independent of the changed shared neural weights. The official scorer evaluates all585 trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E017 — geometric and velocity-constrained context state +- Hypothesis / changed factor: Keep the same shared neural checkpoint and all other operators from parent.json. Change only the active-matter four-frame physical-state encoder: rotate higher angular moments with observed second-moment orientation, scale by observed concentration, use twelve fixed regularized Stokes observation steps for F4, and project F4 into its necessary moment-positivity disk. Fixed observer weight1, ridge0.01 and phase confidence threshold0.1 were selected using train-only full30 diagnostics. No learned or solver parameter changes at inference. All twelve active validation cases rerun; sixty-nine identical parent outputs reused. Exact physical-prefix coverage is in physical_coverage.json. No new random parameters; inherited initialization seed20261021 and sampling seed recorded in the checkpoint manifest. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.802217**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E017b — combine context encoder and native checkpoint +- Hypothesis / changed factor: Combine the E017 fixed physical context encoder with the better shared native-grid checkpoint selected at phase4 update2000. No training, new random parameters or numerical changes. All runtime modules/configuration outside kinetic.py and the shared weights were checked identical between parents. Twelve active outputs are independent of shared weights after verifying all30 steps use the physical operator. Official scoring reuses those12 outputs and69 matching neural-parent outputs. Replayed three neural cases and two novel schemas with exact determinism. Active determinism was already checked for the identical physical encoder. Inherited initialization seed20261021, native-stage sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.803282**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E018 — finer internal advection grids +- Hypothesis / changed factor: Keep the same shared neural checkpoint and active-state encoder. Change only Burgers and Kolmogorov physical updates to a twofold internal Fourier grid in double precision, retaining unresolved state between outputs and using the same finite-cell observation map. Preserve the physical Kolmogorov forcing phase when refining its grid. Three full26 training trajectories per equation supported the change; doubling substeps, using a fourfold grid, or encoding unresolved modes through all four context frames added little. No random parameters or new training. Rerun the six affected validation cases; reuse seventy-five identical parent outputs. All six used26/26 physical updates. Exact repeat checks include Burgers, Kolmogorov, an unchanged Gray-Scott case and two novel schemas. Inherited initialization seed20261021 and sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `970b00933280c87d32c3d322412f89d56ad1969897bbdd0b83dbeccc51231cf8`. +- Validation: fixed 81-case subset; reward **0.806376**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E017 expanded active-matter validation + +Changed factor: expand from twelve to all24 active-matter validation trajectories, all start0. Twelve existing predictions were reused only after exact context equality checks; the other twelve were replayed with the same complete E017 package. Official scorer active-matter VRMSE **0.08948507**, versus 0.19604967 for the earlier physical encoder on the same24 cases. Failed cases 0. Other dataset VRMSE: not reevaluated in this experiment. Deterministic encoder; inherited neural initialization seed20261021, sampling seed 20261023. Checkpoint SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Outcome: expanded group-level validation, no comparison between different-size aggregate rewards. + +## E018a — further fixed-operator train diagnostics + +Hypotheses and changed factors: (1) interpolated velocity/strain can drive hidden moments between context frames; (2) an entropic angular-moment projection can repair the full distribution; (3) moment and velocity filters should differ; (4) unresolved Fourier modes cause Burgers/Kolmogorov error; (5) a finer grid, retained context state, time-step refinement or alternative exponential integration repairs KdV. All diagnostics use TRAIN trajectories only, with exact source indices recorded by scripts. No new random parameters or learned checkpoint. Validation VRMSE per dataset: not evaluated in these diagnostics; E018 separately evaluates the selected grid change. + + +kinetic_forced_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"cubic": false, "observe_initial": false, "params": [3, -2]} | 0.00187798 | 0.00173005 | 4 | +| {"cubic": true, "observe_initial": false, "params": [3, -2]} | 0.00187867 | 0.00173198 | 4 | +| {"cubic": true, "observe_initial": true, "params": [3, -2]} | 0.00413512 | 0.00506158 | 4 | +| {"cubic": false, "observe_initial": false, "params": [9, -3]} | 0.01194177 | 0.01268802 | 4 | +| {"cubic": true, "observe_initial": false, "params": [9, -3]} | 0.01874249 | 0.02191580 | 4 | +| {"cubic": true, "observe_initial": true, "params": [9, -3]} | 0.01873903 | 0.02190757 | 4 | +| {"cubic": false, "observe_initial": false, "params": [17, -5]} | 0.02985521 | 0.04619750 | 4 | +| {"cubic": true, "observe_initial": false, "params": [17, -5]} | 0.04395452 | 0.07376709 | 4 | +| {"cubic": true, "observe_initial": true, "params": [17, -5]} | 0.04395483 | 0.07376773 | 4 | + +kinetic_moment_projection_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187845 | 0.00173138 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043149 | 0.00043489 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00560499 | 0.01007434 | 4 | + +kinetic_moment_projection_long_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00161964 | 0.00254240 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183213 | 0.00254286 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00240461 | 0.00671084 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.48482975 | 1.27747452 | 30 | + +kinetic_unmasked_flow_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187846 | 0.00173107 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00041315 | 0.00041952 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00551630 | 0.01016449 | 4 | + +nonlinear_resolution_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963583 | 0.05441318 | 26 | +| {"double": true, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963618 | 0.05441412 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1} | 0.00572061 | 0.01058393 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 2} | 0.00572059 | 0.01058392 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01216232 | 0.02993135 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01185469 | 0.02870498 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1} | 0.00759149 | 0.02465414 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 2} | 0.00759146 | 0.02465408 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001462 | 0.00006308 | 26 | +| {"double": true, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 2} | 0.00001457 | 0.00006339 | 26 | + +nonlinear_resolution_expanded_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00301647 | 0.00326871 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037459 | 0.00050441 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037497 | 0.00050395 | 26 | +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00124680 | 0.00155330 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012381 | 0.00019031 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012380 | 0.00019021 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.08121000 | 0.15702826 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.02042678 | 0.05910929 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.01992439 | 0.05732524 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.04068160 | 0.08604752 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00718886 | 0.02328656 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00716170 | 0.02330063 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035132 | 0.00209063 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035139 | 0.00208930 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00034709 | 0.00206470 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203683 | 0.01849953 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203362 | 0.01849631 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01204391 | 0.01851219 | 26 | + +nonlinear_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 0} | 0.00572061 | 0.01058393 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 17} | 0.00037297 | 0.00050443 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 93} | 0.00012373 | 0.00019031 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 0} | 0.00759136 | 0.02465325 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 17} | 0.02068291 | 0.05987191 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 93} | 0.00716521 | 0.02321832 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 17} | 0.00035139 | 0.00208930 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 93} | 0.01203363 | 0.01849633 | 26 | + +kdv_numerics_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 0} | 0.00001493 | 0.00006064 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001457 | 0.00006350 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 0} | 0.00001457 | 0.00006347 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00035151 | 0.00209107 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 17} | 0.00019026 | 0.00094255 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00034583 | 0.00205838 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 17} | 0.00034618 | 0.00206044 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01203692 | 0.01849972 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 93} | 0.01360971 | 0.02156036 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01205018 | 0.01851917 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 93} | 0.01204985 | 0.01851889 | 26 | + +Outcomes: interpolated active forcing was harmful at moderate/high activity. Entropic moment projection modestly improved moderate activity but worsened the complete high-activity rollout (0.467750 to0.484830); not selected. Retaining full velocity modes has only a small short-rollout effect; full30 test remains ongoing. Twofold grids robustly improved Burgers and Kolmogorov across three training trajectories. Fourfold grids and more substeps added little; four-frame unresolved-state encoding added little or hurt. KdV grid/integrator variants were not consistently better. A proposed small-domain KdV test found no training cases below domain extent15 and ran no rollouts; replaced it with tests on the existing three training trajectories. Low-frequency differential fits recover nonlinear coefficients near6; fitted dissipative corrections were inconsistent, so no coefficient change was selected. + +The phase5 global-mixing trainer initially stopped before optimizer creation because channels_last was applied to a five-dimensional Fourier parameter. Applying the layout only to Conv2d modules fixed it; restart used the identical checkpoint and seed20261024, with no optimizer updates lost. Exactly one GPU trainer remains. + +Diagnostic source kinetic_forced_context.py: SHA256 `cebf38a2f07fbf2ea664c5b00ac17bd5dd1cef2a128b75a2d7e978fbc102e3a0`. + +Diagnostic source kinetic_moment_projection.py: SHA256 `fc796f88bda7f7802bab82a8e41d76dc1e08e1fc3a98fb9eed20cd45e26fc69a`. + +Diagnostic source kinetic_unmasked_flow.py: SHA256 `4a40d463091ef1dd6a2432e2ce79f0a417c61af1943369711062c6e0113a059b`. + +Diagnostic source spectral_resolution.py: SHA256 `907c567c87c46bb3adf9e11b1282d1a3e1400751d5e5c0bfdd835350d3062420`. + +Diagnostic source spectral_kdv_numerics.py: SHA256 `e3fcb35b99158c156691e324dae512d17f9fca287fb072270e1adbc6437b1ffe`. + +## E018b — distinct active flow and moment filters, negative full-rollout result + +Hypothesis: the stored active velocity/strain retain modes above the moment dealiasing cutoff, so the Stokes solve should retain those modes. Changed only the velocity/strain spectral mask, preserving the moment mask. Four complete30-step TRAIN rollouts, deterministic physical operator, no new seed or checkpoint; validation per dataset not evaluated. + +| Training parameters | Mean VRMSE | +|---|---:| +| [1, -1] | 0.00161508 | +| [3, -2] | 0.00182706 | +| [9, -3] | 0.00258145 | +| [17, -5] | 0.46823457 | + +Outcome: negligible improvement at low/moderate activity and slight worsening at high activity (0.467750 to0.468235); retained the validated operator. The direct training spectral audit found concentration/director power above the2/3 cutoff at float32 roundoff, while high-activity strain retains about0.0022 of its spectral norm there. This numerical distinction is real but does not explain the dominant rollout error. + +## E016 — shared global spectral mixing, update 1000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `3cd1d10c39cf842ae764c3a6390d8af9c784fc4cc2141213e7647256be66ba8c`. +- Validation: fixed 81-case subset; reward **0.803408**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.560153 | + +Outcome: did not improve validation reward; retained current package. + +## E016 — shared global spectral mixing, update 1000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `a0019d25f7e2d44c69f90d0ada4f0459c943fbffa3503dc585227648c836f516`. +- Validation: fixed 81-case subset; reward **0.802480**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.574692 | + +Outcome: did not improve validation reward; retained current package. + +## E020 — pressure uncertainty shrinkage +- Hypothesis / changed factor: Keep the shared neural weights and fixed physical operators. Add pressure-only output shrinkage toward each predicted spatial mean: zero through step10, linear ramp to0.75 at step30. The existing periodic-profile shrinkage remains unchanged. Select this fixed schedule from nine train rollouts for both native and global1000 checkpoints; native training mean VRMSE0.519754 to0.518445, global0.536804 to0.534698. No new random parameters or inference-time fitting; inherited initialization seed20261021 and sampling seed20261023. Rerun all nine neural validation cases, reuse seventy-two identical physical outputs. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `1b8f91e26862343f0238655ab2b43f24ad67da270779efcfa7ef02a841f95ab6`. +- Validation: fixed 81-case subset; reward **0.806391**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Outcome: improvement; published atomically. + +## E020a — pressure schedule training sweep + +Changed factor: pressure-only mixing toward its predicted spatial mean, with start0/5/10, end15/30, strength0/0.25/0.5/0.75/1. Nine training trajectories, one per cooling parameter; train membership was audited again. Native checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4 and global1000 checkpoint from phase5/eval001000_shrink were both evaluated. No random coefficients; inherited initialization20261021, native sampling20261023 and global sampling/new-parameter20261024. Validation per dataset: not evaluated in this training sweep; selected native variant validated separately in E020. + +| Model | Start | End | Strength | Training VRMSE | Pressure VRMSE | +|---|---:|---:|---:|---:|---:| +| native | 0 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 15 | 0.25 | 0.51936945 | 1.08422407 | +| native | 0 | 15 | 0.5 | 0.52051218 | 1.08879500 | +| native | 0 | 15 | 0.75 | 0.52311825 | 1.09921928 | +| native | 0 | 15 | 1.0 | 0.52708753 | 1.11509640 | +| native | 0 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 30 | 0.25 | 0.51915815 | 1.08337887 | +| native | 0 | 30 | 0.5 | 0.51916075 | 1.08338927 | +| native | 0 | 30 | 0.75 | 0.51976061 | 1.08578873 | +| native | 0 | 30 | 1.0 | 0.52095018 | 1.09054701 | +| native | 5 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 15 | 0.25 | 0.51899641 | 1.08273192 | +| native | 5 | 15 | 0.5 | 0.51948753 | 1.08469640 | +| native | 5 | 15 | 0.75 | 0.52120696 | 1.09157413 | +| native | 5 | 15 | 1.0 | 0.52409830 | 1.10313946 | +| native | 5 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 30 | 0.25 | 0.51903559 | 1.08288863 | +| native | 5 | 30 | 0.5 | 0.51874419 | 1.08172304 | +| native | 5 | 30 | 0.75 | 0.51888335 | 1.08227969 | +| native | 5 | 30 | 1.0 | 0.51945286 | 1.08455773 | +| native | 10 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 15 | 0.25 | 0.51861991 | 1.08122590 | +| native | 10 | 15 | 0.5 | 0.51848526 | 1.08068733 | +| native | 10 | 15 | 0.75 | 0.51935090 | 1.08414988 | +| native | 10 | 15 | 1.0 | 0.52118939 | 1.09150382 | +| native | 10 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 30 | 0.25 | 0.51902077 | 1.08282937 | +| native | 10 | 30 | 0.5 | 0.51858337 | 1.08107977 | +| native | 10 | 30 | 0.75 | 0.51844492 | 1.08052597 | +| native | 10 | 30 | 1.0 | 0.51860631 | 1.08117150 | +| global1000 | 0 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 15 | 0.25 | 0.53561169 | 1.15099039 | +| global1000 | 0 | 15 | 0.5 | 0.53658240 | 1.15487321 | +| global1000 | 0 | 15 | 0.75 | 0.53962210 | 1.16703201 | +| global1000 | 0 | 15 | 1.0 | 0.54454053 | 1.18670573 | +| global1000 | 0 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 30 | 0.25 | 0.53562902 | 1.15105970 | +| global1000 | 0 | 30 | 0.5 | 0.53524687 | 1.14953112 | +| global1000 | 0 | 30 | 0.75 | 0.53566622 | 1.15120852 | +| global1000 | 0 | 30 | 1.0 | 0.53687798 | 1.15605554 | +| global1000 | 5 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 15 | 0.25 | 0.53535586 | 1.14996706 | +| global1000 | 5 | 15 | 0.5 | 0.53561634 | 1.15100898 | +| global1000 | 5 | 15 | 0.75 | 0.53755790 | 1.15877521 | +| global1000 | 5 | 15 | 1.0 | 0.54106119 | 1.17278837 | +| global1000 | 5 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 30 | 0.25 | 0.53563652 | 1.15108971 | +| global1000 | 5 | 30 | 0.5 | 0.53500332 | 1.14855691 | +| global1000 | 5 | 30 | 0.75 | 0.53491749 | 1.14821358 | +| global1000 | 5 | 30 | 1.0 | 0.53538269 | 1.15007438 | +| global1000 | 10 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 15 | 0.25 | 0.53513905 | 1.14909983 | +| global1000 | 10 | 15 | 0.5 | 0.53481845 | 1.14781741 | +| global1000 | 10 | 15 | 0.75 | 0.53584572 | 1.15192652 | +| global1000 | 10 | 15 | 1.0 | 0.53814365 | 1.16111824 | +| global1000 | 10 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 30 | 0.25 | 0.53575180 | 1.15155082 | +| global1000 | 10 | 30 | 0.5 | 0.53504651 | 1.14872968 | +| global1000 | 10 | 30 | 0.75 | 0.53469765 | 1.14733423 | +| global1000 | 10 | 30 | 1.0 | 0.53471068 | 1.14738634 | + +Outcome: both models selected start10/end30/strength0.75, but the improvement was small. Earlier/stronger shrinkage is not selected. The global model remains worse than the native model after this correction. + +Global1000 checkpoint SHA256 `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. Study source SHA256 `2d9ddca63b268fbc9b52f5ac1bf800e7400ae7458fbd09f80a21cecaca84ffe9`. + +## E021 preparation — shared global-statistics residual head + +Hypothesis: predicting spatially uniform channel changes explicitly will help the pressure/global-mean dynamics that dominate radiative-layer error. Add one shared MLP head using the existing metadata embedding plus normalized means and standard deviations of each of the four context frames. It emits a uniform residual for every channel on every dataset; there are no dataset-specific specialist weights. The spatial network remains the selected native ConvNeXt body for this prototype. Added56075 parameters, total2815622, seed20261027. Output projection initialized to zero. + +Initialization check on one native TRAIN window: exact parent prediction parity (maximum difference0), finite nonzero output-layer gradient norm 0.043989721685647964. No optimizer updates, no new trained checkpoint, no validation per dataset and no publication. Parent checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4. Consider only after the longer-horizon experiment; regenerate against the actually selected parent architecture if it changes. Prototype source SHA256 `8de97bfff13028b7d33adfd16564d8b7b4f00d2ac50e67d2f051549480cefbe8`. + +## E022a — fixed hidden-state uncertainty quadrature, training diagnostics + +Hypothesis: averaging fixed hidden-state uncertainty may reduce late chaotic active-fluid error. After the four-frame state encoder, evolve center and symmetric smooth higher-moment phase perturbations with the same fixed physical parameters. Perturbations leave observedF0/F2 unchanged and enforce F4 positivity. Fixed latent-noise seed20261026, phase standard deviations0.01 and0.0001. Compare equal quadrature weights and half-center/quarter-endpoint weights. No neural training, no new learned checkpoint; validation per dataset not evaluated. These are physical-state quadrature nodes, not separately trained dataset models. + +kinetic_state_quadrature_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.01 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.01 | 0.00162322 | +| [1, -1] | half_center | 0.01 | 0.00162206 | +| [3, -2] | center | 0.01 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.01 | 0.00183122 | +| [3, -2] | half_center | 0.01 | 0.00183157 | +| [9, -3] | center | 0.01 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.01 | 0.16464736 | +| [9, -3] | half_center | 0.01 | 0.12350380 | +| [17, -5] | center | 0.01 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.01 | 0.44187382 | +| [17, -5] | half_center | 0.01 | 0.42681998 | +kinetic_state_quadrature_small_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.0001 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.0001 | 0.00162011 | +| [1, -1] | half_center | 0.0001 | 0.00161957 | +| [3, -2] | center | 0.0001 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.0001 | 0.00183261 | +| [3, -2] | half_center | 0.0001 | 0.00183262 | +| [9, -3] | center | 0.0001 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.0001 | 0.00258179 | +| [9, -3] | half_center | 0.0001 | 0.00257420 | +| [17, -5] | center | 0.0001 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.0001 | 0.45577160 | +| [17, -5] | half_center | 0.0001 | 0.45775375 | + +Outcome:0.01 is rejected because severe moderate-activity degradation outweighs the high-activity gain. The0.0001 quadrature leaves low/moderate errors nearly unchanged and reduces the hardest training mean from0.467763 to0.455772, but requires three physical integrations. Expand train tests to four additional parameter combinations before considering validation. Center includes an extra FFT roundtrip and positivity projection, so tiny roundoff differences from E017 are expected. + +Additional KdV observation-filter test on train0/17/93, powers0/0.5/1/2: removing deconvolution worsens all three; doubling it helps17 slightly but barely changes93 and hurts0. Keep nominal finite8-cell observation model. No KdV runtime change. + +## Full validation of advection_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775038**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Expanded-validation protocol: all561 non-active cases were replayed with the complete E018 package. All24 active outputs were reused from the expanded E017 validation after checking identical physical assets/configuration, exact input equality for reused quick cases, and thirty physical steps in every case. Hence those outputs are independent of the changed shared neural checkpoint. All585 validation trajectories are scored together by the official scorer. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of pressure_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775053**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Exact reuse protocol: model weights and all runtime operators except pressure regularization are identical to E018. Input metadata confirms only the nine radiative-layer cases contain a pressure field; their inputs match the already replayed E020 quick cases exactly. Reuse those nine new forecasts and576 unchanged E018 full forecasts. The official scorer evaluates all585 validation trajectories together. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E016 — shared global spectral mixing, update 2000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `96a49ac4734af31edd0aa5345f51515a899ef067e36abeb9e2347781c28fc055`. +- Validation: fixed 81-case subset; reward **0.807185**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E022b — expanded state-quadrature training study + +Changed factor: expand the fixed1e-4 hidden-state quadrature test from four to eight TRAIN parameter combinations, same latent seed20261026 and same fixed physical operator. No new learned weights. Validation per dataset not yet evaluated. + +| Additional training parameters | Quadrature | Mean VRMSE | +|---|---|---:| +| [5, -4] | center | 0.20956509 | +| [5, -4] | symmetric_mean | 0.20925570 | +| [5, -4] | half_center | 0.20933180 | +| [7, -3] | center | 0.00780154 | +| [7, -3] | symmetric_mean | 0.00781992 | +| [7, -3] | half_center | 0.00781529 | +| [11, -4] | center | 0.08376674 | +| [11, -4] | symmetric_mean | 0.08358850 | +| [11, -4] | half_center | 0.08358082 | +| [15, -4] | center | 0.26489031 | +| [15, -4] | symmetric_mean | 0.26059937 | +| [15, -4] | half_center | 0.26147711 | + +Across all eight cases, center mean 0.12997754, equal-weight quadrature 0.12788370. Outcome: select equal weights for validation; three physical integrations per active rollout. No publication from training results alone. + +## E016 — shared global spectral mixing, update 2000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `f06079b546194d4243540a3578b4df4e547efc8b64bb67210a4d5fe41648b2bd`. +- Validation: fixed 81-case subset; reward **0.806418**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515442 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775847**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match E018. Parsed per-case logs verify552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E023 — shared spatial-mean forecasting branch + +Hypothesis: remaining radiative-layer mean-pressure drift can be predicted from four-frame global statistics and metadata while the shared autoregressive operator handles spatial structure. Train one common MLP across all19 datasets; every branch will be evaluated on every inference case. No dataset-specific checkpoint selection. Random initialization seed20261028; CPU-only, two threads, batch512, 12000 AdamW updates, cosine learning rate0.0005, half windows start0 and half random, horizons1–30 capped by available train frames, group probabilities0.1/0.1/0.1/0.7. Loss is smooth absolute mean error normalized by target spatial variance; targets never enter model inputs. Uses only the10115 audited training records. Spatial moments derived with float64 accumulation. No pretrained assets. + +Changed factor: add a shared horizon-conditioned mean predictor to the fixed autoregressive spatial forecast. CPU training is concurrent with the one-GPU 16-step experiment. This is distinct from the earlier untrained E021 residual-mean prototype. Validation and checkpoint selection pending. No package change at preparation. + +## E023a — validation mean-branch checkpoint and blend comparison + +Changed factors: shared mean-branch checkpoint at updates2000/6000/12000 and fixed blend0/.25/.5/.75/1. Exact spatial-MSE decomposition on the nine radiative-layer VALID trajectories, keeping the same global2000 spatial prediction. These are validation selection results, not training. Other dataset VRMSE unchanged by the composite physical branch, not reevaluated here. All mean-model weights were randomly initialized here with seed20261028, shared across all19 datasets. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.50464012 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.25 | 0.48902082 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.5 | 0.48371851 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.75 | 0.48967327 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 1.0 | 0.50368029 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 6000 | 0.0 | 0.50464012 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.25 | 0.49150598 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.5 | 0.48589594 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.75 | 0.48943812 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 1.0 | 0.50039246 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 12000 | 0.0 | 0.50464012 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.25 | 0.48902526 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.5 | 0.48202704 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.75 | 0.48522405 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 1.0 | 0.49632276 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | + +Outcome: select update12000, blend0.5; all other tested choices are inferior. Full mean replacement is consistently worse than a partial blend. Actual float32 composite-package replay and determinism checks are required before publication; analytical score alone is not published. + +## E022 — fixed hidden-state quadrature +- Hypothesis / changed factor: Integrate three fixed hidden-angular-state quadrature nodes after the four-frame physical encoder, using one common neural model and identical fixed physical parameters. Nodes share all common computations; no separately trained specialists. Seed20261026, phase scale0.0001, equal weights, observedF0/F2 unchanged, F4 positivity preserved. Eight train trajectories supported this choice; larger perturbations were rejected. Rerun all twelve active validation cases, all30 physical steps; reuse sixty-nine matching non-active outputs. Any validated shared-neural progress during the replay is retained as documented in neural_rebase.json. No inference-time fitting. Inherited neural seeds/checkpoint in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `def27fe183c189d25508bcde6a013f2f6560b6b7d68526b782530bd4b1d7c730`. +- Validation: fixed 81-case subset; reward **0.807374**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E023 — shared mean forecast, composite replay +- Hypothesis / changed factor: Add one shared horizon-conditioned spatial-mean forecasting branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261028; two CPU threads, batch512, 12000 total updates, selected update12000, blend0.5. Inputs are four context means and variances plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in mean_branch_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809033**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809649**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `e7dd6ac71780c453f0a23e995e48d0c94ad68af36df21c2d8993fb7ba914c8a7`. +- Validation: fixed 81-case subset; reward **0.809341**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477935 | + +Outcome: did not improve validation reward; retained current package. + +## E024 — spatial-profile conditioned shared mean branch + +Hypothesis: four-frame channel means and variances omit thermal-layer geometry; coarse profiles along each axis may predict global mean drift more accurately. Changed factor: add eight-bin spatial marginal profiles for both axes, all channels and all four context frames, to the same shared horizon-conditioned MLP. Profiles are centered by context means and normalized by context standard deviations. Random initialization seed20261029; otherwise same shared19-dataset training distribution and12000-update schedule as E023. CPU-only, two threads, no extra GPU. Train data membership audited before deriving profiles; float64 averaging followed by float32 storage. Feature parity between training-cache input and inference from raw TRAIN context passed (feature_checks.json). All parameters initialized from scratch; no pretrained assets. + +Validation pending. Compare against the raw spatial forecast from the currently selected phase6 update1000, obtained by setting only the old mean blend to zero and replaying nine valid contexts. The final candidate, if selected, will replace the existing shared mean branch rather than adding a dataset-specific specialist. No publication from training loss. + +## E024a — profile-branch validation selection + +Changed factors: profile-branch checkpoint2000/6000/12000 and fixed mean blend0/.25/.5/.75/1. Same9 valid radiative-layer cases and phase6 update1000 spatial forecast; the old mean branch was disabled for this ablation replay. Other dataset VRMSE not reevaluated at this selection stage. Shared random initialization seed20261029. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.47627939 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.25 | 0.47072551 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.5 | 0.47161600 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.75 | 0.47837804 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 1.0 | 0.48981324 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 6000 | 0.0 | 0.47627939 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.25 | 0.47073488 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.5 | 0.46949443 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.75 | 0.47282608 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 1.0 | 0.48034141 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 12000 | 0.0 | 0.47627939 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.25 | 0.47050116 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.5 | 0.46970411 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.75 | 0.47404605 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 1.0 | 0.48293819 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | + +Outcome: update6000/blend0.5 selected at0.46949443, versus0.47388509 with E023 and0.47627894 without any mean adjustment. Additional training to12000 improved training loss but slightly worsened validation. Actual composite replay and exact contract checks are running before any publication. + +## E024 — profile-conditioned shared mean forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a profile-conditioned branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261029; two CPU threads, batch512, 12000 total updates, selected update6000, blend0.5. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_mean_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `8dc8554c96b10cffe81e8707063febceb81c18933dd345ac0dfaa329237377d6`. +- Validation: fixed 81-case subset; reward **0.809986**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Outcome: improvement; published atomically. + +## E025 — jointly forecast spatial means and coarse profiles + +Hypothesis: remaining broad thermal-layer structure errors require a forecast of spatial profiles, not only a corrected global mean. Changed factor: shared MLP predicts both global channel means and centered eight-bin marginal profiles on both axes. Enforce agreement between the predicted profile means and global means. Train across all19 datasets from random initialization seed20261030 with the same audited train statistics/profiles,12000 CPU-only updates, batch512, two threads; loss0.25 normalized mean error +0.75 normalized coarse-profile error, with the same smooth absolute form. No new data or pretrained assets. Every branch remains shared and evaluated on every case. + +Inference candidate will interpolate centered coarse-profile residuals with periodic or clamped edges according to metadata, and blend them into the common autoregressive forecast. Physical outputs remain authoritative. Uniform and spatial correction coefficients will be selected on validation, followed by actual package replay and exact determinism checks. Initialization shape, finite-output, zero-correction identity, and mean/profile consistency checks passed on an audited TRAIN context. Validation pending; no publication from training loss. + +## E025a — joint mean/profile validation selection + +Changed factors: joint statistic checkpoint2000/6000/12000, mean correction0/.5/1 and profile correction0/.25/.5/1. Same9 VALID radiative-layer trajectories and phase6 update1000 raw spatial forecasts. MSE decomposition includes the uniform correction, centered profile correction, and all cross terms; actual float32 replay is required next. Shared random seed20261030. Other dataset VRMSE not reevaluated in this selection stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.25 | 0.47635292 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.5 | 0.47958239 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 1.0 | 0.49471505 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.0 | 0.47005735 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.25 | 0.47010562 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.5 | 0.47337008 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 1.0 | 0.48871343 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.0 | 0.48430118 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.25 | 0.48426968 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.5 | 0.48736599 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 1.0 | 0.50220535 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.25 | 0.47518010 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.5 | 0.47612881 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 1.0 | 0.48395044 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.0 | 0.46912443 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.25 | 0.46794480 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.5 | 0.46886089 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 1.0 | 0.47675092 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.0 | 0.47781927 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.25 | 0.47659139 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.5 | 0.47738680 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 1.0 | 0.48486302 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.25 | 0.47462923 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.5 | 0.47500679 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 1.0 | 0.48171365 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.0 | 0.46724263 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.25 | 0.46555366 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.5 | 0.46594474 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 1.0 | 0.47282504 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.0 | 0.47431715 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.25 | 0.47263591 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.5 | 0.47296605 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 1.0 | 0.47954663 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | + +Outcome: select update12000, mean blend0.5, profile blend0.25, VRMSE0.46555366 versus E0240.46949440. At update2000 any profile correction worsened the best mean-only result; profile prediction benefited from further training. Larger profile blends were inferior. Actual replay and contract checks pending; no analytical-only publication. + +## E025 — joint shared mean/profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261030; two CPU threads, batch512, 12000 total updates, selected update12000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `67752d3d72ef3e1cf10485e57abe77cdbc9e99eb48ccb69ce72c8d12d6ab62bd`. +- Validation: fixed 81-case subset; reward **0.810291**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Outcome: improvement; published atomically. + +## E026 — sixteen-bin shared statistical forecast + +Hypothesis: eight-bin profiles smooth thin thermal interfaces; sixteen bins may improve the shared coarse forecast. Changed factor: double the input and output profile resolution on both axes from8 to16 bins, retaining one shared MLP across all19 datasets. Random initialization seed20261031. Same12000-update CPU-only training schedule, two threads, group sampling, loss weighting and horizon distribution as E025; the wider input/output layers increase parameter count. Train-only profile cache membership audited before reading, with float64 averaging. Source snapshot and train/inference feature-parity checks retained in profile16_forecast_v1. No pretrained assets. + +Validation pending; compare on the same9 valid thermal trajectories and the same phase6 update1000 raw spatial forecast. No package change before actual replay and exact contract checks. + +## E027 — pressure schedule after the learned profile decoder + +Hypothesis: the previous pressure uncertainty schedule may no longer be optimal after adding a shared profile forecast. Changed factors: pressure shrink strength0/.25/.5/.75/1, ramp start0/5/10/15 and fixed end30; zero-strength baseline evaluated once. Keep profile shrink0.75 and the trained mean/profile branch fixed. Use nine audited TRAIN trajectories, one per thermal cooling parameter, for schedule selection. No validation targets enter this choice and no new weights are trained. Raw shared CNN predictions replayed with both statistical corrections and regularization disabled, then fixed candidate decoders applied to the same outputs. Inherited neural sampling seed20261025, auxiliary seed20261030; base checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `56b7067dd85f4ba3aa13b34aafc87e1267cf7c14e7a898de9c50044b1f559c04`. Validation per dataset pending. + +## E026a — sixteen-bin validation selection + +Shared initialization seed20261031; same9 valid thermal trajectories and same phase6 update1000 raw spatial forecasts as E025. Changed checkpoint and fixed correction strengths. Other datasets not reevaluated at this stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.25 | 0.47600559 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.5 | 0.47978306 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 1.0 | 0.49835827 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.0 | 0.47140293 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.25 | 0.47106825 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.5 | 0.47484781 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 1.0 | 0.49358365 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.0 | 0.48622849 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.25 | 0.48578095 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.5 | 0.48932182 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 1.0 | 0.50735469 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.25 | 0.47423524 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.5 | 0.47528103 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 1.0 | 0.48625053 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.0 | 0.46620666 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.25 | 0.46415542 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.5 | 0.46526516 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 1.0 | 0.47656498 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.0 | 0.47363337 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.25 | 0.47157736 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.5 | 0.47258672 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 1.0 | 0.48345739 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.25 | 0.47413693 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.5 | 0.47468663 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 1.0 | 0.48361862 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.0 | 0.46656666 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.25 | 0.46436542 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.5 | 0.46493348 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 1.0 | 0.47413113 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.0 | 0.47745171 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.25 | 0.47524133 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.5 | 0.47569852 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 1.0 | 0.48441577 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | + +Outcome: select update6000, mean0.5/profile0.25, VRMSE0.46415542 versus eight-bin E0250.46555365. Additional training to12000 slightly worsened validation. Actual replay and contract checks pending. + +## E027a — training-only pressure schedule results + +No new learned weights or changed seed. Exact parent neural and auxiliary hashes are recorded in E027 and decoder_v1/parent.json. Nine audited TRAIN trajectories only; validation not yet evaluated. + +| Pressure strength | Ramp start | Training radiative-layer VRMSE | +|---:|---:|---:| +| 0.0 | 10 | 0.45035356 | +| 0.25 | 15 | 0.45038955 | +| 0.25 | 10 | 0.45050631 | +| 0.5 | 15 | 0.45065454 | +| 0.25 | 5 | 0.45075753 | +| 0.5 | 10 | 0.45097174 | +| 0.25 | 0 | 0.45108611 | +| 0.75 | 15 | 0.45114587 | +| 0.5 | 5 | 0.45158755 | +| 0.75 | 10 | 0.45174613 | +| 1.0 | 15 | 0.45185799 | +| 0.5 | 0 | 0.45240983 | +| 1.0 | 10 | 0.45282187 | +| 0.75 | 5 | 0.45283721 | +| 0.75 | 0 | 0.45431139 | +| 1.0 | 5 | 0.45449514 | +| 1.0 | 0 | 0.45677051 | + +Outcome: select zero additional pressure smoothing, training VRMSE0.45035356 versus0.45174613 for the inherited0.75/start10 schedule. The earlier pressure schedule helped older checkpoints but is a negative result for the current trained profile decoder. Validate this fixed choice after the sixteen-bin candidate finishes, preserving all latest shared weights. + +## E026 — sixteen-bin shared profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261031; two CPU threads, batch512, 12000 total updates, selected update6000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and sixteen-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile16_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6d988acd1229ae39ec83992e389a01e05d197ab496ab79cbae92cb124f39ad16`. +- Validation: fixed 81-case subset; reward **0.810400**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Outcome: improvement; published atomically. + +## E027 — updated pressure uncertainty schedule +- Hypothesis / changed factor: Replace the inherited pressure-smoothing schedule with train-selected strength0.0, start10, end30. Nine audited TRAIN trajectories selected this fixed schedule with the eight-bin profile decoder; validation then applies it to the latest selected shared weights, including any validated sixteen-bin branch. No new learned weights or inference-time fitting. Parent/checkpoint hashes and all train schedule results are recorded. Rerun nine neural validation cases; reuse72 identical physical outputs. Exact determinism checks before publication. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810428**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Outcome: improvement; published atomically. + +## E028 — statistical correction inside the autoregressive loop + +Hypothesis: applying mean/profile corrections only after the rollout lets internal state drift contaminate later predictions. Changed factor: after each shared neural update, feed a fixed fraction of the precomputed mean/profile correction into the next autoregressive context. Forecast statistics still depend only on the original four frames and metadata; weights remain fixed. Test fractions0/.25/.5/1 on nine audited TRAIN trajectories; the final output decoder remains unchanged. Physical steps remain authoritative and their latent states are unaffected. No new learned weights. Inherited neural seed20261025, auxiliary seed20261031; shared neural checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `3f9e62adca11d1dc83c908b9bbc7da794b5ca7902ecec1401a1519d72b6cd1d2`. Validation pending. + +## E028 — autoregressive statistical feedback +- Hypothesis / changed factor: Feed fraction0.25 of the fixed shared mean/profile correction into each next neural autoregressive state. Final output correction remains unchanged. Statistics are forecast once from the four original context frames and metadata; parameters never change at inference. Train-only fractions0/.25/.5/1 selected this fixed strength on nine audited thermal trajectories. No new learned weights. Physical latent states are unaffected. Rerun nine valid neural cases and reuse72 complete physical outputs, then exact neural, unknown-grid, smoke and diffusion checks before publication. Parent weights and inherited neural/auxiliary seeds are in model_manifest.json and parent records. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `c05aa38ecb979b8d52316d1cf0a5c0b35cb542ab5c79c66f8c73cbe300ee4d9c`. +- Validation: fixed 81-case subset; reward **0.810000**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469308 | + +Outcome: did not improve validation reward; retained current package. + +## E028a — training-only feedback strength + +Same inherited weights and seeds as E028; nine audited TRAIN trajectories only. Changed factor: fixed fraction of the existing output correction inserted into each next neural context. All physical branches retain their complete physical outputs. Validation per dataset pending. + +| Feedback fraction | Training radiative-layer VRMSE | Failed cases | +|---:|---:|---:| +| 0.0 | 0.45007325 | 0 | +| 0.25 | 0.44391940 | 0 | +| 0.5 | 0.44519505 | 0 | +| 1.0 | 0.44886765 | 0 | + +Outcome: select fraction0.25 at0.44391940 versus0.45007325 without feedback. Stronger feedback gives smaller gains; no new learned weights. Actual validation and exact contract checks running. + +## Full validation of quadrature_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.776678**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Expanded fixed state quadrature from12 to all24 active validation trajectories. Replayed the missing12, reused12 quick outputs only after exact context equality, and verified30/30 common physical steps for every active case. The552 corpus/smoke outputs are reused from global2000 full validation after exact physical asset/configuration equality checks and complete physical coverage. The nine neural outputs come from this exact package’s quick validation with matching inputs. All585 cases scored together; no held-out trajectories or inference caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E019 — longer autoregressive training, update 2000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.811223**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 2000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.811195**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.454067 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of mean_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778337**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval001000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778953**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_mean_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779290**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780527**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of decoder_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779732**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779595**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile16_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779704**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E029 — hidden-state quadrature width + +Hypothesis: the selected1e-4 hidden-state spread may understate uncertainty in chaotic active flow. Changed factor: phase widths1e-4/3e-4/1e-3, keeping three equal nodes, fixed latent seed20261026, identical physical encoder/parameters, and identical common-prefix guards. First study four TRAIN parameter combinations (zeta,alpha)=(5,-4),(9,-3),(15,-4),(17,-5), including the moderate case sensitive to excessive perturbation. Reuse the same encoded state and center node within each trajectory. Any improvement must be expanded before validation. No new learned weights. Kinetic source SHA256 `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`; parent package `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Validation pending. + +## E030 — learned coarse-statistic guidance preparation + +Hypothesis motivated by E028: inference-only feedback does not generalize, but the neural operator may learn how to use the same conditional forecast. Add a shared1x1 stem adapter from normalized future mean/profile fields into the common convolutional operator. Its528 weights initialize to zero, exactly preserving the selected parent before training. Fixed auxiliary forecasts consume only the original four context frames and metadata. Main neural parameters after addition3066627; initialization seed20261101. Parent `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Exact CPU forward parity on an audited native TRAIN context: maximum difference0.0; nonzero finite adapter gradient norm3.3356419 on a coarsened TRAIN window. Source in guided_v1/model.py. + +Outcome: untrained architecture preparation only. No optimizer steps, validation selection, package publication, new GPU trainer, or phase7 launcher yet. Current phase6 remains the only GPU trainer. The future training stage must start from the then-selected checkpoint and recheck initialization rather than treat this probe as a trained candidate. + +## E019 — longer autoregressive training, update 1000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809399**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477170 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 2000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.810422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463873 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810313**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465273 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.810189**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466865 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809097**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.481171 | + +Outcome: did not improve validation reward; retained current package. + +## E029a — quadrature-width training results + +Same fixed physical encoder and latent seed20261026. Four TRAIN parameter combinations; every node completed30 physical steps. No new learned weights. + +| Training parameters | Phase width | Active VRMSE | +|---|---:|---:| +| [5, -4] | 0.0001 | 0.20925571 | +| [5, -4] | 0.0003 | 0.20655669 | +| [5, -4] | 0.001 | 0.19077064 | +| [9, -3] | 0.0001 | 0.00258179 | +| [9, -3] | 0.0003 | 0.00410363 | +| [9, -3] | 0.001 | 0.04487926 | +| [15, -4] | 0.0001 | 0.26059940 | +| [15, -4] | 0.0003 | 0.24752074 | +| [15, -4] | 0.001 | 0.23641607 | +| [17, -5] | 0.0001 | 0.45577160 | +| [17, -5] | 0.0003 | 0.42649901 | +| [17, -5] | 0.001 | 0.42283213 | + +Four-case means: {0.0001: 0.23205212625907734, 0.0003: 0.22117002040613443, 0.001: 0.22372452542185783}. Outcome: select3e-4 for an expanded TRAIN check on (1,-1),(3,-2),(7,-3),(11,-4). The1e-3 spread harms the formerly accurate (9,-3) case and is inferior overall. Validation pending. + +## E031 — pressure from momentum balance + +Hypothesis: a fixed momentum-balance projection could recover pressure structure missed by the neural forecast. Independently implement a discrete periodic-x/free-y gradient least-squares inverse with FFT/DCT, finite-time momentum differences and conservative flux divergence. A manufactured scalar-potential check verifies the inverse to roundoff. First use true TRAIN density/velocity only as an oracle feasibility diagnostic; no future truth is supplied at inference. Per-case fitted scalars below are diagnostic only and are never deployed. Nine audited training trajectories. No new learned weights or new seed. + +| Training cooling case | Nominal oracle pressure VRMSE | Pressure correlation | Diagnostic optimal scalar | +|---|---:|---:|---:| +| data/train/turbulent_radiative_layer_tcool_0.03.hdf5 | 0.652605 | 0.753539 | 0.886568 | +| data/train/turbulent_radiative_layer_tcool_0.06.hdf5 | 0.787532 | 0.646115 | 0.683058 | +| data/train/turbulent_radiative_layer_tcool_0.10.hdf5 | 0.835901 | 0.640075 | 0.627110 | +| data/train/turbulent_radiative_layer_tcool_0.18.hdf5 | 1.095653 | 0.426235 | 0.393953 | +| data/train/turbulent_radiative_layer_tcool_0.32.hdf5 | 1.103630 | 0.476517 | 0.405224 | +| data/train/turbulent_radiative_layer_tcool_0.56.hdf5 | 1.789119 | 0.191514 | 0.107686 | +| data/train/turbulent_radiative_layer_tcool_1.00.hdf5 | 1.531407 | 0.145077 | 0.105623 | +| data/train/turbulent_radiative_layer_tcool_1.78.hdf5 | 1.578657 | 0.058558 | 0.042445 | +| data/train/turbulent_radiative_layer_tcool_3.16.hdf5 | 1.457004 | 0.046748 | 0.039365 | + +The oracle is useful mainly for the shortest cooling times and unreliable at long cooling times. Then use only the selected model’s predicted density/velocity, keeping its predicted global pressure mean, and blend the nominal projection at a fixed strength. Parent/checkpoint hashes are in momentum_pressure_v1/forecast_parent.json. + +| Projection blend | Training thermal VRMSE | +|---:|---:| +| 0.0 | 0.43522565 | +| 0.05 | 0.43501140 | +| 0.1 | 0.43541418 | +| 0.25 | 0.44012596 | +| 0.5 | 0.45764482 | +| 1.0 | 0.51557828 | + +Outcome: full-strength projection is harmful; a0.05 blend gives only a small training gain. Test that single train-selected strength on validation before deciding whether a self-contained Torch implementation is warranted. No physical projection is currently installed in the package. + +## E029b — expanded quadrature-width training study + +Changed factor: expand the selected3e-4 width from four to eight TRAIN parameter combinations. Same fixed latent seed20261026 and physical encoder; no new learned weights. Validation pending. + +| Additional training parameters | Phase width | Active VRMSE | Complete physical steps | +|---|---:|---:|---:| +| [1, -1] | 0.0001 | 0.00162011 | 30 | +| [1, -1] | 0.0003 | 0.00162016 | 30 | +| [3, -2] | 0.0001 | 0.00183261 | 30 | +| [3, -2] | 0.0003 | 0.00183292 | 30 | +| [7, -3] | 0.0001 | 0.00781992 | 30 | +| [7, -3] | 0.0003 | 0.00795999 | 30 | +| [11, -4] | 0.0001 | 0.08358850 | 30 | +| [11, -4] | 0.0003 | 0.08545388 | 30 | + +Eight-case means:1e-4=0.12788370,3e-4=0.12269338. Outcome: proceed to validation. + +## E031 — fixed momentum pressure correction +- Hypothesis / changed factor: Fixed0.05 pressure blend from forecast momentum balance, selected on nine audited TRAIN trajectories. Uses context plus predicted density/velocity, finite-time derivatives, conservative flux divergence, and a periodic-x/free-y gradient least-squares projection. Preserves predicted global pressure mean. No fitted diagnostic scalars are deployed and no new learned weights. Larger blends were harmful. A separate validation diagnostic found only a tiny gain; this replay uses the self-contained Torch implementation and official scorer, with exact neural/novel-grid/smoke/diffusion checks before publication. Seventy-two complete physical outputs are unchanged; all model components and inherited seeds/hashes remain in the manifest. No future truth, external files, SciPy, or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `cd7a9a8c2c2038f3be273f65987972072405494507695c64397ef3d3c2e2ae27`. +- Validation: fixed 81-case subset; reward **0.811224**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 4000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810991**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.456640 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `3cfd637325d7e7313d4bfbd48967518b107446b1be900f8859acaa7f67b51793`. +- Validation: fixed 81-case subset; reward **0.810338**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464955 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `8bf82697e735e6de13b39fd00292c1ce9bda307e9a904af1be6a2f4f99333878`. +- Validation: fixed 81-case subset; reward **0.810252**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466053 | + +Outcome: did not improve validation reward; retained current package. + +## E030 training launch + +Phase6 updates3000 and4000 did not beat update2000. Retired the phase6 trainer, its six data workers and watcher after all4000-update validation variants completed; all checkpoints and negative results remain available. Verified those processes exited before launching the next single-GPU trainer. Initialize528 guidance-adapter weights to zero on the latest selected package, preserving its physical operators and fixed auxiliary forecast. Sampling seed20261102, guidance initialization seed20261101; inherited neural initialization seed20261021 and auxiliary seed20261031. Sixteen-step loss capped by available trajectory length, batch6, learning rate0.00003 with cosine decay, noise0.01, group probabilities0.05/0.075/0.075/0.8, native thermal grids, planned6000 updates. Frozen auxiliary weights are copied into phase7/fixed_statistics. Exact parent/checkpoint hashes and stopped PIDs are in phase7/lineage.json. No pretrained assets. Validation pending; the selected package is unchanged. + +## Full validation of momentum_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780529**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E032 — Preserve observed angular constraints in state quadrature (preparation) + +Hypothesis: perturbing F4 changes an angular moment already constrained by the observed flow; rotating F6 and higher may preserve short-term accuracy while averaging latent uncertainty. Changed factor: first perturbed mode, holding the fixed operator and all learned weights fixed. Training study will compare the existing F4+ width3e-4 control with F6+ widths3e-4 and1e-3 on four audited TRAIN parameter regimes. Fixed latent seed20261026. No new trained weights; inherited checkpoint and physical hashes will be saved in latent_constraint_v1/source/model_manifest.json and parent.json. No validation results yet and no package change. Heavy physical study is queued after E029 expanded validation. + +## E033 — Shared coarse two-dimensional interaction forecast (preparation) + +Hypothesis: separate axis profiles cannot represent the spatial interaction component, which accounts for more than half of the remaining pressure and transverse-velocity normalized squared error in the current valid neural rollouts. Changed factor: add a shared horizon-conditioned MLP forecasting the interaction component of4×16 coarse fields. Every model branch will be evaluated on every case; complete physical prefixes remain authoritative. Train across all10115 audited TRAIN records and all19 datasets, random initialization seed20261103,12000 updates, batch512, width192, two CPU threads, group probabilities0.1/0.1/0.1/0.7. Inputs only four context fields/statistics and metadata; loss uses target full-field variance during training. Select updates2000/6000/12000 and fixed output blends0/.25/.5/1 on valid neural rollouts. No pretrained assets. Dataset VRMSE and checkpoint hashes pending. + +Training-cache/runtime feature parity passed exactly on one audited train example from each group. Zero-initialized forecasts equal last-frame coarse interactions exactly. Interpolation preserves global means and marginal profiles within float32 rounding. Sources and checks are retained under coarse_interaction_v1. No inference package change yet. + +## E029 — wider fixed state quadrature +- Hypothesis / changed factor: Increase the fixed hidden-state phase quadrature width from1e-4 to3e-4, retaining three equal nodes, the same fixed physical parameters, and latent seed20261026. Eight audited TRAIN trajectories supported the choice;1e-3 was rejected. No learned weights changed. Replay all twelve active VALID cases and verify30/30 physical steps, reuse69 unchanged outputs, and retain any newer validated neural/decoder package before scoring. Exact determinism and novel-grid checks precede publication. Inherited neural/auxiliary hashes and seeds are in the manifest. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812643**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E034 — late-checkpoint parameter averaging +- Hypothesis / changed factor: Equal parameter average of phase6 updates2000/3000/4000, a single predeclared candidate to reduce late-training optimization variation. No new optimization or data fitting; all three checkpoints were trained together across all19 datasets, initialization seed20261021 and sampling seed20261025. Accumulate in float64 and return original parameter dtype. Same shared averaged weights apply to every case. Preserve current fixed physical operators and shared auxiliary decoders. Source training checkpoint hashes and equal coefficients are in model_manifest.json and weight_average_v1/lineage.json. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact determinism and novel grids if improved. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812766**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.452162 | + +Outcome: improvement; published atomically. + +## E033a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed factors: training checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_interaction_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.25 | 0.45186873 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.5 | 0.45240534 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 1.0 | 0.45590097 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 6000 | 0.0 | 0.45216160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.25 | 0.45148496 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.5 | 0.45166160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 1.0 | 0.45455921 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 12000 | 0.0 | 0.45216160 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.25 | 0.45151015 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.5 | 0.45176523 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 1.0 | 0.45496820 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E035 — Periodic translation augmentation for the coarse branch (preparation) + +Hypothesis: a dense coarse-field predictor should see translated versions of periodic patterns during training. Changed factor: random whole-cell shifts along metadata-declared periodic axes, applied identically to all four context grids and the target. Keep architecture, initialization seed20261103, training sample sequence,12000 updates and all optimizer settings from E033. Use a separate augmentation RNG seed20261104. Train across all19 datasets on two CPU threads; no additional GPU trainer. Marginal means and variances are invariant to these cell permutations. Compare the replacement branch on the current shared neural parent with its previous coarse correction disabled, then replay the full proposed package. Validation results and checkpoint hashes pending. + +## E033 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Add one shared coarse interaction forecast branch, trained from random seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update6000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E033a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `ead6921e9fcb8924118472a881e78c2e0cf57f1a3b23d848b9ddaadfa36bc4ab`. +- Validation: fixed 81-case subset; reward **0.812821**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451485 | + +Outcome: improvement; published atomically. + +## E035a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed training factor: random coarse-cell translations on periodic axes with augmentation seed20261104, preserving initialization seed20261103 and the training sample sequence. Selection factors: checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_equivariant_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.25 | 0.45326987 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.5 | 0.45515343 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 1.0 | 0.46107193 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 6000 | 0.0 | 0.45216160 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.25 | 0.45202987 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.5 | 0.45258554 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 1.0 | 0.45570121 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 12000 | 0.0 | 0.45216160 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.25 | 0.45188915 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.5 | 0.45231514 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 1.0 | 0.45521523 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E030 — learned coarse-statistic guidance, update 1000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `240df3b587dc919b7d002c7fbcc51ab99b3510a1cedd49f0008c6e7a52b82764`. +- Validation: fixed 81-case subset; reward **0.812394**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.456839 | + +Outcome: did not improve validation reward; retained current package. + +## E035 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Replace the shared coarse interaction forecast with one trained using periodic-axis translations, augmentation seed20261104, retaining random initialization seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update12000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E035a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `7d340568043fdd29cfb0382ba3c1065e35f9172396a7f37b5a0809dedecf9f16`. +- Validation: fixed 81-case subset; reward **0.812788**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451889 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `34cea33fb726b0029c2c1734a94ceee72bd4e73bbf17e80cc9bc95a3db954165`. +- Validation: fixed 81-case subset; reward **0.811915**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.462924 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `d52a0fb2cbf3e4000004b0c088c4c1143ac393ce9c9432f1cc68270ac19d5778`. +- Validation: fixed 81-case subset; reward **0.811492**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.468370 | + +Outcome: did not improve validation reward; retained current package. + +## E036 — Cross-channel covariance features with continuation control (preparation) + +Hypothesis: spatial density/pressure correlations and other channel interactions contain dynamical information absent from separate means, variances and marginal profiles. Extend the selected E026 universal mean/profile branch with92928 zero-initialized covariance projection weights. Both a frozen-zero control and a learned-covariance variant continue from the same session-trained checkpoint, using identical samples and sampling seed20261105,6000 updates, batch512, learning rate0.0001 cosine, all19 datasets and two CPU threads each. Inherited random initialization seed20261031. Only difference between variants: whether covariance projection weights can learn. Four-frame context covariances use centered float64 accumulation, normalized by context channel scales. + +Exact parent-output parity passed on one audited TRAIN context from each group. Cache/runtime covariance differences were zero except1.06e-22 in one active-matter element. The cache contains only audited10115 TRAIN records; membership was checked before reading. Compare updates1000/3000/6000 by full neural runtime replay with the selected parent’s fixed output blends. Dataset VRMSE and checkpoint hashes pending. No inference parameter fitting and no pretrained assets. Sources, checks and logs: covariance_statistics_v1. + +## E036a — controlled statistical continuation and covariance feature comparison + +Continue the selected E026 shared auxiliary checkpoint with identical training samples, optimizer schedule and seed20261105 for6000 updates. Inherited random initialization seed20261031. Both variants add a zero-initialized covariance projection; control freezes it at zero, covariance learns it jointly. Added92928 weights. All19 datasets trained together, two CPU threads per variant, batch512, learning rate0.0001 cosine, existing mean/profile loss0.25/0.75. Four-frame spatial cross-channel covariances are normalized by context scales. Fixed parent decoder blends retained; every candidate is replayed through the full neural runtime, including learned guidance if present. No pretrained assets or inference-time fitting. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45146565 | 0.81282209 | `1c547715b94da883224d53fc40a4cc53362a65e59b78328ebb0debc00b7fca73` | +| control | 3000 | 0.45149569 | 0.81281968 | `8b0afaf1338754dccff24b1c7fd292b8b285fd665f7f003514b51403fadb99d9` | +| control | 6000 | 0.45166733 | 0.81280593 | `428e1327e4a4e51ffe0504f9ac9a2179f5c9a6e3c8883981687ab6e45fa45ea2` | +| covariance | 1000 | 0.45160459 | 0.81281096 | `d74d3af566e3c640d1209359df40ec59a9d7209fd9ed169e5186856e534807ef` | +| covariance | 3000 | 0.45153791 | 0.81281630 | `c7414bd840c80574dee726dacf7f41a28277a262da7e4663e779fe4cff50849d` | +| covariance | 6000 | 0.45171225 | 0.81280233 | `1544b181c5f74e3348bd67fd88f1f8154d5110aabf1c8ad62b721a32a60df690` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E036 — shared channel-covariance statistics +- Hypothesis / changed factor: Controlled continuation of the shared statistical branch, selected modecontrol update1000. Start from session-trained E026, add92928 zero-initialized covariance projection weights, and compare freezing them against learning them with the same seed20261105 and sample sequence. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seed20261031. Only the original four context frames supply channel covariances, means, variances and profiles. Same branch evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E036a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `254c11b3fd56624d3b695e25f7d980c48e04d399b76ef8111ce158e26fcaf483`. +- Validation: fixed 81-case subset; reward **0.812822**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Outcome: improvement; published atomically. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/README.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/README.md new file mode 100644 index 0000000000000000000000000000000000000000..91e02121f6708e7bfbaf28e604fe657b9cb1e905 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/README.md @@ -0,0 +1,33 @@ +# Shared neural and spectral surrogate + +A single shared neural operator is evaluated for every case and every autoregressive step. A deterministic spectral branch conditions on PDE metadata and uses independently implemented differential operators where their form was verified on training data. Unsupported equations use the shared neural update. There are no separately trained specialist checkpoints, no pretrained weights and no inference-time fitting. All runtime assets are included. + +Selection uses validation rollouts only. The neural model was trained from random initialization on all train pools, with sampling probabilities documented in EXPERIMENTS.md. See EXPERIMENTS.md for per-dataset results and negative experiments. Run `bash infer.sh INPUT_DIR OUTPUT_DIR`. + +Active matter also uses a fixed Fourier angular-moment Smoluchowski/Stokes operator. A recurrent context encoder inserts the observed low moments into its physical state; no parameter optimization or training occurs at inference. The Bingham initialization table contains mathematical Bessel ratios generated from scratch. Neural forecasts remain available as a fallback if a physical update is non-finite or violates positivity/moment bounds. The entry point pins the package directory through an open file descriptor so atomic publication cannot mix runtime assets. + +The active context encoder transports hidden angular moments with the supplied concentration and orientation, then applies a fixed linear Stokes observation layer and a necessary moment-positivity projection. It consumes only the four supplied frames. This is a fixed state computation; model and solver parameters do not change at inference. + +The active hidden state uses three fixed uncertainty quadrature nodes: its encoded center and symmetric small angular perturbations. Each node uses the same physical parameters and shared neural model; identical computations are shared. Outputs average the common valid physical prefix. A fixed local random generator makes the quadrature reproducible for every input, independent of case IDs and ordering. No separate specialist checkpoints or inference-time training are used. + +Buoyant smoke uses a fixed staggered-grid finite-difference fluid operator with semi-Lagrangian advection, explicit viscosity, and a pressure projection. Numerical spacing and scheme were identified using training trajectories only, with no generator code or initial-condition seeds. The implementation uses only PyTorch. + +Neural forecasts use a fixed, horizon-dependent shrinkage toward profiles averaged over periodic directions, with nonnegative density and concentration. The coefficient was selected on nine training rollouts before validation. The physical operators are unaffected. + +The shared neural operator evaluates long rectangular grids at native resolution. The checkpoint manifest and experiment ledger record training grid sizes, rollout horizon, sampling probabilities and checkpoint lineage. + +Reaction-diffusion updates retain a finer internal Fourier grid and use double-precision second-order exponential integration. The fixed observation operator maps cell averages to and from that state. Numerical choices were selected on training trajectories and tested on validation rollouts. CUDA graphs accelerate the fixed updates without changing their arithmetic. + +The Burgers and forced-vorticity updates also retain a finer internal Fourier grid in double precision. A fixed cell-averaging observation map returns the native output grid. Training-only resolution studies selected a twofold grid increase; smaller time steps and fourfold grids added little. + +The shared neural network includes a metadata-conditioned global Fourier feature layer. It was added with a zero output projection, preserving the parent model at initialization, then trained jointly across all four groups. New parameters were initialized from scratch in this session. + +A shared mean-forecast branch predicts channel spatial means from the four supplied frames, metadata, and requested horizon. It was initialized randomly and trained across all19 datasets. This same branch is evaluated on every case, including cases handled by fixed physical updates; no separately trained dataset specialists are selected. For neural forecasts, a fixed validation-selected blend adjusts the spatial means. Physical updates remain authoritative. The auxiliary checkpoint, parameter count, seed, and source hashes are recorded in model_manifest.json. + +The shared statistical branch also consumes channel profiles with 16 bins along each spatial axis, derived only from the supplied context. These profiles describe broad spatial structure that global means and variances omit. + +The same statistical branch jointly forecasts global means and coarse spatial profiles. A fixed blend corrects the autoregressive forecast’s broad spatial structure, using interpolation consistent with the supplied boundary conditions. Every case uses the same learned statistical weights. + +For density/pressure/velocity fields on a periodic-x and open-or-wall-y grid, a small fixed pressure correction uses the forecast momentum balance. A mixed-boundary gradient projection preserves the predicted global pressure mean. Its blend was selected on training trajectories and checked on validation; it uses only context and the model’s own forecast, with no parameter fitting at inference. + +A shared coarse-field branch forecasts two-dimensional spatial interactions on a4 by16 grid from the four input frames, metadata and horizon. It was initialized randomly and trained across all19 datasets. Its fixed output blend preserves the existing global means and axis profiles, and complete physical prefixes remain authoritative. The same weights are evaluated on every case; no inference-time fitting occurs. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/arena.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/arena.py new file mode 100644 index 0000000000000000000000000000000000000000..370e9af2b204b0326adc20cb9b86a8874a1c8dbe --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/arena.py @@ -0,0 +1,39 @@ +"""Fixed MAC-grid buoyant-fluid update, independently implemented in PyTorch.""" +import math +import torch +import torch.nn.functional as F + +def eligible(meta,channels): + return channels==3 and str(meta.get('pde',''))=='Navier-Stokes: buoyant smoke (PDEArena)' and list(meta.get('fields',[]))==['u','vx','vy'] and 'buoyancy_y' in meta.get('constants',{}) + +class ArenaOperator: + def __init__(self,meta,x): + self.n,self.m=x.shape[-2:];self.device=x.device;self.dtype=torch.float64;x=x.to(self.dtype) + self.hx=float(meta['domain_extent'][0])*(self.n-1)/self.n**2;self.hy=float(meta['domain_extent'][1])*(self.m-1)/self.m**2 + self.interval=float(meta['dt'])*13/14;self.substeps=max(1,math.ceil(self.interval/1.5000001));self.dt=self.interval/self.substeps;self.nu=.01;self.buoyancy=float(meta['constants']['buoyancy_y']) + self.c=x[0].clone();self.u=F.pad(x[1],(0,0,0,1));self.v=F.pad(x[2],(0,1,0,0));device=self.device;dtype=self.dtype + ix=torch.arange(self.n,device=device,dtype=dtype);iy=torch.arange(self.m,device=device,dtype=dtype) + self.eig=(2*torch.cos(math.pi*ix/self.n)-2)[:,None]/self.hx**2+(2*torch.cos(math.pi*iy/self.m)-2)[None,:]/self.hy**2;self.eig[0,0]=1 + self.cx=(2/self.n)**.5*torch.cos(math.pi/self.n*ix[:,None]*(ix[None,:]+.5));self.cy=(2/self.m)**.5*torch.cos(math.pi/self.m*iy[:,None]*(iy[None,:]+.5));self.cx[0]*=2**-.5;self.cy[0]*=2**-.5 + self.coords={} + for name,shape,offset in [('c',(self.n,self.m),(.5,.5)),('u',(self.n+1,self.m),(0.,.5)),('v',(self.n,self.m+1),(.5,0.))]: + xx,yy=torch.meshgrid((torch.arange(shape[0],device=device,dtype=dtype)+offset[0])*self.hx,(torch.arange(shape[1],device=device,dtype=dtype)+offset[1])*self.hy,indexing='ij');self.coords[name]=(xx,yy,offset) + def sample(self,f,xx,yy,offset,mode='border'): + grid=torch.stack([2*(yy/self.hy-offset[1])/(f.shape[1]-1)-1,2*(xx/self.hx-offset[0])/(f.shape[0]-1)-1],-1) + return F.grid_sample(f[None,None],grid[None],mode='bilinear',padding_mode=mode,align_corners=True)[0,0] + def advect(self,f,name): + xx,yy,off=self.coords[name];u=self.sample(self.u,xx,yy,(0,.5));v=self.sample(self.v,xx,yy,(.5,0));return self.sample(f,xx-self.dt*u,yy-self.dt*v,off,'border' if name=='c' else 'zeros') + def diffuse(self,f): + z=F.pad(f,(1,1,1,1));return f+self.dt*self.nu*((z[2:,1:-1]+z[:-2,1:-1]-2*f)/self.hx**2+(z[1:-1,2:]+z[1:-1,:-2]-2*f)/self.hy**2) + def project(self,u,v): + u[0].zero_();u[-1].zero_();v[:,0].zero_();v[:,-1].zero_();div=(u[1:]-u[:-1])/self.hx+(v[:,1:]-v[:,:-1])/self.hy + ph=(self.cx@div@self.cy.T)/self.eig;ph[0,0].zero_();p=self.cx.T@ph@self.cy + u[1:-1]-=(p[1:]-p[:-1])/self.hx;v[:,1:-1]-=(p[:,1:]-p[:,:-1])/self.hy;return u,v + @torch.no_grad() + def advance(self): + for _ in range(self.substeps): + c=self.advect(self.c,'c');u=self.advect(self.u,'u');v=self.advect(self.v,'v');v[:,1:-1]+=self.dt*self.buoyancy*.5*(c[:,1:]+c[:,:-1]);u=self.diffuse(u);v=self.diffuse(v);self.u,self.v=self.project(u,v);self.c=c + return torch.stack([self.c,self.u[:-1],self.v[:,:-1]]).float() + + def step(self,x): + return self.advance() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/coarse_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/coarse_model.py new file mode 100644 index 0000000000000000000000000000000000000000..fc93ff2927589c947bd9a2ee18251f3a3eb5d26d --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/coarse_model.py @@ -0,0 +1,86 @@ +"""A universal coarse spatial interaction forecast, learned from scratch.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + +GRID = (4, 16) + + +def interaction(x): + return x - x.mean(-1, keepdim=True) - x.mean(-2, keepdim=True) + x.mean((-2, -1), keepdim=True) + + +class CoarseForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4 * GRID[0] * GRID[1]) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, MAXC * GRID[0] * GRID[1]), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, grids=None): + b, t, c = means.shape + assert t == 4 and c <= MAXC and grids is not None + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = ((means - last[:, None]) / scale[:, None]).clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + centered = (grids - means[:, :, :, None, None]) / scale[:, None, :, None, None] + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale) * mask / 5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + F.pad(centered.clamp(-100, 100), (0, 0, 0, 0, 0, MAXC-c)).flatten(1), + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + delta = self.net(torch.cat([features, md, self.ds(ids), time], 1)).reshape(b, MAXC, *GRID)[:, :c] + future = grids[:, -1] + scale[:, :, None, None] * delta * h[:, :, None, None] + return interaction(future) + + +@torch.no_grad() +def forecast_coarse(model, context, md, ids, steps): + x = context.double() + mean = x.mean((-2, -1)).float() + variance = x.var((-2, -1), unbiased=False).float() + grid = F.adaptive_avg_pool2d(x, GRID).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + grids=grid[None].expand(steps, -1, -1, -1, -1)) + + +def resize_axis(x, size, axis, periodic): + old = x.shape[axis] + position = (torch.arange(size, device=x.device, dtype=x.dtype) + .5) * old / size - .5 + left = position.floor().long() + weight = position - left + a, b = (left.remainder(old), (left+1).remainder(old)) if periodic else (left.clamp(0, old-1), (left+1).clamp(0, old-1)) + shape = [1] * x.ndim + shape[axis] = size + return x.index_select(axis, a) * (1-weight.reshape(shape)) + x.index_select(axis, b) * weight.reshape(shape) + + +@torch.no_grad() +def correct_coarse(prediction, forecast, weight, bc): + raw = prediction.double() + current = interaction(F.adaptive_avg_pool2d(raw, GRID)) + future = torch.where(torch.isfinite(forecast), forecast.double(), current) + difference = interaction(future - current) + difference = resize_axis(difference, raw.shape[-2], -2, bc[0]) + difference = resize_axis(difference, raw.shape[-1], -1, bc[1]) + # Preserve all existing global means and axis profiles exactly in real arithmetic. + return (raw + weight * interaction(difference)).float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/config.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/config.json new file mode 100644 index 0000000000000000000000000000000000000000..053092cdeecbd3e16ac6c72a52fe9e53333e86bf --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/config.json @@ -0,0 +1,51 @@ +{ + "width": 48, + "depth": 2, + "detail_decay": 0.8, + "physics": true, + "active_tau": 10.0, + "kinetic": { + "modes": 32, + "resolution": 256, + "iterations": 6, + "cuda_graph": true, + "transport": "phase", + "observer_weight": 1.0, + "observer_ridge": 0.01, + "observer_iterations": 12, + "quadrature_scale": 0.0003, + "quadrature_seed": 20261026 + }, + "arena": true, + "neural_regularization": { + "profile_mix": 0.75, + "horizon": 30, + "pressure_mix": 0.0, + "pressure_start": 10, + "pressure_end": 30 + }, + "spectral_options": { + "gs_scale": 2, + "gs_dt": 0.25, + "gs_order": 2, + "gs_double": true, + "cuda_graph": true, + "advection_scale": 2, + "advection_double": true + }, + "native_rectangular": true, + "global_mixing": true, + "mean_forecast": { + "weight": 0.5, + "profile_weight": 0.25, + "width": 192, + "covariance_features": false + }, + "momentum_pressure": { + "weight": 0.05 + }, + "coarse_forecast": { + "width": 192, + "weight": 0.25 + } +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/infer.sh b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..bbf5ac5eb2acfbbad07b871df5e8d072b2304862 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/infer.sh @@ -0,0 +1,3 @@ +#!/bin/bash +set -euo pipefail +exec python "$(dirname "$0")/launch.py" "$@" diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/kinetic.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/kinetic.py new file mode 100644 index 0000000000000000000000000000000000000000..55627fc359e6f3c253b92bd010bad61558052e27 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/kinetic.py @@ -0,0 +1,218 @@ +"""Portable, fixed-parameter angular moment operator; PyTorch JIT accelerates inference.""" +import math +from pathlib import Path +from typing import Tuple +import numpy as np +import torch +torch._C._jit_set_texpr_fuser_enabled(False) +torch._C._jit_override_can_fuse_on_cpu(False) +torch._C._jit_override_can_fuse_on_gpu(False) +torch._C._jit_set_profiling_executor(False) + +@torch.jit.script +def flow(s:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor]: + c=s[0].real;q=s[1].real*.5;r=s[1].imag*.5;sr=s[2].real;si=s[2].imag + base0=2*alpha*q-.8*zeta*((c-sr)*q-si*r);base1=alpha*r-.4*zeta*((c+sr)*r-si*q);pc=.1*c.mean();ps=psi + for _ in range(iterations): + e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]) + sig0=base0+.4*((c+sr)*e0+si*e1);sig1=base1+.2*(si*e0+(c-sr)*e1) + sf=torch.fft.rfft2(torch.stack([sig0,sig1]));raw=(kx*ky*sf[0]+(ky*ky-kx*kx)*sf[1])*inv*inv*mask + ps=(raw+pc*ps)/(1+pc) + vel=torch.fft.irfft2(torch.stack([1j*ky*ps,-1j*kx*ps]),s=[nx,ny]);e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]);om=torch.fft.irfft2((kx*kx+ky*ky)*ps,s=[nx,ny]) + return vel,e0,e1,om,ps + +@torch.jit.script +def rhs(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + s=torch.fft.ifft2(v);vel,e0,e1,om,ps=flow(s,psi,khx,khy,inv,maskh,alpha,zeta,iterations,nx,ny) + dx=torch.fft.ifft2(1j*kx*v);dy=torch.fft.ifft2(1j*ky*v);out=-vel[0]*dx-vel[1]*dy + A=e1+zeta*s[1].imag;B=-e0-zeta*s[1].real + out=out+1j*an*.5*om*s + out[:-1]=out[:-1]+.5j*an[:-1]*(A-1j*B)*s[1:] + out[1:]=out[1:]+.5j*an[1:]*(A+1j*B)*s[:-1] + out[0]=out[0].real.to(out.dtype) + nf=torch.fft.fft2(out)*mask;nf[0,0,0].zero_() + return nf,ps + +@torch.jit.script +def integrate(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,e:torch.Tensor,e2:torch.Tensor,q:torch.Tensor,f1:torch.Tensor,f2:torch.Tensor,f3:torch.Tensor,alpha:float,zeta:float,iterations:int,substeps:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + ps=psi + for _ in range(substeps): + nv,ps=rhs(v,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + a=e2*v+q*nv;na,ps=rhs(a,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + b=e2*v+q*na;nb,ps=rhs(b,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + cc=e2*a+q*(2*nb-nv);nc,ps=rhs(cc,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + v=e*v+f1*nv+2*f2*(na+nb)+f3*nc + return v,ps + +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if n0).all();moment=((y[3]-y[6]).square()+4*y[4].square()<=1.21*y[0].square()).all() + if not (finite and positive and moment):break + out.append(y) + return torch.stack(out) if out else self.context.new_empty((0,11,*self.original_shape)) + +"""Fixed linear Stokes observation layer for physical hidden moments, no parameter fitting.""" +import torch + +class _FlowObservedOperator(_MomentOperator): + def __init__(self,meta,context,observer_weight=1.,observer_ridge=.01,observer_iterations=12,**kwargs): + super().__init__(meta,context,**kwargs);self.observer_weight=observer_weight;self.observer_ridge=observer_ridge;self.observer_iterations=observer_iterations;self.observer_stats=[] + self.r0=self.khx*self.khy*self.inv*self.maskh;self.r1=.5*(self.khy.square()-self.khx.square())*self.inv*self.maskh + def observe(self,x): + super().observe(x) + s=torch.fft.ifft2(self.v);c=x[0];f2=s[1];q=.5*f2.real;r=.5*f2.imag;e0=x[7];e1=x[8] + a=.8*self.zeta*q+.4*e0;b=.8*self.zeta*r+.4*e1 + const0=2*self.alpha*q-.8*self.zeta*c*q+.4*c*e0;const1=self.alpha*r-.4*self.zeta*c*r+.2*c*e1 + def R(z,m):return torch.fft.irfft2(torch.fft.rfft2(z)*m,s=self.shape) + def A(z):return R(a*z[0]+b*z[1],self.r0)+R(-b*z[0]+a*z[1],self.r1) + def AT(z): + u=R(z,self.r0);v=R(z,self.r1);return torch.stack([a*u-b*v,b*u+a*v]) + obs=torch.fft.irfft2((self.khx.square()+self.khy.square())*self.psi*self.maskh,s=self.shape);prior=torch.stack([s[2].real,s[2].imag]);residual=obs-R(const0,self.r0)-2*R(const1,self.r1)-A(prior) + ridge=(.25*(a.square()+b.square()).mean()*self.observer_ridge).clamp_min(1e-8) + solution=torch.zeros_like(residual);rr=residual.clone();direction=rr.clone();norm=rr.square().sum() + for _ in range(self.observer_iterations): + action=A(AT(direction))+ridge*direction;alpha=norm/(direction*action).sum().clamp_min(1e-20);solution=solution+alpha*direction;rr=rr-alpha*action;nextnorm=rr.square().sum();direction=rr+nextnorm/norm.clamp_min(1e-20)*direction;norm=nextnorm + correction=AT(solution)*self.observer_weight;estimate=prior+correction;f4=estimate[0]+1j*estimate[1] + # Necessary Toeplitz moment positivity: |F4-F2^2/c| <= c-|F2|^2/c. + center=f2.square()/c.clamp_min(1e-8);radius=(c-f2.abs().square()/c.clamp_min(1e-8)).clamp_min(0);delta=f4-center;f4=center+delta*torch.minimum(torch.ones_like(radius),radius/delta.abs().clamp_min(1e-10)) + after=residual-A(torch.stack([f4.real,f4.imag])-prior);self.observer_stats.append({'before':float(residual.square().mean().sqrt()),'after':float(after.square().mean().sqrt()),'correction_rms':float(correction.square().mean().sqrt())});s[2]=f4;self.v.copy_(torch.fft.fft2(s)) + +"""Geometric transport of latent angular moments when observed low moments arrive.""" +from pathlib import Path +import numpy as np,torch + +class KineticOperator(_FlowObservedOperator): + def __init__(self,meta,context,transport='phase',**kwargs): + super().__init__(meta,context,**kwargs);self.transport=transport;self.table=torch.as_tensor(np.load(Path(__file__).with_name('bingham_table.npy')),device=context.device) + def ratios(self,mu): + pos=(-torch.log1p(-mu.clamp(0,1-1e-6))).clamp_max(12)*(self.table.shape[1]-1)/12;i=pos.long().clamp_max(self.table.shape[1]-2);w=pos-i;rows=torch.arange(self.M+1,device=mu.device)[:,None,None];return self.table[rows,i]*(1-w)+self.table[rows,i+1]*w + def observe(self,x): + s=torch.fft.ifft2(self.v);c0=s[0].real.clamp_min(1e-8);c1=x[0].clamp_min(1e-8);old=s[1];new=(x[3]-x[6])+1j*(x[4]+x[5]);mu0=old.abs()/c0;mu1=new.abs()/c1 + confidence=(torch.minimum(mu0,mu1)/.1).clamp_max(1);angle=torch.angle(new*old.conj())*confidence + gain=(c1/c0)[None]*torch.exp(1j*(self.an*.5)*angle) + if self.transport=='bingham': + rat0=self.ratios(mu0);rat1=self.ratios(mu1);ratio=(rat1/rat0.clamp_min(1e-8)).clamp(.5,2.);gain=gain*(1+confidence[None]*(ratio-1)) + s[2:]=s[2:]*gain[2:];self.v.copy_(torch.fft.fft2(s));super().observe(x) + +"""Append to a fixed moment operator to integrate three hidden-state quadrature nodes.""" +_EncodedMomentOperator = KineticOperator + +class KineticOperator(_EncodedMomentOperator): + def __init__(self,meta,context,quadrature_scale=0.,quadrature_seed=20261026,**kwargs): + super().__init__(meta,context,**kwargs) + self.quadrature_scale=float(quadrature_scale);self.quadrature_seed=int(quadrature_seed) + self.quadrature_length=float(meta['constants']['L'])/30 + @torch.no_grad() + def rollout(self,steps,guard=False): + if self.quadrature_scale<=0:return super().rollout(steps,guard) + for j in range(1,4): + self.advance() + if guard and not torch.isfinite(self.v).all():return self.context.new_empty((0,11,*self.original_shape)) + self.observe(self.context[j]) + initial=self.v.clone();psi=self.psi.clone();state=torch.fft.ifft2(initial) + rng=torch.Generator(device=self.device);rng.manual_seed(self.quadrature_seed) + noise=torch.randn(self.shape,device=self.device,generator=rng) + filt=torch.exp(-.5*(self.khx.square()+self.khy.square())*self.quadrature_length**2) + noise=torch.fft.irfft2(torch.fft.rfft2(noise)*filt,s=self.shape) + scale=noise.std() if noise.numel()>1 else noise.new_tensor(1.) + noise=noise/scale.clamp_min(1e-12) + forecasts=[] + for sign in [0.,-1.,1.]: + s=state.clone();s[2:]*=torch.exp(1j*(self.an[2:]*.5)*noise[None]*(sign*self.quadrature_scale)) + c=s[0].real.clamp_min(1e-8);center=s[1].square()/c;radius=(c-s[1].abs().square()/c).clamp_min(0) + delta=s[2]-center;s[2]=center+delta*torch.minimum(torch.ones_like(radius),radius/delta.abs().clamp_min(1e-10)) + self.v.copy_(torch.fft.fft2(s));self.psi.copy_(psi);out=[] + for _ in range(steps): + self.advance();y=self.decode() + if guard: + finite=torch.isfinite(y).all();positive=(y[0]>0).all();moment=((y[3]-y[6]).square()+4*y[4].square()<=1.21*y[0].square()).all() + if not (finite and positive and moment):break + out.append(y) + forecasts.append(torch.stack(out) if out else self.context.new_empty((0,11,*self.original_shape))) + used=min(len(y) for y in forecasts) + return (forecasts[0][:used]+forecasts[1][:used]+forecasts[2][:used])/3 diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/make_bingham_table.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/make_bingham_table.py new file mode 100644 index 0000000000000000000000000000000000000000..3cba92fb5f0503306c910a27c10b6e42c7099aa6 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/make_bingham_table.py @@ -0,0 +1,9 @@ +import numpy as np +from scipy.special import ive +from pathlib import Path +s=np.linspace(0,12,8193);mu=-np.expm1(-s);r=mu[1:];k=np.where(r<.53,2*r+r**3+5*r**5/6,np.where(r<.85,-.4+1.39*r+.43/(1-r),1/(r**3-4*r**2+3*r))) +for _ in range(6): + z=ive(1,k)/ive(0,k);k=np.maximum(k+(r-z)/np.maximum(1-z*z-z/k,1e-14),1e-10) +t=np.zeros((65,len(s)),np.float32);t[0]=1 +for j in range(1,65):t[j,1:]=ive(j,k)/ive(0,k) +np.save(Path(__file__).with_name('bingham_table.npy'),t) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/mean_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/mean_model.py new file mode 100644 index 0000000000000000000000000000000000000000..264dc34336192462ea2ddb804fa45875a4129b01 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/mean_model.py @@ -0,0 +1,107 @@ +"""Shared context-statistic forecast; all parameters learned from random initialization.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + + +class MeanForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.covariates = nn.Linear(4*MAXC*MAXC,width,bias=False) + nn.init.zeros_(self.covariates.weight) + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4*2*16) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), nn.Linear(width, MAXC*33), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, profiles=None, covariances=None): + # One independently specified horizon per row. All supplied statistics + # describe only the four context frames; no inference-time fitting. + b, t, c = means.shape + assert t == 4 and c <= MAXC + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = (means - last[:, None]) / scale[:, None] + relative = relative.clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + assert profiles is not None + last_profiles=profiles[:,-1] + profiles=(profiles-means[:,:,:,None,None])/scale[:,None,:,None,None] + profiles=F.pad(profiles.clamp(-100,100),(0,0,0,0,0,MAXC-c)).flatten(1) + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale)*mask/5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + profiles, + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + hidden=self.net[0](torch.cat([features,md,self.ds(ids),time],1)) + if covariances is not None: + covariance=covariances/scale[:,None,:,None]/scale[:,None,None,:] + covariance=F.pad(covariance.clamp(-100,100),(0,MAXC-c,0,MAXC-c)).flatten(1) + hidden=hidden+self.covariates(covariance) + delta=self.net[1:](hidden).reshape(b,MAXC,33)[:,:c] + future_mean=last+scale*delta[:,:,0]*h + future_profiles=last_profiles+scale[:,:,None,None]*delta[:,:,1:].reshape(b,c,2,16)*h[:,:,None,None] + future_profiles=future_profiles-future_profiles.mean(-1,keepdim=True)+future_mean[:,:,None,None] + return future_mean,future_profiles + + +@torch.no_grad() +def forecast_statistics(model, context, md, ids, steps): + mean = context.double().mean((-2, -1)).float() + variance = context.double().var((-2, -1), unbiased=False).float() + profx=F.adaptive_avg_pool2d(context.double(),(16,1)).squeeze(-1) + profy=F.adaptive_avg_pool2d(context.double(),(1,16)).squeeze(-2) + profiles=torch.stack([profx,profy],-2).float() + flattened=context.double().flatten(-2) + centered=flattened-flattened.mean(-1,keepdim=True) + covariance=(centered@centered.transpose(-2,-1)/centered.shape[-1]).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + profiles=profiles[None].expand(steps,-1,-1,-1,-1), + covariances=covariance[None].expand(steps,-1,-1,-1)) + + +@torch.no_grad() +def forecast_means(model,context,md,ids,steps): + return forecast_statistics(model,context,md,ids,steps)[0] + +def interpolate_profile(profile,size,periodic): + position=(torch.arange(size,device=profile.device,dtype=profile.dtype)+.5)*16/size-.5 + left=position.floor().long();weight=position-left + if periodic: + i=left.remainder(16);j=(left+1).remainder(16) + else: + i=left.clamp(0,15);j=(left+1).clamp(0,15) + output=profile[...,i]*(1-weight)+profile[...,j]*weight + return output-output.mean(-1,keepdim=True) + +@torch.no_grad() +def correct_statistics(prediction,means,profiles,mean_weight,profile_weight,bc): + raw=prediction.double();center=raw.mean((-2,-1)) + means=torch.where(torch.isfinite(means),means.double(),center) + output=raw+mean_weight*(means-center)[:,:,None,None] + if profile_weight: + current_x=F.adaptive_avg_pool2d(raw,(16,1)).squeeze(-1) + current_y=F.adaptive_avg_pool2d(raw,(1,16)).squeeze(-2) + current=torch.stack([current_x,current_y],-2) + forecast=torch.where(torch.isfinite(profiles),profiles.double(),current) + difference=(forecast-forecast.mean(-1,keepdim=True))-(current-current.mean(-1,keepdim=True)) + dx=interpolate_profile(difference[:,:,0],raw.shape[-2],bc[0]) + dy=interpolate_profile(difference[:,:,1],raw.shape[-1],bc[1]) + output=output+profile_weight*(dx[:,:,:,None]+dy[:,:,None,:]) + return output.float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/model.py new file mode 100644 index 0000000000000000000000000000000000000000..9d5ab04fb3aab526caad19d46c1fe4b6aef27387 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/model.py @@ -0,0 +1,97 @@ +"""Shared metadata-conditioned convolutional operator, trained from scratch.""" +import math +import numpy as np +import torch +from torch import nn +import torch.nn.functional as F + +DATASETS=['ape2d/burgers','ape2d/decay_turb','ape2d/diff','ape2d/fisher','ape2d/gs_alpha','ape2d/gs_beta','ape2d/gs_delta','ape2d/gs_epsilon','ape2d/gs_gamma','ape2d/gs_iota','ape2d/gs_kappa','ape2d/gs_theta','ape2d/kdv','ape2d/kolm_flow','ape2d/ks','ape2d/sh','pdearena/navier_stokes_2d','well/active_matter','well/turbulent_radiative_layer_2D'] +CONSTANTS=['Viscosity','Viscosity X','Viscosity Y','Diffusivity','Reactivity','Feed Rate','Kill Rate','Critical Number','Domain Extent','tcool','L','zeta','alpha','buoyancy_y'] +MAXC=11 + +def metadata(meta,dataset): + vals=[math.log(max(float(meta.get('dt',1)),1e-12))/5] + vals += [math.log(max(float(v),1e-12))/5 for v in meta.get('domain_extent',[1,1])] + bc=meta.get('boundary_conditions',['periodic']*4) + vals += [float(b=='periodic') for b in bc]+[float(b=='wall') for b in bc] + const=meta.get('constants',{}) + for k in CONSTANTS: + v=float(const.get(k,0)); vals += [math.copysign(math.log1p(abs(v)*100),v)/8,float(k in const)] + return np.array(vals,np.float32),DATASETS.index(dataset)+1 if dataset in DATASETS else 0,(bc[0]=='periodic',bc[2]=='periodic') + +METADIM=3+8+2*len(CONSTANTS) + +def pad(x,n,bc): + x=F.pad(x,(n,n,0,0),mode='circular' if bc[1] else 'replicate') + return F.pad(x,(0,0,n,n),mode='circular' if bc[0] else 'replicate') + +class Block(nn.Module): + def __init__(self,c,cond): + super().__init__(); self.dw=nn.Conv2d(c,c,5,groups=c); self.norm=nn.LayerNorm(c); self.fc1=nn.Linear(c,2*c);self.fc2=nn.Linear(2*c,c) + self.film=nn.Linear(cond,2*c); nn.init.zeros_(self.film.weight);nn.init.zeros_(self.film.bias) + self.scale=nn.Parameter(torch.ones(c)*0.1) + def forward(self,x,e,bc): + h=self.dw(pad(x,2,bc)).permute(0,2,3,1);h=self.norm(h) + g,b=self.film(e).chunk(2,-1);h=h*(1+g[:,None,None,:])+b[:,None,None,:] + h=self.fc2(F.gelu(self.fc1(h)))*self.scale + return x+h.permute(0,3,1,2) + +class GlobalMix(nn.Module): + def __init__(self,width,cond=192,latent=24,modes_x=32,modes_y=96): + super().__init__();self.latent=latent;self.mx=modes_x;self.my=modes_y + self.input=nn.Conv2d(width,latent,1);self.output=nn.Conv2d(latent,width,1) + self.weight=nn.Parameter(torch.randn(2,latent,modes_x,modes_y,2)*0.5) + self.condition=nn.Linear(cond,latent*2);nn.init.zeros_(self.condition.weight);nn.init.zeros_(self.condition.bias) + nn.init.zeros_(self.output.weight);nn.init.zeros_(self.output.bias) + def forward(self,x,e,bc): + h,w=x.shape[-2:];q=self.input(x) + # Even extensions impose the same zero-normal-gradient convention as replicated padding. + if not bc[0]:q=torch.cat([q,q.flip(-2)],-2) + if not bc[1]:q=torch.cat([q,q.flip(-1)],-1) + xf=torch.fft.rfft2(q.float(),norm='ortho');yf=torch.zeros_like(xf) + nx=min(self.mx,q.shape[-2]//2);ny=min(self.my,xf.shape[-1]) + weight=torch.view_as_complex(self.weight.contiguous());gate=torch.view_as_complex(self.condition(e).float().reshape(len(e),self.latent,2).contiguous())[:, :, None, None] + yf[:,:,:nx,:ny]=xf[:,:,:nx,:ny]*weight[0,:,:nx,:ny]*(1+gate) + yf[:,:,-nx:,:ny]=xf[:,:,-nx:,:ny]*weight[1,:,-nx:,:ny]*(1+gate) + z=torch.fft.irfft2(yf,s=q.shape[-2:],norm='ortho')[:,:,:h,:w] + return self.output(F.gelu(z)) + +class Operator(nn.Module): + def __init__(self,width=48,depth=2): + super().__init__(); self.width=width;self.depth=depth;d=192;ws=[width,width*2,width*4,width*6] + self.ds=nn.Embedding(len(DATASETS)+1,32) + self.meta=nn.Sequential(nn.Linear(METADIM+32+MAXC*3,d),nn.SiLU(),nn.Linear(d,d),nn.SiLU()) + self.stem=nn.Conv2d(4*MAXC+2,ws[0],3);self.global_mix=GlobalMix(ws[0],d) + self.enc=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws]) + self.down=nn.ModuleList([nn.Conv2d(ws[i],ws[i+1],2,stride=2) for i in range(3)]) + self.up=nn.ModuleList([nn.Conv2d(ws[i+1]+ws[i],ws[i],1) for i in range(3)]) + self.dec=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws[:-1]]) + self.head=nn.Conv2d(ws[0],MAXC,1);nn.init.zeros_(self.head.weight);nn.init.zeros_(self.head.bias) + def forward(self,x,md,ids,bc=(True,True)): + # Reversible feature normalization is computed from supplied states, with fixed learned parameters. + b,t,c,h,w=x.shape + mean=x[:,-1].mean((-2,-1),keepdim=True) + std=x.std((1,3,4),unbiased=False).clamp_min(1e-3)[:,:,None,None] + z=(x-mean[:,None])/std[:,None] + mask=torch.ones(b,c,device=x.device) + sm=torch.cat([torch.asinh(mean.flatten(1)),torch.log(std.flatten(1))/5,mask],1) + if c0 for v in [dt,lx,ly]):return prediction + state=torch.cat([context,prediction]).double();density,pressure,vx,vy=[fields.index(name) for name in names] + rho=state[:,density];u=state[:,vx];v=state[:,vy];nx,ny=rho.shape[-2:];dx=lx/nx;dy=ly/ny + ddx=lambda value:(torch.roll(value,-1,-2)-torch.roll(value,1,-2))/(2*dx) + force_x=-gradient(rho*u,dt,0)-ddx(rho*u*u)-gradient(rho*u*v,dy,-1) + force_y=-gradient(rho*v,dt,0)-ddx(rho*u*v)-gradient(rho*v*v,dy,-1) + faces_x=.5*(force_x+torch.roll(force_x,-1,-2));faces_y=.5*(force_y[...,:-1]+force_y[...,1:]) + estimate=project_faces(faces_x,faces_y,dx,dy)[len(context):] + original=prediction[:,pressure].double();center=original.mean((-2,-1),keepdim=True) + updated=(1-weight)*original+weight*(estimate+center) + result=prediction.clone();result[:,pressure]=torch.where(torch.isfinite(updated),updated,original).float() + return result diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/numerics.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/numerics.py new file mode 100644 index 0000000000000000000000000000000000000000..3c45e36262df2d382b1be3a53691ec656a37ca9a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S084/numerics.py @@ -0,0 +1,21 @@ +"""Portable periodic Fourier resampling with even-Nyquist handling.""" +import torch +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if nMAXC: + return np.concatenate([predict(model,x[:,i:i+MAXC].cpu().numpy(),meta,ds,steps,device,detail_decay,physics,active_tau,kinetic_cfg,arena,neural_regularization,native_rectangular,spectral_options) for i in range(0,c,MAXC)],1) + factor=2 if min(h,w)>=192 or min(h,w)>=128 and max(h,w)/min(h,w)>2 and not native_rectangular else 1 + z=F.avg_pool2d(x,factor) if factor>1 else x + md,di,bc=metadata(meta,ds);md=torch.tensor(md,device=device)[None];ids=torch.tensor([di],device=device) + future_means,future_profiles=forecast_statistics(model.mean_forecast,x,md,ids,steps) if hasattr(model,'mean_forecast') else (None,None) + future_coarse=forecast_coarse(model.coarse_forecast,x,md,ids,steps) if hasattr(model,'coarse_forecast') else None + kinetic_steps=0 + detail=x[-1]-F.interpolate(z[-1,None],size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else None + ctx=z[None];out=[] + solver=SpectralUpdate(meta,(h,w),device,**(spectral_options or {})) if physics and eligible(meta,c) else None + if arena and arena_eligible(meta,c) and min(h,w)>1:solver=ArenaOperator(meta,x[-1]) + fullstate=x[-1];physical_steps=0;physical_name=type(solver).__name__ if solver is not None else None + for k in range(steps): + with torch.autocast(device_type=device,dtype=torch.bfloat16,enabled=device=='cuda'):y=model(ctx,md,ids,bc) + if not torch.isfinite(y).all():y=ctx[:,-1].clone() + if solver is not None: + candidate=solver.step(fullstate) + if torch.isfinite(candidate).all(): + physical_steps+=1 + fullstate=candidate;y=(F.avg_pool2d(candidate,factor) if factor>1 else candidate)[None] + else:solver=None + ctx=torch.cat([ctx[:,1:],y[:,None]],1) + yh=F.interpolate(y,size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else y[0] + if detail is not None:yh=yh+detail*(detail_decay**(k+1)) + if solver is not None:yh=fullstate + yh=torch.where(torch.isfinite(yh),yh,x[-1]) + out.append(yh.cpu().numpy()) + pred=np.stack(out).astype(np.float32) + if physical_name:print(f'physical steps: {physical_name} {physical_steps}/{steps}',flush=True) + if neural_regularization is not None and solver is None: + raw=torch.as_tensor(pred,device=device);pred=regularize(raw,meta,**neural_regularization).cpu().numpy() + if future_means is not None and solver is None: + pred=correct_statistics(torch.as_tensor(pred,device=device),future_means,future_profiles,model.mean_weight,model.profile_weight,bc).cpu().numpy() + if active_tau is not None: + raw=torch.as_tensor(pred,device=device);projected=project_active(x,raw,meta,active_tau) + pred=torch.where(torch.isfinite(projected),projected,raw).cpu().numpy() + if kinetic_cfg and str(meta.get('pde',''))=='active_matter' and c==11 and list(meta.get('fields',[]))==['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3'] and all(k in meta.get('constants',{}) for k in ['L','alpha','zeta']): + from kinetic import KineticOperator + cfg=dict(kinetic_cfg) + try: + op=KineticOperator(meta,x,**cfg) + except RuntimeError: + if not cfg.get('cuda_graph',False):raise + cfg['cuda_graph']=False;op=KineticOperator(meta,x,**cfg) + physical=op.rollout(steps,guard=True);used=len(physical);kinetic_steps=used + if used:pred[:used]=physical.cpu().numpy() + print(f'kinetic frames: {used}/{steps}',flush=True) + if getattr(model,'momentum_pressure_weight',0): + pred=pressure_projection(x,torch.as_tensor(pred,device=device),meta,model.momentum_pressure_weight).cpu().numpy() + if future_coarse is not None and solver is None: + prefix=max(physical_steps,kinetic_steps) + if prefix1;self.initialized=False + self.double=gs_double if meta['pde']=='Gray Scott' else advection_double if advective else False + self.real_dtype=torch.float64 if self.double else torch.float32 + self.complex_dtype=torch.complex128 if self.double else torch.complex64 + if self.fine:shape=tuple(int(scale*s) for s in shape) + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.to(self.real_dtype);self.ky=ky.to(self.real_dtype);k2=kx*kx+ky*ky;self.k2=k2.to(self.real_dtype) + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if self.fine: + ox,oy=self.original_shape;fx=torch.fft.fftfreq(ox,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(oy,device=device,dtype=torch.float64)[None,:];self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=gs_dt;self.order=gs_order + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=torch.arange(ny,device=device,dtype=self.real_dtype)*ly/ny+7/16*ly/self.original_shape[1] if self.fine else (torch.arange(ny,device=device)+7/16)*ly/ny + self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(self.complex_dtype if v.is_complex() else self.real_dtype) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + self.graph=None + if cuda_graph and torch.device(device).type=='cuda' and name!='Diffusion': + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + self.buffer=torch.zeros((nc,nx,ny//2+1),device=device,dtype=self.complex_dtype) + try: + def body():self.buffer.copy_(self.substep(self.buffer)) + stream=torch.cuda.Stream();stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3):body() + torch.cuda.current_stream().wait_stream(stream);torch.cuda.synchronize();graph=torch.cuda.CUDAGraph() + with torch.cuda.graph(graph):body() + self.graph=graph + except RuntimeError: + self.graph=None + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + if self.fine: + if not self.initialized: + latent=torch.fft.irfft2(torch.fft.rfft2(u.to(self.real_dtype))/self.observation,s=self.original_shape);v=self.fft(resample(latent,self.shape));self.initialized=True + else:v=self.internal + else:v=self.fft(u.to(self.real_dtype))/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + if self.graph is not None: + self.buffer.copy_(v) + for _ in range(self.substeps):self.graph.replay() + v=self.buffer + else: + for _ in range(self.substeps):v=self.substep(v) + if self.fine: + self.internal=v + low=resample(self.ifft(v),self.original_shape) + return torch.fft.irfft2(torch.fft.rfft2(low)*self.observation,s=self.original_shape).float() + return self.ifft(v*self.observation).float() + def substep(self,v): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;return stage+self.etd2*(self.nonlinear(stage)-nv) + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + return self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/EXPERIMENTS.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/EXPERIMENTS.md new file mode 100644 index 0000000000000000000000000000000000000000..1351fb6cc82171bc6f58e7da7daffe35fdcff8d4 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/EXPERIMENTS.md @@ -0,0 +1,3720 @@ +# Experiment ledger + +All development uses train/valid pools of `/workspace/data/split.json`. No held-out targets or case definitions are used. One H100 GPU is available. No pretrained assets are used. + +## E000 — package initialization +- Hypothesis: a spatial-mean prior provides a finite, portable starting package while training begins. +- Changed factor: parameter-free spatial-mean prediction. +- Seed: not applicable (deterministic). +- Validation: fixed 81-case subset (seed 71, all start at frame 0), reward 0.313512; 0 failed cases. +- Checkpoint hash: no checkpoint; predictor SHA256 307c11c9f4f3f186b96a34597822e1e40e95d47d6159b3de3fe0934ebae13d7d. +- Outcome: initial package published; not a trained candidate or a reproduced baseline run. + +## E001 — coupled, metadata-conditioned multiscale model +- Hypothesis: joint channel modeling, physical metadata, scale-aware normalization and periodic boundaries improve cross-equation prediction relative to the documented starter weaknesses. +- Changed factors: new shared convolutional architecture; all 19 dataset identifiers represented in one parameter set; group-balanced training. +- Initialization seed: 20261021. +- Validation: pending. +- Checkpoint hash: pending. +- Outcome: preparing train-only reduced-resolution cache and validation data. + +### E000 validation VRMSE by dataset + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 1.000634 | +| ape2d/decay_turb | 0.999991 | +| ape2d/diff | 0.999973 | +| ape2d/fisher | 1.581309 | +| ape2d/gs_alpha | 1.005787 | +| ape2d/gs_beta | 1.023021 | +| ape2d/gs_delta | 1.591578 | +| ape2d/gs_epsilon | 1.004624 | +| ape2d/gs_gamma | 1.000439 | +| ape2d/gs_iota | 1.633860 | +| ape2d/gs_kappa | 1.079744 | +| ape2d/gs_theta | 1.574410 | +| ape2d/kdv | 1.000191 | +| ape2d/kolm_flow | 0.999992 | +| ape2d/ks | 0.999992 | +| ape2d/sh | 1.041517 | +| pdearena/navier_stokes_2d | 1.000318 | +| well/active_matter | 0.992894 | +| well/turbulent_radiative_layer_2D | 1.502768 | + +## E001-smoke — 100 updates +- Hypothesis / changed factor: Short pipeline check of new coupled operator with 40 base channels, 12 samples per batch, one-step objective, float16 cache; no baseline retraining. Float16 cache will be replaced to reduce quantization error. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Validation: fixed 81-case subset; reward **0.322278**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.779417 | +| ape2d/decay_turb | 0.919250 | +| ape2d/diff | 0.573262 | +| ape2d/fisher | 1.128033 | +| ape2d/gs_alpha | 0.999278 | +| ape2d/gs_beta | 1.093050 | +| ape2d/gs_delta | 1.567459 | +| ape2d/gs_epsilon | 0.717784 | +| ape2d/gs_gamma | 0.761942 | +| ape2d/gs_iota | 1.560647 | +| ape2d/gs_kappa | 0.956638 | +| ape2d/gs_theta | 1.511911 | +| ape2d/kdv | 0.764748 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.939004 | +| ape2d/sh | 0.919993 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — intermediate 2,000 updates +- Hypothesis / changed factor: Intermediate checkpoint of the same 8,000-update one-step run; all other factors unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `cc58e45b55743bae890d3c3f212be97b831186431591d2501186f69378ce6f37`. +- Validation: fixed 81-case subset; reward **0.263620**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.817172 | +| ape2d/decay_turb | 2.389786 | +| ape2d/diff | 0.837767 | +| ape2d/fisher | 1.272807 | +| ape2d/gs_alpha | 2.153748 | +| ape2d/gs_beta | 8.179170 | +| ape2d/gs_delta | 1.369924 | +| ape2d/gs_epsilon | 1.198089 | +| ape2d/gs_gamma | 1.805734 | +| ape2d/gs_iota | 1.964417 | +| ape2d/gs_kappa | 0.904690 | +| ape2d/gs_theta | 1.609477 | +| ape2d/kdv | 1.151686 | +| ape2d/kolm_flow | 2.257709 | +| ape2d/ks | 2.152634 | +| ape2d/sh | 0.836066 | +| pdearena/navier_stokes_2d | 1.003965 | +| well/active_matter | 2.135917 | +| well/turbulent_radiative_layer_2D | 1.657301 | + +Outcome: did not improve validation reward; retained current package. + +## E003 — metadata-conditioned spectral branch +- Hypothesis / changed factor: Add independently implemented ETDRK4 differential operators to the shared model. All neural weights unchanged from E001-smoke; neural branch still applied every step on every case. Equation parameters come from metadata; solver forms selected using train-only diagnostics. Kolmogorov remains with the neural update pending forcing/drag diagnosis. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Runtime code SHA256: `9e848b7d80d919c0b32a15d07bf4643a1b56b5235004827d0950660afa2e4952`. +- Validation: fixed 81-case subset; reward **0.401067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — one-step operator, 8,000 updates +- Hypothesis / changed factor: Joint channel ConvNeXt U-Net, metadata conditioning, group-balanced batches, float32 reduced-grid cache, one-step log variance-normalized MSE. Random initialization; 48 base channels and 16 samples per batch. Development inference wrapper relative-path error was corrected before scoring. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `9a645f05c3a17872093f7f461376b1fedd663ceefbbd879ef8f1cd4b8d8e58fb`. +- Validation: fixed 81-case subset; reward **0.350858**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: did not improve validation reward; retained current package. + +## E004 — trained neural branch and forced turbulence operator +- Hypothesis / changed factor: Replace smoke neural weights with E001 step 8,000, still one shared network across all cases. Add Kolmogorov forcing -4 cos(4y) and drag 0.2, identified only in training diagnostics. Both effects are recorded together; full group scores isolate transfer changes from corpus changes. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `d8dd1c803a436415851c119241900367c593b0acfb904f6f5b3f3b17b7444478`. +- Validation: fixed 81-case subset; reward **0.437086**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## E005A — physical output constraints and horizon decay (validation ablations) + +Hypothesis: exact tensor identities and reduced confidence at longer horizons stabilize active matter. All experiments reuse E001 8,000-update predictions; seed 20261021. No new weights or training data. Checkpoint SHA256: df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d. + +The diffusion coefficient 0.05, trace(D)=concentration and E=sym(gradient velocity) were checked on training pools. Horizon decay candidates 10, 20 and 40 steps are hyperparameters selected on validation, not parameters fitted to validation targets. These ablations are not published until integrated into a portable package and replayed. + +### active_diffusion: Analytic concentration diffusion only + +Reward 0.351796; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.620833 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_constraints: Concentration diffusion and algebraic tensor projection + +Reward 0.352944; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.570211 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_strain: Recompute strain from predicted velocity + +Reward 0.338342; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 2.350641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: negative; strain derivatives amplify velocity errors. + +### active_projection: Weighted Fourier projection of velocity and strain + +Reward 0.352917; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.571374 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_10: Weighted projection and exponential shrinkage, tau=10 + +Reward 0.375912; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832421 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_20: Weighted projection and exponential shrinkage, tau=20 + +Reward 0.372622; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.911641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_40: Weighted projection and exponential shrinkage, tau=40 + +Reward 0.365053; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.123714 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +## Train-only solver diagnostics (no validation selection) + +Initial Navier–Stokes sign and step-size hypotheses failed (non-finite or high training error); smaller steps and the opposite vorticity convention succeeded. KS requires separate fourth derivatives, not the squared Laplacian. Kolmogorov forcing and drag were identified using training trajectories only. Doubling the latent solver grid from 256 to 512 barely changed Gray–Scott/KS training errors and was rejected as a general fix. Grid averaging, integrator order and timestep are under investigation. These are diagnostic tests, not held-out experiments; validation VRMSE is not applicable. No checkpoint was created. Source, configurations and train results are retained under research/physics_diagnostics. + +## E005B — integrated physical projection and horizon regularization +- Hypothesis / changed factor: E004 model plus concentration diffusion, weighted Fourier projection of velocity and strain, tensor identities and exponential relaxation toward context means for active-matter fluctuations, tau=10 steps. All weights unchanged; tau selected from fixed validation ablations. The neural model is still applied to every case. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `83bec07505f714951064c32e8cb4cc20720921d9b62a1e4c1ae3bba353048798`. +- Validation: fixed 81-case subset; reward **0.462140**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Full validation of physics_smoke + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.400925**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.060324 | +| ape2d/decay_turb | 0.000387 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.002465 | +| ape2d/gs_alpha | 0.022938 | +| ape2d/gs_beta | 0.012380 | +| ape2d/gs_delta | 0.146653 | +| ape2d/gs_epsilon | 0.026477 | +| ape2d/gs_gamma | 0.222599 | +| ape2d/gs_iota | 0.036049 | +| ape2d/gs_kappa | 0.109056 | +| ape2d/gs_theta | 0.135853 | +| ape2d/kdv | 0.025460 | +| ape2d/kolm_flow | 0.979667 | +| ape2d/ks | 0.080605 | +| ape2d/sh | 0.002222 | +| pdearena/navier_stokes_2d | 0.923463 | +| well/active_matter | 1.120708 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E002 — intermediate 2,000 multi-step updates +- Hypothesis / changed factor: E005B inference architecture and physical operators; replace only shared neural weights with the 2,000-update checkpoint of four-step fine-tuning with 0.01 input noise. +- Seed: 20261021. +- Checkpoint SHA256: `d720a4dbeee0c16a125aca0cee22ee060e1e80c6b863efbd169fd69b5f2c9efc`. +- Runtime code SHA256: `f4b3be2c3e227fcfa3e8cb9774970780928d6635210a425fa2c6b07e8f08e8a5`. +- Validation: fixed 81-case subset; reward **0.473422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.559579 | +| well/active_matter | 0.827134 | +| well/turbulent_radiative_layer_2D | 0.628780 | + +Outcome: improvement; published atomically. + +## E006 — observation and integration operators +- Hypothesis / changed factor: Same E005B neural weights, projection and horizon regularization. Spectral branch now models finite 8-cell averaging, the corresponding Kolmogorov forcing location, and ETDRK2 substeps for Fisher/Swift–Hohenberg/KS/KdV confirmed on training trajectories. Gray–Scott uses ETDRK4 max substep 2. No reference generator code or seeds used. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.490722**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Contract and determinism verification + +Seed 0 inference checks on three validation inputs spanning scalar periodic, 11-channel periodic and rectangular mixed-boundary data, plus independently constructed unseen 2-channel 7×11 and 13-channel 129×177 inputs. All shapes/dtypes were correct, every output finite, maximum rerun difference exactly 0. Small-grid padding and non-finite-step fallback added without changing any predictions on the three validation cases (bitwise parity). No synthetic training trajectories were generated. Evidence: research/universal_v1/contract_checks.json. + +## E007 — combine improved physics and multi-step training +- Hypothesis / changed factor: E006 spectral observation/integration model and E005B output projection, with shared neural weights after 4,000 four-step fine-tuning updates. Corpus and transfer gains combined; all weights originate from E001 random initialization. +- Seed: 20261021. +- Checkpoint SHA256: `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.506925**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.531697 | +| well/active_matter | 0.821690 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: improvement; published atomically. + +## E002 — four-step training with input perturbations +- Hypothesis / changed factor: Fine-tune E001 for 8,000 updates with four autoregressive steps in the loss, 0.01 normalized input noise, lr 0.0003, batch 12. Test whether recovery from model errors improves long rollouts. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `1d2b51ab44d7b8212b22c4c1894b02b03697b7ed1d30f756f216e6076006e48d`. +- Validation: fixed 81-case subset; reward **0.398333**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.411539 | +| ape2d/decay_turb | 0.688351 | +| ape2d/diff | 0.194736 | +| ape2d/fisher | 0.729932 | +| ape2d/gs_alpha | 0.660633 | +| ape2d/gs_beta | 0.856659 | +| ape2d/gs_delta | 0.790159 | +| ape2d/gs_epsilon | 0.457965 | +| ape2d/gs_gamma | 0.625919 | +| ape2d/gs_iota | 0.527543 | +| ape2d/gs_kappa | 0.587907 | +| ape2d/gs_theta | 0.768902 | +| ape2d/kdv | 0.263418 | +| ape2d/kolm_flow | 0.973194 | +| ape2d/ks | 0.959533 | +| ape2d/sh | 0.611700 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.929633 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: did not improve validation reward; retained current package. + +## E002 — final multi-step checkpoint with validated inference operators +- Hypothesis / changed factor: Replace only the E007 shared neural weights with E002 after all 8,000 four-step fine-tuning updates. Physical operators and active-matter horizon regularization unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.511096**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.801125 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## Full validation of phase2_4000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.500153**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045446 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.527867 | +| well/active_matter | 0.839802 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E009 — active-fluid moment physics, train-only diagnostics + +Hypothesis: the transfer error is partly due to missing orientation-distribution moments. Implemented independent Bingham closure and a Fourier angular-moment Smoluchowski/Stokes system, based on public equation descriptions (the local The Well active_matter README) and training-only checks. No generator code, seeds, pretrained weights or normalization statistics used. Translation and rotation diffusivities 0.05 and beta 0.8 were verified against train fields; alpha and zeta come from supplied metadata. + +Changed factors: Bingham closure vs retained angular moments; spatial grids 128/256; angular resolution 12/24/40; zero vs Bingham initialization; fixed recurrent context encoder that inserts observed low moments. The encoder updates physical state only; no parameters or gradients are fitted at inference. + +Seed: not applicable (deterministic analytical operators). Checkpoint: none; these are not published candidates. Validation VRMSE: not yet run for any dataset. Four selected TRAIN trajectories only. Bingham 30-step train mean VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5): 0.0261, 0.5173, 0.8618, 1.0525. Doubling spatial resolution did not resolve closure drift. At (3,-2), 12 angular modes with context encoding reduced four-step train VRMSE to about 0.0044, versus 0.153 for Bingham. Twelve modes were inadequate at high activity; increasing angular resolution is being tested. A GPU prototype initially failed due to TorchScript complex-number optimization; this is a portability failure, not a scored model. + +Outcome: promising mechanism, still under development; retained the validated shared neural/spectral model. Detailed train results and source variants remain in research/physics_diagnostics. + + +## E009b — active kinetic resolution and implementation diagnostics + +Hypothesis: angular and spatial discretization limit the kinetic closure. Changed factors: M=32/40/64 even angular modes, 128 versus native256 spatial grid, spatial2/3 filtering enabled/disabled, eager versus CUDA graph execution. Seed: none (deterministic operators); neural checkpoint unchanged. Validation: not evaluated for these train-only numerical diagnostics. Runtime source and all numerical results are in physics_diagnostics. + +Native256/M32 thirty-step TRAIN VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5) was 0.025519, 0.014763, 0.100166, 0.880100. M40 and M64 gave nearly equal short-rollout errors, whereas 128 spatial resolution remained inaccurate or unstable at high activity. Removing the spatial filter did not help. Four-step errors at native256/M64 were approximately 0.0024/0.0031/0.0041/0.0055 for (9,-3), and 0.0233/0.0483/0.0937/0.1448 for (17,-5). Outcome: select native256/M32 for validation, retaining a finite physical-prefix fallback. TorchScript fusion failures and CUDA graph scalar-copy failures were fixed before scoring. CUDA graphs reduce launch overhead; all state storage addresses remain fixed. + +## E011a — buoyant smoke discretization, train-only identification + +Hypothesis: a fixed incompressible-fluid operator can reduce PDEArena rollout error once the observed grid layout is represented correctly. Independent MAC solver with semi-Lagrangian advection, explicit viscosity and discrete Neumann pressure projection. Training data reveal that saved velocities are lower faces with the final zero normal face omitted. Correct discrete divergence is about 0.00016 RMS. Physical domain32 and stored interval6 account for endpoint metadata conventions. + +Changed factors on three TRAIN trajectories: stored interval6 versus6.461538, internal time steps0.25/0.5/1/1.5, viscosity0/0.001/0.005/0.008/0.01/0.012, and linear versus MacCormack scalar advection. Seed: none; checkpoint: none (analytic prototype). Validation per dataset: not yet evaluated. At interval6, linear scalar advection and viscosity0.01, one-step mean train VRMSE was 0.0000243. MacCormack with the same viscosity gave0.0981; reducing the internal step to0.25 with MacCormack gave0.3391. Thus closer continuum integration did not reproduce the observed numerical dynamics. Eight additional complete TRAIN rollouts with the selected scheme had mean VRMSE below0.00011. A separate torch-only double-precision implementation reproduced the SciPy prototype. Outcome: proceed to validation; no held-out data or generator code used. Detailed tested values and errors are retained in arena_initial_results.json, arena_scheme_results.json, arena_long_results.json and arena_torch_results.json. + +## E010 — fixed angular-moment active-fluid operator +- Hypothesis / changed factor: Add the independently implemented Smoluchowski/Stokes angular-moment operator to the same phase2-final shared neural model. M32 even angular modes, native256 grid, six Stokes iterations, ETDRK4, deterministic CUDA graph; fixed Bingham initialization and observed low-moment context encoding. No inference parameter fitting. All twelve active validation cases used all thirty physical steps. Exact repeat determinism verified on low/high-activity validation inputs and a diffusion input; two novel grid/channel schemas also passed. Runtime pins the complete package directory during atomic updates. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `d11ea64e646593878f730df43933d3f1e64f481a34083bb7687be973e51e2d33`. +- Validation: fixed 81-case subset; reward **0.572674**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E011 — fixed MAC-grid buoyant-fluid operator +- Hypothesis / changed factor: Keep the E010 shared neural checkpoint and all other inference components; add a metadata-conditioned MAC fluid update for the buoyant-smoke schema. The selected numerical scheme was identified on train pools only. Quick validation reran all twelve PDEArena cases and reused E010 predictions for unchanged operators with identical weights. PDEArena mean VRMSE is 0.00008349. Exact repeat determinism passed for two smoke cases, diffusion, and two novel grid/channel schemas. Expanded end-to-end validation is next. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `397cac1ec1b4314c9e5832071d0cc8a3451a85a8d891ba469a64722ac4d69f65`. +- Validation: fixed 81-case subset; reward **0.713334**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E012a — neural forecast uncertainty, train-only ablation + +Hypothesis: spatial averaging reduces uncertain late-time turbulent fluctuations while retaining vertical mean profiles. Same phase2-final checkpoint (SHA256 d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420), seed20261021. Nine TRAIN trajectories, one for each cooling time; no validation used to select the coefficient. Changed factors: nonnegative material scalars; Gaussian smoothing with final width1/2/4/8/16 pixels; linear horizon-dependent shrinkage toward the mean over periodic directions with final weight0.25/0.5/0.75/1.0 at step30. + +| Variant | Train VRMSE | +|---|---:| +| raw | 0.538547 | +| positive | 0.538100 | +| blur_1 | 0.537242 | +| blur_2 | 0.536386 | +| blur_4 | 0.534177 | +| blur_8 | 0.529554 | +| blur_16 | 0.526478 | +| profile_0.25 | 0.530485 | +| profile_0.5 | 0.525987 | +| profile_0.75 | 0.524946 | +| profile_1.0 | 0.527495 | + +Validation per dataset: not evaluated in this ablation. Outcome: coefficient0.75 selected for subsequent validation. Positivity alone helped very little; smoothing cannot resolve most pressure error. Roughly one-fifth of pressure MSE comes from spatial-mean error, so a correction to the mean alone is insufficient. + +## E012 — fixed spatial uncertainty shrinkage +- Hypothesis / changed factor: The same shared neural weights and physical operators as E011. Add horizon-dependent shrinkage toward means over periodic directions (weight 0.75 by step30), plus nonnegative density/concentration. The coefficient was selected on nine training trajectories before validation. All nine radiative-layer validation rollouts were rerun; unchanged physical outputs were reused for the other72 cases. Exact determinism passed on three radiative-layer inputs and two novel grid/channel schemas. Validation radiative-layer VRMSE improved from0.579723 to0.546835. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715448**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546835 | + +Outcome: improvement; published atomically. + +## E013 — native rectangular inference before fine-tuning +- Hypothesis / changed factor: Change only the shared neural inference grid for rectangular inputs from64x192 to native128x384, keeping the E012 checkpoint and all other operators/regularization fixed. This tests resolution transfer before native-grid fine-tuning. Radiative-layer VRMSE worsened from0.546835 to0.769883; retain E012. All other validation predictions are unchanged physical outputs. The result motivates retraining the shared operator on native grids rather than changing inference resolution alone. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `cc9804d49168dff3c2a3868118de9b3b3bccbd2b0982814f94617a3115b7b6fa`. +- Validation: fixed 81-case subset; reward **0.703067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.769883 | + +Outcome: did not improve validation reward; retained current package. + +## E008 — larger-batch transfer-focused training, update 5000 +- Hypothesis / changed factor: Continue the same shared network from E002 final weights: batch 48, four-step loss, 0.01 input noise, corpus probability 0.1 and 0.3 for each transfer group, lr 0.0003 cosine, sampling seed 20261022; initialization seed remains 20261021. Inference operators frozen to E012 (angular-moment active matter, MAC buoyant smoke and fixed neural uncertainty shrinkage). Quick validation reruns all nine radiative-layer trajectories and reuses fixed physical predictions for the other seventy-two cases, which are independent of the neural weights; expanded validation runs end-to-end. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715555**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + + +## E015a — reaction-diffusion spatial/time convergence, training only + +Hypothesis: unresolved reaction harmonics cause remaining Gray–Scott drift. Changed factors: grid256/512/1024 after finite8-cell observation deconvolution; ETDRK4 maxdt2/0.5 vs ETDRK2 maxdt0.2/0.25/0.3/0.5; optional physical context encoding preserving unresolved modes; finite-averaging ratio2/4/8/16 versus its continuum limit. Deterministic numerical operators, no random seed or learned checkpoint. Validation per dataset: not yet evaluated. + +On selected TRAIN gamma/delta trajectories, grid256/RK4dm2 three-step VRMSE was gamma .000764/.000799/.001016 and delta .000817/.001631/.002567. Grid512/RK2dm.25 gave gamma .000197/.000268/.000394 and delta .000105/.000222/.000335. Grid1024 and the recurrent physical-state encoder gave no additional improvement. Ratio8 remained best among tested observation filters. Thus prior negative grid-only tests did not rule out improvement when resolution and observation inversion were changed together. A portable double-precision GPU implementation is undergoing full train rollouts. Numerical results and all variants are retained in gs_resolution_results.json, gs_resolution_time_results.json, gs_context_results.json and gs_observation_results.json. + +A separate spectral CUDA-graph implementation reproduced ten different corpus training cases exactly over three forecast steps (max absolute difference0 for each). Some runtime reduction was measured while shared training was active, e.g. three Navier–Stokes steps .70s to .47s. This changes execution only; no numerical parameters are fitted at inference. + +## E016 design — shared global spectral feature mixing, not yet trained + +Hypothesis: acoustic pressure patterns need communication across the native rectangular grid. Prepared a shared Fourier feature-mixing residual at the first neural stage, with even extensions in nonperiodic directions and metadata-conditioned complex gains. All weights will be trained jointly on all pools; this is not a dataset specialist. Adds306552 trainable parameters to the2.759547M shared net, total3066099. New output projection starts at zero. Seed20261024 for the new parameters. Initial model matches the existing parent exactly, and a training-only backward pass gives finite nonzero gradients. No optimizer updates or validation results yet; not a publishable candidate. Source and initialization checks in research/global_v1. Continue native-grid training first to establish a comparison. + +## E015b — offline numerical coefficient calibration, negative train result + +Hypothesis: small global corrections to Gray–Scott diffusivities and elapsed time reduce residual error. Fitted three fixed coefficients offline to one-step errors on TRAIN gamma/delta trajectories, with the fine512/RK2dm.25 operator. No inference-time fitting. Seed: none; checkpoint: none (analytic calibration); validation per dataset: not run because full TRAIN rollouts worsened. + +Fitted options: `{"gs_scale": 2, "gs_dt": 0.25, "gs_order": 2, "gs_double": true, "gs_diffusion_u": 1.9999732723556725e-05, "gs_diffusion_v": 1.0000209495791168e-05, "gs_time_scale": 0.9999851457782819}`. One-step normalized MSE decreased from 2.3365036901186555e-08 to 1.9611423235582817e-08. + +| Train dataset | Variant | 26-step VRMSE | +|---|---|---:| +| gs_gamma | uncalibrated | 0.029923 | +| gs_gamma | calibrated | 0.031195 | +| gs_delta | uncalibrated | 0.013152 | +| gs_delta | calibrated | 0.014885 | +| gs_alpha | uncalibrated | 0.002711 | +| gs_alpha | calibrated | 0.002466 | +| gs_theta | uncalibrated | 0.010527 | +| gs_theta | calibrated | 0.049919 | + +Outcome: rejected. The tiny one-step improvement did not carry over to long rollouts, particularly on the unused theta training trajectory. Retain nominal physical coefficients. + +## E015 — finer reaction-diffusion state with matched integration +- Hypothesis / changed factor: Keep the same phase3/update5000 shared neural checkpoint. Gray–Scott now retains a2x internal Fourier grid, uses double-precision ETDRK2 with max internal dt0.25, and observes it through the finite8-cell averaging filter. Nominal physical coefficients retained after calibration failed full train rollouts. Enable deterministic CUDA graphs for spectral updates. Reran all48 corpus validation cases; corpus mean improved from0.032509 to0.010893. Other33 cases reuse identical parent outputs. Exact repeat checks passed on two Gray–Scott cases, Kolmogorov flow and novel schemas; all24 unchanged corpus cases matched parent outputs bit for bit. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `3962bd91a5cf0d9b11b901a10dda99623d4d995b034f57f75c81e62bc41c8aff`. +- Validation: fixed 81-case subset; reward **0.755127**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + +## Full validation of arena_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.696465**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045295 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E014 — native-grid eight-step training, update 1000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.756053**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 1000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755034**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546620 | + +Outcome: did not improve validation reward; retained current package. + +## E017a — active-state observation diagnostics on training trajectories + +Hypotheses: remaining active-fluid drift arises either from numerical resolution or inconsistent hidden angular moments after observed low moments are inserted. Changed factors: time step, spatial grid, Stokes iterations, finite-difference moment inversion, fixed velocity-constrained moment observation, positivity projection, rigid angular transport, and Bingham amplitude transport. All diagnostics use the first trajectory in the corresponding TRAIN pool, starting at frame zero; no validation or held-out targets. The context encoder uses only four observed frames and updates physical state; solver and encoder parameters are fixed. No parameter optimization at inference. + +Seed: deterministic physical operators, no new random parameters. Shared checkpoint unchanged; current selected SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Validation VRMSE per dataset: not evaluated for these train-only diagnostics. + + +kinetic_accuracy_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"name": "baseline", "params": [17, -5]} | 0.07398549 | 0.13735820 | 4 | +| {"name": "smaller_time_step", "params": [17, -5]} | 0.07397478 | 0.13734564 | 4 | +| {"name": "stokes_iterations", "params": [17, -5]} | 0.07398360 | 0.13735427 | 4 | +| {"name": "upsample_384", "params": [17, -5]} | 0.07628622 | 0.14166540 | 4 | +| {"name": "upsample_512", "params": [17, -5]} | 0.07628392 | 0.14165913 | 4 | + +kinetic_encoder_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"last": false, "params": [3, -2], "weight": 0} | 0.00370120 | 0.00440354 | 4 | +| {"last": false, "params": [3, -2], "weight": 1} | 0.01032324 | 0.01005470 | 4 | +| {"last": true, "params": [3, -2], "weight": 0.25} | 0.00641056 | 0.00683838 | 4 | +| {"last": true, "params": [3, -2], "weight": 1} | 0.02690719 | 0.02838808 | 4 | +| {"last": false, "params": [9, -3], "weight": 0} | 0.00375097 | 0.00547713 | 4 | +| {"last": false, "params": [9, -3], "weight": 1} | 0.14332058 | 0.16554202 | 4 | +| {"last": true, "params": [9, -3], "weight": 0.25} | 0.04367409 | 0.04571036 | 4 | +| {"last": true, "params": [9, -3], "weight": 1} | 0.12102679 | 0.12214902 | 4 | +| {"last": false, "params": [17, -5], "weight": 0} | 0.07398282 | 0.13735300 | 4 | +| {"last": false, "params": [17, -5], "weight": 1} | 0.35820127 | 0.58563930 | 4 | +| {"last": true, "params": [17, -5], "weight": 0.25} | 0.05379023 | 0.08525104 | 4 | +| {"last": true, "params": [17, -5], "weight": 1} | 0.20885072 | 0.28744033 | 4 | + +kinetic_flow_encoder_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "weight": 0.0} | 0.02551891 | 0.05217000 | 30 | +| {"params": [1, -1], "weight": 0.25} | 0.01215404 | 0.02445865 | 30 | +| {"params": [1, -1], "weight": 1.0} | 0.00162011 | 0.00254364 | 30 | +| {"params": [3, -2], "weight": 0.0} | 0.01476343 | 0.04135585 | 30 | +| {"params": [3, -2], "weight": 0.25} | 0.00774478 | 0.02111043 | 30 | +| {"params": [3, -2], "weight": 1.0} | 0.00182436 | 0.00253463 | 30 | +| {"params": [9, -3], "weight": 0.0} | 0.10013896 | 0.45469648 | 30 | +| {"params": [9, -3], "weight": 0.25} | 0.03468125 | 0.16266733 | 30 | +| {"params": [9, -3], "weight": 1.0} | 0.04499391 | 0.18140420 | 30 | +| {"params": [17, -5], "weight": 0.0} | 0.54988378 | 1.29260397 | 30 | +| {"params": [17, -5], "weight": 0.25} | 0.48679241 | 1.30647564 | 30 | +| {"params": [17, -5], "weight": 1.0} | 0.62024931 | 1.32728279 | 30 | + +kinetic_transport_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.00372676 | 0.00443898 | 4 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187841 | 0.00173137 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 0.0} | 0.00375795 | 0.00446493 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 1.0} | 0.00187924 | 0.00173272 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00041423 | 0.00039727 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043372 | 0.00043870 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 0.0} | 0.00041337 | 0.00038921 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 1.0} | 0.00039989 | 0.00040295 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.00621011 | 0.01090280 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00566252 | 0.01021467 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 0.0} | 0.00627246 | 0.01102048 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 1.0} | 0.00555848 | 0.01004915 | 4 | + +kinetic_transport_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 0.0} | 0.02551866 | 0.05216906 | 30 | +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00162010 | 0.00254361 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.01480152 | 0.04140588 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183204 | 0.00254354 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00303750 | 0.01372149 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00258502 | 0.00757161 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.45993094 | 1.23510826 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.46775045 | 1.23808050 | 30 | + +Outcome: finer numerics did not improve forecasts. Temporal-derivative inversion was mostly harmful. Fixed velocity observation improves low-activity cases; rigid transport additionally improves moderate/high activity. Bingham amplitude transport adds little in short tests. Select phase transport plus observation weight1 for validation based on complete 30-step training evidence. No candidate from this entry alone was published. + +- Diagnostic source kinetic_torch.py: SHA256 `1082bbf868d579e4c101b86d5fccffc9e43ba1c45e569e1c1dcbc4a19c5ddd6a`. +- Diagnostic source kinetic_flow_encoder.py: SHA256 `e0d1e5022cc8e0595c1c9fd43dc15ba4e77c7998e6fef139c5b58e2df30dde22`. +- Diagnostic source kinetic_transport_encoder.py: SHA256 `4bbb953264d06a28cd3832509172ae68240a24ec779116a3e0377346818e26e9`. + +## E014 — native-grid eight-step training, update 2000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.757118**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 2000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755962**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.532789 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of fine_gs_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.730834**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261022; checkpoint SHA256 `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.049812 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Expanded-validation protocol: all561 non-active cases were rerun with this complete package. The24 active-matter outputs were reused from the earlier end-to-end E011 run after verifying identical active solver assets/configuration and thirty physical steps for every case, so those outputs are independent of the changed shared neural weights. The official scorer evaluates all585 trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E017 — geometric and velocity-constrained context state +- Hypothesis / changed factor: Keep the same shared neural checkpoint and all other operators from parent.json. Change only the active-matter four-frame physical-state encoder: rotate higher angular moments with observed second-moment orientation, scale by observed concentration, use twelve fixed regularized Stokes observation steps for F4, and project F4 into its necessary moment-positivity disk. Fixed observer weight1, ridge0.01 and phase confidence threshold0.1 were selected using train-only full30 diagnostics. No learned or solver parameter changes at inference. All twelve active validation cases rerun; sixty-nine identical parent outputs reused. Exact physical-prefix coverage is in physical_coverage.json. No new random parameters; inherited initialization seed20261021 and sampling seed recorded in the checkpoint manifest. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.802217**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E017b — combine context encoder and native checkpoint +- Hypothesis / changed factor: Combine the E017 fixed physical context encoder with the better shared native-grid checkpoint selected at phase4 update2000. No training, new random parameters or numerical changes. All runtime modules/configuration outside kinetic.py and the shared weights were checked identical between parents. Twelve active outputs are independent of shared weights after verifying all30 steps use the physical operator. Official scoring reuses those12 outputs and69 matching neural-parent outputs. Replayed three neural cases and two novel schemas with exact determinism. Active determinism was already checked for the identical physical encoder. Inherited initialization seed20261021, native-stage sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.803282**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E018 — finer internal advection grids +- Hypothesis / changed factor: Keep the same shared neural checkpoint and active-state encoder. Change only Burgers and Kolmogorov physical updates to a twofold internal Fourier grid in double precision, retaining unresolved state between outputs and using the same finite-cell observation map. Preserve the physical Kolmogorov forcing phase when refining its grid. Three full26 training trajectories per equation supported the change; doubling substeps, using a fourfold grid, or encoding unresolved modes through all four context frames added little. No random parameters or new training. Rerun the six affected validation cases; reuse seventy-five identical parent outputs. All six used26/26 physical updates. Exact repeat checks include Burgers, Kolmogorov, an unchanged Gray-Scott case and two novel schemas. Inherited initialization seed20261021 and sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `970b00933280c87d32c3d322412f89d56ad1969897bbdd0b83dbeccc51231cf8`. +- Validation: fixed 81-case subset; reward **0.806376**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E017 expanded active-matter validation + +Changed factor: expand from twelve to all24 active-matter validation trajectories, all start0. Twelve existing predictions were reused only after exact context equality checks; the other twelve were replayed with the same complete E017 package. Official scorer active-matter VRMSE **0.08948507**, versus 0.19604967 for the earlier physical encoder on the same24 cases. Failed cases 0. Other dataset VRMSE: not reevaluated in this experiment. Deterministic encoder; inherited neural initialization seed20261021, sampling seed 20261023. Checkpoint SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Outcome: expanded group-level validation, no comparison between different-size aggregate rewards. + +## E018a — further fixed-operator train diagnostics + +Hypotheses and changed factors: (1) interpolated velocity/strain can drive hidden moments between context frames; (2) an entropic angular-moment projection can repair the full distribution; (3) moment and velocity filters should differ; (4) unresolved Fourier modes cause Burgers/Kolmogorov error; (5) a finer grid, retained context state, time-step refinement or alternative exponential integration repairs KdV. All diagnostics use TRAIN trajectories only, with exact source indices recorded by scripts. No new random parameters or learned checkpoint. Validation VRMSE per dataset: not evaluated in these diagnostics; E018 separately evaluates the selected grid change. + + +kinetic_forced_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"cubic": false, "observe_initial": false, "params": [3, -2]} | 0.00187798 | 0.00173005 | 4 | +| {"cubic": true, "observe_initial": false, "params": [3, -2]} | 0.00187867 | 0.00173198 | 4 | +| {"cubic": true, "observe_initial": true, "params": [3, -2]} | 0.00413512 | 0.00506158 | 4 | +| {"cubic": false, "observe_initial": false, "params": [9, -3]} | 0.01194177 | 0.01268802 | 4 | +| {"cubic": true, "observe_initial": false, "params": [9, -3]} | 0.01874249 | 0.02191580 | 4 | +| {"cubic": true, "observe_initial": true, "params": [9, -3]} | 0.01873903 | 0.02190757 | 4 | +| {"cubic": false, "observe_initial": false, "params": [17, -5]} | 0.02985521 | 0.04619750 | 4 | +| {"cubic": true, "observe_initial": false, "params": [17, -5]} | 0.04395452 | 0.07376709 | 4 | +| {"cubic": true, "observe_initial": true, "params": [17, -5]} | 0.04395483 | 0.07376773 | 4 | + +kinetic_moment_projection_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187845 | 0.00173138 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043149 | 0.00043489 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00560499 | 0.01007434 | 4 | + +kinetic_moment_projection_long_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00161964 | 0.00254240 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183213 | 0.00254286 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00240461 | 0.00671084 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.48482975 | 1.27747452 | 30 | + +kinetic_unmasked_flow_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187846 | 0.00173107 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00041315 | 0.00041952 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00551630 | 0.01016449 | 4 | + +nonlinear_resolution_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963583 | 0.05441318 | 26 | +| {"double": true, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963618 | 0.05441412 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1} | 0.00572061 | 0.01058393 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 2} | 0.00572059 | 0.01058392 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01216232 | 0.02993135 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01185469 | 0.02870498 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1} | 0.00759149 | 0.02465414 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 2} | 0.00759146 | 0.02465408 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001462 | 0.00006308 | 26 | +| {"double": true, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 2} | 0.00001457 | 0.00006339 | 26 | + +nonlinear_resolution_expanded_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00301647 | 0.00326871 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037459 | 0.00050441 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037497 | 0.00050395 | 26 | +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00124680 | 0.00155330 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012381 | 0.00019031 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012380 | 0.00019021 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.08121000 | 0.15702826 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.02042678 | 0.05910929 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.01992439 | 0.05732524 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.04068160 | 0.08604752 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00718886 | 0.02328656 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00716170 | 0.02330063 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035132 | 0.00209063 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035139 | 0.00208930 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00034709 | 0.00206470 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203683 | 0.01849953 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203362 | 0.01849631 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01204391 | 0.01851219 | 26 | + +nonlinear_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 0} | 0.00572061 | 0.01058393 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 17} | 0.00037297 | 0.00050443 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 93} | 0.00012373 | 0.00019031 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 0} | 0.00759136 | 0.02465325 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 17} | 0.02068291 | 0.05987191 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 93} | 0.00716521 | 0.02321832 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 17} | 0.00035139 | 0.00208930 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 93} | 0.01203363 | 0.01849633 | 26 | + +kdv_numerics_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 0} | 0.00001493 | 0.00006064 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001457 | 0.00006350 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 0} | 0.00001457 | 0.00006347 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00035151 | 0.00209107 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 17} | 0.00019026 | 0.00094255 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00034583 | 0.00205838 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 17} | 0.00034618 | 0.00206044 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01203692 | 0.01849972 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 93} | 0.01360971 | 0.02156036 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01205018 | 0.01851917 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 93} | 0.01204985 | 0.01851889 | 26 | + +Outcomes: interpolated active forcing was harmful at moderate/high activity. Entropic moment projection modestly improved moderate activity but worsened the complete high-activity rollout (0.467750 to0.484830); not selected. Retaining full velocity modes has only a small short-rollout effect; full30 test remains ongoing. Twofold grids robustly improved Burgers and Kolmogorov across three training trajectories. Fourfold grids and more substeps added little; four-frame unresolved-state encoding added little or hurt. KdV grid/integrator variants were not consistently better. A proposed small-domain KdV test found no training cases below domain extent15 and ran no rollouts; replaced it with tests on the existing three training trajectories. Low-frequency differential fits recover nonlinear coefficients near6; fitted dissipative corrections were inconsistent, so no coefficient change was selected. + +The phase5 global-mixing trainer initially stopped before optimizer creation because channels_last was applied to a five-dimensional Fourier parameter. Applying the layout only to Conv2d modules fixed it; restart used the identical checkpoint and seed20261024, with no optimizer updates lost. Exactly one GPU trainer remains. + +Diagnostic source kinetic_forced_context.py: SHA256 `cebf38a2f07fbf2ea664c5b00ac17bd5dd1cef2a128b75a2d7e978fbc102e3a0`. + +Diagnostic source kinetic_moment_projection.py: SHA256 `fc796f88bda7f7802bab82a8e41d76dc1e08e1fc3a98fb9eed20cd45e26fc69a`. + +Diagnostic source kinetic_unmasked_flow.py: SHA256 `4a40d463091ef1dd6a2432e2ce79f0a417c61af1943369711062c6e0113a059b`. + +Diagnostic source spectral_resolution.py: SHA256 `907c567c87c46bb3adf9e11b1282d1a3e1400751d5e5c0bfdd835350d3062420`. + +Diagnostic source spectral_kdv_numerics.py: SHA256 `e3fcb35b99158c156691e324dae512d17f9fca287fb072270e1adbc6437b1ffe`. + +## E018b — distinct active flow and moment filters, negative full-rollout result + +Hypothesis: the stored active velocity/strain retain modes above the moment dealiasing cutoff, so the Stokes solve should retain those modes. Changed only the velocity/strain spectral mask, preserving the moment mask. Four complete30-step TRAIN rollouts, deterministic physical operator, no new seed or checkpoint; validation per dataset not evaluated. + +| Training parameters | Mean VRMSE | +|---|---:| +| [1, -1] | 0.00161508 | +| [3, -2] | 0.00182706 | +| [9, -3] | 0.00258145 | +| [17, -5] | 0.46823457 | + +Outcome: negligible improvement at low/moderate activity and slight worsening at high activity (0.467750 to0.468235); retained the validated operator. The direct training spectral audit found concentration/director power above the2/3 cutoff at float32 roundoff, while high-activity strain retains about0.0022 of its spectral norm there. This numerical distinction is real but does not explain the dominant rollout error. + +## E016 — shared global spectral mixing, update 1000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `3cd1d10c39cf842ae764c3a6390d8af9c784fc4cc2141213e7647256be66ba8c`. +- Validation: fixed 81-case subset; reward **0.803408**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.560153 | + +Outcome: did not improve validation reward; retained current package. + +## E016 — shared global spectral mixing, update 1000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `a0019d25f7e2d44c69f90d0ada4f0459c943fbffa3503dc585227648c836f516`. +- Validation: fixed 81-case subset; reward **0.802480**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.574692 | + +Outcome: did not improve validation reward; retained current package. + +## E020 — pressure uncertainty shrinkage +- Hypothesis / changed factor: Keep the shared neural weights and fixed physical operators. Add pressure-only output shrinkage toward each predicted spatial mean: zero through step10, linear ramp to0.75 at step30. The existing periodic-profile shrinkage remains unchanged. Select this fixed schedule from nine train rollouts for both native and global1000 checkpoints; native training mean VRMSE0.519754 to0.518445, global0.536804 to0.534698. No new random parameters or inference-time fitting; inherited initialization seed20261021 and sampling seed20261023. Rerun all nine neural validation cases, reuse seventy-two identical physical outputs. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `1b8f91e26862343f0238655ab2b43f24ad67da270779efcfa7ef02a841f95ab6`. +- Validation: fixed 81-case subset; reward **0.806391**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Outcome: improvement; published atomically. + +## E020a — pressure schedule training sweep + +Changed factor: pressure-only mixing toward its predicted spatial mean, with start0/5/10, end15/30, strength0/0.25/0.5/0.75/1. Nine training trajectories, one per cooling parameter; train membership was audited again. Native checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4 and global1000 checkpoint from phase5/eval001000_shrink were both evaluated. No random coefficients; inherited initialization20261021, native sampling20261023 and global sampling/new-parameter20261024. Validation per dataset: not evaluated in this training sweep; selected native variant validated separately in E020. + +| Model | Start | End | Strength | Training VRMSE | Pressure VRMSE | +|---|---:|---:|---:|---:|---:| +| native | 0 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 15 | 0.25 | 0.51936945 | 1.08422407 | +| native | 0 | 15 | 0.5 | 0.52051218 | 1.08879500 | +| native | 0 | 15 | 0.75 | 0.52311825 | 1.09921928 | +| native | 0 | 15 | 1.0 | 0.52708753 | 1.11509640 | +| native | 0 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 30 | 0.25 | 0.51915815 | 1.08337887 | +| native | 0 | 30 | 0.5 | 0.51916075 | 1.08338927 | +| native | 0 | 30 | 0.75 | 0.51976061 | 1.08578873 | +| native | 0 | 30 | 1.0 | 0.52095018 | 1.09054701 | +| native | 5 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 15 | 0.25 | 0.51899641 | 1.08273192 | +| native | 5 | 15 | 0.5 | 0.51948753 | 1.08469640 | +| native | 5 | 15 | 0.75 | 0.52120696 | 1.09157413 | +| native | 5 | 15 | 1.0 | 0.52409830 | 1.10313946 | +| native | 5 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 30 | 0.25 | 0.51903559 | 1.08288863 | +| native | 5 | 30 | 0.5 | 0.51874419 | 1.08172304 | +| native | 5 | 30 | 0.75 | 0.51888335 | 1.08227969 | +| native | 5 | 30 | 1.0 | 0.51945286 | 1.08455773 | +| native | 10 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 15 | 0.25 | 0.51861991 | 1.08122590 | +| native | 10 | 15 | 0.5 | 0.51848526 | 1.08068733 | +| native | 10 | 15 | 0.75 | 0.51935090 | 1.08414988 | +| native | 10 | 15 | 1.0 | 0.52118939 | 1.09150382 | +| native | 10 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 30 | 0.25 | 0.51902077 | 1.08282937 | +| native | 10 | 30 | 0.5 | 0.51858337 | 1.08107977 | +| native | 10 | 30 | 0.75 | 0.51844492 | 1.08052597 | +| native | 10 | 30 | 1.0 | 0.51860631 | 1.08117150 | +| global1000 | 0 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 15 | 0.25 | 0.53561169 | 1.15099039 | +| global1000 | 0 | 15 | 0.5 | 0.53658240 | 1.15487321 | +| global1000 | 0 | 15 | 0.75 | 0.53962210 | 1.16703201 | +| global1000 | 0 | 15 | 1.0 | 0.54454053 | 1.18670573 | +| global1000 | 0 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 30 | 0.25 | 0.53562902 | 1.15105970 | +| global1000 | 0 | 30 | 0.5 | 0.53524687 | 1.14953112 | +| global1000 | 0 | 30 | 0.75 | 0.53566622 | 1.15120852 | +| global1000 | 0 | 30 | 1.0 | 0.53687798 | 1.15605554 | +| global1000 | 5 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 15 | 0.25 | 0.53535586 | 1.14996706 | +| global1000 | 5 | 15 | 0.5 | 0.53561634 | 1.15100898 | +| global1000 | 5 | 15 | 0.75 | 0.53755790 | 1.15877521 | +| global1000 | 5 | 15 | 1.0 | 0.54106119 | 1.17278837 | +| global1000 | 5 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 30 | 0.25 | 0.53563652 | 1.15108971 | +| global1000 | 5 | 30 | 0.5 | 0.53500332 | 1.14855691 | +| global1000 | 5 | 30 | 0.75 | 0.53491749 | 1.14821358 | +| global1000 | 5 | 30 | 1.0 | 0.53538269 | 1.15007438 | +| global1000 | 10 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 15 | 0.25 | 0.53513905 | 1.14909983 | +| global1000 | 10 | 15 | 0.5 | 0.53481845 | 1.14781741 | +| global1000 | 10 | 15 | 0.75 | 0.53584572 | 1.15192652 | +| global1000 | 10 | 15 | 1.0 | 0.53814365 | 1.16111824 | +| global1000 | 10 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 30 | 0.25 | 0.53575180 | 1.15155082 | +| global1000 | 10 | 30 | 0.5 | 0.53504651 | 1.14872968 | +| global1000 | 10 | 30 | 0.75 | 0.53469765 | 1.14733423 | +| global1000 | 10 | 30 | 1.0 | 0.53471068 | 1.14738634 | + +Outcome: both models selected start10/end30/strength0.75, but the improvement was small. Earlier/stronger shrinkage is not selected. The global model remains worse than the native model after this correction. + +Global1000 checkpoint SHA256 `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. Study source SHA256 `2d9ddca63b268fbc9b52f5ac1bf800e7400ae7458fbd09f80a21cecaca84ffe9`. + +## E021 preparation — shared global-statistics residual head + +Hypothesis: predicting spatially uniform channel changes explicitly will help the pressure/global-mean dynamics that dominate radiative-layer error. Add one shared MLP head using the existing metadata embedding plus normalized means and standard deviations of each of the four context frames. It emits a uniform residual for every channel on every dataset; there are no dataset-specific specialist weights. The spatial network remains the selected native ConvNeXt body for this prototype. Added56075 parameters, total2815622, seed20261027. Output projection initialized to zero. + +Initialization check on one native TRAIN window: exact parent prediction parity (maximum difference0), finite nonzero output-layer gradient norm 0.043989721685647964. No optimizer updates, no new trained checkpoint, no validation per dataset and no publication. Parent checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4. Consider only after the longer-horizon experiment; regenerate against the actually selected parent architecture if it changes. Prototype source SHA256 `8de97bfff13028b7d33adfd16564d8b7b4f00d2ac50e67d2f051549480cefbe8`. + +## E022a — fixed hidden-state uncertainty quadrature, training diagnostics + +Hypothesis: averaging fixed hidden-state uncertainty may reduce late chaotic active-fluid error. After the four-frame state encoder, evolve center and symmetric smooth higher-moment phase perturbations with the same fixed physical parameters. Perturbations leave observedF0/F2 unchanged and enforce F4 positivity. Fixed latent-noise seed20261026, phase standard deviations0.01 and0.0001. Compare equal quadrature weights and half-center/quarter-endpoint weights. No neural training, no new learned checkpoint; validation per dataset not evaluated. These are physical-state quadrature nodes, not separately trained dataset models. + +kinetic_state_quadrature_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.01 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.01 | 0.00162322 | +| [1, -1] | half_center | 0.01 | 0.00162206 | +| [3, -2] | center | 0.01 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.01 | 0.00183122 | +| [3, -2] | half_center | 0.01 | 0.00183157 | +| [9, -3] | center | 0.01 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.01 | 0.16464736 | +| [9, -3] | half_center | 0.01 | 0.12350380 | +| [17, -5] | center | 0.01 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.01 | 0.44187382 | +| [17, -5] | half_center | 0.01 | 0.42681998 | +kinetic_state_quadrature_small_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.0001 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.0001 | 0.00162011 | +| [1, -1] | half_center | 0.0001 | 0.00161957 | +| [3, -2] | center | 0.0001 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.0001 | 0.00183261 | +| [3, -2] | half_center | 0.0001 | 0.00183262 | +| [9, -3] | center | 0.0001 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.0001 | 0.00258179 | +| [9, -3] | half_center | 0.0001 | 0.00257420 | +| [17, -5] | center | 0.0001 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.0001 | 0.45577160 | +| [17, -5] | half_center | 0.0001 | 0.45775375 | + +Outcome:0.01 is rejected because severe moderate-activity degradation outweighs the high-activity gain. The0.0001 quadrature leaves low/moderate errors nearly unchanged and reduces the hardest training mean from0.467763 to0.455772, but requires three physical integrations. Expand train tests to four additional parameter combinations before considering validation. Center includes an extra FFT roundtrip and positivity projection, so tiny roundoff differences from E017 are expected. + +Additional KdV observation-filter test on train0/17/93, powers0/0.5/1/2: removing deconvolution worsens all three; doubling it helps17 slightly but barely changes93 and hurts0. Keep nominal finite8-cell observation model. No KdV runtime change. + +## Full validation of advection_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775038**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Expanded-validation protocol: all561 non-active cases were replayed with the complete E018 package. All24 active outputs were reused from the expanded E017 validation after checking identical physical assets/configuration, exact input equality for reused quick cases, and thirty physical steps in every case. Hence those outputs are independent of the changed shared neural checkpoint. All585 validation trajectories are scored together by the official scorer. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of pressure_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775053**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Exact reuse protocol: model weights and all runtime operators except pressure regularization are identical to E018. Input metadata confirms only the nine radiative-layer cases contain a pressure field; their inputs match the already replayed E020 quick cases exactly. Reuse those nine new forecasts and576 unchanged E018 full forecasts. The official scorer evaluates all585 validation trajectories together. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E016 — shared global spectral mixing, update 2000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `96a49ac4734af31edd0aa5345f51515a899ef067e36abeb9e2347781c28fc055`. +- Validation: fixed 81-case subset; reward **0.807185**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E022b — expanded state-quadrature training study + +Changed factor: expand the fixed1e-4 hidden-state quadrature test from four to eight TRAIN parameter combinations, same latent seed20261026 and same fixed physical operator. No new learned weights. Validation per dataset not yet evaluated. + +| Additional training parameters | Quadrature | Mean VRMSE | +|---|---|---:| +| [5, -4] | center | 0.20956509 | +| [5, -4] | symmetric_mean | 0.20925570 | +| [5, -4] | half_center | 0.20933180 | +| [7, -3] | center | 0.00780154 | +| [7, -3] | symmetric_mean | 0.00781992 | +| [7, -3] | half_center | 0.00781529 | +| [11, -4] | center | 0.08376674 | +| [11, -4] | symmetric_mean | 0.08358850 | +| [11, -4] | half_center | 0.08358082 | +| [15, -4] | center | 0.26489031 | +| [15, -4] | symmetric_mean | 0.26059937 | +| [15, -4] | half_center | 0.26147711 | + +Across all eight cases, center mean 0.12997754, equal-weight quadrature 0.12788370. Outcome: select equal weights for validation; three physical integrations per active rollout. No publication from training results alone. + +## E016 — shared global spectral mixing, update 2000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `f06079b546194d4243540a3578b4df4e547efc8b64bb67210a4d5fe41648b2bd`. +- Validation: fixed 81-case subset; reward **0.806418**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515442 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775847**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match E018. Parsed per-case logs verify552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E023 — shared spatial-mean forecasting branch + +Hypothesis: remaining radiative-layer mean-pressure drift can be predicted from four-frame global statistics and metadata while the shared autoregressive operator handles spatial structure. Train one common MLP across all19 datasets; every branch will be evaluated on every inference case. No dataset-specific checkpoint selection. Random initialization seed20261028; CPU-only, two threads, batch512, 12000 AdamW updates, cosine learning rate0.0005, half windows start0 and half random, horizons1–30 capped by available train frames, group probabilities0.1/0.1/0.1/0.7. Loss is smooth absolute mean error normalized by target spatial variance; targets never enter model inputs. Uses only the10115 audited training records. Spatial moments derived with float64 accumulation. No pretrained assets. + +Changed factor: add a shared horizon-conditioned mean predictor to the fixed autoregressive spatial forecast. CPU training is concurrent with the one-GPU 16-step experiment. This is distinct from the earlier untrained E021 residual-mean prototype. Validation and checkpoint selection pending. No package change at preparation. + +## E023a — validation mean-branch checkpoint and blend comparison + +Changed factors: shared mean-branch checkpoint at updates2000/6000/12000 and fixed blend0/.25/.5/.75/1. Exact spatial-MSE decomposition on the nine radiative-layer VALID trajectories, keeping the same global2000 spatial prediction. These are validation selection results, not training. Other dataset VRMSE unchanged by the composite physical branch, not reevaluated here. All mean-model weights were randomly initialized here with seed20261028, shared across all19 datasets. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.50464012 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.25 | 0.48902082 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.5 | 0.48371851 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.75 | 0.48967327 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 1.0 | 0.50368029 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 6000 | 0.0 | 0.50464012 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.25 | 0.49150598 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.5 | 0.48589594 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.75 | 0.48943812 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 1.0 | 0.50039246 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 12000 | 0.0 | 0.50464012 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.25 | 0.48902526 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.5 | 0.48202704 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.75 | 0.48522405 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 1.0 | 0.49632276 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | + +Outcome: select update12000, blend0.5; all other tested choices are inferior. Full mean replacement is consistently worse than a partial blend. Actual float32 composite-package replay and determinism checks are required before publication; analytical score alone is not published. + +## E022 — fixed hidden-state quadrature +- Hypothesis / changed factor: Integrate three fixed hidden-angular-state quadrature nodes after the four-frame physical encoder, using one common neural model and identical fixed physical parameters. Nodes share all common computations; no separately trained specialists. Seed20261026, phase scale0.0001, equal weights, observedF0/F2 unchanged, F4 positivity preserved. Eight train trajectories supported this choice; larger perturbations were rejected. Rerun all twelve active validation cases, all30 physical steps; reuse sixty-nine matching non-active outputs. Any validated shared-neural progress during the replay is retained as documented in neural_rebase.json. No inference-time fitting. Inherited neural seeds/checkpoint in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `def27fe183c189d25508bcde6a013f2f6560b6b7d68526b782530bd4b1d7c730`. +- Validation: fixed 81-case subset; reward **0.807374**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E023 — shared mean forecast, composite replay +- Hypothesis / changed factor: Add one shared horizon-conditioned spatial-mean forecasting branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261028; two CPU threads, batch512, 12000 total updates, selected update12000, blend0.5. Inputs are four context means and variances plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in mean_branch_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809033**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809649**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `e7dd6ac71780c453f0a23e995e48d0c94ad68af36df21c2d8993fb7ba914c8a7`. +- Validation: fixed 81-case subset; reward **0.809341**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477935 | + +Outcome: did not improve validation reward; retained current package. + +## E024 — spatial-profile conditioned shared mean branch + +Hypothesis: four-frame channel means and variances omit thermal-layer geometry; coarse profiles along each axis may predict global mean drift more accurately. Changed factor: add eight-bin spatial marginal profiles for both axes, all channels and all four context frames, to the same shared horizon-conditioned MLP. Profiles are centered by context means and normalized by context standard deviations. Random initialization seed20261029; otherwise same shared19-dataset training distribution and12000-update schedule as E023. CPU-only, two threads, no extra GPU. Train data membership audited before deriving profiles; float64 averaging followed by float32 storage. Feature parity between training-cache input and inference from raw TRAIN context passed (feature_checks.json). All parameters initialized from scratch; no pretrained assets. + +Validation pending. Compare against the raw spatial forecast from the currently selected phase6 update1000, obtained by setting only the old mean blend to zero and replaying nine valid contexts. The final candidate, if selected, will replace the existing shared mean branch rather than adding a dataset-specific specialist. No publication from training loss. + +## E024a — profile-branch validation selection + +Changed factors: profile-branch checkpoint2000/6000/12000 and fixed mean blend0/.25/.5/.75/1. Same9 valid radiative-layer cases and phase6 update1000 spatial forecast; the old mean branch was disabled for this ablation replay. Other dataset VRMSE not reevaluated at this selection stage. Shared random initialization seed20261029. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.47627939 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.25 | 0.47072551 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.5 | 0.47161600 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.75 | 0.47837804 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 1.0 | 0.48981324 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 6000 | 0.0 | 0.47627939 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.25 | 0.47073488 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.5 | 0.46949443 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.75 | 0.47282608 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 1.0 | 0.48034141 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 12000 | 0.0 | 0.47627939 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.25 | 0.47050116 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.5 | 0.46970411 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.75 | 0.47404605 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 1.0 | 0.48293819 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | + +Outcome: update6000/blend0.5 selected at0.46949443, versus0.47388509 with E023 and0.47627894 without any mean adjustment. Additional training to12000 improved training loss but slightly worsened validation. Actual composite replay and exact contract checks are running before any publication. + +## E024 — profile-conditioned shared mean forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a profile-conditioned branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261029; two CPU threads, batch512, 12000 total updates, selected update6000, blend0.5. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_mean_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `8dc8554c96b10cffe81e8707063febceb81c18933dd345ac0dfaa329237377d6`. +- Validation: fixed 81-case subset; reward **0.809986**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Outcome: improvement; published atomically. + +## E025 — jointly forecast spatial means and coarse profiles + +Hypothesis: remaining broad thermal-layer structure errors require a forecast of spatial profiles, not only a corrected global mean. Changed factor: shared MLP predicts both global channel means and centered eight-bin marginal profiles on both axes. Enforce agreement between the predicted profile means and global means. Train across all19 datasets from random initialization seed20261030 with the same audited train statistics/profiles,12000 CPU-only updates, batch512, two threads; loss0.25 normalized mean error +0.75 normalized coarse-profile error, with the same smooth absolute form. No new data or pretrained assets. Every branch remains shared and evaluated on every case. + +Inference candidate will interpolate centered coarse-profile residuals with periodic or clamped edges according to metadata, and blend them into the common autoregressive forecast. Physical outputs remain authoritative. Uniform and spatial correction coefficients will be selected on validation, followed by actual package replay and exact determinism checks. Initialization shape, finite-output, zero-correction identity, and mean/profile consistency checks passed on an audited TRAIN context. Validation pending; no publication from training loss. + +## E025a — joint mean/profile validation selection + +Changed factors: joint statistic checkpoint2000/6000/12000, mean correction0/.5/1 and profile correction0/.25/.5/1. Same9 VALID radiative-layer trajectories and phase6 update1000 raw spatial forecasts. MSE decomposition includes the uniform correction, centered profile correction, and all cross terms; actual float32 replay is required next. Shared random seed20261030. Other dataset VRMSE not reevaluated in this selection stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.25 | 0.47635292 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.5 | 0.47958239 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 1.0 | 0.49471505 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.0 | 0.47005735 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.25 | 0.47010562 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.5 | 0.47337008 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 1.0 | 0.48871343 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.0 | 0.48430118 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.25 | 0.48426968 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.5 | 0.48736599 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 1.0 | 0.50220535 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.25 | 0.47518010 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.5 | 0.47612881 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 1.0 | 0.48395044 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.0 | 0.46912443 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.25 | 0.46794480 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.5 | 0.46886089 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 1.0 | 0.47675092 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.0 | 0.47781927 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.25 | 0.47659139 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.5 | 0.47738680 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 1.0 | 0.48486302 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.25 | 0.47462923 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.5 | 0.47500679 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 1.0 | 0.48171365 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.0 | 0.46724263 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.25 | 0.46555366 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.5 | 0.46594474 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 1.0 | 0.47282504 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.0 | 0.47431715 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.25 | 0.47263591 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.5 | 0.47296605 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 1.0 | 0.47954663 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | + +Outcome: select update12000, mean blend0.5, profile blend0.25, VRMSE0.46555366 versus E0240.46949440. At update2000 any profile correction worsened the best mean-only result; profile prediction benefited from further training. Larger profile blends were inferior. Actual replay and contract checks pending; no analytical-only publication. + +## E025 — joint shared mean/profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261030; two CPU threads, batch512, 12000 total updates, selected update12000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `67752d3d72ef3e1cf10485e57abe77cdbc9e99eb48ccb69ce72c8d12d6ab62bd`. +- Validation: fixed 81-case subset; reward **0.810291**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Outcome: improvement; published atomically. + +## E026 — sixteen-bin shared statistical forecast + +Hypothesis: eight-bin profiles smooth thin thermal interfaces; sixteen bins may improve the shared coarse forecast. Changed factor: double the input and output profile resolution on both axes from8 to16 bins, retaining one shared MLP across all19 datasets. Random initialization seed20261031. Same12000-update CPU-only training schedule, two threads, group sampling, loss weighting and horizon distribution as E025; the wider input/output layers increase parameter count. Train-only profile cache membership audited before reading, with float64 averaging. Source snapshot and train/inference feature-parity checks retained in profile16_forecast_v1. No pretrained assets. + +Validation pending; compare on the same9 valid thermal trajectories and the same phase6 update1000 raw spatial forecast. No package change before actual replay and exact contract checks. + +## E027 — pressure schedule after the learned profile decoder + +Hypothesis: the previous pressure uncertainty schedule may no longer be optimal after adding a shared profile forecast. Changed factors: pressure shrink strength0/.25/.5/.75/1, ramp start0/5/10/15 and fixed end30; zero-strength baseline evaluated once. Keep profile shrink0.75 and the trained mean/profile branch fixed. Use nine audited TRAIN trajectories, one per thermal cooling parameter, for schedule selection. No validation targets enter this choice and no new weights are trained. Raw shared CNN predictions replayed with both statistical corrections and regularization disabled, then fixed candidate decoders applied to the same outputs. Inherited neural sampling seed20261025, auxiliary seed20261030; base checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `56b7067dd85f4ba3aa13b34aafc87e1267cf7c14e7a898de9c50044b1f559c04`. Validation per dataset pending. + +## E026a — sixteen-bin validation selection + +Shared initialization seed20261031; same9 valid thermal trajectories and same phase6 update1000 raw spatial forecasts as E025. Changed checkpoint and fixed correction strengths. Other datasets not reevaluated at this stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.25 | 0.47600559 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.5 | 0.47978306 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 1.0 | 0.49835827 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.0 | 0.47140293 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.25 | 0.47106825 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.5 | 0.47484781 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 1.0 | 0.49358365 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.0 | 0.48622849 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.25 | 0.48578095 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.5 | 0.48932182 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 1.0 | 0.50735469 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.25 | 0.47423524 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.5 | 0.47528103 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 1.0 | 0.48625053 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.0 | 0.46620666 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.25 | 0.46415542 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.5 | 0.46526516 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 1.0 | 0.47656498 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.0 | 0.47363337 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.25 | 0.47157736 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.5 | 0.47258672 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 1.0 | 0.48345739 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.25 | 0.47413693 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.5 | 0.47468663 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 1.0 | 0.48361862 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.0 | 0.46656666 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.25 | 0.46436542 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.5 | 0.46493348 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 1.0 | 0.47413113 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.0 | 0.47745171 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.25 | 0.47524133 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.5 | 0.47569852 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 1.0 | 0.48441577 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | + +Outcome: select update6000, mean0.5/profile0.25, VRMSE0.46415542 versus eight-bin E0250.46555365. Additional training to12000 slightly worsened validation. Actual replay and contract checks pending. + +## E027a — training-only pressure schedule results + +No new learned weights or changed seed. Exact parent neural and auxiliary hashes are recorded in E027 and decoder_v1/parent.json. Nine audited TRAIN trajectories only; validation not yet evaluated. + +| Pressure strength | Ramp start | Training radiative-layer VRMSE | +|---:|---:|---:| +| 0.0 | 10 | 0.45035356 | +| 0.25 | 15 | 0.45038955 | +| 0.25 | 10 | 0.45050631 | +| 0.5 | 15 | 0.45065454 | +| 0.25 | 5 | 0.45075753 | +| 0.5 | 10 | 0.45097174 | +| 0.25 | 0 | 0.45108611 | +| 0.75 | 15 | 0.45114587 | +| 0.5 | 5 | 0.45158755 | +| 0.75 | 10 | 0.45174613 | +| 1.0 | 15 | 0.45185799 | +| 0.5 | 0 | 0.45240983 | +| 1.0 | 10 | 0.45282187 | +| 0.75 | 5 | 0.45283721 | +| 0.75 | 0 | 0.45431139 | +| 1.0 | 5 | 0.45449514 | +| 1.0 | 0 | 0.45677051 | + +Outcome: select zero additional pressure smoothing, training VRMSE0.45035356 versus0.45174613 for the inherited0.75/start10 schedule. The earlier pressure schedule helped older checkpoints but is a negative result for the current trained profile decoder. Validate this fixed choice after the sixteen-bin candidate finishes, preserving all latest shared weights. + +## E026 — sixteen-bin shared profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261031; two CPU threads, batch512, 12000 total updates, selected update6000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and sixteen-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile16_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6d988acd1229ae39ec83992e389a01e05d197ab496ab79cbae92cb124f39ad16`. +- Validation: fixed 81-case subset; reward **0.810400**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Outcome: improvement; published atomically. + +## E027 — updated pressure uncertainty schedule +- Hypothesis / changed factor: Replace the inherited pressure-smoothing schedule with train-selected strength0.0, start10, end30. Nine audited TRAIN trajectories selected this fixed schedule with the eight-bin profile decoder; validation then applies it to the latest selected shared weights, including any validated sixteen-bin branch. No new learned weights or inference-time fitting. Parent/checkpoint hashes and all train schedule results are recorded. Rerun nine neural validation cases; reuse72 identical physical outputs. Exact determinism checks before publication. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810428**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Outcome: improvement; published atomically. + +## E028 — statistical correction inside the autoregressive loop + +Hypothesis: applying mean/profile corrections only after the rollout lets internal state drift contaminate later predictions. Changed factor: after each shared neural update, feed a fixed fraction of the precomputed mean/profile correction into the next autoregressive context. Forecast statistics still depend only on the original four frames and metadata; weights remain fixed. Test fractions0/.25/.5/1 on nine audited TRAIN trajectories; the final output decoder remains unchanged. Physical steps remain authoritative and their latent states are unaffected. No new learned weights. Inherited neural seed20261025, auxiliary seed20261031; shared neural checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `3f9e62adca11d1dc83c908b9bbc7da794b5ca7902ecec1401a1519d72b6cd1d2`. Validation pending. + +## E028 — autoregressive statistical feedback +- Hypothesis / changed factor: Feed fraction0.25 of the fixed shared mean/profile correction into each next neural autoregressive state. Final output correction remains unchanged. Statistics are forecast once from the four original context frames and metadata; parameters never change at inference. Train-only fractions0/.25/.5/1 selected this fixed strength on nine audited thermal trajectories. No new learned weights. Physical latent states are unaffected. Rerun nine valid neural cases and reuse72 complete physical outputs, then exact neural, unknown-grid, smoke and diffusion checks before publication. Parent weights and inherited neural/auxiliary seeds are in model_manifest.json and parent records. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `c05aa38ecb979b8d52316d1cf0a5c0b35cb542ab5c79c66f8c73cbe300ee4d9c`. +- Validation: fixed 81-case subset; reward **0.810000**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469308 | + +Outcome: did not improve validation reward; retained current package. + +## E028a — training-only feedback strength + +Same inherited weights and seeds as E028; nine audited TRAIN trajectories only. Changed factor: fixed fraction of the existing output correction inserted into each next neural context. All physical branches retain their complete physical outputs. Validation per dataset pending. + +| Feedback fraction | Training radiative-layer VRMSE | Failed cases | +|---:|---:|---:| +| 0.0 | 0.45007325 | 0 | +| 0.25 | 0.44391940 | 0 | +| 0.5 | 0.44519505 | 0 | +| 1.0 | 0.44886765 | 0 | + +Outcome: select fraction0.25 at0.44391940 versus0.45007325 without feedback. Stronger feedback gives smaller gains; no new learned weights. Actual validation and exact contract checks running. + +## Full validation of quadrature_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.776678**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Expanded fixed state quadrature from12 to all24 active validation trajectories. Replayed the missing12, reused12 quick outputs only after exact context equality, and verified30/30 common physical steps for every active case. The552 corpus/smoke outputs are reused from global2000 full validation after exact physical asset/configuration equality checks and complete physical coverage. The nine neural outputs come from this exact package’s quick validation with matching inputs. All585 cases scored together; no held-out trajectories or inference caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E019 — longer autoregressive training, update 2000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.811223**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 2000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.811195**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.454067 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of mean_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778337**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval001000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778953**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_mean_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779290**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780527**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of decoder_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779732**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779595**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile16_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779704**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E029 — hidden-state quadrature width + +Hypothesis: the selected1e-4 hidden-state spread may understate uncertainty in chaotic active flow. Changed factor: phase widths1e-4/3e-4/1e-3, keeping three equal nodes, fixed latent seed20261026, identical physical encoder/parameters, and identical common-prefix guards. First study four TRAIN parameter combinations (zeta,alpha)=(5,-4),(9,-3),(15,-4),(17,-5), including the moderate case sensitive to excessive perturbation. Reuse the same encoded state and center node within each trajectory. Any improvement must be expanded before validation. No new learned weights. Kinetic source SHA256 `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`; parent package `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Validation pending. + +## E030 — learned coarse-statistic guidance preparation + +Hypothesis motivated by E028: inference-only feedback does not generalize, but the neural operator may learn how to use the same conditional forecast. Add a shared1x1 stem adapter from normalized future mean/profile fields into the common convolutional operator. Its528 weights initialize to zero, exactly preserving the selected parent before training. Fixed auxiliary forecasts consume only the original four context frames and metadata. Main neural parameters after addition3066627; initialization seed20261101. Parent `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Exact CPU forward parity on an audited native TRAIN context: maximum difference0.0; nonzero finite adapter gradient norm3.3356419 on a coarsened TRAIN window. Source in guided_v1/model.py. + +Outcome: untrained architecture preparation only. No optimizer steps, validation selection, package publication, new GPU trainer, or phase7 launcher yet. Current phase6 remains the only GPU trainer. The future training stage must start from the then-selected checkpoint and recheck initialization rather than treat this probe as a trained candidate. + +## E019 — longer autoregressive training, update 1000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809399**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477170 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 2000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.810422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463873 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810313**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465273 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.810189**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466865 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809097**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.481171 | + +Outcome: did not improve validation reward; retained current package. + +## E029a — quadrature-width training results + +Same fixed physical encoder and latent seed20261026. Four TRAIN parameter combinations; every node completed30 physical steps. No new learned weights. + +| Training parameters | Phase width | Active VRMSE | +|---|---:|---:| +| [5, -4] | 0.0001 | 0.20925571 | +| [5, -4] | 0.0003 | 0.20655669 | +| [5, -4] | 0.001 | 0.19077064 | +| [9, -3] | 0.0001 | 0.00258179 | +| [9, -3] | 0.0003 | 0.00410363 | +| [9, -3] | 0.001 | 0.04487926 | +| [15, -4] | 0.0001 | 0.26059940 | +| [15, -4] | 0.0003 | 0.24752074 | +| [15, -4] | 0.001 | 0.23641607 | +| [17, -5] | 0.0001 | 0.45577160 | +| [17, -5] | 0.0003 | 0.42649901 | +| [17, -5] | 0.001 | 0.42283213 | + +Four-case means: {0.0001: 0.23205212625907734, 0.0003: 0.22117002040613443, 0.001: 0.22372452542185783}. Outcome: select3e-4 for an expanded TRAIN check on (1,-1),(3,-2),(7,-3),(11,-4). The1e-3 spread harms the formerly accurate (9,-3) case and is inferior overall. Validation pending. + +## E031 — pressure from momentum balance + +Hypothesis: a fixed momentum-balance projection could recover pressure structure missed by the neural forecast. Independently implement a discrete periodic-x/free-y gradient least-squares inverse with FFT/DCT, finite-time momentum differences and conservative flux divergence. A manufactured scalar-potential check verifies the inverse to roundoff. First use true TRAIN density/velocity only as an oracle feasibility diagnostic; no future truth is supplied at inference. Per-case fitted scalars below are diagnostic only and are never deployed. Nine audited training trajectories. No new learned weights or new seed. + +| Training cooling case | Nominal oracle pressure VRMSE | Pressure correlation | Diagnostic optimal scalar | +|---|---:|---:|---:| +| data/train/turbulent_radiative_layer_tcool_0.03.hdf5 | 0.652605 | 0.753539 | 0.886568 | +| data/train/turbulent_radiative_layer_tcool_0.06.hdf5 | 0.787532 | 0.646115 | 0.683058 | +| data/train/turbulent_radiative_layer_tcool_0.10.hdf5 | 0.835901 | 0.640075 | 0.627110 | +| data/train/turbulent_radiative_layer_tcool_0.18.hdf5 | 1.095653 | 0.426235 | 0.393953 | +| data/train/turbulent_radiative_layer_tcool_0.32.hdf5 | 1.103630 | 0.476517 | 0.405224 | +| data/train/turbulent_radiative_layer_tcool_0.56.hdf5 | 1.789119 | 0.191514 | 0.107686 | +| data/train/turbulent_radiative_layer_tcool_1.00.hdf5 | 1.531407 | 0.145077 | 0.105623 | +| data/train/turbulent_radiative_layer_tcool_1.78.hdf5 | 1.578657 | 0.058558 | 0.042445 | +| data/train/turbulent_radiative_layer_tcool_3.16.hdf5 | 1.457004 | 0.046748 | 0.039365 | + +The oracle is useful mainly for the shortest cooling times and unreliable at long cooling times. Then use only the selected model’s predicted density/velocity, keeping its predicted global pressure mean, and blend the nominal projection at a fixed strength. Parent/checkpoint hashes are in momentum_pressure_v1/forecast_parent.json. + +| Projection blend | Training thermal VRMSE | +|---:|---:| +| 0.0 | 0.43522565 | +| 0.05 | 0.43501140 | +| 0.1 | 0.43541418 | +| 0.25 | 0.44012596 | +| 0.5 | 0.45764482 | +| 1.0 | 0.51557828 | + +Outcome: full-strength projection is harmful; a0.05 blend gives only a small training gain. Test that single train-selected strength on validation before deciding whether a self-contained Torch implementation is warranted. No physical projection is currently installed in the package. + +## E029b — expanded quadrature-width training study + +Changed factor: expand the selected3e-4 width from four to eight TRAIN parameter combinations. Same fixed latent seed20261026 and physical encoder; no new learned weights. Validation pending. + +| Additional training parameters | Phase width | Active VRMSE | Complete physical steps | +|---|---:|---:|---:| +| [1, -1] | 0.0001 | 0.00162011 | 30 | +| [1, -1] | 0.0003 | 0.00162016 | 30 | +| [3, -2] | 0.0001 | 0.00183261 | 30 | +| [3, -2] | 0.0003 | 0.00183292 | 30 | +| [7, -3] | 0.0001 | 0.00781992 | 30 | +| [7, -3] | 0.0003 | 0.00795999 | 30 | +| [11, -4] | 0.0001 | 0.08358850 | 30 | +| [11, -4] | 0.0003 | 0.08545388 | 30 | + +Eight-case means:1e-4=0.12788370,3e-4=0.12269338. Outcome: proceed to validation. + +## E031 — fixed momentum pressure correction +- Hypothesis / changed factor: Fixed0.05 pressure blend from forecast momentum balance, selected on nine audited TRAIN trajectories. Uses context plus predicted density/velocity, finite-time derivatives, conservative flux divergence, and a periodic-x/free-y gradient least-squares projection. Preserves predicted global pressure mean. No fitted diagnostic scalars are deployed and no new learned weights. Larger blends were harmful. A separate validation diagnostic found only a tiny gain; this replay uses the self-contained Torch implementation and official scorer, with exact neural/novel-grid/smoke/diffusion checks before publication. Seventy-two complete physical outputs are unchanged; all model components and inherited seeds/hashes remain in the manifest. No future truth, external files, SciPy, or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `cd7a9a8c2c2038f3be273f65987972072405494507695c64397ef3d3c2e2ae27`. +- Validation: fixed 81-case subset; reward **0.811224**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 4000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810991**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.456640 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `3cfd637325d7e7313d4bfbd48967518b107446b1be900f8859acaa7f67b51793`. +- Validation: fixed 81-case subset; reward **0.810338**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464955 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `8bf82697e735e6de13b39fd00292c1ce9bda307e9a904af1be6a2f4f99333878`. +- Validation: fixed 81-case subset; reward **0.810252**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466053 | + +Outcome: did not improve validation reward; retained current package. + +## E030 training launch + +Phase6 updates3000 and4000 did not beat update2000. Retired the phase6 trainer, its six data workers and watcher after all4000-update validation variants completed; all checkpoints and negative results remain available. Verified those processes exited before launching the next single-GPU trainer. Initialize528 guidance-adapter weights to zero on the latest selected package, preserving its physical operators and fixed auxiliary forecast. Sampling seed20261102, guidance initialization seed20261101; inherited neural initialization seed20261021 and auxiliary seed20261031. Sixteen-step loss capped by available trajectory length, batch6, learning rate0.00003 with cosine decay, noise0.01, group probabilities0.05/0.075/0.075/0.8, native thermal grids, planned6000 updates. Frozen auxiliary weights are copied into phase7/fixed_statistics. Exact parent/checkpoint hashes and stopped PIDs are in phase7/lineage.json. No pretrained assets. Validation pending; the selected package is unchanged. + +## Full validation of momentum_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780529**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E032 — Preserve observed angular constraints in state quadrature (preparation) + +Hypothesis: perturbing F4 changes an angular moment already constrained by the observed flow; rotating F6 and higher may preserve short-term accuracy while averaging latent uncertainty. Changed factor: first perturbed mode, holding the fixed operator and all learned weights fixed. Training study will compare the existing F4+ width3e-4 control with F6+ widths3e-4 and1e-3 on four audited TRAIN parameter regimes. Fixed latent seed20261026. No new trained weights; inherited checkpoint and physical hashes will be saved in latent_constraint_v1/source/model_manifest.json and parent.json. No validation results yet and no package change. Heavy physical study is queued after E029 expanded validation. + +## E033 — Shared coarse two-dimensional interaction forecast (preparation) + +Hypothesis: separate axis profiles cannot represent the spatial interaction component, which accounts for more than half of the remaining pressure and transverse-velocity normalized squared error in the current valid neural rollouts. Changed factor: add a shared horizon-conditioned MLP forecasting the interaction component of4×16 coarse fields. Every model branch will be evaluated on every case; complete physical prefixes remain authoritative. Train across all10115 audited TRAIN records and all19 datasets, random initialization seed20261103,12000 updates, batch512, width192, two CPU threads, group probabilities0.1/0.1/0.1/0.7. Inputs only four context fields/statistics and metadata; loss uses target full-field variance during training. Select updates2000/6000/12000 and fixed output blends0/.25/.5/1 on valid neural rollouts. No pretrained assets. Dataset VRMSE and checkpoint hashes pending. + +Training-cache/runtime feature parity passed exactly on one audited train example from each group. Zero-initialized forecasts equal last-frame coarse interactions exactly. Interpolation preserves global means and marginal profiles within float32 rounding. Sources and checks are retained under coarse_interaction_v1. No inference package change yet. + +## E029 — wider fixed state quadrature +- Hypothesis / changed factor: Increase the fixed hidden-state phase quadrature width from1e-4 to3e-4, retaining three equal nodes, the same fixed physical parameters, and latent seed20261026. Eight audited TRAIN trajectories supported the choice;1e-3 was rejected. No learned weights changed. Replay all twelve active VALID cases and verify30/30 physical steps, reuse69 unchanged outputs, and retain any newer validated neural/decoder package before scoring. Exact determinism and novel-grid checks precede publication. Inherited neural/auxiliary hashes and seeds are in the manifest. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812643**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E034 — late-checkpoint parameter averaging +- Hypothesis / changed factor: Equal parameter average of phase6 updates2000/3000/4000, a single predeclared candidate to reduce late-training optimization variation. No new optimization or data fitting; all three checkpoints were trained together across all19 datasets, initialization seed20261021 and sampling seed20261025. Accumulate in float64 and return original parameter dtype. Same shared averaged weights apply to every case. Preserve current fixed physical operators and shared auxiliary decoders. Source training checkpoint hashes and equal coefficients are in model_manifest.json and weight_average_v1/lineage.json. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact determinism and novel grids if improved. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812766**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.452162 | + +Outcome: improvement; published atomically. + +## E033a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed factors: training checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_interaction_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.25 | 0.45186873 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.5 | 0.45240534 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 1.0 | 0.45590097 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 6000 | 0.0 | 0.45216160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.25 | 0.45148496 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.5 | 0.45166160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 1.0 | 0.45455921 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 12000 | 0.0 | 0.45216160 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.25 | 0.45151015 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.5 | 0.45176523 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 1.0 | 0.45496820 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E035 — Periodic translation augmentation for the coarse branch (preparation) + +Hypothesis: a dense coarse-field predictor should see translated versions of periodic patterns during training. Changed factor: random whole-cell shifts along metadata-declared periodic axes, applied identically to all four context grids and the target. Keep architecture, initialization seed20261103, training sample sequence,12000 updates and all optimizer settings from E033. Use a separate augmentation RNG seed20261104. Train across all19 datasets on two CPU threads; no additional GPU trainer. Marginal means and variances are invariant to these cell permutations. Compare the replacement branch on the current shared neural parent with its previous coarse correction disabled, then replay the full proposed package. Validation results and checkpoint hashes pending. + +## E033 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Add one shared coarse interaction forecast branch, trained from random seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update6000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E033a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `ead6921e9fcb8924118472a881e78c2e0cf57f1a3b23d848b9ddaadfa36bc4ab`. +- Validation: fixed 81-case subset; reward **0.812821**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451485 | + +Outcome: improvement; published atomically. + +## E035a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed training factor: random coarse-cell translations on periodic axes with augmentation seed20261104, preserving initialization seed20261103 and the training sample sequence. Selection factors: checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_equivariant_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.25 | 0.45326987 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.5 | 0.45515343 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 1.0 | 0.46107193 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 6000 | 0.0 | 0.45216160 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.25 | 0.45202987 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.5 | 0.45258554 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 1.0 | 0.45570121 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 12000 | 0.0 | 0.45216160 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.25 | 0.45188915 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.5 | 0.45231514 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 1.0 | 0.45521523 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E030 — learned coarse-statistic guidance, update 1000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `240df3b587dc919b7d002c7fbcc51ab99b3510a1cedd49f0008c6e7a52b82764`. +- Validation: fixed 81-case subset; reward **0.812394**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.456839 | + +Outcome: did not improve validation reward; retained current package. + +## E035 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Replace the shared coarse interaction forecast with one trained using periodic-axis translations, augmentation seed20261104, retaining random initialization seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update12000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E035a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `7d340568043fdd29cfb0382ba3c1065e35f9172396a7f37b5a0809dedecf9f16`. +- Validation: fixed 81-case subset; reward **0.812788**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451889 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `34cea33fb726b0029c2c1734a94ceee72bd4e73bbf17e80cc9bc95a3db954165`. +- Validation: fixed 81-case subset; reward **0.811915**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.462924 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `d52a0fb2cbf3e4000004b0c088c4c1143ac393ce9c9432f1cc68270ac19d5778`. +- Validation: fixed 81-case subset; reward **0.811492**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.468370 | + +Outcome: did not improve validation reward; retained current package. + +## E036 — Cross-channel covariance features with continuation control (preparation) + +Hypothesis: spatial density/pressure correlations and other channel interactions contain dynamical information absent from separate means, variances and marginal profiles. Extend the selected E026 universal mean/profile branch with92928 zero-initialized covariance projection weights. Both a frozen-zero control and a learned-covariance variant continue from the same session-trained checkpoint, using identical samples and sampling seed20261105,6000 updates, batch512, learning rate0.0001 cosine, all19 datasets and two CPU threads each. Inherited random initialization seed20261031. Only difference between variants: whether covariance projection weights can learn. Four-frame context covariances use centered float64 accumulation, normalized by context channel scales. + +Exact parent-output parity passed on one audited TRAIN context from each group. Cache/runtime covariance differences were zero except1.06e-22 in one active-matter element. The cache contains only audited10115 TRAIN records; membership was checked before reading. Compare updates1000/3000/6000 by full neural runtime replay with the selected parent’s fixed output blends. Dataset VRMSE and checkpoint hashes pending. No inference parameter fitting and no pretrained assets. Sources, checks and logs: covariance_statistics_v1. + +## E036a — controlled statistical continuation and covariance feature comparison + +Continue the selected E026 shared auxiliary checkpoint with identical training samples, optimizer schedule and seed20261105 for6000 updates. Inherited random initialization seed20261031. Both variants add a zero-initialized covariance projection; control freezes it at zero, covariance learns it jointly. Added92928 weights. All19 datasets trained together, two CPU threads per variant, batch512, learning rate0.0001 cosine, existing mean/profile loss0.25/0.75. Four-frame spatial cross-channel covariances are normalized by context scales. Fixed parent decoder blends retained; every candidate is replayed through the full neural runtime, including learned guidance if present. No pretrained assets or inference-time fitting. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45146565 | 0.81282209 | `1c547715b94da883224d53fc40a4cc53362a65e59b78328ebb0debc00b7fca73` | +| control | 3000 | 0.45149569 | 0.81281968 | `8b0afaf1338754dccff24b1c7fd292b8b285fd665f7f003514b51403fadb99d9` | +| control | 6000 | 0.45166733 | 0.81280593 | `428e1327e4a4e51ffe0504f9ac9a2179f5c9a6e3c8883981687ab6e45fa45ea2` | +| covariance | 1000 | 0.45160459 | 0.81281096 | `d74d3af566e3c640d1209359df40ec59a9d7209fd9ed169e5186856e534807ef` | +| covariance | 3000 | 0.45153791 | 0.81281630 | `c7414bd840c80574dee726dacf7f41a28277a262da7e4663e779fe4cff50849d` | +| covariance | 6000 | 0.45171225 | 0.81280233 | `1544b181c5f74e3348bd67fd88f1f8154d5110aabf1c8ad62b721a32a60df690` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E036 — shared channel-covariance statistics +- Hypothesis / changed factor: Controlled continuation of the shared statistical branch, selected modecontrol update1000. Start from session-trained E026, add92928 zero-initialized covariance projection weights, and compare freezing them against learning them with the same seed20261105 and sample sequence. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seed20261031. Only the original four context frames supply channel covariances, means, variances and profiles. Same branch evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E036a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `254c11b3fd56624d3b695e25f7d980c48e04d399b76ef8111ce158e26fcaf483`. +- Validation: fixed 81-case subset; reward **0.812822**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Outcome: improvement; published atomically. + +## E036b — covariance hypothesis rejected by the matched control + +At every paired checkpoint, learned covariance features scored worse than the frozen-zero control: update1000 control0.451465647 vs covariance0.451604591;3000 control0.451495692 vs covariance0.451537907;6000 control0.451667333 vs covariance0.451712249 on radiative-layer valid VRMSE. The feature hypothesis is rejected. The control’s1000-update continuation has only a tiny improvement over the prior0.451484957 and was selected after exact neural/unknown/smoke/diffusion checks. Its covariance projection remains identically zero; no gain is attributed to the new features. Other dataset VRMSE and all checkpoint hashes are listed in E036a. No separate specialist models are used. + +## E037 — Resolved continuity feasibility diagnostic (preparation) + +Hypothesis: a fixed conservative density transport may explain or reduce residual density error. First test a second-order MUSCL/minmod, SSP-RK2 continuity update using temporally interpolated velocities on three audited TRAIN trajectories with low/medium/high cooling time. X periodic, Y open; metadata grid spacing/time, maximum summed Courant number0.4. Future TRAIN velocities are used only as an oracle feasibility diagnostic, never as an inference input or proposed package. No random seed or learned checkpoint; validation per dataset not evaluated. Results and source retained in continuity_v1. Proceed to forecast-velocity tests only if this diagnostic supports the operator. + +## E037a — continuity feasibility result: reject direct transport + +Using true future velocities from TRAIN trajectories only, the fixed conservative density update gives density-channel VRMSE: + +| Cooling time | Density VRMSE | +|---|---:| +| 0.03 | 0.16781393 | +| 0.32 | 0.18198222 | +| 3.16 | 0.41145309 | + +The high-cooling-time case remains inaccurate even with oracle velocities, so no forecast-velocity candidate or validation tuning is pursued. Additional TRAIN differential diagnostics found weak continuity alignment: correlations0.20–0.30 over early frames and0.37–0.54 over the full window. Swapping velocity components did not repair it. Fitted diagnostic scalars were not deployed. These results do not distinguish temporal sampling, numerical discretization and unresolved dynamics as causes. No package operator change; no learned checkpoint or random seed; other datasets were not evaluated. + +## E038 — complete rollout training preparation + +Hypothesis: matching the full scored horizon reduces the training/inference rollout gap. Prepare up to30 autoregressive steps, capped by corpus26/Arena10 available frames. Three microbatches of two accumulate gradients into effective batch6, freeing each graph before the next to fit one H100. Keep existing log1p variance-normalized loss, learning rate0.00003, noise0.01, group probabilities0.05/0.075/0.075/0.8 and25% initial-window sampling. Sampling seed20261106,4000 planned optimizer updates. Preserve whichever shared architecture and complete package is selected at launch. + +Queued transition policy: after all phase7/update3000 variants finish, move on if phase7 has supplied no selected main checkpoint; otherwise allow phase7 to complete6000 updates. Stop the previous trainer and workers and verify exit before starting phase8. No second GPU trainer is running. The new trainer supports both guided and unguided selected architectures. CPU guidance/runtime parity passed on an audited train window and its periodic translation, including covariance-capable statistical modules. No optimizer step or new model checkpoint has been produced yet. Sources: full_rollout_v1 and launch_phase8.py/watch_phase8.py. + +## Full validation of quadrature_wide_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779384**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Expanded fixed state quadrature from12 to all24 active validation trajectories. Replayed the missing12, reused12 quick outputs only after exact context equality, and verified30/30 common physical steps for every active case. The552 corpus/smoke outputs are reused from global2000 full validation after exact physical asset/configuration equality checks and complete physical coverage. The nine neural outputs come from this exact package’s quick validation with matching inputs. All585 cases scored together; no held-out trajectories or inference caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +E038 preparation check detail: the initial combined training-batch/horizon statistical call differed from separate inference calls by up to1.53e-5 in float32 physical units. No model was trained or published with that new trainer. Evaluating each case with the same horizon-batch shape as inference removed the rounding difference; the repeated two-case TRAIN check now has exact maximum difference0.0 for means and profiles. Evidence: full_rollout_v1/guidance_checks.json. + +## Full validation of weight_average_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779507**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.452162 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of coarse_interaction_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779561**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.451485 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of control_001000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779562**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E029c — restore narrow quadrature after full validation +- Hypothesis / changed factor: The larger VALID cohort reverses E029: width3e-4 activeVRMSE0.090264899 versus width1e-4 0.087454288 across all24 trajectories, despite improving the first12. Restore width1e-4 while preserving the current averaged main weights and all selected auxiliary branches. No retraining or new random parameters; fixed latent seed20261026. Exact physical source/config parity and complete576-prefix coverage justify reuse of the previously validated narrow physical outputs; nine neural outputs are unchanged, with exact quick/full context equality. All runtime assets remain self-contained. Select using the full585 reward rather than the misleading quick81 ranking; full group metrics and provenance are retained. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `81a0ca9ff1e18280e3f4ad44b69ab7cf2d0588dabff10f45b147f4b4049c852b`. +- Validation: fixed 81-case subset; reward **0.811403**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Selection basis: full585 physical-operator comparison: candidate0.78070730, current0.77956248 + +Outcome: improvement; published atomically. + +## Full validation of quadrature_revert_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780707**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Restore width1e-4 after all24 active VALID trajectories reversed the first12 ranking. Exact physical source/configuration equality with quadrature_candidate proves all576 complete physical outputs reusable. Copy24 narrower active outputs plus unchanged552 corpus/Arena outputs. Preserve all current main/auxiliary weights, code and neural output settings; copy their nine neural predictions only after quick/full context equality. No inference cache is packaged. Selection uses the larger585-cohort result, explicitly overriding the misleading81-case subset ranking. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +Selection-policy update after E029c: future physical-operator changes require full585 reports for both candidate and current model before record.py can publish them. Neural-only changes with identical physical assets/configuration may use the81-case comparison, since its nine radiative-layer cases are the entire valid pool and the physical outputs remain fixed. The active-state study E032 must compare against restored width1e-4 as well as its repeated wider control, and must validate all24 active cases before any publication. + +## E030 — learned coarse-statistic guidance, update 2000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `f2909b2ac6992e7c5328bfb609461f7733e7f9feba6bb41605dfed8d9fba67b5`. +- Validation: fixed 81-case subset; reward **0.810702**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.460297 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 2000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `6671cf5bf2c01b1a1d198b389bf3a10df15cd3a08319580a934432fce2caa979`. +- Validation: fixed 81-case subset; reward **0.810151**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.467359 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 2000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `8770d29e341f163d118460a6dfc5a5153ff09a1456fbb4b16399d8b1883e0443`. +- Validation: fixed 81-case subset; reward **0.810022**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469027 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E039 — equal averaging across the guidance extension +- Hypothesis / changed factor: Equal parameter average of the current selected shared main model and phase7/update1000, the better individual guided checkpoint so far. Extend the unguided parent with528 zero guide weights before averaging, preserving its function; accumulate in float64 and return original dtypes. This remains a single shared network evaluated on every case. No new optimization, pretrained assets or inference-time fitting. Source checkpoint hashes, equal coefficients and zero extension are in the manifest and mixed_average_v1/lineage.json. Inherited main initialization20261021, phase6 sampling20261025, guidance initialization20261101 and phase7 sampling20261102. Keep all selected physical operators and auxiliary branches. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact neural/novel-grid determinism before publication. +- Seed: 20261102. +- Checkpoint SHA256: `e4d1863c759fba48fd588f3e382c94733af1d529db1753d4f095d0134df7d643`. +- Runtime code SHA256: `d6ba1f84a29f2a2e52dbc9debda8f6a703fc3a12bb687fbeb6b69b8adb9b68c3`. +- Validation: fixed 81-case subset; reward **0.811432**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451106 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/README.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/README.md new file mode 100644 index 0000000000000000000000000000000000000000..2b68cfb94528369ea67d56b7ad84a3d4e0948ec0 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/README.md @@ -0,0 +1,35 @@ +# Shared neural and spectral surrogate + +A single shared neural operator is evaluated for every case and every autoregressive step. A deterministic spectral branch conditions on PDE metadata and uses independently implemented differential operators where their form was verified on training data. Unsupported equations use the shared neural update. There are no separately trained specialist checkpoints, no pretrained weights and no inference-time fitting. All runtime assets are included. + +Selection uses validation rollouts only. The neural model was trained from random initialization on all train pools, with sampling probabilities documented in EXPERIMENTS.md. See EXPERIMENTS.md for per-dataset results and negative experiments. Run `bash infer.sh INPUT_DIR OUTPUT_DIR`. + +Active matter also uses a fixed Fourier angular-moment Smoluchowski/Stokes operator. A recurrent context encoder inserts the observed low moments into its physical state; no parameter optimization or training occurs at inference. The Bingham initialization table contains mathematical Bessel ratios generated from scratch. Neural forecasts remain available as a fallback if a physical update is non-finite or violates positivity/moment bounds. The entry point pins the package directory through an open file descriptor so atomic publication cannot mix runtime assets. + +The active context encoder transports hidden angular moments with the supplied concentration and orientation, then applies a fixed linear Stokes observation layer and a necessary moment-positivity projection. It consumes only the four supplied frames. This is a fixed state computation; model and solver parameters do not change at inference. + +The active hidden state uses three fixed uncertainty quadrature nodes: its encoded center and symmetric small angular perturbations. Each node uses the same physical parameters and shared neural model; identical computations are shared. Outputs average the common valid physical prefix. A fixed local random generator makes the quadrature reproducible for every input, independent of case IDs and ordering. No separate specialist checkpoints or inference-time training are used. + +Buoyant smoke uses a fixed staggered-grid finite-difference fluid operator with semi-Lagrangian advection, explicit viscosity, and a pressure projection. Numerical spacing and scheme were identified using training trajectories only, with no generator code or initial-condition seeds. The implementation uses only PyTorch. + +Neural forecasts use a fixed, horizon-dependent shrinkage toward profiles averaged over periodic directions, with nonnegative density and concentration. The coefficient was selected on nine training rollouts before validation. The physical operators are unaffected. + +The shared neural operator evaluates long rectangular grids at native resolution. The checkpoint manifest and experiment ledger record training grid sizes, rollout horizon, sampling probabilities and checkpoint lineage. + +Reaction-diffusion updates retain a finer internal Fourier grid and use double-precision second-order exponential integration. The fixed observation operator maps cell averages to and from that state. Numerical choices were selected on training trajectories and tested on validation rollouts. CUDA graphs accelerate the fixed updates without changing their arithmetic. + +The Burgers and forced-vorticity updates also retain a finer internal Fourier grid in double precision. A fixed cell-averaging observation map returns the native output grid. Training-only resolution studies selected a twofold grid increase; smaller time steps and fourfold grids added little. + +The shared neural network includes a metadata-conditioned global Fourier feature layer. It was added with a zero output projection, preserving the parent model at initialization, then trained jointly across all four groups. New parameters were initialized from scratch in this session. + +A shared mean-forecast branch predicts channel spatial means from the four supplied frames, metadata, and requested horizon. It was initialized randomly and trained across all19 datasets. This same branch is evaluated on every case, including cases handled by fixed physical updates; no separately trained dataset specialists are selected. For neural forecasts, a fixed validation-selected blend adjusts the spatial means. Physical updates remain authoritative. The auxiliary checkpoint, parameter count, seed, and source hashes are recorded in model_manifest.json. + +The shared statistical branch also consumes channel profiles with 16 bins along each spatial axis, derived only from the supplied context. These profiles describe broad spatial structure that global means and variances omit. + +The same statistical branch jointly forecasts global means and coarse spatial profiles. A fixed blend corrects the autoregressive forecast’s broad spatial structure, using interpolation consistent with the supplied boundary conditions. Every case uses the same learned statistical weights. + +For density/pressure/velocity fields on a periodic-x and open-or-wall-y grid, a small fixed pressure correction uses the forecast momentum balance. A mixed-boundary gradient projection preserves the predicted global pressure mean. Its blend was selected on training trajectories and checked on validation; it uses only context and the model’s own forecast, with no parameter fitting at inference. + +The autoregressive network consumes fixed future mean/profile forecasts through a jointly trained guidance adapter. Those guide fields are computed only from the original four supplied frames. The auxiliary parameters remain fixed during inference. + +A shared coarse-field branch forecasts two-dimensional spatial interactions on a4 by16 grid from the four input frames, metadata and horizon. It was initialized randomly and trained across all19 datasets. Its fixed output blend preserves the existing global means and axis profiles, and complete physical prefixes remain authoritative. The same weights are evaluated on every case; no inference-time fitting occurs. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/arena.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/arena.py new file mode 100644 index 0000000000000000000000000000000000000000..370e9af2b204b0326adc20cb9b86a8874a1c8dbe --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/arena.py @@ -0,0 +1,39 @@ +"""Fixed MAC-grid buoyant-fluid update, independently implemented in PyTorch.""" +import math +import torch +import torch.nn.functional as F + +def eligible(meta,channels): + return channels==3 and str(meta.get('pde',''))=='Navier-Stokes: buoyant smoke (PDEArena)' and list(meta.get('fields',[]))==['u','vx','vy'] and 'buoyancy_y' in meta.get('constants',{}) + +class ArenaOperator: + def __init__(self,meta,x): + self.n,self.m=x.shape[-2:];self.device=x.device;self.dtype=torch.float64;x=x.to(self.dtype) + self.hx=float(meta['domain_extent'][0])*(self.n-1)/self.n**2;self.hy=float(meta['domain_extent'][1])*(self.m-1)/self.m**2 + self.interval=float(meta['dt'])*13/14;self.substeps=max(1,math.ceil(self.interval/1.5000001));self.dt=self.interval/self.substeps;self.nu=.01;self.buoyancy=float(meta['constants']['buoyancy_y']) + self.c=x[0].clone();self.u=F.pad(x[1],(0,0,0,1));self.v=F.pad(x[2],(0,1,0,0));device=self.device;dtype=self.dtype + ix=torch.arange(self.n,device=device,dtype=dtype);iy=torch.arange(self.m,device=device,dtype=dtype) + self.eig=(2*torch.cos(math.pi*ix/self.n)-2)[:,None]/self.hx**2+(2*torch.cos(math.pi*iy/self.m)-2)[None,:]/self.hy**2;self.eig[0,0]=1 + self.cx=(2/self.n)**.5*torch.cos(math.pi/self.n*ix[:,None]*(ix[None,:]+.5));self.cy=(2/self.m)**.5*torch.cos(math.pi/self.m*iy[:,None]*(iy[None,:]+.5));self.cx[0]*=2**-.5;self.cy[0]*=2**-.5 + self.coords={} + for name,shape,offset in [('c',(self.n,self.m),(.5,.5)),('u',(self.n+1,self.m),(0.,.5)),('v',(self.n,self.m+1),(.5,0.))]: + xx,yy=torch.meshgrid((torch.arange(shape[0],device=device,dtype=dtype)+offset[0])*self.hx,(torch.arange(shape[1],device=device,dtype=dtype)+offset[1])*self.hy,indexing='ij');self.coords[name]=(xx,yy,offset) + def sample(self,f,xx,yy,offset,mode='border'): + grid=torch.stack([2*(yy/self.hy-offset[1])/(f.shape[1]-1)-1,2*(xx/self.hx-offset[0])/(f.shape[0]-1)-1],-1) + return F.grid_sample(f[None,None],grid[None],mode='bilinear',padding_mode=mode,align_corners=True)[0,0] + def advect(self,f,name): + xx,yy,off=self.coords[name];u=self.sample(self.u,xx,yy,(0,.5));v=self.sample(self.v,xx,yy,(.5,0));return self.sample(f,xx-self.dt*u,yy-self.dt*v,off,'border' if name=='c' else 'zeros') + def diffuse(self,f): + z=F.pad(f,(1,1,1,1));return f+self.dt*self.nu*((z[2:,1:-1]+z[:-2,1:-1]-2*f)/self.hx**2+(z[1:-1,2:]+z[1:-1,:-2]-2*f)/self.hy**2) + def project(self,u,v): + u[0].zero_();u[-1].zero_();v[:,0].zero_();v[:,-1].zero_();div=(u[1:]-u[:-1])/self.hx+(v[:,1:]-v[:,:-1])/self.hy + ph=(self.cx@div@self.cy.T)/self.eig;ph[0,0].zero_();p=self.cx.T@ph@self.cy + u[1:-1]-=(p[1:]-p[:-1])/self.hx;v[:,1:-1]-=(p[:,1:]-p[:,:-1])/self.hy;return u,v + @torch.no_grad() + def advance(self): + for _ in range(self.substeps): + c=self.advect(self.c,'c');u=self.advect(self.u,'u');v=self.advect(self.v,'v');v[:,1:-1]+=self.dt*self.buoyancy*.5*(c[:,1:]+c[:,:-1]);u=self.diffuse(u);v=self.diffuse(v);self.u,self.v=self.project(u,v);self.c=c + return torch.stack([self.c,self.u[:-1],self.v[:,:-1]]).float() + + def step(self,x): + return self.advance() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/coarse_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/coarse_model.py new file mode 100644 index 0000000000000000000000000000000000000000..fc93ff2927589c947bd9a2ee18251f3a3eb5d26d --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/coarse_model.py @@ -0,0 +1,86 @@ +"""A universal coarse spatial interaction forecast, learned from scratch.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + +GRID = (4, 16) + + +def interaction(x): + return x - x.mean(-1, keepdim=True) - x.mean(-2, keepdim=True) + x.mean((-2, -1), keepdim=True) + + +class CoarseForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4 * GRID[0] * GRID[1]) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, MAXC * GRID[0] * GRID[1]), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, grids=None): + b, t, c = means.shape + assert t == 4 and c <= MAXC and grids is not None + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = ((means - last[:, None]) / scale[:, None]).clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + centered = (grids - means[:, :, :, None, None]) / scale[:, None, :, None, None] + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale) * mask / 5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + F.pad(centered.clamp(-100, 100), (0, 0, 0, 0, 0, MAXC-c)).flatten(1), + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + delta = self.net(torch.cat([features, md, self.ds(ids), time], 1)).reshape(b, MAXC, *GRID)[:, :c] + future = grids[:, -1] + scale[:, :, None, None] * delta * h[:, :, None, None] + return interaction(future) + + +@torch.no_grad() +def forecast_coarse(model, context, md, ids, steps): + x = context.double() + mean = x.mean((-2, -1)).float() + variance = x.var((-2, -1), unbiased=False).float() + grid = F.adaptive_avg_pool2d(x, GRID).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + grids=grid[None].expand(steps, -1, -1, -1, -1)) + + +def resize_axis(x, size, axis, periodic): + old = x.shape[axis] + position = (torch.arange(size, device=x.device, dtype=x.dtype) + .5) * old / size - .5 + left = position.floor().long() + weight = position - left + a, b = (left.remainder(old), (left+1).remainder(old)) if periodic else (left.clamp(0, old-1), (left+1).clamp(0, old-1)) + shape = [1] * x.ndim + shape[axis] = size + return x.index_select(axis, a) * (1-weight.reshape(shape)) + x.index_select(axis, b) * weight.reshape(shape) + + +@torch.no_grad() +def correct_coarse(prediction, forecast, weight, bc): + raw = prediction.double() + current = interaction(F.adaptive_avg_pool2d(raw, GRID)) + future = torch.where(torch.isfinite(forecast), forecast.double(), current) + difference = interaction(future - current) + difference = resize_axis(difference, raw.shape[-2], -2, bc[0]) + difference = resize_axis(difference, raw.shape[-1], -1, bc[1]) + # Preserve all existing global means and axis profiles exactly in real arithmetic. + return (raw + weight * interaction(difference)).float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/config.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/config.json new file mode 100644 index 0000000000000000000000000000000000000000..fbbc8b083c22ebf7f050859fb19079f62bc69522 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/config.json @@ -0,0 +1,52 @@ +{ + "width": 48, + "depth": 2, + "detail_decay": 0.8, + "physics": true, + "active_tau": 10.0, + "kinetic": { + "modes": 32, + "resolution": 256, + "iterations": 6, + "cuda_graph": true, + "transport": "phase", + "observer_weight": 1.0, + "observer_ridge": 0.01, + "observer_iterations": 12, + "quadrature_scale": 0.0001, + "quadrature_seed": 20261026 + }, + "arena": true, + "neural_regularization": { + "profile_mix": 0.75, + "horizon": 30, + "pressure_mix": 0.0, + "pressure_start": 10, + "pressure_end": 30 + }, + "spectral_options": { + "gs_scale": 2, + "gs_dt": 0.25, + "gs_order": 2, + "gs_double": true, + "cuda_graph": true, + "advection_scale": 2, + "advection_double": true + }, + "native_rectangular": true, + "global_mixing": true, + "mean_forecast": { + "weight": 0.5, + "profile_weight": 0.25, + "width": 192, + "covariance_features": false + }, + "momentum_pressure": { + "weight": 0.05 + }, + "coarse_forecast": { + "width": 192, + "weight": 0.25 + }, + "statistic_guidance": true +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/infer.sh b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/infer.sh new file mode 100644 index 0000000000000000000000000000000000000000..bbf5ac5eb2acfbbad07b871df5e8d072b2304862 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/infer.sh @@ -0,0 +1,3 @@ +#!/bin/bash +set -euo pipefail +exec python "$(dirname "$0")/launch.py" "$@" diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/kinetic.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/kinetic.py new file mode 100644 index 0000000000000000000000000000000000000000..55627fc359e6f3c253b92bd010bad61558052e27 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/kinetic.py @@ -0,0 +1,218 @@ +"""Portable, fixed-parameter angular moment operator; PyTorch JIT accelerates inference.""" +import math +from pathlib import Path +from typing import Tuple +import numpy as np +import torch +torch._C._jit_set_texpr_fuser_enabled(False) +torch._C._jit_override_can_fuse_on_cpu(False) +torch._C._jit_override_can_fuse_on_gpu(False) +torch._C._jit_set_profiling_executor(False) + +@torch.jit.script +def flow(s:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor,torch.Tensor]: + c=s[0].real;q=s[1].real*.5;r=s[1].imag*.5;sr=s[2].real;si=s[2].imag + base0=2*alpha*q-.8*zeta*((c-sr)*q-si*r);base1=alpha*r-.4*zeta*((c+sr)*r-si*q);pc=.1*c.mean();ps=psi + for _ in range(iterations): + e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]) + sig0=base0+.4*((c+sr)*e0+si*e1);sig1=base1+.2*(si*e0+(c-sr)*e1) + sf=torch.fft.rfft2(torch.stack([sig0,sig1]));raw=(kx*ky*sf[0]+(ky*ky-kx*kx)*sf[1])*inv*inv*mask + ps=(raw+pc*ps)/(1+pc) + vel=torch.fft.irfft2(torch.stack([1j*ky*ps,-1j*kx*ps]),s=[nx,ny]);e0=torch.fft.irfft2(-kx*ky*ps,s=[nx,ny]);e1=torch.fft.irfft2(.5*(kx*kx-ky*ky)*ps,s=[nx,ny]);om=torch.fft.irfft2((kx*kx+ky*ky)*ps,s=[nx,ny]) + return vel,e0,e1,om,ps + +@torch.jit.script +def rhs(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,alpha:float,zeta:float,iterations:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + s=torch.fft.ifft2(v);vel,e0,e1,om,ps=flow(s,psi,khx,khy,inv,maskh,alpha,zeta,iterations,nx,ny) + dx=torch.fft.ifft2(1j*kx*v);dy=torch.fft.ifft2(1j*ky*v);out=-vel[0]*dx-vel[1]*dy + A=e1+zeta*s[1].imag;B=-e0-zeta*s[1].real + out=out+1j*an*.5*om*s + out[:-1]=out[:-1]+.5j*an[:-1]*(A-1j*B)*s[1:] + out[1:]=out[1:]+.5j*an[1:]*(A+1j*B)*s[:-1] + out[0]=out[0].real.to(out.dtype) + nf=torch.fft.fft2(out)*mask;nf[0,0,0].zero_() + return nf,ps + +@torch.jit.script +def integrate(v:torch.Tensor,psi:torch.Tensor,kx:torch.Tensor,ky:torch.Tensor,khx:torch.Tensor,khy:torch.Tensor,inv:torch.Tensor,mask:torch.Tensor,maskh:torch.Tensor,an:torch.Tensor,e:torch.Tensor,e2:torch.Tensor,q:torch.Tensor,f1:torch.Tensor,f2:torch.Tensor,f3:torch.Tensor,alpha:float,zeta:float,iterations:int,substeps:int,nx:int,ny:int)->Tuple[torch.Tensor,torch.Tensor]: + ps=psi + for _ in range(substeps): + nv,ps=rhs(v,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + a=e2*v+q*nv;na,ps=rhs(a,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + b=e2*v+q*na;nb,ps=rhs(b,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + cc=e2*a+q*(2*nb-nv);nc,ps=rhs(cc,ps,kx,ky,khx,khy,inv,mask,maskh,an,alpha,zeta,iterations,nx,ny) + v=e*v+f1*nv+2*f2*(na+nb)+f3*nc + return v,ps + +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if n0).all();moment=((y[3]-y[6]).square()+4*y[4].square()<=1.21*y[0].square()).all() + if not (finite and positive and moment):break + out.append(y) + return torch.stack(out) if out else self.context.new_empty((0,11,*self.original_shape)) + +"""Fixed linear Stokes observation layer for physical hidden moments, no parameter fitting.""" +import torch + +class _FlowObservedOperator(_MomentOperator): + def __init__(self,meta,context,observer_weight=1.,observer_ridge=.01,observer_iterations=12,**kwargs): + super().__init__(meta,context,**kwargs);self.observer_weight=observer_weight;self.observer_ridge=observer_ridge;self.observer_iterations=observer_iterations;self.observer_stats=[] + self.r0=self.khx*self.khy*self.inv*self.maskh;self.r1=.5*(self.khy.square()-self.khx.square())*self.inv*self.maskh + def observe(self,x): + super().observe(x) + s=torch.fft.ifft2(self.v);c=x[0];f2=s[1];q=.5*f2.real;r=.5*f2.imag;e0=x[7];e1=x[8] + a=.8*self.zeta*q+.4*e0;b=.8*self.zeta*r+.4*e1 + const0=2*self.alpha*q-.8*self.zeta*c*q+.4*c*e0;const1=self.alpha*r-.4*self.zeta*c*r+.2*c*e1 + def R(z,m):return torch.fft.irfft2(torch.fft.rfft2(z)*m,s=self.shape) + def A(z):return R(a*z[0]+b*z[1],self.r0)+R(-b*z[0]+a*z[1],self.r1) + def AT(z): + u=R(z,self.r0);v=R(z,self.r1);return torch.stack([a*u-b*v,b*u+a*v]) + obs=torch.fft.irfft2((self.khx.square()+self.khy.square())*self.psi*self.maskh,s=self.shape);prior=torch.stack([s[2].real,s[2].imag]);residual=obs-R(const0,self.r0)-2*R(const1,self.r1)-A(prior) + ridge=(.25*(a.square()+b.square()).mean()*self.observer_ridge).clamp_min(1e-8) + solution=torch.zeros_like(residual);rr=residual.clone();direction=rr.clone();norm=rr.square().sum() + for _ in range(self.observer_iterations): + action=A(AT(direction))+ridge*direction;alpha=norm/(direction*action).sum().clamp_min(1e-20);solution=solution+alpha*direction;rr=rr-alpha*action;nextnorm=rr.square().sum();direction=rr+nextnorm/norm.clamp_min(1e-20)*direction;norm=nextnorm + correction=AT(solution)*self.observer_weight;estimate=prior+correction;f4=estimate[0]+1j*estimate[1] + # Necessary Toeplitz moment positivity: |F4-F2^2/c| <= c-|F2|^2/c. + center=f2.square()/c.clamp_min(1e-8);radius=(c-f2.abs().square()/c.clamp_min(1e-8)).clamp_min(0);delta=f4-center;f4=center+delta*torch.minimum(torch.ones_like(radius),radius/delta.abs().clamp_min(1e-10)) + after=residual-A(torch.stack([f4.real,f4.imag])-prior);self.observer_stats.append({'before':float(residual.square().mean().sqrt()),'after':float(after.square().mean().sqrt()),'correction_rms':float(correction.square().mean().sqrt())});s[2]=f4;self.v.copy_(torch.fft.fft2(s)) + +"""Geometric transport of latent angular moments when observed low moments arrive.""" +from pathlib import Path +import numpy as np,torch + +class KineticOperator(_FlowObservedOperator): + def __init__(self,meta,context,transport='phase',**kwargs): + super().__init__(meta,context,**kwargs);self.transport=transport;self.table=torch.as_tensor(np.load(Path(__file__).with_name('bingham_table.npy')),device=context.device) + def ratios(self,mu): + pos=(-torch.log1p(-mu.clamp(0,1-1e-6))).clamp_max(12)*(self.table.shape[1]-1)/12;i=pos.long().clamp_max(self.table.shape[1]-2);w=pos-i;rows=torch.arange(self.M+1,device=mu.device)[:,None,None];return self.table[rows,i]*(1-w)+self.table[rows,i+1]*w + def observe(self,x): + s=torch.fft.ifft2(self.v);c0=s[0].real.clamp_min(1e-8);c1=x[0].clamp_min(1e-8);old=s[1];new=(x[3]-x[6])+1j*(x[4]+x[5]);mu0=old.abs()/c0;mu1=new.abs()/c1 + confidence=(torch.minimum(mu0,mu1)/.1).clamp_max(1);angle=torch.angle(new*old.conj())*confidence + gain=(c1/c0)[None]*torch.exp(1j*(self.an*.5)*angle) + if self.transport=='bingham': + rat0=self.ratios(mu0);rat1=self.ratios(mu1);ratio=(rat1/rat0.clamp_min(1e-8)).clamp(.5,2.);gain=gain*(1+confidence[None]*(ratio-1)) + s[2:]=s[2:]*gain[2:];self.v.copy_(torch.fft.fft2(s));super().observe(x) + +"""Append to a fixed moment operator to integrate three hidden-state quadrature nodes.""" +_EncodedMomentOperator = KineticOperator + +class KineticOperator(_EncodedMomentOperator): + def __init__(self,meta,context,quadrature_scale=0.,quadrature_seed=20261026,**kwargs): + super().__init__(meta,context,**kwargs) + self.quadrature_scale=float(quadrature_scale);self.quadrature_seed=int(quadrature_seed) + self.quadrature_length=float(meta['constants']['L'])/30 + @torch.no_grad() + def rollout(self,steps,guard=False): + if self.quadrature_scale<=0:return super().rollout(steps,guard) + for j in range(1,4): + self.advance() + if guard and not torch.isfinite(self.v).all():return self.context.new_empty((0,11,*self.original_shape)) + self.observe(self.context[j]) + initial=self.v.clone();psi=self.psi.clone();state=torch.fft.ifft2(initial) + rng=torch.Generator(device=self.device);rng.manual_seed(self.quadrature_seed) + noise=torch.randn(self.shape,device=self.device,generator=rng) + filt=torch.exp(-.5*(self.khx.square()+self.khy.square())*self.quadrature_length**2) + noise=torch.fft.irfft2(torch.fft.rfft2(noise)*filt,s=self.shape) + scale=noise.std() if noise.numel()>1 else noise.new_tensor(1.) + noise=noise/scale.clamp_min(1e-12) + forecasts=[] + for sign in [0.,-1.,1.]: + s=state.clone();s[2:]*=torch.exp(1j*(self.an[2:]*.5)*noise[None]*(sign*self.quadrature_scale)) + c=s[0].real.clamp_min(1e-8);center=s[1].square()/c;radius=(c-s[1].abs().square()/c).clamp_min(0) + delta=s[2]-center;s[2]=center+delta*torch.minimum(torch.ones_like(radius),radius/delta.abs().clamp_min(1e-10)) + self.v.copy_(torch.fft.fft2(s));self.psi.copy_(psi);out=[] + for _ in range(steps): + self.advance();y=self.decode() + if guard: + finite=torch.isfinite(y).all();positive=(y[0]>0).all();moment=((y[3]-y[6]).square()+4*y[4].square()<=1.21*y[0].square()).all() + if not (finite and positive and moment):break + out.append(y) + forecasts.append(torch.stack(out) if out else self.context.new_empty((0,11,*self.original_shape))) + used=min(len(y) for y in forecasts) + return (forecasts[0][:used]+forecasts[1][:used]+forecasts[2][:used])/3 diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/make_bingham_table.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/make_bingham_table.py new file mode 100644 index 0000000000000000000000000000000000000000..3cba92fb5f0503306c910a27c10b6e42c7099aa6 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/make_bingham_table.py @@ -0,0 +1,9 @@ +import numpy as np +from scipy.special import ive +from pathlib import Path +s=np.linspace(0,12,8193);mu=-np.expm1(-s);r=mu[1:];k=np.where(r<.53,2*r+r**3+5*r**5/6,np.where(r<.85,-.4+1.39*r+.43/(1-r),1/(r**3-4*r**2+3*r))) +for _ in range(6): + z=ive(1,k)/ive(0,k);k=np.maximum(k+(r-z)/np.maximum(1-z*z-z/k,1e-14),1e-10) +t=np.zeros((65,len(s)),np.float32);t[0]=1 +for j in range(1,65):t[j,1:]=ive(j,k)/ive(0,k) +np.save(Path(__file__).with_name('bingham_table.npy'),t) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/mean_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/mean_model.py new file mode 100644 index 0000000000000000000000000000000000000000..264dc34336192462ea2ddb804fa45875a4129b01 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/mean_model.py @@ -0,0 +1,107 @@ +"""Shared context-statistic forecast; all parameters learned from random initialization.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + + +class MeanForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.covariates = nn.Linear(4*MAXC*MAXC,width,bias=False) + nn.init.zeros_(self.covariates.weight) + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4*2*16) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), nn.Linear(width, MAXC*33), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, profiles=None, covariances=None): + # One independently specified horizon per row. All supplied statistics + # describe only the four context frames; no inference-time fitting. + b, t, c = means.shape + assert t == 4 and c <= MAXC + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = (means - last[:, None]) / scale[:, None] + relative = relative.clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + assert profiles is not None + last_profiles=profiles[:,-1] + profiles=(profiles-means[:,:,:,None,None])/scale[:,None,:,None,None] + profiles=F.pad(profiles.clamp(-100,100),(0,0,0,0,0,MAXC-c)).flatten(1) + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale)*mask/5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + profiles, + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + hidden=self.net[0](torch.cat([features,md,self.ds(ids),time],1)) + if covariances is not None: + covariance=covariances/scale[:,None,:,None]/scale[:,None,None,:] + covariance=F.pad(covariance.clamp(-100,100),(0,MAXC-c,0,MAXC-c)).flatten(1) + hidden=hidden+self.covariates(covariance) + delta=self.net[1:](hidden).reshape(b,MAXC,33)[:,:c] + future_mean=last+scale*delta[:,:,0]*h + future_profiles=last_profiles+scale[:,:,None,None]*delta[:,:,1:].reshape(b,c,2,16)*h[:,:,None,None] + future_profiles=future_profiles-future_profiles.mean(-1,keepdim=True)+future_mean[:,:,None,None] + return future_mean,future_profiles + + +@torch.no_grad() +def forecast_statistics(model, context, md, ids, steps): + mean = context.double().mean((-2, -1)).float() + variance = context.double().var((-2, -1), unbiased=False).float() + profx=F.adaptive_avg_pool2d(context.double(),(16,1)).squeeze(-1) + profy=F.adaptive_avg_pool2d(context.double(),(1,16)).squeeze(-2) + profiles=torch.stack([profx,profy],-2).float() + flattened=context.double().flatten(-2) + centered=flattened-flattened.mean(-1,keepdim=True) + covariance=(centered@centered.transpose(-2,-1)/centered.shape[-1]).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + profiles=profiles[None].expand(steps,-1,-1,-1,-1), + covariances=covariance[None].expand(steps,-1,-1,-1)) + + +@torch.no_grad() +def forecast_means(model,context,md,ids,steps): + return forecast_statistics(model,context,md,ids,steps)[0] + +def interpolate_profile(profile,size,periodic): + position=(torch.arange(size,device=profile.device,dtype=profile.dtype)+.5)*16/size-.5 + left=position.floor().long();weight=position-left + if periodic: + i=left.remainder(16);j=(left+1).remainder(16) + else: + i=left.clamp(0,15);j=(left+1).clamp(0,15) + output=profile[...,i]*(1-weight)+profile[...,j]*weight + return output-output.mean(-1,keepdim=True) + +@torch.no_grad() +def correct_statistics(prediction,means,profiles,mean_weight,profile_weight,bc): + raw=prediction.double();center=raw.mean((-2,-1)) + means=torch.where(torch.isfinite(means),means.double(),center) + output=raw+mean_weight*(means-center)[:,:,None,None] + if profile_weight: + current_x=F.adaptive_avg_pool2d(raw,(16,1)).squeeze(-1) + current_y=F.adaptive_avg_pool2d(raw,(1,16)).squeeze(-2) + current=torch.stack([current_x,current_y],-2) + forecast=torch.where(torch.isfinite(profiles),profiles.double(),current) + difference=(forecast-forecast.mean(-1,keepdim=True))-(current-current.mean(-1,keepdim=True)) + dx=interpolate_profile(difference[:,:,0],raw.shape[-2],bc[0]) + dy=interpolate_profile(difference[:,:,1],raw.shape[-1],bc[1]) + output=output+profile_weight*(dx[:,:,:,None]+dy[:,:,None,:]) + return output.float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/model.py new file mode 100644 index 0000000000000000000000000000000000000000..9b07dcabb694baa1a87e687419c75951dcacbb75 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/model.py @@ -0,0 +1,116 @@ +"""Shared metadata-conditioned convolutional operator, trained from scratch.""" +import math +import numpy as np +import torch +from torch import nn +import torch.nn.functional as F + +DATASETS=['ape2d/burgers','ape2d/decay_turb','ape2d/diff','ape2d/fisher','ape2d/gs_alpha','ape2d/gs_beta','ape2d/gs_delta','ape2d/gs_epsilon','ape2d/gs_gamma','ape2d/gs_iota','ape2d/gs_kappa','ape2d/gs_theta','ape2d/kdv','ape2d/kolm_flow','ape2d/ks','ape2d/sh','pdearena/navier_stokes_2d','well/active_matter','well/turbulent_radiative_layer_2D'] +CONSTANTS=['Viscosity','Viscosity X','Viscosity Y','Diffusivity','Reactivity','Feed Rate','Kill Rate','Critical Number','Domain Extent','tcool','L','zeta','alpha','buoyancy_y'] +MAXC=11 + +def metadata(meta,dataset): + vals=[math.log(max(float(meta.get('dt',1)),1e-12))/5] + vals += [math.log(max(float(v),1e-12))/5 for v in meta.get('domain_extent',[1,1])] + bc=meta.get('boundary_conditions',['periodic']*4) + vals += [float(b=='periodic') for b in bc]+[float(b=='wall') for b in bc] + const=meta.get('constants',{}) + for k in CONSTANTS: + v=float(const.get(k,0)); vals += [math.copysign(math.log1p(abs(v)*100),v)/8,float(k in const)] + return np.array(vals,np.float32),DATASETS.index(dataset)+1 if dataset in DATASETS else 0,(bc[0]=='periodic',bc[2]=='periodic') + +METADIM=3+8+2*len(CONSTANTS) + +def pad(x,n,bc): + x=F.pad(x,(n,n,0,0),mode='circular' if bc[1] else 'replicate') + return F.pad(x,(0,0,n,n),mode='circular' if bc[0] else 'replicate') + +class Block(nn.Module): + def __init__(self,c,cond): + super().__init__(); self.dw=nn.Conv2d(c,c,5,groups=c); self.norm=nn.LayerNorm(c); self.fc1=nn.Linear(c,2*c);self.fc2=nn.Linear(2*c,c) + self.film=nn.Linear(cond,2*c); nn.init.zeros_(self.film.weight);nn.init.zeros_(self.film.bias) + self.scale=nn.Parameter(torch.ones(c)*0.1) + def forward(self,x,e,bc): + h=self.dw(pad(x,2,bc)).permute(0,2,3,1);h=self.norm(h) + g,b=self.film(e).chunk(2,-1);h=h*(1+g[:,None,None,:])+b[:,None,None,:] + h=self.fc2(F.gelu(self.fc1(h)))*self.scale + return x+h.permute(0,3,1,2) + +class GlobalMix(nn.Module): + def __init__(self,width,cond=192,latent=24,modes_x=32,modes_y=96): + super().__init__();self.latent=latent;self.mx=modes_x;self.my=modes_y + self.input=nn.Conv2d(width,latent,1);self.output=nn.Conv2d(latent,width,1) + self.weight=nn.Parameter(torch.randn(2,latent,modes_x,modes_y,2)*0.5) + self.condition=nn.Linear(cond,latent*2);nn.init.zeros_(self.condition.weight);nn.init.zeros_(self.condition.bias) + nn.init.zeros_(self.output.weight);nn.init.zeros_(self.output.bias) + def forward(self,x,e,bc): + h,w=x.shape[-2:];q=self.input(x) + # Even extensions impose the same zero-normal-gradient convention as replicated padding. + if not bc[0]:q=torch.cat([q,q.flip(-2)],-2) + if not bc[1]:q=torch.cat([q,q.flip(-1)],-1) + xf=torch.fft.rfft2(q.float(),norm='ortho');yf=torch.zeros_like(xf) + nx=min(self.mx,q.shape[-2]//2);ny=min(self.my,xf.shape[-1]) + weight=torch.view_as_complex(self.weight.contiguous());gate=torch.view_as_complex(self.condition(e).float().reshape(len(e),self.latent,2).contiguous())[:, :, None, None] + yf[:,:,:nx,:ny]=xf[:,:,:nx,:ny]*weight[0,:,:nx,:ny]*(1+gate) + yf[:,:,-nx:,:ny]=xf[:,:,-nx:,:ny]*weight[1,:,-nx:,:ny]*(1+gate) + z=torch.fft.irfft2(yf,s=q.shape[-2:],norm='ortho')[:,:,:h,:w] + return self.output(F.gelu(z)) + +class Operator(nn.Module): + def __init__(self,width=48,depth=2): + super().__init__(); self.width=width;self.depth=depth;d=192;ws=[width,width*2,width*4,width*6] + self.ds=nn.Embedding(len(DATASETS)+1,32) + self.meta=nn.Sequential(nn.Linear(METADIM+32+MAXC*3,d),nn.SiLU(),nn.Linear(d,d),nn.SiLU()) + self.stem=nn.Conv2d(4*MAXC+2,ws[0],3);self.global_mix=GlobalMix(ws[0],d) + self.guide_stem=nn.Conv2d(MAXC,ws[0],1,bias=False);nn.init.zeros_(self.guide_stem.weight) + self.enc=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws]) + self.down=nn.ModuleList([nn.Conv2d(ws[i],ws[i+1],2,stride=2) for i in range(3)]) + self.up=nn.ModuleList([nn.Conv2d(ws[i+1]+ws[i],ws[i],1) for i in range(3)]) + self.dec=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws[:-1]]) + self.head=nn.Conv2d(ws[0],MAXC,1);nn.init.zeros_(self.head.weight);nn.init.zeros_(self.head.bias) + def forward(self,x,md,ids,bc=(True,True),guidance=None): + # Reversible feature normalization is computed from supplied states, with fixed learned parameters. + b,t,c,h,w=x.shape + mean=x[:,-1].mean((-2,-1),keepdim=True) + std=x.std((1,3,4),unbiased=False).clamp_min(1e-3)[:,:,None,None] + z=(x-mean[:,None])/std[:,None] + mask=torch.ones(b,c,device=x.device) + sm=torch.cat([torch.asinh(mean.flatten(1)),torch.log(std.flatten(1))/5,mask],1) + if c0 for v in [dt,lx,ly]):return prediction + state=torch.cat([context,prediction]).double();density,pressure,vx,vy=[fields.index(name) for name in names] + rho=state[:,density];u=state[:,vx];v=state[:,vy];nx,ny=rho.shape[-2:];dx=lx/nx;dy=ly/ny + ddx=lambda value:(torch.roll(value,-1,-2)-torch.roll(value,1,-2))/(2*dx) + force_x=-gradient(rho*u,dt,0)-ddx(rho*u*u)-gradient(rho*u*v,dy,-1) + force_y=-gradient(rho*v,dt,0)-ddx(rho*u*v)-gradient(rho*v*v,dy,-1) + faces_x=.5*(force_x+torch.roll(force_x,-1,-2));faces_y=.5*(force_y[...,:-1]+force_y[...,1:]) + estimate=project_faces(faces_x,faces_y,dx,dy)[len(context):] + original=prediction[:,pressure].double();center=original.mean((-2,-1),keepdim=True) + updated=(1-weight)*original+weight*(estimate+center) + result=prediction.clone();result[:,pressure]=torch.where(torch.isfinite(updated),updated,original).float() + return result diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/numerics.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/numerics.py new file mode 100644 index 0000000000000000000000000000000000000000..3c45e36262df2d382b1be3a53691ec656a37ca9a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S091/numerics.py @@ -0,0 +1,21 @@ +"""Portable periodic Fourier resampling with even-Nyquist handling.""" +import torch +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if nMAXC: + return np.concatenate([predict(model,x[:,i:i+MAXC].cpu().numpy(),meta,ds,steps,device,detail_decay,physics,active_tau,kinetic_cfg,arena,neural_regularization,native_rectangular,spectral_options) for i in range(0,c,MAXC)],1) + factor=2 if min(h,w)>=192 or min(h,w)>=128 and max(h,w)/min(h,w)>2 and not native_rectangular else 1 + z=F.avg_pool2d(x,factor) if factor>1 else x + md,di,bc=metadata(meta,ds);md=torch.tensor(md,device=device)[None];ids=torch.tensor([di],device=device) + future_means,future_profiles=forecast_statistics(model.mean_forecast,x,md,ids,steps) if hasattr(model,'mean_forecast') else (None,None) + future_coarse=forecast_coarse(model.coarse_forecast,x,md,ids,steps) if hasattr(model,'coarse_forecast') else None + kinetic_steps=0 + detail=x[-1]-F.interpolate(z[-1,None],size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else None + ctx=z[None];out=[] + solver=SpectralUpdate(meta,(h,w),device,**(spectral_options or {})) if physics and eligible(meta,c) else None + if arena and arena_eligible(meta,c) and min(h,w)>1:solver=ArenaOperator(meta,x[-1]) + fullstate=x[-1];physical_steps=0;physical_name=type(solver).__name__ if solver is not None else None + for k in range(steps): + with torch.autocast(device_type=device,dtype=torch.bfloat16,enabled=device=='cuda'):y=model(ctx,md,ids,bc,guidance=(future_means[k:k+1],future_profiles[k:k+1]) if future_means is not None else None) + if not torch.isfinite(y).all():y=ctx[:,-1].clone() + if solver is not None: + candidate=solver.step(fullstate) + if torch.isfinite(candidate).all(): + physical_steps+=1 + fullstate=candidate;y=(F.avg_pool2d(candidate,factor) if factor>1 else candidate)[None] + else:solver=None + ctx=torch.cat([ctx[:,1:],y[:,None]],1) + yh=F.interpolate(y,size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else y[0] + if detail is not None:yh=yh+detail*(detail_decay**(k+1)) + if solver is not None:yh=fullstate + yh=torch.where(torch.isfinite(yh),yh,x[-1]) + out.append(yh.cpu().numpy()) + pred=np.stack(out).astype(np.float32) + if physical_name:print(f'physical steps: {physical_name} {physical_steps}/{steps}',flush=True) + if neural_regularization is not None and solver is None: + raw=torch.as_tensor(pred,device=device);pred=regularize(raw,meta,**neural_regularization).cpu().numpy() + if future_means is not None and solver is None: + pred=correct_statistics(torch.as_tensor(pred,device=device),future_means,future_profiles,model.mean_weight,model.profile_weight,bc).cpu().numpy() + if active_tau is not None: + raw=torch.as_tensor(pred,device=device);projected=project_active(x,raw,meta,active_tau) + pred=torch.where(torch.isfinite(projected),projected,raw).cpu().numpy() + if kinetic_cfg and str(meta.get('pde',''))=='active_matter' and c==11 and list(meta.get('fields',[]))==['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3'] and all(k in meta.get('constants',{}) for k in ['L','alpha','zeta']): + from kinetic import KineticOperator + cfg=dict(kinetic_cfg) + try: + op=KineticOperator(meta,x,**cfg) + except RuntimeError: + if not cfg.get('cuda_graph',False):raise + cfg['cuda_graph']=False;op=KineticOperator(meta,x,**cfg) + physical=op.rollout(steps,guard=True);used=len(physical);kinetic_steps=used + if used:pred[:used]=physical.cpu().numpy() + print(f'kinetic frames: {used}/{steps}',flush=True) + if getattr(model,'momentum_pressure_weight',0): + pred=pressure_projection(x,torch.as_tensor(pred,device=device),meta,model.momentum_pressure_weight).cpu().numpy() + if future_coarse is not None and solver is None: + prefix=max(physical_steps,kinetic_steps) + if prefix1;self.initialized=False + self.double=gs_double if meta['pde']=='Gray Scott' else advection_double if advective else False + self.real_dtype=torch.float64 if self.double else torch.float32 + self.complex_dtype=torch.complex128 if self.double else torch.complex64 + if self.fine:shape=tuple(int(scale*s) for s in shape) + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.to(self.real_dtype);self.ky=ky.to(self.real_dtype);k2=kx*kx+ky*ky;self.k2=k2.to(self.real_dtype) + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if self.fine: + ox,oy=self.original_shape;fx=torch.fft.fftfreq(ox,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(oy,device=device,dtype=torch.float64)[None,:];self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=gs_dt;self.order=gs_order + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=torch.arange(ny,device=device,dtype=self.real_dtype)*ly/ny+7/16*ly/self.original_shape[1] if self.fine else (torch.arange(ny,device=device)+7/16)*ly/ny + self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(self.complex_dtype if v.is_complex() else self.real_dtype) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + self.graph=None + if cuda_graph and torch.device(device).type=='cuda' and name!='Diffusion': + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + self.buffer=torch.zeros((nc,nx,ny//2+1),device=device,dtype=self.complex_dtype) + try: + def body():self.buffer.copy_(self.substep(self.buffer)) + stream=torch.cuda.Stream();stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3):body() + torch.cuda.current_stream().wait_stream(stream);torch.cuda.synchronize();graph=torch.cuda.CUDAGraph() + with torch.cuda.graph(graph):body() + self.graph=graph + except RuntimeError: + self.graph=None + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + if self.fine: + if not self.initialized: + latent=torch.fft.irfft2(torch.fft.rfft2(u.to(self.real_dtype))/self.observation,s=self.original_shape);v=self.fft(resample(latent,self.shape));self.initialized=True + else:v=self.internal + else:v=self.fft(u.to(self.real_dtype))/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + if self.graph is not None: + self.buffer.copy_(v) + for _ in range(self.substeps):self.graph.replay() + v=self.buffer + else: + for _ in range(self.substeps):v=self.substep(v) + if self.fine: + self.internal=v + low=resample(self.ifft(v),self.original_shape) + return torch.fft.irfft2(torch.fft.rfft2(low)*self.observation,s=self.original_shape).float() + return self.ifft(v*self.observation).float() + def substep(self,v): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;return stage+self.etd2*(self.nonlinear(stage)-nv) + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + return self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/EXPERIMENTS.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/EXPERIMENTS.md new file mode 100644 index 0000000000000000000000000000000000000000..3967825ad46e6b12970f2841f443171d67e7f342 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/EXPERIMENTS.md @@ -0,0 +1,8669 @@ +# Experiment ledger + +All development uses train/valid pools of `/workspace/data/split.json`. No held-out targets or case definitions are used. One H100 GPU is available. No pretrained assets are used. + +## E000 — package initialization +- Hypothesis: a spatial-mean prior provides a finite, portable starting package while training begins. +- Changed factor: parameter-free spatial-mean prediction. +- Seed: not applicable (deterministic). +- Validation: fixed 81-case subset (seed 71, all start at frame 0), reward 0.313512; 0 failed cases. +- Checkpoint hash: no checkpoint; predictor SHA256 307c11c9f4f3f186b96a34597822e1e40e95d47d6159b3de3fe0934ebae13d7d. +- Outcome: initial package published; not a trained candidate or a reproduced baseline run. + +## E001 — coupled, metadata-conditioned multiscale model +- Hypothesis: joint channel modeling, physical metadata, scale-aware normalization and periodic boundaries improve cross-equation prediction relative to the documented starter weaknesses. +- Changed factors: new shared convolutional architecture; all 19 dataset identifiers represented in one parameter set; group-balanced training. +- Initialization seed: 20261021. +- Validation: pending. +- Checkpoint hash: pending. +- Outcome: preparing train-only reduced-resolution cache and validation data. + +### E000 validation VRMSE by dataset + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 1.000634 | +| ape2d/decay_turb | 0.999991 | +| ape2d/diff | 0.999973 | +| ape2d/fisher | 1.581309 | +| ape2d/gs_alpha | 1.005787 | +| ape2d/gs_beta | 1.023021 | +| ape2d/gs_delta | 1.591578 | +| ape2d/gs_epsilon | 1.004624 | +| ape2d/gs_gamma | 1.000439 | +| ape2d/gs_iota | 1.633860 | +| ape2d/gs_kappa | 1.079744 | +| ape2d/gs_theta | 1.574410 | +| ape2d/kdv | 1.000191 | +| ape2d/kolm_flow | 0.999992 | +| ape2d/ks | 0.999992 | +| ape2d/sh | 1.041517 | +| pdearena/navier_stokes_2d | 1.000318 | +| well/active_matter | 0.992894 | +| well/turbulent_radiative_layer_2D | 1.502768 | + +## E001-smoke — 100 updates +- Hypothesis / changed factor: Short pipeline check of new coupled operator with 40 base channels, 12 samples per batch, one-step objective, float16 cache; no baseline retraining. Float16 cache will be replaced to reduce quantization error. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Validation: fixed 81-case subset; reward **0.322278**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.779417 | +| ape2d/decay_turb | 0.919250 | +| ape2d/diff | 0.573262 | +| ape2d/fisher | 1.128033 | +| ape2d/gs_alpha | 0.999278 | +| ape2d/gs_beta | 1.093050 | +| ape2d/gs_delta | 1.567459 | +| ape2d/gs_epsilon | 0.717784 | +| ape2d/gs_gamma | 0.761942 | +| ape2d/gs_iota | 1.560647 | +| ape2d/gs_kappa | 0.956638 | +| ape2d/gs_theta | 1.511911 | +| ape2d/kdv | 0.764748 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.939004 | +| ape2d/sh | 0.919993 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — intermediate 2,000 updates +- Hypothesis / changed factor: Intermediate checkpoint of the same 8,000-update one-step run; all other factors unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `cc58e45b55743bae890d3c3f212be97b831186431591d2501186f69378ce6f37`. +- Validation: fixed 81-case subset; reward **0.263620**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.817172 | +| ape2d/decay_turb | 2.389786 | +| ape2d/diff | 0.837767 | +| ape2d/fisher | 1.272807 | +| ape2d/gs_alpha | 2.153748 | +| ape2d/gs_beta | 8.179170 | +| ape2d/gs_delta | 1.369924 | +| ape2d/gs_epsilon | 1.198089 | +| ape2d/gs_gamma | 1.805734 | +| ape2d/gs_iota | 1.964417 | +| ape2d/gs_kappa | 0.904690 | +| ape2d/gs_theta | 1.609477 | +| ape2d/kdv | 1.151686 | +| ape2d/kolm_flow | 2.257709 | +| ape2d/ks | 2.152634 | +| ape2d/sh | 0.836066 | +| pdearena/navier_stokes_2d | 1.003965 | +| well/active_matter | 2.135917 | +| well/turbulent_radiative_layer_2D | 1.657301 | + +Outcome: did not improve validation reward; retained current package. + +## E003 — metadata-conditioned spectral branch +- Hypothesis / changed factor: Add independently implemented ETDRK4 differential operators to the shared model. All neural weights unchanged from E001-smoke; neural branch still applied every step on every case. Equation parameters come from metadata; solver forms selected using train-only diagnostics. Kolmogorov remains with the neural update pending forcing/drag diagnosis. +- Seed: 20261021. +- Checkpoint SHA256: `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. +- Runtime code SHA256: `9e848b7d80d919c0b32a15d07bf4643a1b56b5235004827d0950660afa2e4952`. +- Validation: fixed 81-case subset; reward **0.401067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.974662 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.927664 | +| well/active_matter | 1.129510 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: improvement; published atomically. + +## E001 — one-step operator, 8,000 updates +- Hypothesis / changed factor: Joint channel ConvNeXt U-Net, metadata conditioning, group-balanced batches, float32 reduced-grid cache, one-step log variance-normalized MSE. Random initialization; 48 base channels and 16 samples per batch. Development inference wrapper relative-path error was corrected before scoring. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `9a645f05c3a17872093f7f461376b1fedd663ceefbbd879ef8f1cd4b8d8e58fb`. +- Validation: fixed 81-case subset; reward **0.350858**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: did not improve validation reward; retained current package. + +## E004 — trained neural branch and forced turbulence operator +- Hypothesis / changed factor: Replace smoke neural weights with E001 step 8,000, still one shared network across all cases. Add Kolmogorov forcing -4 cos(4y) and drag 0.2, identified only in training diagnostics. Both effects are recorded together; full group scores isolate transfer changes from corpus changes. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `d8dd1c803a436415851c119241900367c593b0acfb904f6f5b3f3b17b7444478`. +- Validation: fixed 81-case subset; reward **0.437086**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.663369 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## E005A — physical output constraints and horizon decay (validation ablations) + +Hypothesis: exact tensor identities and reduced confidence at longer horizons stabilize active matter. All experiments reuse E001 8,000-update predictions; seed 20261021. No new weights or training data. Checkpoint SHA256: df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d. + +The diffusion coefficient 0.05, trace(D)=concentration and E=sym(gradient velocity) were checked on training pools. Horizon decay candidates 10, 20 and 40 steps are hyperparameters selected on validation, not parameters fitted to validation targets. These ablations are not published until integrated into a portable package and replayed. + +### active_diffusion: Analytic concentration diffusion only + +Reward 0.351796; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.620833 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_constraints: Concentration diffusion and algebraic tensor projection + +Reward 0.352944; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.570211 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_strain: Recompute strain from predicted velocity + +Reward 0.338342; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 2.350641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: negative; strain derivatives amplify velocity errors. + +### active_projection: Weighted Fourier projection of velocity and strain + +Reward 0.352917; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.571374 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_10: Weighted projection and exponential shrinkage, tau=10 + +Reward 0.375912; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832421 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_20: Weighted projection and exponential shrinkage, tau=20 + +Reward 0.372622; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.911641 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +### active_projection_40: Weighted projection and exponential shrinkage, tau=40 + +Reward 0.365053; failed cases 0. Changed factor is stated above; other predictions unchanged. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.507140 | +| ape2d/decay_turb | 0.829406 | +| ape2d/diff | 0.320034 | +| ape2d/fisher | 0.481280 | +| ape2d/gs_alpha | 0.911294 | +| ape2d/gs_beta | 1.310238 | +| ape2d/gs_delta | 0.908035 | +| ape2d/gs_epsilon | 0.727216 | +| ape2d/gs_gamma | 0.760257 | +| ape2d/gs_iota | 0.676685 | +| ape2d/gs_kappa | 0.623630 | +| ape2d/gs_theta | 0.829690 | +| ape2d/kdv | 0.398326 | +| ape2d/kolm_flow | 1.315148 | +| ape2d/ks | 0.888742 | +| ape2d/sh | 0.925032 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 1.123714 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: candidate ablation; compare integrated model before publication. + +## Train-only solver diagnostics (no validation selection) + +Initial Navier–Stokes sign and step-size hypotheses failed (non-finite or high training error); smaller steps and the opposite vorticity convention succeeded. KS requires separate fourth derivatives, not the squared Laplacian. Kolmogorov forcing and drag were identified using training trajectories only. Doubling the latent solver grid from 256 to 512 barely changed Gray–Scott/KS training errors and was rejected as a general fix. Grid averaging, integrator order and timestep are under investigation. These are diagnostic tests, not held-out experiments; validation VRMSE is not applicable. No checkpoint was created. Source, configurations and train results are retained under research/physics_diagnostics. + +## E005B — integrated physical projection and horizon regularization +- Hypothesis / changed factor: E004 model plus concentration diffusion, weighted Fourier projection of velocity and strain, tensor identities and exponential relaxation toward context means for active-matter fluctuations, tau=10 steps. All weights unchanged; tau selected from fixed validation ablations. The neural model is still applied to every case. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `83bec07505f714951064c32e8cb4cc20720921d9b62a1e4c1ae3bba353048798`. +- Validation: fixed 81-case subset; reward **0.462140**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Full validation of physics_smoke + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.400925**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `5ac6054ef153f01dfda3d736e0642014f3057337eaa20e891aea01fe6d706e16`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.060324 | +| ape2d/decay_turb | 0.000387 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.002465 | +| ape2d/gs_alpha | 0.022938 | +| ape2d/gs_beta | 0.012380 | +| ape2d/gs_delta | 0.146653 | +| ape2d/gs_epsilon | 0.026477 | +| ape2d/gs_gamma | 0.222599 | +| ape2d/gs_iota | 0.036049 | +| ape2d/gs_kappa | 0.109056 | +| ape2d/gs_theta | 0.135853 | +| ape2d/kdv | 0.025460 | +| ape2d/kolm_flow | 0.979667 | +| ape2d/ks | 0.080605 | +| ape2d/sh | 0.002222 | +| pdearena/navier_stokes_2d | 0.923463 | +| well/active_matter | 1.120708 | +| well/turbulent_radiative_layer_2D | 1.281937 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E002 — intermediate 2,000 multi-step updates +- Hypothesis / changed factor: E005B inference architecture and physical operators; replace only shared neural weights with the 2,000-update checkpoint of four-step fine-tuning with 0.01 input noise. +- Seed: 20261021. +- Checkpoint SHA256: `d720a4dbeee0c16a125aca0cee22ee060e1e80c6b863efbd169fd69b5f2c9efc`. +- Runtime code SHA256: `f4b3be2c3e227fcfa3e8cb9774970780928d6635210a425fa2c6b07e8f08e8a5`. +- Validation: fixed 81-case subset; reward **0.473422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.044875 | +| ape2d/decay_turb | 0.000741 | +| ape2d/diff | 0.000002 | +| ape2d/fisher | 0.001230 | +| ape2d/gs_alpha | 0.019048 | +| ape2d/gs_beta | 0.010848 | +| ape2d/gs_delta | 0.146363 | +| ape2d/gs_epsilon | 0.044519 | +| ape2d/gs_gamma | 0.223558 | +| ape2d/gs_iota | 0.020980 | +| ape2d/gs_kappa | 0.119925 | +| ape2d/gs_theta | 0.119511 | +| ape2d/kdv | 0.017155 | +| ape2d/kolm_flow | 0.287456 | +| ape2d/ks | 0.087151 | +| ape2d/sh | 0.002458 | +| pdearena/navier_stokes_2d | 0.559579 | +| well/active_matter | 0.827134 | +| well/turbulent_radiative_layer_2D | 0.628780 | + +Outcome: improvement; published atomically. + +## E006 — observation and integration operators +- Hypothesis / changed factor: Same E005B neural weights, projection and horizon regularization. Spectral branch now models finite 8-cell averaging, the corresponding Kolmogorov forcing location, and ETDRK2 substeps for Fisher/Swift–Hohenberg/KS/KdV confirmed on training trajectories. Gray–Scott uses ETDRK4 max substep 2. No reference generator code or seeds used. +- Seed: 20261021. +- Checkpoint SHA256: `df9d763b100dd434cf97cc21c15525f38e87fbedc56e8172342cb38d5d2cbf3d`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.490722**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.611519 | +| well/active_matter | 0.832422 | +| well/turbulent_radiative_layer_2D | 0.780865 | + +Outcome: improvement; published atomically. + +## Contract and determinism verification + +Seed 0 inference checks on three validation inputs spanning scalar periodic, 11-channel periodic and rectangular mixed-boundary data, plus independently constructed unseen 2-channel 7×11 and 13-channel 129×177 inputs. All shapes/dtypes were correct, every output finite, maximum rerun difference exactly 0. Small-grid padding and non-finite-step fallback added without changing any predictions on the three validation cases (bitwise parity). No synthetic training trajectories were generated. Evidence: research/universal_v1/contract_checks.json. + +## E007 — combine improved physics and multi-step training +- Hypothesis / changed factor: E006 spectral observation/integration model and E005B output projection, with shared neural weights after 4,000 four-step fine-tuning updates. Corpus and transfer gains combined; all weights originate from E001 random initialization. +- Seed: 20261021. +- Checkpoint SHA256: `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.506925**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.531697 | +| well/active_matter | 0.821690 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: improvement; published atomically. + +## E002 — four-step training with input perturbations +- Hypothesis / changed factor: Fine-tune E001 for 8,000 updates with four autoregressive steps in the loss, 0.01 normalized input noise, lr 0.0003, batch 12. Test whether recovery from model errors improves long rollouts. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `1d2b51ab44d7b8212b22c4c1894b02b03697b7ed1d30f756f216e6076006e48d`. +- Validation: fixed 81-case subset; reward **0.398333**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.411539 | +| ape2d/decay_turb | 0.688351 | +| ape2d/diff | 0.194736 | +| ape2d/fisher | 0.729932 | +| ape2d/gs_alpha | 0.660633 | +| ape2d/gs_beta | 0.856659 | +| ape2d/gs_delta | 0.790159 | +| ape2d/gs_epsilon | 0.457965 | +| ape2d/gs_gamma | 0.625919 | +| ape2d/gs_iota | 0.527543 | +| ape2d/gs_kappa | 0.587907 | +| ape2d/gs_theta | 0.768902 | +| ape2d/kdv | 0.263418 | +| ape2d/kolm_flow | 0.973194 | +| ape2d/ks | 0.959533 | +| ape2d/sh | 0.611700 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.929633 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: did not improve validation reward; retained current package. + +## E002 — final multi-step checkpoint with validated inference operators +- Hypothesis / changed factor: Replace only the E007 shared neural weights with E002 after all 8,000 four-step fine-tuning updates. Physical operators and active-matter horizon regularization unchanged. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ef7fb353326d3a6210fc85eb7ee6c8c8438873f1b1736c764562e74f63642cef`. +- Validation: fixed 81-case subset; reward **0.511096**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.801125 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## Full validation of phase2_4000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.500153**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `b0075d614087b0d8ff2c4da046ab09098ff0004a8fa4664be868246f6991e1b8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045446 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.527867 | +| well/active_matter | 0.839802 | +| well/turbulent_radiative_layer_2D | 0.581463 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E009 — active-fluid moment physics, train-only diagnostics + +Hypothesis: the transfer error is partly due to missing orientation-distribution moments. Implemented independent Bingham closure and a Fourier angular-moment Smoluchowski/Stokes system, based on public equation descriptions (the local The Well active_matter README) and training-only checks. No generator code, seeds, pretrained weights or normalization statistics used. Translation and rotation diffusivities 0.05 and beta 0.8 were verified against train fields; alpha and zeta come from supplied metadata. + +Changed factors: Bingham closure vs retained angular moments; spatial grids 128/256; angular resolution 12/24/40; zero vs Bingham initialization; fixed recurrent context encoder that inserts observed low moments. The encoder updates physical state only; no parameters or gradients are fitted at inference. + +Seed: not applicable (deterministic analytical operators). Checkpoint: none; these are not published candidates. Validation VRMSE: not yet run for any dataset. Four selected TRAIN trajectories only. Bingham 30-step train mean VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5): 0.0261, 0.5173, 0.8618, 1.0525. Doubling spatial resolution did not resolve closure drift. At (3,-2), 12 angular modes with context encoding reduced four-step train VRMSE to about 0.0044, versus 0.153 for Bingham. Twelve modes were inadequate at high activity; increasing angular resolution is being tested. A GPU prototype initially failed due to TorchScript complex-number optimization; this is a portability failure, not a scored model. + +Outcome: promising mechanism, still under development; retained the validated shared neural/spectral model. Detailed train results and source variants remain in research/physics_diagnostics. + + +## E009b — active kinetic resolution and implementation diagnostics + +Hypothesis: angular and spatial discretization limit the kinetic closure. Changed factors: M=32/40/64 even angular modes, 128 versus native256 spatial grid, spatial2/3 filtering enabled/disabled, eager versus CUDA graph execution. Seed: none (deterministic operators); neural checkpoint unchanged. Validation: not evaluated for these train-only numerical diagnostics. Runtime source and all numerical results are in physics_diagnostics. + +Native256/M32 thirty-step TRAIN VRMSE at (zeta,alpha)=(1,-1),(3,-2),(9,-3),(17,-5) was 0.025519, 0.014763, 0.100166, 0.880100. M40 and M64 gave nearly equal short-rollout errors, whereas 128 spatial resolution remained inaccurate or unstable at high activity. Removing the spatial filter did not help. Four-step errors at native256/M64 were approximately 0.0024/0.0031/0.0041/0.0055 for (9,-3), and 0.0233/0.0483/0.0937/0.1448 for (17,-5). Outcome: select native256/M32 for validation, retaining a finite physical-prefix fallback. TorchScript fusion failures and CUDA graph scalar-copy failures were fixed before scoring. CUDA graphs reduce launch overhead; all state storage addresses remain fixed. + +## E011a — buoyant smoke discretization, train-only identification + +Hypothesis: a fixed incompressible-fluid operator can reduce PDEArena rollout error once the observed grid layout is represented correctly. Independent MAC solver with semi-Lagrangian advection, explicit viscosity and discrete Neumann pressure projection. Training data reveal that saved velocities are lower faces with the final zero normal face omitted. Correct discrete divergence is about 0.00016 RMS. Physical domain32 and stored interval6 account for endpoint metadata conventions. + +Changed factors on three TRAIN trajectories: stored interval6 versus6.461538, internal time steps0.25/0.5/1/1.5, viscosity0/0.001/0.005/0.008/0.01/0.012, and linear versus MacCormack scalar advection. Seed: none; checkpoint: none (analytic prototype). Validation per dataset: not yet evaluated. At interval6, linear scalar advection and viscosity0.01, one-step mean train VRMSE was 0.0000243. MacCormack with the same viscosity gave0.0981; reducing the internal step to0.25 with MacCormack gave0.3391. Thus closer continuum integration did not reproduce the observed numerical dynamics. Eight additional complete TRAIN rollouts with the selected scheme had mean VRMSE below0.00011. A separate torch-only double-precision implementation reproduced the SciPy prototype. Outcome: proceed to validation; no held-out data or generator code used. Detailed tested values and errors are retained in arena_initial_results.json, arena_scheme_results.json, arena_long_results.json and arena_torch_results.json. + +## E010 — fixed angular-moment active-fluid operator +- Hypothesis / changed factor: Add the independently implemented Smoluchowski/Stokes angular-moment operator to the same phase2-final shared neural model. M32 even angular modes, native256 grid, six Stokes iterations, ETDRK4, deterministic CUDA graph; fixed Bingham initialization and observed low-moment context encoding. No inference parameter fitting. All twelve active validation cases used all thirty physical steps. Exact repeat determinism verified on low/high-activity validation inputs and a diffusion input; two novel grid/channel schemas also passed. Runtime pins the complete package directory during atomic updates. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `d11ea64e646593878f730df43933d3f1e64f481a34083bb7687be973e51e2d33`. +- Validation: fixed 81-case subset; reward **0.572674**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.487448 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E011 — fixed MAC-grid buoyant-fluid operator +- Hypothesis / changed factor: Keep the E010 shared neural checkpoint and all other inference components; add a metadata-conditioned MAC fluid update for the buoyant-smoke schema. The selected numerical scheme was identified on train pools only. Quick validation reran all twelve PDEArena cases and reused E010 predictions for unchanged operators with identical weights. PDEArena mean VRMSE is 0.00008349. Exact repeat determinism passed for two smoke cases, diffusion, and two novel grid/channel schemas. Expanded end-to-end validation is next. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `397cac1ec1b4314c9e5832071d0cc8a3451a85a8d891ba469a64722ac4d69f65`. +- Validation: fixed 81-case subset; reward **0.713334**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: improvement; published atomically. + +## E012a — neural forecast uncertainty, train-only ablation + +Hypothesis: spatial averaging reduces uncertain late-time turbulent fluctuations while retaining vertical mean profiles. Same phase2-final checkpoint (SHA256 d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420), seed20261021. Nine TRAIN trajectories, one for each cooling time; no validation used to select the coefficient. Changed factors: nonnegative material scalars; Gaussian smoothing with final width1/2/4/8/16 pixels; linear horizon-dependent shrinkage toward the mean over periodic directions with final weight0.25/0.5/0.75/1.0 at step30. + +| Variant | Train VRMSE | +|---|---:| +| raw | 0.538547 | +| positive | 0.538100 | +| blur_1 | 0.537242 | +| blur_2 | 0.536386 | +| blur_4 | 0.534177 | +| blur_8 | 0.529554 | +| blur_16 | 0.526478 | +| profile_0.25 | 0.530485 | +| profile_0.5 | 0.525987 | +| profile_0.75 | 0.524946 | +| profile_1.0 | 0.527495 | + +Validation per dataset: not evaluated in this ablation. Outcome: coefficient0.75 selected for subsequent validation. Positivity alone helped very little; smoothing cannot resolve most pressure error. Roughly one-fifth of pressure MSE comes from spatial-mean error, so a correction to the mean alone is insufficient. + +## E012 — fixed spatial uncertainty shrinkage +- Hypothesis / changed factor: The same shared neural weights and physical operators as E011. Add horizon-dependent shrinkage toward means over periodic directions (weight 0.75 by step30), plus nonnegative density/concentration. The coefficient was selected on nine training trajectories before validation. All nine radiative-layer validation rollouts were rerun; unchanged physical outputs were reused for the other72 cases. Exact determinism passed on three radiative-layer inputs and two novel grid/channel schemas. Validation radiative-layer VRMSE improved from0.579723 to0.546835. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715448**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546835 | + +Outcome: improvement; published atomically. + +## E013 — native rectangular inference before fine-tuning +- Hypothesis / changed factor: Change only the shared neural inference grid for rectangular inputs from64x192 to native128x384, keeping the E012 checkpoint and all other operators/regularization fixed. This tests resolution transfer before native-grid fine-tuning. Radiative-layer VRMSE worsened from0.546835 to0.769883; retain E012. All other validation predictions are unchanged physical outputs. The result motivates retraining the shared operator on native grids rather than changing inference resolution alone. +- Seed: 20261021. +- Checkpoint SHA256: `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. +- Runtime code SHA256: `cc9804d49168dff3c2a3868118de9b3b3bccbd2b0982814f94617a3115b7b6fa`. +- Validation: fixed 81-case subset; reward **0.703067**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.769883 | + +Outcome: did not improve validation reward; retained current package. + +## E008 — larger-batch transfer-focused training, update 5000 +- Hypothesis / changed factor: Continue the same shared network from E002 final weights: batch 48, four-step loss, 0.01 input noise, corpus probability 0.1 and 0.3 for each transfer group, lr 0.0003 cosine, sampling seed 20261022; initialization seed remains 20261021. Inference operators frozen to E012 (angular-moment active matter, MAC buoyant smoke and fixed neural uncertainty shrinkage). Quick validation reruns all nine radiative-layer trajectories and reuses fixed physical predictions for the other seventy-two cases, which are independent of the neural weights; expanded validation runs end-to-end. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `ba79305f6b0d6631b1fcea9b289b1b08335df7ab26afe685790aa2a7d4d03d98`. +- Validation: fixed 81-case subset; reward **0.715555**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.015600 | +| ape2d/gs_beta | 0.005591 | +| ape2d/gs_delta | 0.140335 | +| ape2d/gs_epsilon | 0.002787 | +| ape2d/gs_gamma | 0.066528 | +| ape2d/gs_iota | 0.018343 | +| ape2d/gs_kappa | 0.084217 | +| ape2d/gs_theta | 0.101542 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + + +## E015a — reaction-diffusion spatial/time convergence, training only + +Hypothesis: unresolved reaction harmonics cause remaining Gray–Scott drift. Changed factors: grid256/512/1024 after finite8-cell observation deconvolution; ETDRK4 maxdt2/0.5 vs ETDRK2 maxdt0.2/0.25/0.3/0.5; optional physical context encoding preserving unresolved modes; finite-averaging ratio2/4/8/16 versus its continuum limit. Deterministic numerical operators, no random seed or learned checkpoint. Validation per dataset: not yet evaluated. + +On selected TRAIN gamma/delta trajectories, grid256/RK4dm2 three-step VRMSE was gamma .000764/.000799/.001016 and delta .000817/.001631/.002567. Grid512/RK2dm.25 gave gamma .000197/.000268/.000394 and delta .000105/.000222/.000335. Grid1024 and the recurrent physical-state encoder gave no additional improvement. Ratio8 remained best among tested observation filters. Thus prior negative grid-only tests did not rule out improvement when resolution and observation inversion were changed together. A portable double-precision GPU implementation is undergoing full train rollouts. Numerical results and all variants are retained in gs_resolution_results.json, gs_resolution_time_results.json, gs_context_results.json and gs_observation_results.json. + +A separate spectral CUDA-graph implementation reproduced ten different corpus training cases exactly over three forecast steps (max absolute difference0 for each). Some runtime reduction was measured while shared training was active, e.g. three Navier–Stokes steps .70s to .47s. This changes execution only; no numerical parameters are fitted at inference. + +## E016 design — shared global spectral feature mixing, not yet trained + +Hypothesis: acoustic pressure patterns need communication across the native rectangular grid. Prepared a shared Fourier feature-mixing residual at the first neural stage, with even extensions in nonperiodic directions and metadata-conditioned complex gains. All weights will be trained jointly on all pools; this is not a dataset specialist. Adds306552 trainable parameters to the2.759547M shared net, total3066099. New output projection starts at zero. Seed20261024 for the new parameters. Initial model matches the existing parent exactly, and a training-only backward pass gives finite nonzero gradients. No optimizer updates or validation results yet; not a publishable candidate. Source and initialization checks in research/global_v1. Continue native-grid training first to establish a comparison. + +## E015b — offline numerical coefficient calibration, negative train result + +Hypothesis: small global corrections to Gray–Scott diffusivities and elapsed time reduce residual error. Fitted three fixed coefficients offline to one-step errors on TRAIN gamma/delta trajectories, with the fine512/RK2dm.25 operator. No inference-time fitting. Seed: none; checkpoint: none (analytic calibration); validation per dataset: not run because full TRAIN rollouts worsened. + +Fitted options: `{"gs_scale": 2, "gs_dt": 0.25, "gs_order": 2, "gs_double": true, "gs_diffusion_u": 1.9999732723556725e-05, "gs_diffusion_v": 1.0000209495791168e-05, "gs_time_scale": 0.9999851457782819}`. One-step normalized MSE decreased from 2.3365036901186555e-08 to 1.9611423235582817e-08. + +| Train dataset | Variant | 26-step VRMSE | +|---|---|---:| +| gs_gamma | uncalibrated | 0.029923 | +| gs_gamma | calibrated | 0.031195 | +| gs_delta | uncalibrated | 0.013152 | +| gs_delta | calibrated | 0.014885 | +| gs_alpha | uncalibrated | 0.002711 | +| gs_alpha | calibrated | 0.002466 | +| gs_theta | uncalibrated | 0.010527 | +| gs_theta | calibrated | 0.049919 | + +Outcome: rejected. The tiny one-step improvement did not carry over to long rollouts, particularly on the unused theta training trajectory. Retain nominal physical coefficients. + +## E015 — finer reaction-diffusion state with matched integration +- Hypothesis / changed factor: Keep the same phase3/update5000 shared neural checkpoint. Gray–Scott now retains a2x internal Fourier grid, uses double-precision ETDRK2 with max internal dt0.25, and observes it through the finite8-cell averaging filter. Nominal physical coefficients retained after calibration failed full train rollouts. Enable deterministic CUDA graphs for spectral updates. Reran all48 corpus validation cases; corpus mean improved from0.032509 to0.010893. Other33 cases reuse identical parent outputs. Exact repeat checks passed on two Gray–Scott cases, Kolmogorov flow and novel schemas; all24 unchanged corpus cases matched parent outputs bit for bit. +- Seed: 20261022. +- Checkpoint SHA256: `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. +- Runtime code SHA256: `3962bd91a5cf0d9b11b901a10dda99623d4d995b034f57f75c81e62bc41c8aff`. +- Validation: fixed 81-case subset; reward **0.755127**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Outcome: improvement; published atomically. + +## Full validation of arena_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.696465**, failed cases 0. Neural initialization seed 20261021; checkpoint SHA256 `d804de9d9c22ff0eb451f83ee2e6b9aa9bbf3f887e99b5f8dd887e9b2c414420`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.045295 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.016217 | +| ape2d/gs_beta | 0.006001 | +| ape2d/gs_delta | 0.140847 | +| ape2d/gs_epsilon | 0.002213 | +| ape2d/gs_gamma | 0.093321 | +| ape2d/gs_iota | 0.031320 | +| ape2d/gs_kappa | 0.092346 | +| ape2d/gs_theta | 0.115251 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.579723 | + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E014 — native-grid eight-step training, update 1000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.756053**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 1000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755034**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.546620 | + +Outcome: did not improve validation reward; retained current package. + +## E017a — active-state observation diagnostics on training trajectories + +Hypotheses: remaining active-fluid drift arises either from numerical resolution or inconsistent hidden angular moments after observed low moments are inserted. Changed factors: time step, spatial grid, Stokes iterations, finite-difference moment inversion, fixed velocity-constrained moment observation, positivity projection, rigid angular transport, and Bingham amplitude transport. All diagnostics use the first trajectory in the corresponding TRAIN pool, starting at frame zero; no validation or held-out targets. The context encoder uses only four observed frames and updates physical state; solver and encoder parameters are fixed. No parameter optimization at inference. + +Seed: deterministic physical operators, no new random parameters. Shared checkpoint unchanged; current selected SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Validation VRMSE per dataset: not evaluated for these train-only diagnostics. + + +kinetic_accuracy_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"name": "baseline", "params": [17, -5]} | 0.07398549 | 0.13735820 | 4 | +| {"name": "smaller_time_step", "params": [17, -5]} | 0.07397478 | 0.13734564 | 4 | +| {"name": "stokes_iterations", "params": [17, -5]} | 0.07398360 | 0.13735427 | 4 | +| {"name": "upsample_384", "params": [17, -5]} | 0.07628622 | 0.14166540 | 4 | +| {"name": "upsample_512", "params": [17, -5]} | 0.07628392 | 0.14165913 | 4 | + +kinetic_encoder_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"last": false, "params": [3, -2], "weight": 0} | 0.00370120 | 0.00440354 | 4 | +| {"last": false, "params": [3, -2], "weight": 1} | 0.01032324 | 0.01005470 | 4 | +| {"last": true, "params": [3, -2], "weight": 0.25} | 0.00641056 | 0.00683838 | 4 | +| {"last": true, "params": [3, -2], "weight": 1} | 0.02690719 | 0.02838808 | 4 | +| {"last": false, "params": [9, -3], "weight": 0} | 0.00375097 | 0.00547713 | 4 | +| {"last": false, "params": [9, -3], "weight": 1} | 0.14332058 | 0.16554202 | 4 | +| {"last": true, "params": [9, -3], "weight": 0.25} | 0.04367409 | 0.04571036 | 4 | +| {"last": true, "params": [9, -3], "weight": 1} | 0.12102679 | 0.12214902 | 4 | +| {"last": false, "params": [17, -5], "weight": 0} | 0.07398282 | 0.13735300 | 4 | +| {"last": false, "params": [17, -5], "weight": 1} | 0.35820127 | 0.58563930 | 4 | +| {"last": true, "params": [17, -5], "weight": 0.25} | 0.05379023 | 0.08525104 | 4 | +| {"last": true, "params": [17, -5], "weight": 1} | 0.20885072 | 0.28744033 | 4 | + +kinetic_flow_encoder_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "weight": 0.0} | 0.02551891 | 0.05217000 | 30 | +| {"params": [1, -1], "weight": 0.25} | 0.01215404 | 0.02445865 | 30 | +| {"params": [1, -1], "weight": 1.0} | 0.00162011 | 0.00254364 | 30 | +| {"params": [3, -2], "weight": 0.0} | 0.01476343 | 0.04135585 | 30 | +| {"params": [3, -2], "weight": 0.25} | 0.00774478 | 0.02111043 | 30 | +| {"params": [3, -2], "weight": 1.0} | 0.00182436 | 0.00253463 | 30 | +| {"params": [9, -3], "weight": 0.0} | 0.10013896 | 0.45469648 | 30 | +| {"params": [9, -3], "weight": 0.25} | 0.03468125 | 0.16266733 | 30 | +| {"params": [9, -3], "weight": 1.0} | 0.04499391 | 0.18140420 | 30 | +| {"params": [17, -5], "weight": 0.0} | 0.54988378 | 1.29260397 | 30 | +| {"params": [17, -5], "weight": 0.25} | 0.48679241 | 1.30647564 | 30 | +| {"params": [17, -5], "weight": 1.0} | 0.62024931 | 1.32728279 | 30 | + +kinetic_transport_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.00372676 | 0.00443898 | 4 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187841 | 0.00173137 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 0.0} | 0.00375795 | 0.00446493 | 4 | +| {"params": [3, -2], "transport": "bingham", "weight": 1.0} | 0.00187924 | 0.00173272 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00041423 | 0.00039727 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043372 | 0.00043870 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 0.0} | 0.00041337 | 0.00038921 | 4 | +| {"params": [9, -3], "transport": "bingham", "weight": 1.0} | 0.00039989 | 0.00040295 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.00621011 | 0.01090280 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00566252 | 0.01021467 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 0.0} | 0.00627246 | 0.01102048 | 4 | +| {"params": [17, -5], "transport": "bingham", "weight": 1.0} | 0.00555848 | 0.01004915 | 4 | + +kinetic_transport_long_results.json + +| Training parameters and variant | Mean VRMSE | Last-step VRMSE | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 0.0} | 0.02551866 | 0.05216906 | 30 | +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00162010 | 0.00254361 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 0.0} | 0.01480152 | 0.04140588 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183204 | 0.00254354 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 0.0} | 0.00303750 | 0.01372149 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00258502 | 0.00757161 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 0.0} | 0.45993094 | 1.23510826 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.46775045 | 1.23808050 | 30 | + +Outcome: finer numerics did not improve forecasts. Temporal-derivative inversion was mostly harmful. Fixed velocity observation improves low-activity cases; rigid transport additionally improves moderate/high activity. Bingham amplitude transport adds little in short tests. Select phase transport plus observation weight1 for validation based on complete 30-step training evidence. No candidate from this entry alone was published. + +- Diagnostic source kinetic_torch.py: SHA256 `1082bbf868d579e4c101b86d5fccffc9e43ba1c45e569e1c1dcbc4a19c5ddd6a`. +- Diagnostic source kinetic_flow_encoder.py: SHA256 `e0d1e5022cc8e0595c1c9fd43dc15ba4e77c7998e6fef139c5b58e2df30dde22`. +- Diagnostic source kinetic_transport_encoder.py: SHA256 `4bbb953264d06a28cd3832509172ae68240a24ec779116a3e0377346818e26e9`. + +## E014 — native-grid eight-step training, update 2000, shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `3c1a0c1b744a5ed45e470052d1cc0220ded8ea97ab6e41bed3abc2c752dcc53f`. +- Validation: fixed 81-case subset; reward **0.757118**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E014 — native-grid eight-step training, update 2000, no_shrink +- Hypothesis / changed factor: Continue one shared model from the best checkpoint selected at phase3 update5000. Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling seed20261023, initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses E015 fixed physical predictions for the other72 cases. See phase4/lineage.json for exact parent selection. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `54a042282c9138ed9a7f385c2e2f2c8915d954d997f18d89ad1cc8da03b6230a`. +- Validation: fixed 81-case subset; reward **0.755962**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.146138 | +| well/turbulent_radiative_layer_2D | 0.532789 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of fine_gs_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.730834**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261022; checkpoint SHA256 `b15d7793357ba38253d0c90c465eef87c96a46c963d7347caa803ccfc19f11a8`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.049812 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.050146 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.196050 | +| well/turbulent_radiative_layer_2D | 0.545224 | + +Expanded-validation protocol: all561 non-active cases were rerun with this complete package. The24 active-matter outputs were reused from the earlier end-to-end E011 run after verifying identical active solver assets/configuration and thirty physical steps for every case, so those outputs are independent of the changed shared neural weights. The official scorer evaluates all585 trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E017 — geometric and velocity-constrained context state +- Hypothesis / changed factor: Keep the same shared neural checkpoint and all other operators from parent.json. Change only the active-matter four-frame physical-state encoder: rotate higher angular moments with observed second-moment orientation, scale by observed concentration, use twelve fixed regularized Stokes observation steps for F4, and project F4 into its necessary moment-positivity disk. Fixed observer weight1, ridge0.01 and phase confidence threshold0.1 were selected using train-only full30 diagnostics. No learned or solver parameter changes at inference. All twelve active validation cases rerun; sixty-nine identical parent outputs reused. Exact physical-prefix coverage is in physical_coverage.json. No new random parameters; inherited initialization seed20261021 and sampling seed recorded in the checkpoint manifest. +- Seed: 20261023. +- Checkpoint SHA256: `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.802217**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.531451 | + +Outcome: improvement; published atomically. + +## E017b — combine context encoder and native checkpoint +- Hypothesis / changed factor: Combine the E017 fixed physical context encoder with the better shared native-grid checkpoint selected at phase4 update2000. No training, new random parameters or numerical changes. All runtime modules/configuration outside kinetic.py and the shared weights were checked identical between parents. Twelve active outputs are independent of shared weights after verifying all30 steps use the physical operator. Official scoring reuses those12 outputs and69 matching neural-parent outputs. Replayed three neural cases and two novel schemas with exact determinism. Active determinism was already checked for the identical physical encoder. Inherited initialization seed20261021, native-stage sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `4122e82bf4f0ada06b25c2d0aec4b0b27ba60c97f8a5db63eedceace35062d55`. +- Validation: fixed 81-case subset; reward **0.803282**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.035969 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.032076 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E018 — finer internal advection grids +- Hypothesis / changed factor: Keep the same shared neural checkpoint and active-state encoder. Change only Burgers and Kolmogorov physical updates to a twofold internal Fourier grid in double precision, retaining unresolved state between outputs and using the same finite-cell observation map. Preserve the physical Kolmogorov forcing phase when refining its grid. Three full26 training trajectories per equation supported the change; doubling substeps, using a fourfold grid, or encoding unresolved modes through all four context frames added little. No random parameters or new training. Rerun the six affected validation cases; reuse seventy-five identical parent outputs. All six used26/26 physical updates. Exact repeat checks include Burgers, Kolmogorov, an unchanged Gray-Scott case and two novel schemas. Inherited initialization seed20261021 and sampling seed20261023. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `970b00933280c87d32c3d322412f89d56ad1969897bbdd0b83dbeccc51231cf8`. +- Validation: fixed 81-case subset; reward **0.806376**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Outcome: improvement; published atomically. + +## E017 expanded active-matter validation + +Changed factor: expand from twelve to all24 active-matter validation trajectories, all start0. Twelve existing predictions were reused only after exact context equality checks; the other twelve were replayed with the same complete E017 package. Official scorer active-matter VRMSE **0.08948507**, versus 0.19604967 for the earlier physical encoder on the same24 cases. Failed cases 0. Other dataset VRMSE: not reevaluated in this experiment. Deterministic encoder; inherited neural initialization seed20261021, sampling seed 20261023. Checkpoint SHA256 `896ab17892d5add120c2b53ea96bab6a92e5a8125b1f40f1c349007ab56aba67`. Outcome: expanded group-level validation, no comparison between different-size aggregate rewards. + +## E018a — further fixed-operator train diagnostics + +Hypotheses and changed factors: (1) interpolated velocity/strain can drive hidden moments between context frames; (2) an entropic angular-moment projection can repair the full distribution; (3) moment and velocity filters should differ; (4) unresolved Fourier modes cause Burgers/Kolmogorov error; (5) a finer grid, retained context state, time-step refinement or alternative exponential integration repairs KdV. All diagnostics use TRAIN trajectories only, with exact source indices recorded by scripts. No new random parameters or learned checkpoint. Validation VRMSE per dataset: not evaluated in these diagnostics; E018 separately evaluates the selected grid change. + + +kinetic_forced_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"cubic": false, "observe_initial": false, "params": [3, -2]} | 0.00187798 | 0.00173005 | 4 | +| {"cubic": true, "observe_initial": false, "params": [3, -2]} | 0.00187867 | 0.00173198 | 4 | +| {"cubic": true, "observe_initial": true, "params": [3, -2]} | 0.00413512 | 0.00506158 | 4 | +| {"cubic": false, "observe_initial": false, "params": [9, -3]} | 0.01194177 | 0.01268802 | 4 | +| {"cubic": true, "observe_initial": false, "params": [9, -3]} | 0.01874249 | 0.02191580 | 4 | +| {"cubic": true, "observe_initial": true, "params": [9, -3]} | 0.01873903 | 0.02190757 | 4 | +| {"cubic": false, "observe_initial": false, "params": [17, -5]} | 0.02985521 | 0.04619750 | 4 | +| {"cubic": true, "observe_initial": false, "params": [17, -5]} | 0.04395452 | 0.07376709 | 4 | +| {"cubic": true, "observe_initial": true, "params": [17, -5]} | 0.04395483 | 0.07376773 | 4 | + +kinetic_moment_projection_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187845 | 0.00173138 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00043149 | 0.00043489 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00560499 | 0.01007434 | 4 | + +kinetic_moment_projection_long_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [1, -1], "transport": "phase", "weight": 1.0} | 0.00161964 | 0.00254240 | 30 | +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00183213 | 0.00254286 | 30 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00240461 | 0.00671084 | 30 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.48482975 | 1.27747452 | 30 | + +kinetic_unmasked_flow_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"params": [3, -2], "transport": "phase", "weight": 1.0} | 0.00187846 | 0.00173107 | 4 | +| {"params": [9, -3], "transport": "phase", "weight": 1.0} | 0.00041315 | 0.00041952 | 4 | +| {"params": [17, -5], "transport": "phase", "weight": 1.0} | 0.00551630 | 0.01016449 | 4 | + +nonlinear_resolution_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963583 | 0.05441318 | 26 | +| {"double": true, "name": "burgers", "scale": 1, "time_factor": 1} | 0.02963618 | 0.05441412 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1} | 0.00572061 | 0.01058393 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 2} | 0.00572059 | 0.01058392 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01216232 | 0.02993135 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 1, "time_factor": 1} | 0.01185469 | 0.02870498 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1} | 0.00759149 | 0.02465414 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 2} | 0.00759146 | 0.02465408 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001462 | 0.00006308 | 26 | +| {"double": true, "name": "kdv", "scale": 1, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1} | 0.00001473 | 0.00006323 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 2} | 0.00001457 | 0.00006339 | 26 | + +nonlinear_resolution_expanded_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00301647 | 0.00326871 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037459 | 0.00050441 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 8.994104268544953e-05} | 0.00037497 | 0.00050395 | 26 | +| {"double": false, "name": "burgers", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00124680 | 0.00155330 | 26 | +| {"double": true, "name": "burgers", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012381 | 0.00019031 | 26 | +| {"double": true, "name": "burgers", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0002315303592548704} | 0.00012380 | 0.00019021 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.08121000 | 0.15702826 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.02042678 | 0.05910929 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.00020737974572554083} | 0.01992439 | 0.05732524 | 26 | +| {"double": false, "name": "kolm_flow", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.04068160 | 0.08604752 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00718886 | 0.02328656 | 26 | +| {"double": true, "name": "kolm_flow", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.00035763457551476617} | 0.00716170 | 0.02330063 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035132 | 0.00209063 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00035139 | 0.00208930 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 17, "viscosity": 0.0009525995873512995} | 0.00034709 | 0.00206470 | 26 | +| {"double": false, "name": "kdv", "scale": 1, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203683 | 0.01849953 | 26 | +| {"double": true, "name": "kdv", "scale": 2, "time_factor": 1, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01203362 | 0.01849631 | 26 | +| {"double": true, "name": "kdv", "scale": 4, "time_factor": 2, "trajectory": 93, "viscosity": 0.0006360816950437411} | 0.01204391 | 0.01851219 | 26 | + +nonlinear_context_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 0} | 0.00572061 | 0.01058393 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 17} | 0.00037297 | 0.00050443 | 26 | +| {"context": true, "name": "burgers", "scale": 2, "trajectory": 93} | 0.00012373 | 0.00019031 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 0} | 0.00759136 | 0.02465325 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 17} | 0.02068291 | 0.05987191 | 26 | +| {"context": true, "name": "kolm_flow", "scale": 2, "trajectory": 93} | 0.00716521 | 0.02321832 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 17} | 0.00035139 | 0.00208930 | 26 | +| {"context": true, "name": "kdv", "scale": 2, "trajectory": 93} | 0.01203363 | 0.01849633 | 26 | + +kdv_numerics_results.json + +| Variant | Mean training VRMSE | Final step | Steps | +|---|---:|---:|---:| +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001473 | 0.00006323 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 0} | 0.00001493 | 0.00006064 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 0} | 0.00001457 | 0.00006350 | 26 | +| {"constants": {"Domain Extent": 78.47792260107708, "Viscosity": 0.0002475354258754664}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 0} | 0.00001457 | 0.00006347 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00035151 | 0.00209107 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 17} | 0.00019026 | 0.00094255 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 17} | 0.00034583 | 0.00205838 | 26 | +| {"constants": {"Domain Extent": 97.97852532879602, "Viscosity": 0.0009525995873512995}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 17} | 0.00034618 | 0.00206044 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01203692 | 0.01849972 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": true, "time_factor": 1, "trajectory": 93} | 0.01360971 | 0.02156036 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 4, "real_nl": false, "time_factor": 1, "trajectory": 93} | 0.01205018 | 0.01851917 | 26 | +| {"constants": {"Domain Extent": 100.87555517480047, "Viscosity": 0.0006360816950437411}, "order": 2, "real_nl": false, "time_factor": 4, "trajectory": 93} | 0.01204985 | 0.01851889 | 26 | + +Outcomes: interpolated active forcing was harmful at moderate/high activity. Entropic moment projection modestly improved moderate activity but worsened the complete high-activity rollout (0.467750 to0.484830); not selected. Retaining full velocity modes has only a small short-rollout effect; full30 test remains ongoing. Twofold grids robustly improved Burgers and Kolmogorov across three training trajectories. Fourfold grids and more substeps added little; four-frame unresolved-state encoding added little or hurt. KdV grid/integrator variants were not consistently better. A proposed small-domain KdV test found no training cases below domain extent15 and ran no rollouts; replaced it with tests on the existing three training trajectories. Low-frequency differential fits recover nonlinear coefficients near6; fitted dissipative corrections were inconsistent, so no coefficient change was selected. + +The phase5 global-mixing trainer initially stopped before optimizer creation because channels_last was applied to a five-dimensional Fourier parameter. Applying the layout only to Conv2d modules fixed it; restart used the identical checkpoint and seed20261024, with no optimizer updates lost. Exactly one GPU trainer remains. + +Diagnostic source kinetic_forced_context.py: SHA256 `cebf38a2f07fbf2ea664c5b00ac17bd5dd1cef2a128b75a2d7e978fbc102e3a0`. + +Diagnostic source kinetic_moment_projection.py: SHA256 `fc796f88bda7f7802bab82a8e41d76dc1e08e1fc3a98fb9eed20cd45e26fc69a`. + +Diagnostic source kinetic_unmasked_flow.py: SHA256 `4a40d463091ef1dd6a2432e2ce79f0a417c61af1943369711062c6e0113a059b`. + +Diagnostic source spectral_resolution.py: SHA256 `907c567c87c46bb3adf9e11b1282d1a3e1400751d5e5c0bfdd835350d3062420`. + +Diagnostic source spectral_kdv_numerics.py: SHA256 `e3fcb35b99158c156691e324dae512d17f9fca287fb072270e1adbc6437b1ffe`. + +## E018b — distinct active flow and moment filters, negative full-rollout result + +Hypothesis: the stored active velocity/strain retain modes above the moment dealiasing cutoff, so the Stokes solve should retain those modes. Changed only the velocity/strain spectral mask, preserving the moment mask. Four complete30-step TRAIN rollouts, deterministic physical operator, no new seed or checkpoint; validation per dataset not evaluated. + +| Training parameters | Mean VRMSE | +|---|---:| +| [1, -1] | 0.00161508 | +| [3, -2] | 0.00182706 | +| [9, -3] | 0.00258145 | +| [17, -5] | 0.46823457 | + +Outcome: negligible improvement at low/moderate activity and slight worsening at high activity (0.467750 to0.468235); retained the validated operator. The direct training spectral audit found concentration/director power above the2/3 cutoff at float32 roundoff, while high-activity strain retains about0.0022 of its spectral norm there. This numerical distinction is real but does not explain the dominant rollout error. + +## E016 — shared global spectral mixing, update 1000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `3cd1d10c39cf842ae764c3a6390d8af9c784fc4cc2141213e7647256be66ba8c`. +- Validation: fixed 81-case subset; reward **0.803408**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.560153 | + +Outcome: did not improve validation reward; retained current package. + +## E016 — shared global spectral mixing, update 1000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. +- Runtime code SHA256: `a0019d25f7e2d44c69f90d0ada4f0459c943fbffa3503dc585227648c836f516`. +- Validation: fixed 81-case subset; reward **0.802480**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.574692 | + +Outcome: did not improve validation reward; retained current package. + +## E020 — pressure uncertainty shrinkage +- Hypothesis / changed factor: Keep the shared neural weights and fixed physical operators. Add pressure-only output shrinkage toward each predicted spatial mean: zero through step10, linear ramp to0.75 at step30. The existing periodic-profile shrinkage remains unchanged. Select this fixed schedule from nine train rollouts for both native and global1000 checkpoints; native training mean VRMSE0.519754 to0.518445, global0.536804 to0.534698. No new random parameters or inference-time fitting; inherited initialization seed20261021 and sampling seed20261023. Rerun all nine neural validation cases, reuse seventy-two identical physical outputs. +- Seed: 20261023. +- Checkpoint SHA256: `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. +- Runtime code SHA256: `1b8f91e26862343f0238655ab2b43f24ad67da270779efcfa7ef02a841f95ab6`. +- Validation: fixed 81-case subset; reward **0.806391**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Outcome: improvement; published atomically. + +## E020a — pressure schedule training sweep + +Changed factor: pressure-only mixing toward its predicted spatial mean, with start0/5/10, end15/30, strength0/0.25/0.5/0.75/1. Nine training trajectories, one per cooling parameter; train membership was audited again. Native checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4 and global1000 checkpoint from phase5/eval001000_shrink were both evaluated. No random coefficients; inherited initialization20261021, native sampling20261023 and global sampling/new-parameter20261024. Validation per dataset: not evaluated in this training sweep; selected native variant validated separately in E020. + +| Model | Start | End | Strength | Training VRMSE | Pressure VRMSE | +|---|---:|---:|---:|---:|---:| +| native | 0 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 15 | 0.25 | 0.51936945 | 1.08422407 | +| native | 0 | 15 | 0.5 | 0.52051218 | 1.08879500 | +| native | 0 | 15 | 0.75 | 0.52311825 | 1.09921928 | +| native | 0 | 15 | 1.0 | 0.52708753 | 1.11509640 | +| native | 0 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 0 | 30 | 0.25 | 0.51915815 | 1.08337887 | +| native | 0 | 30 | 0.5 | 0.51916075 | 1.08338927 | +| native | 0 | 30 | 0.75 | 0.51976061 | 1.08578873 | +| native | 0 | 30 | 1.0 | 0.52095018 | 1.09054701 | +| native | 5 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 15 | 0.25 | 0.51899641 | 1.08273192 | +| native | 5 | 15 | 0.5 | 0.51948753 | 1.08469640 | +| native | 5 | 15 | 0.75 | 0.52120696 | 1.09157413 | +| native | 5 | 15 | 1.0 | 0.52409830 | 1.10313946 | +| native | 5 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 5 | 30 | 0.25 | 0.51903559 | 1.08288863 | +| native | 5 | 30 | 0.5 | 0.51874419 | 1.08172304 | +| native | 5 | 30 | 0.75 | 0.51888335 | 1.08227969 | +| native | 5 | 30 | 1.0 | 0.51945286 | 1.08455773 | +| native | 10 | 15 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 15 | 0.25 | 0.51861991 | 1.08122590 | +| native | 10 | 15 | 0.5 | 0.51848526 | 1.08068733 | +| native | 10 | 15 | 0.75 | 0.51935090 | 1.08414988 | +| native | 10 | 15 | 1.0 | 0.52118939 | 1.09150382 | +| native | 10 | 30 | 0.0 | 0.51975400 | 1.08576229 | +| native | 10 | 30 | 0.25 | 0.51902077 | 1.08282937 | +| native | 10 | 30 | 0.5 | 0.51858337 | 1.08107977 | +| native | 10 | 30 | 0.75 | 0.51844492 | 1.08052597 | +| native | 10 | 30 | 1.0 | 0.51860631 | 1.08117150 | +| global1000 | 0 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 15 | 0.25 | 0.53561169 | 1.15099039 | +| global1000 | 0 | 15 | 0.5 | 0.53658240 | 1.15487321 | +| global1000 | 0 | 15 | 0.75 | 0.53962210 | 1.16703201 | +| global1000 | 0 | 15 | 1.0 | 0.54454053 | 1.18670573 | +| global1000 | 0 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 0 | 30 | 0.25 | 0.53562902 | 1.15105970 | +| global1000 | 0 | 30 | 0.5 | 0.53524687 | 1.14953112 | +| global1000 | 0 | 30 | 0.75 | 0.53566622 | 1.15120852 | +| global1000 | 0 | 30 | 1.0 | 0.53687798 | 1.15605554 | +| global1000 | 5 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 15 | 0.25 | 0.53535586 | 1.14996706 | +| global1000 | 5 | 15 | 0.5 | 0.53561634 | 1.15100898 | +| global1000 | 5 | 15 | 0.75 | 0.53755790 | 1.15877521 | +| global1000 | 5 | 15 | 1.0 | 0.54106119 | 1.17278837 | +| global1000 | 5 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 5 | 30 | 0.25 | 0.53563652 | 1.15108971 | +| global1000 | 5 | 30 | 0.5 | 0.53500332 | 1.14855691 | +| global1000 | 5 | 30 | 0.75 | 0.53491749 | 1.14821358 | +| global1000 | 5 | 30 | 1.0 | 0.53538269 | 1.15007438 | +| global1000 | 10 | 15 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 15 | 0.25 | 0.53513905 | 1.14909983 | +| global1000 | 10 | 15 | 0.5 | 0.53481845 | 1.14781741 | +| global1000 | 10 | 15 | 0.75 | 0.53584572 | 1.15192652 | +| global1000 | 10 | 15 | 1.0 | 0.53814365 | 1.16111824 | +| global1000 | 10 | 30 | 0.0 | 0.53680422 | 1.15576050 | +| global1000 | 10 | 30 | 0.25 | 0.53575180 | 1.15155082 | +| global1000 | 10 | 30 | 0.5 | 0.53504651 | 1.14872968 | +| global1000 | 10 | 30 | 0.75 | 0.53469765 | 1.14733423 | +| global1000 | 10 | 30 | 1.0 | 0.53471068 | 1.14738634 | + +Outcome: both models selected start10/end30/strength0.75, but the improvement was small. Earlier/stronger shrinkage is not selected. The global model remains worse than the native model after this correction. + +Global1000 checkpoint SHA256 `5ab33ed446947b26133a4c48cd684232da8c6cd13f6dc7fda32ce2c794a340cf`. Study source SHA256 `2d9ddca63b268fbc9b52f5ac1bf800e7400ae7458fbd09f80a21cecaca84ffe9`. + +## E021 preparation — shared global-statistics residual head + +Hypothesis: predicting spatially uniform channel changes explicitly will help the pressure/global-mean dynamics that dominate radiative-layer error. Add one shared MLP head using the existing metadata embedding plus normalized means and standard deviations of each of the four context frames. It emits a uniform residual for every channel on every dataset; there are no dataset-specific specialist weights. The spatial network remains the selected native ConvNeXt body for this prototype. Added56075 parameters, total2815622, seed20261027. Output projection initialized to zero. + +Initialization check on one native TRAIN window: exact parent prediction parity (maximum difference0), finite nonzero output-layer gradient norm 0.043989721685647964. No optimizer updates, no new trained checkpoint, no validation per dataset and no publication. Parent checkpoint d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4. Consider only after the longer-horizon experiment; regenerate against the actually selected parent architecture if it changes. Prototype source SHA256 `8de97bfff13028b7d33adfd16564d8b7b4f00d2ac50e67d2f051549480cefbe8`. + +## E022a — fixed hidden-state uncertainty quadrature, training diagnostics + +Hypothesis: averaging fixed hidden-state uncertainty may reduce late chaotic active-fluid error. After the four-frame state encoder, evolve center and symmetric smooth higher-moment phase perturbations with the same fixed physical parameters. Perturbations leave observedF0/F2 unchanged and enforce F4 positivity. Fixed latent-noise seed20261026, phase standard deviations0.01 and0.0001. Compare equal quadrature weights and half-center/quarter-endpoint weights. No neural training, no new learned checkpoint; validation per dataset not evaluated. These are physical-state quadrature nodes, not separately trained dataset models. + +kinetic_state_quadrature_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.01 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.01 | 0.00162322 | +| [1, -1] | half_center | 0.01 | 0.00162206 | +| [3, -2] | center | 0.01 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.01 | 0.00183122 | +| [3, -2] | half_center | 0.01 | 0.00183157 | +| [9, -3] | center | 0.01 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.01 | 0.16464736 | +| [9, -3] | half_center | 0.01 | 0.12350380 | +| [17, -5] | center | 0.01 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.01 | 0.44187382 | +| [17, -5] | half_center | 0.01 | 0.42681998 | +kinetic_state_quadrature_small_results.json + +| Training parameters | Quadrature | Phase std | Mean VRMSE | +|---|---|---:|---:| +| [1, -1] | center | 0.0001 | 0.00161962 | +| [1, -1] | symmetric_mean | 0.0001 | 0.00162011 | +| [1, -1] | half_center | 0.0001 | 0.00161957 | +| [3, -2] | center | 0.0001 | 0.00183264 | +| [3, -2] | symmetric_mean | 0.0001 | 0.00183261 | +| [3, -2] | half_center | 0.0001 | 0.00183262 | +| [9, -3] | center | 0.0001 | 0.00258143 | +| [9, -3] | symmetric_mean | 0.0001 | 0.00258179 | +| [9, -3] | half_center | 0.0001 | 0.00257420 | +| [17, -5] | center | 0.0001 | 0.46776295 | +| [17, -5] | symmetric_mean | 0.0001 | 0.45577160 | +| [17, -5] | half_center | 0.0001 | 0.45775375 | + +Outcome:0.01 is rejected because severe moderate-activity degradation outweighs the high-activity gain. The0.0001 quadrature leaves low/moderate errors nearly unchanged and reduces the hardest training mean from0.467763 to0.455772, but requires three physical integrations. Expand train tests to four additional parameter combinations before considering validation. Center includes an extra FFT roundtrip and positivity projection, so tiny roundoff differences from E017 are expected. + +Additional KdV observation-filter test on train0/17/93, powers0/0.5/1/2: removing deconvolution worsens all three; doubling it helps17 slightly but barely changes93 and hurts0. Keep nominal finite8-cell observation model. No KdV runtime change. + +## Full validation of advection_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775038**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.516039 | + +Expanded-validation protocol: all561 non-active cases were replayed with the complete E018 package. All24 active outputs were reused from the expanded E017 validation after checking identical physical assets/configuration, exact input equality for reused quick cases, and thirty physical steps in every case. Hence those outputs are independent of the changed shared neural checkpoint. All585 validation trajectories are scored together by the official scorer. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of pressure_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775053**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261023; checkpoint SHA256 `d42dea4d135de34b46edad7bbdb797b897a289e4251e15d21185f7eebef093f4`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.515831 | + +Exact reuse protocol: model weights and all runtime operators except pressure regularization are identical to E018. Input metadata confirms only the nine radiative-layer cases contain a pressure field; their inputs match the already replayed E020 quick cases exactly. Reuse those nine new forecasts and576 unchanged E018 full forecasts. The official scorer evaluates all585 validation trajectories together. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E016 — shared global spectral mixing, update 2000, shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.75, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `96a49ac4734af31edd0aa5345f51515a899ef067e36abeb9e2347781c28fc055`. +- Validation: fixed 81-case subset; reward **0.807185**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E022b — expanded state-quadrature training study + +Changed factor: expand the fixed1e-4 hidden-state quadrature test from four to eight TRAIN parameter combinations, same latent seed20261026 and same fixed physical operator. No new learned weights. Validation per dataset not yet evaluated. + +| Additional training parameters | Quadrature | Mean VRMSE | +|---|---|---:| +| [5, -4] | center | 0.20956509 | +| [5, -4] | symmetric_mean | 0.20925570 | +| [5, -4] | half_center | 0.20933180 | +| [7, -3] | center | 0.00780154 | +| [7, -3] | symmetric_mean | 0.00781992 | +| [7, -3] | half_center | 0.00781529 | +| [11, -4] | center | 0.08376674 | +| [11, -4] | symmetric_mean | 0.08358850 | +| [11, -4] | half_center | 0.08358082 | +| [15, -4] | center | 0.26489031 | +| [15, -4] | symmetric_mean | 0.26059937 | +| [15, -4] | half_center | 0.26147711 | + +Across all eight cases, center mean 0.12997754, equal-weight quadrature 0.12788370. Outcome: select equal weights for validation; three physical integrations per active rollout. No publication from training results alone. + +## E016 — shared global spectral mixing, update 2000, no_shrink +- Hypothesis / changed factor: Continue the same shared model from the best checkpoint selected after phase4 update2000. Add a zero-initialized global Fourier feature-mixing residual (306552 new parameters). Native128x384 rectangular grids, eight-step loss, batch12, lr0.0001 cosine, input noise0.01, sampling probabilities0.05/0.075/0.075/0.8 across the four groups, sampling and added-parameter seed20261024; inherited convolutional initialization seed20261021. Periodic axes are independently translated for augmentation. Fixed profile shrinkage coefficient 0.0, with any train-selected pressure schedule preserved in config; material scalars remain nonnegative. Quick validation reruns all nine affected neural cases and reuses the matching fixed physical predictions for the other72 cases; the exact source is in physical_cache_source.json. See phase5/lineage.json for exact parent selection. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `f06079b546194d4243540a3578b4df4e547efc8b64bb67210a4d5fe41648b2bd`. +- Validation: fixed 81-case subset; reward **0.806418**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040813 | +| well/turbulent_radiative_layer_2D | 0.515442 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.775847**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.089485 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match E018. Parsed per-case logs verify552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E023 — shared spatial-mean forecasting branch + +Hypothesis: remaining radiative-layer mean-pressure drift can be predicted from four-frame global statistics and metadata while the shared autoregressive operator handles spatial structure. Train one common MLP across all19 datasets; every branch will be evaluated on every inference case. No dataset-specific checkpoint selection. Random initialization seed20261028; CPU-only, two threads, batch512, 12000 AdamW updates, cosine learning rate0.0005, half windows start0 and half random, horizons1–30 capped by available train frames, group probabilities0.1/0.1/0.1/0.7. Loss is smooth absolute mean error normalized by target spatial variance; targets never enter model inputs. Uses only the10115 audited training records. Spatial moments derived with float64 accumulation. No pretrained assets. + +Changed factor: add a shared horizon-conditioned mean predictor to the fixed autoregressive spatial forecast. CPU training is concurrent with the one-GPU 16-step experiment. This is distinct from the earlier untrained E021 residual-mean prototype. Validation and checkpoint selection pending. No package change at preparation. + +## E023a — validation mean-branch checkpoint and blend comparison + +Changed factors: shared mean-branch checkpoint at updates2000/6000/12000 and fixed blend0/.25/.5/.75/1. Exact spatial-MSE decomposition on the nine radiative-layer VALID trajectories, keeping the same global2000 spatial prediction. These are validation selection results, not training. Other dataset VRMSE unchanged by the composite physical branch, not reevaluated here. All mean-model weights were randomly initialized here with seed20261028, shared across all19 datasets. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.50464012 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.25 | 0.48902082 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.5 | 0.48371851 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 0.75 | 0.48967327 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 2000 | 1.0 | 0.50368029 | `badacb4c2f94a501aa92926036494dc9aa5687955c10185dd2e854cd27c80b01` | +| 6000 | 0.0 | 0.50464012 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.25 | 0.49150598 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.5 | 0.48589594 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 0.75 | 0.48943812 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 6000 | 1.0 | 0.50039246 | `23c0ccd4d870f645a506c1d03389a1dc68a5bdec98e7c5db2452af7b8534cf45` | +| 12000 | 0.0 | 0.50464012 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.25 | 0.48902526 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.5 | 0.48202704 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 0.75 | 0.48522405 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | +| 12000 | 1.0 | 0.49632276 | `966bbd7fbba235780d5d92d61715526f643d45192821195ef2dc5a322a5d1708` | + +Outcome: select update12000, blend0.5; all other tested choices are inferior. Full mean replacement is consistently worse than a partial blend. Actual float32 composite-package replay and determinism checks are required before publication; analytical score alone is not published. + +## E022 — fixed hidden-state quadrature +- Hypothesis / changed factor: Integrate three fixed hidden-angular-state quadrature nodes after the four-frame physical encoder, using one common neural model and identical fixed physical parameters. Nodes share all common computations; no separately trained specialists. Seed20261026, phase scale0.0001, equal weights, observedF0/F2 unchanged, F4 positivity preserved. Eight train trajectories supported this choice; larger perturbations were rejected. Rerun all twelve active validation cases, all30 physical steps; reuse sixty-nine matching non-active outputs. Any validated shared-neural progress during the replay is retained as documented in neural_rebase.json. No inference-time fitting. Inherited neural seeds/checkpoint in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `def27fe183c189d25508bcde6a013f2f6560b6b7d68526b782530bd4b1d7c730`. +- Validation: fixed 81-case subset; reward **0.807374**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Outcome: improvement; published atomically. + +## E023 — shared mean forecast, composite replay +- Hypothesis / changed factor: Add one shared horizon-conditioned spatial-mean forecasting branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261028; two CPU threads, batch512, 12000 total updates, selected update12000, blend0.5. Inputs are four context means and variances plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in mean_branch_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261024. +- Checkpoint SHA256: `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809033**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `79792621208d875ac021e5d2697bfbe1ac5bdd40b4c046982806e4fc785930a9`. +- Validation: fixed 81-case subset; reward **0.809649**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 1000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `e7dd6ac71780c453f0a23e995e48d0c94ad68af36df21c2d8993fb7ba914c8a7`. +- Validation: fixed 81-case subset; reward **0.809341**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477935 | + +Outcome: did not improve validation reward; retained current package. + +## E024 — spatial-profile conditioned shared mean branch + +Hypothesis: four-frame channel means and variances omit thermal-layer geometry; coarse profiles along each axis may predict global mean drift more accurately. Changed factor: add eight-bin spatial marginal profiles for both axes, all channels and all four context frames, to the same shared horizon-conditioned MLP. Profiles are centered by context means and normalized by context standard deviations. Random initialization seed20261029; otherwise same shared19-dataset training distribution and12000-update schedule as E023. CPU-only, two threads, no extra GPU. Train data membership audited before deriving profiles; float64 averaging followed by float32 storage. Feature parity between training-cache input and inference from raw TRAIN context passed (feature_checks.json). All parameters initialized from scratch; no pretrained assets. + +Validation pending. Compare against the raw spatial forecast from the currently selected phase6 update1000, obtained by setting only the old mean blend to zero and replaying nine valid contexts. The final candidate, if selected, will replace the existing shared mean branch rather than adding a dataset-specific specialist. No publication from training loss. + +## E024a — profile-branch validation selection + +Changed factors: profile-branch checkpoint2000/6000/12000 and fixed mean blend0/.25/.5/.75/1. Same9 valid radiative-layer cases and phase6 update1000 spatial forecast; the old mean branch was disabled for this ablation replay. Other dataset VRMSE not reevaluated at this selection stage. Shared random initialization seed20261029. + +| Update | Blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---| +| 2000 | 0.0 | 0.47627939 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.25 | 0.47072551 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.5 | 0.47161600 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 0.75 | 0.47837804 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 2000 | 1.0 | 0.48981324 | `e9409e9f47aa36cd88ad7baf9ba736fa759d68bdfc6e3afae18aa018e2d90b0d` | +| 6000 | 0.0 | 0.47627939 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.25 | 0.47073488 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.5 | 0.46949443 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 0.75 | 0.47282608 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 6000 | 1.0 | 0.48034141 | `277eb5b8ad2e959d77d56ed2ad765b7d1074dbe9929cb1380a637cc5275207b5` | +| 12000 | 0.0 | 0.47627939 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.25 | 0.47050116 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.5 | 0.46970411 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 0.75 | 0.47404605 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | +| 12000 | 1.0 | 0.48293819 | `35ca12ee9f289fa2bb62c4010aaf8677f9f1ad01e600383d82e4a85400a801e9` | + +Outcome: update6000/blend0.5 selected at0.46949443, versus0.47388509 with E023 and0.47627894 without any mean adjustment. Additional training to12000 improved training loss but slightly worsened validation. Actual composite replay and exact contract checks are running before any publication. + +## E024 — profile-conditioned shared mean forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a profile-conditioned branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261029; two CPU threads, batch512, 12000 total updates, selected update6000, blend0.5. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000 and blends0/.25/.5/.75/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_mean_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `8dc8554c96b10cffe81e8707063febceb81c18933dd345ac0dfaa329237377d6`. +- Validation: fixed 81-case subset; reward **0.809986**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Outcome: improvement; published atomically. + +## E025 — jointly forecast spatial means and coarse profiles + +Hypothesis: remaining broad thermal-layer structure errors require a forecast of spatial profiles, not only a corrected global mean. Changed factor: shared MLP predicts both global channel means and centered eight-bin marginal profiles on both axes. Enforce agreement between the predicted profile means and global means. Train across all19 datasets from random initialization seed20261030 with the same audited train statistics/profiles,12000 CPU-only updates, batch512, two threads; loss0.25 normalized mean error +0.75 normalized coarse-profile error, with the same smooth absolute form. No new data or pretrained assets. Every branch remains shared and evaluated on every case. + +Inference candidate will interpolate centered coarse-profile residuals with periodic or clamped edges according to metadata, and blend them into the common autoregressive forecast. Physical outputs remain authoritative. Uniform and spatial correction coefficients will be selected on validation, followed by actual package replay and exact determinism checks. Initialization shape, finite-output, zero-correction identity, and mean/profile consistency checks passed on an audited TRAIN context. Validation pending; no publication from training loss. + +## E025a — joint mean/profile validation selection + +Changed factors: joint statistic checkpoint2000/6000/12000, mean correction0/.5/1 and profile correction0/.25/.5/1. Same9 VALID radiative-layer trajectories and phase6 update1000 raw spatial forecasts. MSE decomposition includes the uniform correction, centered profile correction, and all cross terms; actual float32 replay is required next. Shared random seed20261030. Other dataset VRMSE not reevaluated in this selection stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.25 | 0.47635292 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 0.5 | 0.47958239 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.0 | 1.0 | 0.49471505 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.0 | 0.47005735 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.25 | 0.47010562 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 0.5 | 0.47337008 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 0.5 | 1.0 | 0.48871343 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.0 | 0.48430118 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.25 | 0.48426968 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 0.5 | 0.48736599 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 2000 | 1.0 | 1.0 | 0.50220535 | `41b75c10287fe9d72930cee5710078584a93995b8f778d379463590a08606946` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.25 | 0.47518010 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 0.5 | 0.47612881 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.0 | 1.0 | 0.48395044 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.0 | 0.46912443 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.25 | 0.46794480 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 0.5 | 0.46886089 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 0.5 | 1.0 | 0.47675092 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.0 | 0.47781927 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.25 | 0.47659139 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 0.5 | 0.47738680 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 6000 | 1.0 | 1.0 | 0.48486302 | `da88489f1d47caf35fb6ce42372af4de5d94e481b0c3a2cca498a6312de3cc3e` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.25 | 0.47462923 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 0.5 | 0.47500679 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.0 | 1.0 | 0.48171365 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.0 | 0.46724263 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.25 | 0.46555366 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 0.5 | 0.46594474 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 0.5 | 1.0 | 0.47282504 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.0 | 0.47431715 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.25 | 0.47263591 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 0.5 | 0.47296605 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | +| 12000 | 1.0 | 1.0 | 0.47954663 | `c314a10729fb3e715b8042be2a35d27bc7bc7d19ae49f3b8dda090076af9e916` | + +Outcome: select update12000, mean blend0.5, profile blend0.25, VRMSE0.46555366 versus E0240.46949440. At update2000 any profile correction worsened the best mean-only result; profile prediction benefited from further training. Larger profile blends were inferior. Actual replay and contract checks pending; no analytical-only publication. + +## E025 — joint shared mean/profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261030; two CPU threads, batch512, 12000 total updates, selected update12000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and eight-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `67752d3d72ef3e1cf10485e57abe77cdbc9e99eb48ccb69ce72c8d12d6ab62bd`. +- Validation: fixed 81-case subset; reward **0.810291**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Outcome: improvement; published atomically. + +## E026 — sixteen-bin shared statistical forecast + +Hypothesis: eight-bin profiles smooth thin thermal interfaces; sixteen bins may improve the shared coarse forecast. Changed factor: double the input and output profile resolution on both axes from8 to16 bins, retaining one shared MLP across all19 datasets. Random initialization seed20261031. Same12000-update CPU-only training schedule, two threads, group sampling, loss weighting and horizon distribution as E025; the wider input/output layers increase parameter count. Train-only profile cache membership audited before reading, with float64 averaging. Source snapshot and train/inference feature-parity checks retained in profile16_forecast_v1. No pretrained assets. + +Validation pending; compare on the same9 valid thermal trajectories and the same phase6 update1000 raw spatial forecast. No package change before actual replay and exact contract checks. + +## E027 — pressure schedule after the learned profile decoder + +Hypothesis: the previous pressure uncertainty schedule may no longer be optimal after adding a shared profile forecast. Changed factors: pressure shrink strength0/.25/.5/.75/1, ramp start0/5/10/15 and fixed end30; zero-strength baseline evaluated once. Keep profile shrink0.75 and the trained mean/profile branch fixed. Use nine audited TRAIN trajectories, one per thermal cooling parameter, for schedule selection. No validation targets enter this choice and no new weights are trained. Raw shared CNN predictions replayed with both statistical corrections and regularization disabled, then fixed candidate decoders applied to the same outputs. Inherited neural sampling seed20261025, auxiliary seed20261030; base checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `56b7067dd85f4ba3aa13b34aafc87e1267cf7c14e7a898de9c50044b1f559c04`. Validation per dataset pending. + +## E026a — sixteen-bin validation selection + +Shared initialization seed20261031; same9 valid thermal trajectories and same phase6 update1000 raw spatial forecasts as E025. Changed checkpoint and fixed correction strengths. Other datasets not reevaluated at this stage. + +| Update | Mean blend | Profile blend | Radiative-layer VRMSE | Training checkpoint SHA256 | +|---:|---:|---:|---:|---| +| 2000 | 0.0 | 0.0 | 0.47627939 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.25 | 0.47600559 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 0.5 | 0.47978306 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.0 | 1.0 | 0.49835827 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.0 | 0.47140293 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.25 | 0.47106825 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 0.5 | 0.47484781 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 0.5 | 1.0 | 0.49358365 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.0 | 0.48622849 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.25 | 0.48578095 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 0.5 | 0.48932182 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 2000 | 1.0 | 1.0 | 0.50735469 | `a9ba14c74680bd13117a9e9c85477489824833ab96ad48c8755dbe581951f79c` | +| 6000 | 0.0 | 0.0 | 0.47627939 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.25 | 0.47423524 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 0.5 | 0.47528103 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.0 | 1.0 | 0.48625053 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.0 | 0.46620666 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.25 | 0.46415542 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 0.5 | 0.46526516 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 0.5 | 1.0 | 0.47656498 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.0 | 0.47363337 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.25 | 0.47157736 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 0.5 | 0.47258672 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 6000 | 1.0 | 1.0 | 0.48345739 | `bdcfcc729a257d0aec15b8e7b2d37214f0f6e267245c4d6c359e0b6663534919` | +| 12000 | 0.0 | 0.0 | 0.47627939 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.25 | 0.47413693 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 0.5 | 0.47468663 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.0 | 1.0 | 0.48361862 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.0 | 0.46656666 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.25 | 0.46436542 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 0.5 | 0.46493348 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 0.5 | 1.0 | 0.47413113 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.0 | 0.47745171 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.25 | 0.47524133 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 0.5 | 0.47569852 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | +| 12000 | 1.0 | 1.0 | 0.48441577 | `ef40c373b4bd0c8aa3239760ecc6982841db6644a79578c1cd5c5736daf286f2` | + +Outcome: select update6000, mean0.5/profile0.25, VRMSE0.46415542 versus eight-bin E0250.46555365. Additional training to12000 slightly worsened validation. Actual replay and contract checks pending. + +## E027a — training-only pressure schedule results + +No new learned weights or changed seed. Exact parent neural and auxiliary hashes are recorded in E027 and decoder_v1/parent.json. Nine audited TRAIN trajectories only; validation not yet evaluated. + +| Pressure strength | Ramp start | Training radiative-layer VRMSE | +|---:|---:|---:| +| 0.0 | 10 | 0.45035356 | +| 0.25 | 15 | 0.45038955 | +| 0.25 | 10 | 0.45050631 | +| 0.5 | 15 | 0.45065454 | +| 0.25 | 5 | 0.45075753 | +| 0.5 | 10 | 0.45097174 | +| 0.25 | 0 | 0.45108611 | +| 0.75 | 15 | 0.45114587 | +| 0.5 | 5 | 0.45158755 | +| 0.75 | 10 | 0.45174613 | +| 1.0 | 15 | 0.45185799 | +| 0.5 | 0 | 0.45240983 | +| 1.0 | 10 | 0.45282187 | +| 0.75 | 5 | 0.45283721 | +| 0.75 | 0 | 0.45431139 | +| 1.0 | 5 | 0.45449514 | +| 1.0 | 0 | 0.45677051 | + +Outcome: select zero additional pressure smoothing, training VRMSE0.45035356 versus0.45174613 for the inherited0.75/start10 schedule. The earlier pressure schedule helped older checkpoints but is a negative result for the current trained profile decoder. Validate this fixed choice after the sixteen-bin candidate finishes, preserving all latest shared weights. + +## E026 — sixteen-bin shared profile forecast, composite replay +- Hypothesis / changed factor: Replace the shared horizon-conditioned spatial-mean forecasting branch with a joint mean-and-profile forecast branch to the common autoregressive model. All19 datasets trained together from random initialization seed20261031; two CPU threads, batch512, 12000 total updates, selected update6000, mean blend0.5, profile blend0.25. Inputs are four context means, variances, and sixteen-bin spatial profiles on both axes, plus metadata and horizon; the same branch is evaluated on every case. Physical updates remain authoritative. Compare updates2000/6000/12000, mean blends0/.5/1, and profile blends0/.25/.5/1 on the nine valid neural cases; all analytical results, including regressions, are retained in profile16_forecast_v1/validation/results.json and the ledger. Replay the complete candidate on nine neural validation cases, reuse72 identical physical outputs from the documented parent, then exact determinism and novel-grid checks. Auxiliary checkpoint hash and source are in model_manifest.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6d988acd1229ae39ec83992e389a01e05d197ab496ab79cbae92cb124f39ad16`. +- Validation: fixed 81-case subset; reward **0.810400**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Outcome: improvement; published atomically. + +## E027 — updated pressure uncertainty schedule +- Hypothesis / changed factor: Replace the inherited pressure-smoothing schedule with train-selected strength0.0, start10, end30. Nine audited TRAIN trajectories selected this fixed schedule with the eight-bin profile decoder; validation then applies it to the latest selected shared weights, including any validated sixteen-bin branch. No new learned weights or inference-time fitting. Parent/checkpoint hashes and all train schedule results are recorded. Rerun nine neural validation cases; reuse72 identical physical outputs. Exact determinism checks before publication. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810428**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Outcome: improvement; published atomically. + +## E028 — statistical correction inside the autoregressive loop + +Hypothesis: applying mean/profile corrections only after the rollout lets internal state drift contaminate later predictions. Changed factor: after each shared neural update, feed a fixed fraction of the precomputed mean/profile correction into the next autoregressive context. Forecast statistics still depend only on the original four frames and metadata; weights remain fixed. Test fractions0/.25/.5/1 on nine audited TRAIN trajectories; the final output decoder remains unchanged. Physical steps remain authoritative and their latent states are unaffected. No new learned weights. Inherited neural seed20261025, auxiliary seed20261031; shared neural checkpoint `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`, auxiliary checkpoint `3f9e62adca11d1dc83c908b9bbc7da794b5ca7902ecec1401a1519d72b6cd1d2`. Validation pending. + +## E028 — autoregressive statistical feedback +- Hypothesis / changed factor: Feed fraction0.25 of the fixed shared mean/profile correction into each next neural autoregressive state. Final output correction remains unchanged. Statistics are forecast once from the four original context frames and metadata; parameters never change at inference. Train-only fractions0/.25/.5/1 selected this fixed strength on nine audited thermal trajectories. No new learned weights. Physical latent states are unaffected. Rerun nine valid neural cases and reuse72 complete physical outputs, then exact neural, unknown-grid, smoke and diffusion checks before publication. Parent weights and inherited neural/auxiliary seeds are in model_manifest.json and parent records. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `c05aa38ecb979b8d52316d1cf0a5c0b35cb542ab5c79c66f8c73cbe300ee4d9c`. +- Validation: fixed 81-case subset; reward **0.810000**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469308 | + +Outcome: did not improve validation reward; retained current package. + +## E028a — training-only feedback strength + +Same inherited weights and seeds as E028; nine audited TRAIN trajectories only. Changed factor: fixed fraction of the existing output correction inserted into each next neural context. All physical branches retain their complete physical outputs. Validation per dataset pending. + +| Feedback fraction | Training radiative-layer VRMSE | Failed cases | +|---:|---:|---:| +| 0.0 | 0.45007325 | 0 | +| 0.25 | 0.44391940 | 0 | +| 0.5 | 0.44519505 | 0 | +| 1.0 | 0.44886765 | 0 | + +Outcome: select fraction0.25 at0.44391940 versus0.45007325 without feedback. Stronger feedback gives smaller gains; no new learned weights. Actual validation and exact contract checks running. + +## Full validation of quadrature_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.776678**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.504640 | + +Expanded fixed state quadrature from12 to all24 active validation trajectories. Replayed the missing12, reused12 quick outputs only after exact context equality, and verified30/30 common physical steps for every active case. The552 corpus/smoke outputs are reused from global2000 full validation after exact physical asset/configuration equality checks and complete physical coverage. The nine neural outputs come from this exact package’s quick validation with matching inputs. All585 cases scored together; no held-out trajectories or inference caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E019 — longer autoregressive training, update 2000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.811223**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 2000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.811195**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.454067 | + +Outcome: did not improve validation reward; retained current package. + +## Full validation of mean_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778337**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261024; checkpoint SHA256 `64257b71008a23332418862181c0ae9dd80dfd31ee9f9227fc60a44af4bc0b7b`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.482027 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval001000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.778953**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.473885 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_mean_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779290**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.469494 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of eval002000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780527**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453721 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of decoder_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779732**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.463803 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779595**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.465554 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of profile16_forecast_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779704**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.464155 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E029 — hidden-state quadrature width + +Hypothesis: the selected1e-4 hidden-state spread may understate uncertainty in chaotic active flow. Changed factor: phase widths1e-4/3e-4/1e-3, keeping three equal nodes, fixed latent seed20261026, identical physical encoder/parameters, and identical common-prefix guards. First study four TRAIN parameter combinations (zeta,alpha)=(5,-4),(9,-3),(15,-4),(17,-5), including the moderate case sensitive to excessive perturbation. Reuse the same encoded state and center node within each trajectory. Any improvement must be expanded before validation. No new learned weights. Kinetic source SHA256 `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`; parent package `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Validation pending. + +## E030 — learned coarse-statistic guidance preparation + +Hypothesis motivated by E028: inference-only feedback does not generalize, but the neural operator may learn how to use the same conditional forecast. Add a shared1x1 stem adapter from normalized future mean/profile fields into the common convolutional operator. Its528 weights initialize to zero, exactly preserving the selected parent before training. Fixed auxiliary forecasts consume only the original four context frames and metadata. Main neural parameters after addition3066627; initialization seed20261101. Parent `/run-state/workspace/research/universal_v1/phase6/eval002000_shrink/package`. Exact CPU forward parity on an audited native TRAIN context: maximum difference0.0; nonzero finite adapter gradient norm3.3356419 on a coarsened TRAIN window. Source in guided_v1/model.py. + +Outcome: untrained architecture preparation only. No optimizer steps, validation selection, package publication, new GPU trainer, or phase7 launcher yet. Current phase6 remains the only GPU trainer. The future training stage must start from the then-selected checkpoint and recheck initialization rather than treat this probe as a trained candidate. + +## E019 — longer autoregressive training, update 1000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `63077f85913445ad106cd0005c404f053416c65c5a52adb2440f49b37be827c6`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809399**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.477170 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 2000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.810422**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463873 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810313**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.465273 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `f8c141326b7752d6baaf02c28f1669d7f2d29f17935105920c3443f2db164ff2`. +- Validation: fixed 81-case subset; reward **0.810189**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466865 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 3000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `a9c1ed98932bf2a9a1d43fb0883e39ffcc8e185c44fafedb212a54901a3513a0`. +- Runtime code SHA256: `13d5d53dffec34063c41000601dff2c99f98729d80c1fae03c707d8e8206436b`. +- Validation: fixed 81-case subset; reward **0.809097**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.481171 | + +Outcome: did not improve validation reward; retained current package. + +## E029a — quadrature-width training results + +Same fixed physical encoder and latent seed20261026. Four TRAIN parameter combinations; every node completed30 physical steps. No new learned weights. + +| Training parameters | Phase width | Active VRMSE | +|---|---:|---:| +| [5, -4] | 0.0001 | 0.20925571 | +| [5, -4] | 0.0003 | 0.20655669 | +| [5, -4] | 0.001 | 0.19077064 | +| [9, -3] | 0.0001 | 0.00258179 | +| [9, -3] | 0.0003 | 0.00410363 | +| [9, -3] | 0.001 | 0.04487926 | +| [15, -4] | 0.0001 | 0.26059940 | +| [15, -4] | 0.0003 | 0.24752074 | +| [15, -4] | 0.001 | 0.23641607 | +| [17, -5] | 0.0001 | 0.45577160 | +| [17, -5] | 0.0003 | 0.42649901 | +| [17, -5] | 0.001 | 0.42283213 | + +Four-case means: {0.0001: 0.23205212625907734, 0.0003: 0.22117002040613443, 0.001: 0.22372452542185783}. Outcome: select3e-4 for an expanded TRAIN check on (1,-1),(3,-2),(7,-3),(11,-4). The1e-3 spread harms the formerly accurate (9,-3) case and is inferior overall. Validation pending. + +## E031 — pressure from momentum balance + +Hypothesis: a fixed momentum-balance projection could recover pressure structure missed by the neural forecast. Independently implement a discrete periodic-x/free-y gradient least-squares inverse with FFT/DCT, finite-time momentum differences and conservative flux divergence. A manufactured scalar-potential check verifies the inverse to roundoff. First use true TRAIN density/velocity only as an oracle feasibility diagnostic; no future truth is supplied at inference. Per-case fitted scalars below are diagnostic only and are never deployed. Nine audited training trajectories. No new learned weights or new seed. + +| Training cooling case | Nominal oracle pressure VRMSE | Pressure correlation | Diagnostic optimal scalar | +|---|---:|---:|---:| +| data/train/turbulent_radiative_layer_tcool_0.03.hdf5 | 0.652605 | 0.753539 | 0.886568 | +| data/train/turbulent_radiative_layer_tcool_0.06.hdf5 | 0.787532 | 0.646115 | 0.683058 | +| data/train/turbulent_radiative_layer_tcool_0.10.hdf5 | 0.835901 | 0.640075 | 0.627110 | +| data/train/turbulent_radiative_layer_tcool_0.18.hdf5 | 1.095653 | 0.426235 | 0.393953 | +| data/train/turbulent_radiative_layer_tcool_0.32.hdf5 | 1.103630 | 0.476517 | 0.405224 | +| data/train/turbulent_radiative_layer_tcool_0.56.hdf5 | 1.789119 | 0.191514 | 0.107686 | +| data/train/turbulent_radiative_layer_tcool_1.00.hdf5 | 1.531407 | 0.145077 | 0.105623 | +| data/train/turbulent_radiative_layer_tcool_1.78.hdf5 | 1.578657 | 0.058558 | 0.042445 | +| data/train/turbulent_radiative_layer_tcool_3.16.hdf5 | 1.457004 | 0.046748 | 0.039365 | + +The oracle is useful mainly for the shortest cooling times and unreliable at long cooling times. Then use only the selected model’s predicted density/velocity, keeping its predicted global pressure mean, and blend the nominal projection at a fixed strength. Parent/checkpoint hashes are in momentum_pressure_v1/forecast_parent.json. + +| Projection blend | Training thermal VRMSE | +|---:|---:| +| 0.0 | 0.43522565 | +| 0.05 | 0.43501140 | +| 0.1 | 0.43541418 | +| 0.25 | 0.44012596 | +| 0.5 | 0.45764482 | +| 1.0 | 0.51557828 | + +Outcome: full-strength projection is harmful; a0.05 blend gives only a small training gain. Test that single train-selected strength on validation before deciding whether a self-contained Torch implementation is warranted. No physical projection is currently installed in the package. + +## E029b — expanded quadrature-width training study + +Changed factor: expand the selected3e-4 width from four to eight TRAIN parameter combinations. Same fixed latent seed20261026 and physical encoder; no new learned weights. Validation pending. + +| Additional training parameters | Phase width | Active VRMSE | Complete physical steps | +|---|---:|---:|---:| +| [1, -1] | 0.0001 | 0.00162011 | 30 | +| [1, -1] | 0.0003 | 0.00162016 | 30 | +| [3, -2] | 0.0001 | 0.00183261 | 30 | +| [3, -2] | 0.0003 | 0.00183292 | 30 | +| [7, -3] | 0.0001 | 0.00781992 | 30 | +| [7, -3] | 0.0003 | 0.00795999 | 30 | +| [11, -4] | 0.0001 | 0.08358850 | 30 | +| [11, -4] | 0.0003 | 0.08545388 | 30 | + +Eight-case means:1e-4=0.12788370,3e-4=0.12269338. Outcome: proceed to validation. + +## E031 — fixed momentum pressure correction +- Hypothesis / changed factor: Fixed0.05 pressure blend from forecast momentum balance, selected on nine audited TRAIN trajectories. Uses context plus predicted density/velocity, finite-time derivatives, conservative flux divergence, and a periodic-x/free-y gradient least-squares projection. Preserves predicted global pressure mean. No fitted diagnostic scalars are deployed and no new learned weights. Larger blends were harmful. A separate validation diagnostic found only a tiny gain; this replay uses the self-contained Torch implementation and official scorer, with exact neural/novel-grid/smoke/diffusion checks before publication. Seventy-two complete physical outputs are unchanged; all model components and inherited seeds/hashes remain in the manifest. No future truth, external files, SciPy, or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `cd7a9a8c2c2038f3be273f65987972072405494507695c64397ef3d3c2e2ae27`. +- Validation: fixed 81-case subset; reward **0.811224**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E019 — longer autoregressive training, update 4000, shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `6e1bf0b3ad3887096d1f9cba6a00fed3d9e03bcd8f7e7956ae8bbb856aa6f0b9`. +- Validation: fixed 81-case subset; reward **0.810991**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.456640 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_shrink +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.0; statistical output corrections enabled=True. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `3cfd637325d7e7313d4bfbd48967518b107446b1be900f8859acaa7f67b51793`. +- Validation: fixed 81-case subset; reward **0.810338**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.464955 | + +Outcome: did not improve validation reward; retained current package. + +## E019 — longer autoregressive training, update 4000, no_statistics +- Hypothesis / changed factor: Continue the validation-selected shared model after phase5 update2000. Preserve its architecture and fixed physical operators. Sixteen-step fully backpropagated autoregressive loss, capped at ten steps for the shorter Arena trajectories; batch6, lr0.00005 cosine, native rectangular grids, input noise0.01, group sampling probabilities0.05/0.075/0.075/0.8. Sampling seed20261025; inherited initialization and any global-parameter seed are in phase6/lineage.json. Fixed profile shrinkage coefficient 0.75; statistical output corrections enabled=False. The statistical branch is still evaluated on every case. Any train-selected pressure schedule is preserved in config. Rerun nine neural validation cases; reuse seventy-two matching physical outputs with exact source in physical_cache_source.json. +- Seed: 20261025. +- Checkpoint SHA256: `d5cb68e953191a4ef430c992e5d52d8983b315e57badf4c9f389c0f4858d452e`. +- Runtime code SHA256: `8bf82697e735e6de13b39fd00292c1ce9bda307e9a904af1be6a2f4f99333878`. +- Validation: fixed 81-case subset; reward **0.810252**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.466053 | + +Outcome: did not improve validation reward; retained current package. + +## E030 training launch + +Phase6 updates3000 and4000 did not beat update2000. Retired the phase6 trainer, its six data workers and watcher after all4000-update validation variants completed; all checkpoints and negative results remain available. Verified those processes exited before launching the next single-GPU trainer. Initialize528 guidance-adapter weights to zero on the latest selected package, preserving its physical operators and fixed auxiliary forecast. Sampling seed20261102, guidance initialization seed20261101; inherited neural initialization seed20261021 and auxiliary seed20261031. Sixteen-step loss capped by available trajectory length, batch6, learning rate0.00003 with cosine decay, noise0.01, group probabilities0.05/0.075/0.075/0.8, native thermal grids, planned6000 updates. Frozen auxiliary weights are copied into phase7/fixed_statistics. Exact parent/checkpoint hashes and stopped PIDs are in phase7/lineage.json. No pretrained assets. Validation pending; the selected package is unchanged. + +## Full validation of momentum_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780529**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E032 — Preserve observed angular constraints in state quadrature (preparation) + +Hypothesis: perturbing F4 changes an angular moment already constrained by the observed flow; rotating F6 and higher may preserve short-term accuracy while averaging latent uncertainty. Changed factor: first perturbed mode, holding the fixed operator and all learned weights fixed. Training study will compare the existing F4+ width3e-4 control with F6+ widths3e-4 and1e-3 on four audited TRAIN parameter regimes. Fixed latent seed20261026. No new trained weights; inherited checkpoint and physical hashes will be saved in latent_constraint_v1/source/model_manifest.json and parent.json. No validation results yet and no package change. Heavy physical study is queued after E029 expanded validation. + +## E033 — Shared coarse two-dimensional interaction forecast (preparation) + +Hypothesis: separate axis profiles cannot represent the spatial interaction component, which accounts for more than half of the remaining pressure and transverse-velocity normalized squared error in the current valid neural rollouts. Changed factor: add a shared horizon-conditioned MLP forecasting the interaction component of4×16 coarse fields. Every model branch will be evaluated on every case; complete physical prefixes remain authoritative. Train across all10115 audited TRAIN records and all19 datasets, random initialization seed20261103,12000 updates, batch512, width192, two CPU threads, group probabilities0.1/0.1/0.1/0.7. Inputs only four context fields/statistics and metadata; loss uses target full-field variance during training. Select updates2000/6000/12000 and fixed output blends0/.25/.5/1 on valid neural rollouts. No pretrained assets. Dataset VRMSE and checkpoint hashes pending. + +Training-cache/runtime feature parity passed exactly on one audited train example from each group. Zero-initialized forecasts equal last-frame coarse interactions exactly. Interpolation preserves global means and marginal profiles within float32 rounding. Sources and checks are retained under coarse_interaction_v1. No inference package change yet. + +## E029 — wider fixed state quadrature +- Hypothesis / changed factor: Increase the fixed hidden-state phase quadrature width from1e-4 to3e-4, retaining three equal nodes, the same fixed physical parameters, and latent seed20261026. Eight audited TRAIN trajectories supported the choice;1e-3 was rejected. No learned weights changed. Replay all twelve active VALID cases and verify30/30 physical steps, reuse69 unchanged outputs, and retain any newer validated neural/decoder package before scoring. Exact determinism and novel-grid checks precede publication. Inherited neural/auxiliary hashes and seeds are in the manifest. +- Seed: 20261025. +- Checkpoint SHA256: `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812643**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Outcome: improvement; published atomically. + +## E034 — late-checkpoint parameter averaging +- Hypothesis / changed factor: Equal parameter average of phase6 updates2000/3000/4000, a single predeclared candidate to reduce late-training optimization variation. No new optimization or data fitting; all three checkpoints were trained together across all19 datasets, initialization seed20261021 and sampling seed20261025. Accumulate in float64 and return original parameter dtype. Same shared averaged weights apply to every case. Preserve current fixed physical operators and shared auxiliary decoders. Source training checkpoint hashes and equal coefficients are in model_manifest.json and weight_average_v1/lineage.json. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact determinism and novel grids if improved. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `b550a87bfb3428c51ba9c4aea3c27cafa1a714de17372e387be9dc041d9ffdad`. +- Validation: fixed 81-case subset; reward **0.812766**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.452162 | + +Outcome: improvement; published atomically. + +## E033a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed factors: training checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_interaction_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.25 | 0.45186873 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 0.5 | 0.45240534 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 2000 | 1.0 | 0.45590097 | `547715dd80dc733986dc756fe92d96fed0f68926438f1a80ffb2a99c987b8488` | +| 6000 | 0.0 | 0.45216160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.25 | 0.45148496 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 0.5 | 0.45166160 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 6000 | 1.0 | 0.45455921 | `4998f5dd3904a32be09f01f715a28acaa9b4c548c64a6bc9146659691f7c09c1` | +| 12000 | 0.0 | 0.45216160 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.25 | 0.45151015 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 0.5 | 0.45176523 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | +| 12000 | 1.0 | 0.45496820 | `6c23c212f450cc9ddfdfda94aa0f524f76c0ef33f5374e26004914cc321d06e4` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E035 — Periodic translation augmentation for the coarse branch (preparation) + +Hypothesis: a dense coarse-field predictor should see translated versions of periodic patterns during training. Changed factor: random whole-cell shifts along metadata-declared periodic axes, applied identically to all four context grids and the target. Keep architecture, initialization seed20261103, training sample sequence,12000 updates and all optimizer settings from E033. Use a separate augmentation RNG seed20261104. Train across all19 datasets on two CPU threads; no additional GPU trainer. Marginal means and variances are invariant to these cell permutations. Compare the replacement branch on the current shared neural parent with its previous coarse correction disabled, then replay the full proposed package. Validation results and checkpoint hashes pending. + +## E033 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Add one shared coarse interaction forecast branch, trained from random seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update6000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E033a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `ead6921e9fcb8924118472a881e78c2e0cf57f1a3b23d848b9ddaadfa36bc4ab`. +- Validation: fixed 81-case subset; reward **0.812821**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451485 | + +Outcome: improvement; published atomically. + +## E035a — coarse interaction checkpoint and blend selection + +Same shared branch trained across all19 datasets, random seed20261103. Changed training factor: random coarse-cell translations on periodic axes with augmentation seed20261104, preserving initialization seed20261103 and the training sample sequence. Selection factors: checkpoint and fixed output blend. All nine neural validation trajectories; physical outputs are unchanged. Each checkpoint is identified by its training-file SHA256 below. Full source and training configuration are in coarse_equivariant_v1/phase1/source and config.json. + +| Update | Blend | Radiative-layer VRMSE | Checkpoint SHA256 | +|---|---:|---:|---| +| 2000 | 0.0 | 0.45216160 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.25 | 0.45326987 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 0.5 | 0.45515343 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 2000 | 1.0 | 0.46107193 | `f7eccd5c589dddd1f05e1be14378e442c89524323bc4c2e1ed4fd6584b61c456` | +| 6000 | 0.0 | 0.45216160 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.25 | 0.45202987 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 0.5 | 0.45258554 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 6000 | 1.0 | 0.45570121 | `7b7d3f8e517702b18a518497fdab3648b02b01940b79111045de2e1393114044` | +| 12000 | 0.0 | 0.45216160 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.25 | 0.45188915 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 0.5 | 0.45231514 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | +| 12000 | 1.0 | 0.45521523 | `d45b11f7789c5250563751c49b12a63b80f8c8efddfd3c543e4af006eb2f8601` | + +Unchanged validation datasets for every analytical ablation: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: proceed to complete package replay. + +## E030 — learned coarse-statistic guidance, update 1000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `240df3b587dc919b7d002c7fbcc51ab99b3510a1cedd49f0008c6e7a52b82764`. +- Validation: fixed 81-case subset; reward **0.812394**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.456839 | + +Outcome: did not improve validation reward; retained current package. + +## E035 — shared coarse interaction forecast, composite replay +- Hypothesis / changed factor: Replace the shared coarse interaction forecast with one trained using periodic-axis translations, augmentation seed20261104, retaining random initialization seed20261103 on all10115 audited training records and all19 datasets. Two CPU threads,12000updates, batch512, width192, selected update12000 and fixed blend0.25. Four context coarse fields, means, variances, metadata and horizon predict the two-dimensional interaction component at4 by16 resolution. The fixed correction preserves existing spatial means and marginal profiles; every branch is evaluated on every case, while complete physical prefixes remain authoritative. Replay nine neural valid cases, reuse72 unchanged full physical outputs, and check exact neural/active/diffusion/novel-grid determinism before publication. All checkpoint/blend results including negative ablations appear in E035a. Training and packaged checkpoint hashes are in model_manifest.json. No pretrained assets or inference-time fitting. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `7d340568043fdd29cfb0382ba3c1065e35f9172396a7f37b5a0809dedecf9f16`. +- Validation: fixed 81-case subset; reward **0.812788**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451889 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `34cea33fb726b0029c2c1734a94ceee72bd4e73bbf17e80cc9bc95a3db954165`. +- Validation: fixed 81-case subset; reward **0.811915**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.462924 | + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 1000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `1ecd27a33702b3ea517ebae81477612153fd68e0043448960fb16c98ac3b12a1`. +- Runtime code SHA256: `d52a0fb2cbf3e4000004b0c088c4c1143ac393ce9c9432f1cc68270ac19d5778`. +- Validation: fixed 81-case subset; reward **0.811492**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.468370 | + +Outcome: did not improve validation reward; retained current package. + +## E036 — Cross-channel covariance features with continuation control (preparation) + +Hypothesis: spatial density/pressure correlations and other channel interactions contain dynamical information absent from separate means, variances and marginal profiles. Extend the selected E026 universal mean/profile branch with92928 zero-initialized covariance projection weights. Both a frozen-zero control and a learned-covariance variant continue from the same session-trained checkpoint, using identical samples and sampling seed20261105,6000 updates, batch512, learning rate0.0001 cosine, all19 datasets and two CPU threads each. Inherited random initialization seed20261031. Only difference between variants: whether covariance projection weights can learn. Four-frame context covariances use centered float64 accumulation, normalized by context channel scales. + +Exact parent-output parity passed on one audited TRAIN context from each group. Cache/runtime covariance differences were zero except1.06e-22 in one active-matter element. The cache contains only audited10115 TRAIN records; membership was checked before reading. Compare updates1000/3000/6000 by full neural runtime replay with the selected parent’s fixed output blends. Dataset VRMSE and checkpoint hashes pending. No inference parameter fitting and no pretrained assets. Sources, checks and logs: covariance_statistics_v1. + +## E036a — controlled statistical continuation and covariance feature comparison + +Continue the selected E026 shared auxiliary checkpoint with identical training samples, optimizer schedule and seed20261105 for6000 updates. Inherited random initialization seed20261031. Both variants add a zero-initialized covariance projection; control freezes it at zero, covariance learns it jointly. Added92928 weights. All19 datasets trained together, two CPU threads per variant, batch512, learning rate0.0001 cosine, existing mean/profile loss0.25/0.75. Four-frame spatial cross-channel covariances are normalized by context scales. Fixed parent decoder blends retained; every candidate is replayed through the full neural runtime, including learned guidance if present. No pretrained assets or inference-time fitting. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45146565 | 0.81282209 | `1c547715b94da883224d53fc40a4cc53362a65e59b78328ebb0debc00b7fca73` | +| control | 3000 | 0.45149569 | 0.81281968 | `8b0afaf1338754dccff24b1c7fd292b8b285fd665f7f003514b51403fadb99d9` | +| control | 6000 | 0.45166733 | 0.81280593 | `428e1327e4a4e51ffe0504f9ac9a2179f5c9a6e3c8883981687ab6e45fa45ea2` | +| covariance | 1000 | 0.45160459 | 0.81281096 | `d74d3af566e3c640d1209359df40ec59a9d7209fd9ed169e5186856e534807ef` | +| covariance | 3000 | 0.45153791 | 0.81281630 | `c7414bd840c80574dee726dacf7f41a28277a262da7e4663e779fe4cff50849d` | +| covariance | 6000 | 0.45171225 | 0.81280233 | `1544b181c5f74e3348bd67fd88f1f8154d5110aabf1c8ad62b721a32a60df690` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.03904003 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E036 — shared channel-covariance statistics +- Hypothesis / changed factor: Controlled continuation of the shared statistical branch, selected modecontrol update1000. Start from session-trained E026, add92928 zero-initialized covariance projection weights, and compare freezing them against learning them with the same seed20261105 and sample sequence. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seed20261031. Only the original four context frames supply channel covariances, means, variances and profiles. Same branch evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E036a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `254c11b3fd56624d3b695e25f7d980c48e04d399b76ef8111ce158e26fcaf483`. +- Validation: fixed 81-case subset; reward **0.812822**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039040 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Outcome: improvement; published atomically. + +## E036b — covariance hypothesis rejected by the matched control + +At every paired checkpoint, learned covariance features scored worse than the frozen-zero control: update1000 control0.451465647 vs covariance0.451604591;3000 control0.451495692 vs covariance0.451537907;6000 control0.451667333 vs covariance0.451712249 on radiative-layer valid VRMSE. The feature hypothesis is rejected. The control’s1000-update continuation has only a tiny improvement over the prior0.451484957 and was selected after exact neural/unknown/smoke/diffusion checks. Its covariance projection remains identically zero; no gain is attributed to the new features. Other dataset VRMSE and all checkpoint hashes are listed in E036a. No separate specialist models are used. + +## E037 — Resolved continuity feasibility diagnostic (preparation) + +Hypothesis: a fixed conservative density transport may explain or reduce residual density error. First test a second-order MUSCL/minmod, SSP-RK2 continuity update using temporally interpolated velocities on three audited TRAIN trajectories with low/medium/high cooling time. X periodic, Y open; metadata grid spacing/time, maximum summed Courant number0.4. Future TRAIN velocities are used only as an oracle feasibility diagnostic, never as an inference input or proposed package. No random seed or learned checkpoint; validation per dataset not evaluated. Results and source retained in continuity_v1. Proceed to forecast-velocity tests only if this diagnostic supports the operator. + +## E037a — continuity feasibility result: reject direct transport + +Using true future velocities from TRAIN trajectories only, the fixed conservative density update gives density-channel VRMSE: + +| Cooling time | Density VRMSE | +|---|---:| +| 0.03 | 0.16781393 | +| 0.32 | 0.18198222 | +| 3.16 | 0.41145309 | + +The high-cooling-time case remains inaccurate even with oracle velocities, so no forecast-velocity candidate or validation tuning is pursued. Additional TRAIN differential diagnostics found weak continuity alignment: correlations0.20–0.30 over early frames and0.37–0.54 over the full window. Swapping velocity components did not repair it. Fitted diagnostic scalars were not deployed. These results do not distinguish temporal sampling, numerical discretization and unresolved dynamics as causes. No package operator change; no learned checkpoint or random seed; other datasets were not evaluated. + +## E038 — complete rollout training preparation + +Hypothesis: matching the full scored horizon reduces the training/inference rollout gap. Prepare up to30 autoregressive steps, capped by corpus26/Arena10 available frames. Three microbatches of two accumulate gradients into effective batch6, freeing each graph before the next to fit one H100. Keep existing log1p variance-normalized loss, learning rate0.00003, noise0.01, group probabilities0.05/0.075/0.075/0.8 and25% initial-window sampling. Sampling seed20261106,4000 planned optimizer updates. Preserve whichever shared architecture and complete package is selected at launch. + +Queued transition policy: after all phase7/update3000 variants finish, move on if phase7 has supplied no selected main checkpoint; otherwise allow phase7 to complete6000 updates. Stop the previous trainer and workers and verify exit before starting phase8. No second GPU trainer is running. The new trainer supports both guided and unguided selected architectures. CPU guidance/runtime parity passed on an audited train window and its periodic translation, including covariance-capable statistical modules. No optimizer step or new model checkpoint has been produced yet. Sources: full_rollout_v1 and launch_phase8.py/watch_phase8.py. + +## Full validation of quadrature_wide_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779384**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `f2428e6dc68d46a87431091b096a2643398cf16b020e4f408af1debbea3c7b39`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.453701 | + +Expanded fixed state quadrature from12 to all24 active validation trajectories. Replayed the missing12, reused12 quick outputs only after exact context equality, and verified30/30 common physical steps for every active case. The552 corpus/smoke outputs are reused from global2000 full validation after exact physical asset/configuration equality checks and complete physical coverage. The nine neural outputs come from this exact package’s quick validation with matching inputs. All585 cases scored together; no held-out trajectories or inference caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +E038 preparation check detail: the initial combined training-batch/horizon statistical call differed from separate inference calls by up to1.53e-5 in float32 physical units. No model was trained or published with that new trainer. Evaluating each case with the same horizon-batch shape as inference removed the rounding difference; the repeated two-case TRAIN check now has exact maximum difference0.0 for means and profiles. Evidence: full_rollout_v1/guidance_checks.json. + +## Full validation of weight_average_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779507**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.452162 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of coarse_interaction_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779561**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.451485 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## Full validation of control_001000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.779562**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.090265 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_wide_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E029c — restore narrow quadrature after full validation +- Hypothesis / changed factor: The larger VALID cohort reverses E029: width3e-4 activeVRMSE0.090264899 versus width1e-4 0.087454288 across all24 trajectories, despite improving the first12. Restore width1e-4 while preserving the current averaged main weights and all selected auxiliary branches. No retraining or new random parameters; fixed latent seed20261026. Exact physical source/config parity and complete576-prefix coverage justify reuse of the previously validated narrow physical outputs; nine neural outputs are unchanged, with exact quick/full context equality. All runtime assets remain self-contained. Select using the full585 reward rather than the misleading quick81 ranking; full group metrics and provenance are retained. +- Seed: 20261025. +- Checkpoint SHA256: `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. +- Runtime code SHA256: `81a0ca9ff1e18280e3f4ad44b69ab7cf2d0588dabff10f45b147f4b4049c852b`. +- Validation: fixed 81-case subset; reward **0.811403**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Selection basis: full585 physical-operator comparison: candidate0.78070730, current0.77956248 + +Outcome: improvement; published atomically. + +## Full validation of quadrature_revert_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780707**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261025; checkpoint SHA256 `fd0a824960ba9ea91ca6de3b25c8030d2a0c665bfe08910dcd5dae25fcbb8ff7`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.451466 | + +Restore width1e-4 after all24 active VALID trajectories reversed the first12 ranking. Exact physical source/configuration equality with quadrature_candidate proves all576 complete physical outputs reusable. Copy24 narrower active outputs plus unchanged552 corpus/Arena outputs. Preserve all current main/auxiliary weights, code and neural output settings; copy their nine neural predictions only after quick/full context equality. No inference cache is packaged. Selection uses the larger585-cohort result, explicitly overriding the misleading81-case subset ranking. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +Selection-policy update after E029c: future physical-operator changes require full585 reports for both candidate and current model before record.py can publish them. Neural-only changes with identical physical assets/configuration may use the81-case comparison, since its nine radiative-layer cases are the entire valid pool and the physical outputs remain fixed. The active-state study E032 must compare against restored width1e-4 as well as its repeated wider control, and must validate all24 active cases before any publication. + +## E030 — learned coarse-statistic guidance, update 2000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `f2909b2ac6992e7c5328bfb609461f7733e7f9feba6bb41605dfed8d9fba67b5`. +- Validation: fixed 81-case subset; reward **0.810702**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.460297 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 2000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `6671cf5bf2c01b1a1d198b389bf3a10df15cd3a08319580a934432fce2caa979`. +- Validation: fixed 81-case subset; reward **0.810151**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.467359 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 2000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `c7289b09d8f0531dd577ba65498292c6ab5b865887d4830e1d044c021ccc41d7`. +- Runtime code SHA256: `8770d29e341f163d118460a6dfc5a5153ff09a1456fbb4b16399d8b1883e0443`. +- Validation: fixed 81-case subset; reward **0.810022**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469027 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E039 — equal averaging across the guidance extension +- Hypothesis / changed factor: Equal parameter average of the current selected shared main model and phase7/update1000, the better individual guided checkpoint so far. Extend the unguided parent with528 zero guide weights before averaging, preserving its function; accumulate in float64 and return original dtypes. This remains a single shared network evaluated on every case. No new optimization, pretrained assets or inference-time fitting. Source checkpoint hashes, equal coefficients and zero extension are in the manifest and mixed_average_v1/lineage.json. Inherited main initialization20261021, phase6 sampling20261025, guidance initialization20261101 and phase7 sampling20261102. Keep all selected physical operators and auxiliary branches. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact neural/novel-grid determinism before publication. +- Seed: 20261102. +- Checkpoint SHA256: `e4d1863c759fba48fd588f3e382c94733af1d529db1753d4f095d0134df7d643`. +- Runtime code SHA256: `d6ba1f84a29f2a2e52dbc9debda8f6a703fc3a12bb687fbeb6b69b8adb9b68c3`. +- Validation: fixed 81-case subset; reward **0.811432**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451106 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of mixed_average_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780736**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `e4d1863c759fba48fd588f3e382c94733af1d529db1753d4f095d0134df7d643`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.451106 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E040 — Dispersion-scaled hidden-state quadrature on fresh training trajectories + +Hypothesis: when the observed alignment|F2|/c approaches one, higher angular moments become tightly constrained, so fixed-amplitude hidden phase perturbations overstate uncertainty. Multiply an F6-and-higher perturbation field by1-|F2|/c, preserving F0,F2 and the flow-constrained F4. This is a fixed state computation; no parameters are fitted at inference. Training-input inspection found median alignment0.94/0.967/0.981/0.989 for representative zeta9/11/15/17 states, supporting a test of this constraint. + +Pilot uses the second audited TRAIN trajectory in the zeta9/alpha-3 and zeta17/alpha-5 pools, distinct from previous first-trajectory pilots. Rerun the current F4+width1e-4 baseline, F6+constant3e-4 control, and dispersion-scaled F6+base widths0.003/0.01. Three equal state nodes, latent seed20261026, unchanged kinetic parameters. Queue after E032 so only one heavy active-state study runs at once alongside the single neural GPU trainer. No new learned checkpoint or package change; validation per dataset pending. Any proposed physical change must subsequently pass all24 active validation cases and the full585 publication gate. Source and future results: uncertainty_cone_v1. + +## E032a — F6-only training pilot completed + +Four audited first TRAIN trajectories, fixed latent seed20261026 and unchanged physical source/configuration. Exact source hash/config/seed and trajectory membership match the earlier narrow baseline study. No new learned checkpoint. + +| Training parameters | First perturbed angular mode | Width | Active VRMSE | +|---|---:|---:|---:| +| [1, -1] | F4 |0.0001| 0.00162011 | +| [1, -1] | F4 | 0.0003 | 0.00162016 | +| [1, -1] | F6 | 0.0003 | 0.00162014 | +| [1, -1] | F6 | 0.001 | 0.00162015 | +| [9, -3] | F4 |0.0001| 0.00258179 | +| [9, -3] | F4 | 0.0003 | 0.00410363 | +| [9, -3] | F6 | 0.0003 | 0.00296343 | +| [9, -3] | F6 | 0.001 | 0.01561814 | +| [15, -4] | F4 |0.0001| 0.26059940 | +| [15, -4] | F4 | 0.0003 | 0.24752071 | +| [15, -4] | F6 | 0.0003 | 0.25210917 | +| [15, -4] | F6 | 0.001 | 0.23472451 | +| [17, -5] | F4 |0.0001| 0.45577160 | +| [17, -5] | F4 | 0.0003 | 0.42649895 | +| [17, -5] | F6 | 0.0003 | 0.43343046 | +| [17, -5] | F6 | 0.001 | 0.42344326 | + +All variants completed30/30 physical steps. Exploratory four-case means: {"narrow_baseline_mean": 0.18014322529779747, "candidate_means": {"(2, 0.0003)": 0.16993586576427333, "(3, 0.0003)": 0.17253080147202127, "(3, 0.001)": 0.16885151350288652}, "scope": "four first TRAIN trajectories only; no validation selection"} + +Outcome: both F6-only widths improve this small TRAIN mean relative to narrow F4, but the moderate zeta9 case worsens, especially at width0.001. These four trajectories overrepresent strong-activity regimes and cannot justify publication after E029’s cohort reversal. No validation dataset was evaluated for this operator. E040 tests fresh second training trajectories and a dispersion-scaled perturbation; any later physical proposal must pass all24 active VALID trajectories and full585 selection. + +## E030 — learned coarse-statistic guidance, update 3000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. +- Runtime code SHA256: `f2909b2ac6992e7c5328bfb609461f7733e7f9feba6bb41605dfed8d9fba67b5`. +- Validation: fixed 81-case subset; reward **0.811484**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450464 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E041 — fixed periodic translation average of the shared coarse branch + +Hypothesis: averaging the coarse MLP over four cyclic translations on each periodic coarse axis reduces phase sensitivity. Use the SAME selected E033/update6000 weights, with four periodic-x rotations for thermal cases and up to sixteen combinations when both axes are periodic; never translate across a nonperiodic boundary. Mean/profile and main neural branches are unchanged. No new training, parameters, or random seed; inherited coarse initialization20261103. Changed selection factors are only group averaging and fixed output blend. Pin E039 as parent, disable its previous coarse output correction, replay all nine neural valid trajectories, then evaluate the exact quadratic error of each linear correction using the official unbiased variance. This is validation-only selection. + +Main checkpoint SHA256 `e4d1863c759fba48fd588f3e382c94733af1d529db1753d4f095d0134df7d643`; coarse packaged checkpoint SHA256 `ad9a9996912c63eaaed7aa01423bc8598c9a8c966915b277ab64ce1774027dd3`. Exact native/coarse periodic equivariance checks passed on all nine thermal contexts within floating point tolerance. Original single-branch analytical VRMSE agrees with official package scoring to8e-11. + +| Coarse computation | Blend | Radiative-layer VRMSE | +|---|---:|---:| +| single | 0.0 | 0.45183754 | +| single | 0.25 | 0.45110587 | +| single | 0.5 | 0.45123235 | +| single | 1.0 | 0.45404672 | +| orbit | 0.0 | 0.45183754 | +| orbit | 0.25 | 0.45266324 | +| orbit | 0.5 | 0.45440332 | +| orbit | 1.0 | 0.46051732 | + +Other validation datasets are unchanged in every variant: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: reject periodic averaging. Every nonzero orbit blend is worse than both the current correction and the uncorrected parent. No candidate is published. This tests the inference projection of existing weights, not whether a jointly trained equivariant architecture can help. Full source, parent hashes, all per-case/channel results and equivariance checks are in coarse_orbit_v1. + +## E030 — learned coarse-statistic guidance, update 3000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. +- Runtime code SHA256: `6671cf5bf2c01b1a1d198b389bf3a10df15cd3a08319580a934432fce2caa979`. +- Validation: fixed 81-case subset; reward **0.811423**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451222 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 3000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. +- Runtime code SHA256: `8770d29e341f163d118460a6dfc5a5153ff09a1456fbb4b16399d8b1883e0443`. +- Validation: fixed 81-case subset; reward **0.810998**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.456544 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of eval003000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780788**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.450464 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E042 — orthogonal radiative-layer error diagnostic + +Analyze the selected guided update3000, with no model change, on nine audited TRAIN rollouts and all nine thermal VALID rollouts. Decompose each error exactly into global mean, two axis profiles, and two-dimensional interaction. Unbiased target variance normalization; components sum to total squared error within1e-10. Source/results and parent hash are in error_components_v1. Inherited main seed20261021, guide seed20261101, sampling20261102; checkpoint SHA256 `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. + +| Cohort | Channel | VRMSE | Mean error share | X profile | Y profile | Interaction | +|---|---|---:|---:|---:|---:|---:| +| train9 | density | 0.159511 | 0.001 | 0.037 | 0.082 | 0.880 | +| train9 | pressure | 0.849566 | 0.139 | 0.023 | 0.193 | 0.645 | +| train9 | velocity_0 | 0.184589 | 0.014 | 0.010 | 0.290 | 0.686 | +| train9 | velocity_1 | 0.485618 | 0.030 | 0.111 | 0.050 | 0.808 | +| valid9 | density | 0.180323 | 0.001 | 0.035 | 0.087 | 0.877 | +| valid9 | pressure | 0.899158 | 0.229 | 0.021 | 0.231 | 0.519 | +| valid9 | velocity_0 | 0.207556 | 0.015 | 0.015 | 0.277 | 0.694 | +| valid9 | velocity_1 | 0.514820 | 0.039 | 0.096 | 0.059 | 0.807 | + +Outcome: most squared error is in the two-dimensional interaction component. Pressure also has material mean/profile error, especially in validation. This supports the planned complete-rollout training and cautions that auxiliary mean-only refinements cannot solve the dominant error. No model or inference parameter changed. Other dataset validation metrics remain those of E030/update3000 shrink. + +## E043 — hidden angular-moment consistency diagnostic (preparation) + +Hypothesis: replacing observed F0/F2 and flow-constrained F4 can leave the retained higher angular moments inconsistent with any nonnegative orientation distribution. Inspect the positive-semidefinite Toeplitz moment condition on fixed spatial subsamples of two second TRAIN trajectories, zeta9/alpha-3 and zeta17/alpha-5. Test a fixed Dykstra state projection between nonnegative128-bin doubled-angle densities and the affine first-three-moment constraint,32 or128 passes. Preserve encoded F0/F2/F4 exactly and measure the necessary PSD condition, hidden-moment change and remaining negative mass before considering rollout experiments. No parameters are optimized or trained; it is a fixed state constraint layer. No new checkpoint or validation results. Source/model/config hashes will be pinned in positive_moment_v1/parent.json. Queue after E040 so only one heavy physical study runs alongside the sole neural trainer. An initial launch named inspect.py shadowed Python’s inspect module and failed before computation; renamed analyze_state.py and relaunched. + +## E040a — dispersion-scaled pilot results + +Fresh second TRAIN trajectories; fixed latent seed20261026 and identical kinetic parameters. All variants completed30/30 physical steps. Source/config hashes are in uncertainty_cone_v1/parent.json. No new learned checkpoint; no validation datasets evaluated. + +| Zeta,alpha | First mode | Base width | Dispersion scaling | Active VRMSE | +|---|---|---:|---|---:| +| [9, -3] | F4 | 0.0001 | False | 0.00571887 | +| [9, -3] | F6 | 0.0003 | False | 0.00578518 | +| [9, -3] | F6 | 0.003 | True | 0.00579862 | +| [9, -3] | F6 | 0.01 | True | 0.00923840 | +| [17, -5] | F4 | 0.0001 | False | 0.73280036 | +| [17, -5] | F6 | 0.0003 | False | 0.70843041 | +| [17, -5] | F6 | 0.003 | True | 0.73443890 | +| [17, -5] | F6 | 0.01 | True | 0.70559788 | + +Outcome: the smaller dispersion-scaled width worsens both cases; the larger improves the strongly active case but worsens the moderate case by62%. Its paired mean0.357418 is also slightly worse than constant F6+width3e-4 at0.357108, versus narrow baseline0.369260. Thus dispersion scaling does not solve the moderate-regime regression and is not advanced to validation. Constant F6+ remains an exploratory alternative, but the present evidence does not justify another physical publication. The next diagnostic examines actual moment consistency rather than another width schedule. + +## E043a — moment consistency inspection and rollout pilot + +Inspect4096 fixed spatial points from each of two audited second TRAIN trajectories. Eigenvalue diagnostics initially used complex64; one highly aligned batch did not converge, so diagnostics were rerun in complex128. Physical computation and projection remain float32. No training or new checkpoint. + +| Parameters | Projection passes | Minimum normalized Toeplitz eigenvalue | Fraction below -1e-5 | Hidden relative RMS change | +|---|---:|---:|---:|---:| +| [9, -3] | 0 | -0.018840 | 0.085205 | 0 | +| [9, -3] | 32 | 0.000002 | 0.000000 | 0.000340 | +| [9, -3] | 128 | 0.000002 | 0.000000 | 0.000340 | +| [17, -5] | 0 | -0.415949 | 0.953613 | 0 | +| [17, -5] | 32 | -0.018204 | 0.800537 | 0.018390 | +| [17, -5] | 128 | -0.012869 | 0.705078 | 0.018852 | + +Outcome: inconsistency is measurable, especially in the strongly active state.128 passes reduce the mean negative minimum eigenvalue from0.08443 to0.000800 while preserving F0/F2/F4 exactly. Residual violations remain: the layer improves consistency but does not certify exact positivity after a finite number of passes. Start a two-case full30 TRAIN rollout pilot with128 fixed passes on each of the three initial quadrature nodes. Keep current width1e-4, seed20261026 and all physical parameters. Baselines are reused from E040 only after exact source/config/seed/trajectory parity assertions. This is a state transformation with a fixed pass count, not fitting model parameters. Pilot validation per dataset pending; any physical publication must later pass all24 active VALID and full585 comparison. + +## E044 — shared statistical objective preparation + +Hypothesis: separate pointwise absolute losses for global means and profile bins do not optimize field VRMSE. Test the exact normalized RMSE of a piecewise-constant additive mean/axis-profile reconstruction. Orthogonal decomposition adds mean-error squared, centered axis-profile error squared, and unresolved target spatial variance before the square root. Future target statistics appear only in the training loss; all inference features remain four-frame context and metadata. A native-grid TRAIN reconstruction check matched the sufficient-statistic formula within2.8e-14. + +Matched control and new objective both resume the selected E036/control1000 auxiliary weights, preserving544587 parameters and the frozen zero covariance projection. All10115 audited TRAIN records and all19 datasets, seed20261108, same sample sequence, batch512,6000 updates, learning rate0.0001 cosine, two CPU threads each. No GPU trainer is added. Save1000/3000/6000; evaluate all six complete packages with fixed decoder blends and the same main-model parent. Inherited random initialization20261031; parent manifest and weights copied to statistic_objective_v1 with hashes in each training config. Validation per dataset pending; no package change. + +## E043b — consistency projection pilot and TRAIN expansion + +128 fixed passes,128 angular bins, current three-node width1e-4 and seed20261026. No learned parameter changes or new checkpoints. Both cases completed30/30 physical steps; early12-step error improved in the strongly active case, but long-horizon aggregate improvement is small. + +| Training parameters | Baseline VRMSE | Projected VRMSE | +|---|---:|---:| +| [9, -3] | 0.00571887 | 0.00592186 | +| [17, -5] | 0.73280036 | 0.72618914 | + +Outcome: mixed pilot, with a small mean improvement and a moderate-regime regression. Expand paired TRAIN tests to zeta1/alpha-1,5/-4,11/-4,15/-4 before any validation. Use the second training trajectory where available;11/-4 has exactly one training trajectory, so use only that permitted trajectory0. Both baseline and projected nodes are rerun on each context. Pool bounds are asserted before reading. Combined study will cover six parameter regimes; no physical publication based on the two-case pilot. Validation per dataset remains pending. + +## E044a — matched continuation and reconstructed-field objective comparison + +Continue the selected E036/control1000 shared auxiliary weights with identical samples, optimizer schedule and seed20261108 for6000 updates. Inherited random initialization seed20261031. No new parameters; covariance projection stays frozen at zero in both variants. All19 datasets trained together, two CPU threads per variant, batch512, learning rate0.0001 cosine. Control retains smooth absolute mean/profile loss0.25/0.75. Field objective uses VRMSE of a piecewise-constant additive reconstruction from predicted global means and sixteen-bin axis profiles, including unresolved target variance inside the square root. The analytical objective matched explicit TRAIN field reconstruction within2.8e-14. Fixed parent decoder blends retained; every candidate is replayed through the complete neural runtime, including guide values. No pretrained assets or inference-time fitting. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45090301 | 0.81144841 | `1910418f3394b026abf31ee21ff7bd05c78b791cc65cad8051b3174c775762ad` | +| control | 3000 | 0.45086799 | 0.81145122 | `2f82749ebc5ad7bf045621462de93b7ba8df476ebbbe38f48af81728ed7b5f74` | +| control | 6000 | 0.45057396 | 0.81147483 | `7860e8464d4ba66e31f6abcf74afec49bce8542ffed82fcb16ec5fc0cf64886d` | +| field | 1000 | 0.45046986 | 0.81148320 | `89cab23fc7018265a91990b16d5a6aeb0f785c9628bd7ffccafe17d6a500f9f2` | +| field | 3000 | 0.45074759 | 0.81146089 | `c452dd5fcfd80fec1df3e008d05495cd516c7f20ee2a7dbede0b8fc8f1fbe7b4` | +| field | 6000 | 0.45012636 | 0.81151080 | `e268751b6d0d1fa8cfafb6c8b3c61cbe7b926ea08c4468c69ed1ef767b08a9ce` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E044 — shared statistics objective, composite replay +- Hypothesis / changed factor: Controlled continuation of the shared statistical branch, selected modefield update6000. Start from session-trained E036/control1000 with no new parameters, and compare existing smooth absolute objective against reconstructed-field VRMSE using the same seed20261108 and sample sequence. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seed20261031. Only the original four context frames supply means, variances and profiles. The unused covariance projection remains frozen at zero. Same branch evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E044a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. +- Runtime code SHA256: `2f545d9f81f9b609775cc0073d181a99b24fcfdbd823aaf70c0997a445b0c366`. +- Validation: fixed 81-case subset; reward **0.811511**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450126 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E045 — guided late-checkpoint averaging preparation + +Predeclare one equal parameter average of phase7 updates4000/5000/6000, before those checkpoints are selected. This tests whether averaging late optimization noise helps the guided shared model, as earlier phase6 averaging did. Accumulate in float64, restore original dtypes, preserve current physical/auxiliary components and replay the complete model on all nine thermal valid trajectories. No new optimization, parameters or inference ensemble; same resulting weights for every case. Inherited main initialization20261021, guide initialization20261101 and phase7 sampling20261102. Source checkpoint hashes will be in guided_average_v1/lineage.json. + +The queued phase8 launcher will evaluate this single candidate after the full phase7 schedule and individual checkpoint selection, before freezing its starting model. Restarted only the waiting launcher; the sole GPU trainer and its workers were not interrupted. Full585 derivation and determinism checks will accompany a winning average. Validation per dataset pending. + +## Full validation of field_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780815**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `9914b50a288baa532bf558ada319048667c41f2a3163f8c7a9177432c2373431`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.450126 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E030 — learned coarse-statistic guidance, update 4000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `2f545d9f81f9b609775cc0073d181a99b24fcfdbd823aaf70c0997a445b0c366`. +- Validation: fixed 81-case subset; reward **0.811518**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450032 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E030 — learned coarse-statistic guidance, update 4000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `121b5a37b6db742769502c11e350f3490338916078a76e88a98511a363f1b6a6`. +- Validation: fixed 81-case subset; reward **0.811379**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451770 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 4000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `75cf5b74461f47b5b43d9822e62c9cacfb24c8e0bd5dec4431759c9e2257b146`. +- Validation: fixed 81-case subset; reward **0.810911**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.457652 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E046 — joint auxiliary field objective preparation + +Hypothesis: couple the universal mean/profile and coarse interaction branches through the VRMSE of their combined reconstructed field, instead of optimizing each branch’s component loss independently. The target’s unresolved variance belongs inside the same square root. With16-bin marginal profiles and a4x16 zero-marginal interaction grid, these components are orthogonal; sufficient-statistic loss agrees with explicit native TRAIN reconstruction within3.6e-14. Future target statistics are used only in training loss. + +Both variants continue the selected E044 mean/profile and E033 coarse weights with no new parameters, identical seed20261109/sample sequence,6000 updates, batch512, learning rate0.0001 cosine and two CPU threads each. Train all10115 authorized records across all19 datasets. The independent control sums the two selected component objectives; the joint variant uses reconstructed-field VRMSE. Freeze the unused covariance projection at zero. Inherited mean/coarse initialization seeds20261031/20261103; parent weights and manifests are pinned in joint_statistics_v1. Evaluate all1000/3000/6000 checkpoints through the complete model with fixed blends. No extra GPU trainer, pretrained weights or inference-time fitting. Validation per dataset pending. + +## Full validation of eval004000_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780822**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.450032 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E047 — capacity expansion preparation + +Hypothesis for a later training stage: the shared main model has insufficient capacity for the dominant two-dimensional thermal structure. Prepare a width48-to96 conversion while preserving its initial function. Duplicate every learned feature channel, divide outgoing pair weights by two, and add tiny zero-sum outgoing pair perturbations so identical activations retain the same sum but training gradients can break symmetry. Keep the Fourier latent width and metadata embedding fixed. No new pretrained assets; all original parameters were trained from scratch in this session. Perturbation seed20261110, relative RMS scale0.01. + +Main parameter count grows from3066627 to9526707; auxiliary branches remain unchanged. This is a preparation check only: no wider optimizer run or validation targets, and no package change. Checkpoint parent SHA256 `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. On four audited TRAIN contexts resized to32x64, comparing the exact same guided forward computation: + +| Dataset | Normalized RMS difference | Normalized maximum difference | +|---|---:|---:| +| ape2d/burgers | 5.731e-08 | 3.858e-07 | +| pdearena/navier_stokes_2d | 1.460e-07 | 7.728e-07 | +| well/active_matter | 5.009e-07 | 1.366e-05 | +| well/turbulent_radiative_layer_2D | 5.590e-07 | 3.958e-06 | + +Outcome: numerical parity and symmetry-breaking checks passed. Conversion source and audit are in wider_v1. Consider this after the existing complete-rollout experiment has been selected; no second GPU trainer is launched. A prospective full-horizon wider run would use microbatch1 and accumulation6 to fit the single H100. Validation per dataset pending. + +## E046a — independent and joint auxiliary field-objective comparison + +Continue both selected universal auxiliary branches, E044 mean/profile and E033 coarse interaction, with identical samples, optimizer schedule and seed20261109 for6000 updates. Inherited initialization seeds20261031/20261103. No new parameters; covariance projection stays frozen at zero. All19 datasets, two CPU threads per variant, batch512, learning rate0.0001 cosine. Independent control sums the selected mean/profile field objective and coarse interaction objective. Joint loss is the VRMSE of the combined mean,16-bin profiles and4x16 interaction reconstruction, including unresolved target variance inside the square root. The analytical objective matches explicit TRAIN reconstruction within3.6e-14. Fixed decoder blends retained; every candidate is replayed through the complete neural runtime, including guidance. No pretrained assets or inference-time fitting. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| independent | 1000 | 0.45021534 | 0.81150365 | `dc0556f9fb9aa536df9483542500ad6b24b5d7c47f00fa7172f289d9fbc97f02` | +| independent | 3000 | 0.45030356 | 0.81149656 | `c58996befc6158c232310086006748f118c0ff8c4d0a45336d3406fe643de2f4` | +| independent | 6000 | 0.45048595 | 0.81148190 | `18abbb2e719fae79306aa632d8959db9c48a723c8d9a734c63e9ed4725d6d081` | +| joint | 1000 | 0.45019307 | 0.81150544 | `07d8f4abfe8c48bd06e8275a1cdcb2c37482eb799c8640ad22d19aef4b8afe27` | +| joint | 3000 | 0.45004713 | 0.81151717 | `d0c87d21a0ae94fb1f09006de9151fb1a032a754825203c808619c1ed1b3a91e` | +| joint | 6000 | 0.45012994 | 0.81151051 | `c4874c7c4cb379999df2147349c74e9cc0226964ebfd34fa465ac592a863716b` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E046 — joint auxiliary objective, composite replay +- Hypothesis / changed factor: Controlled continuation of both shared auxiliary branches, selected modejoint update3000. Start from session-trained E044 and E033 weights with no new parameters; compare independent component objectives against joint reconstructed-field VRMSE using the same seed20261109 and sample sequence. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seeds20261031/20261103. Only the original four context frames supply means, variances, profiles and coarse fields. The unused covariance projection stays frozen at zero. Both branches evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E046a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `073eec78a834161996e478502f0debca7ebd4e1ea4b27dabfed049fa079d01c2`. +- Validation: fixed 81-case subset; reward **0.811517**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450047 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E043c — six-regime TRAIN result and full validation launch + +Same fixed128-pass/128-bin projection, three uncertainty nodes at width1e-4, latent seed20261026 and unchanged physical parameters. All six regimes and both variants completed30/30 physical steps. No new learned checkpoint. + +| Training parameters | Trajectory | Baseline VRMSE | Projected VRMSE | +|---|---:|---:|---:| +| [1, -1] | 1 | 0.00068666 | 0.00068664 | +| [5, -4] | 1 | 0.09479659 | 0.09478904 | +| [9, -3] | 1 | 0.00571887 | 0.00592186 | +| [11, -4] | 0 | 0.08358849 | 0.08178639 | +| [15, -4] | 1 | 0.25770625 | 0.24436744 | +| [17, -5] | 1 | 0.73280036 | 0.72618914 | + +Six-case mean 0.19588287 to 0.19229008. Outcome: five improvements, one moderate-case regression; advance this single fixed operator to all24 active VALID trajectories. This cohort includes the sole permitted11/-4 training trajectory, never another trajectory from that pool. + +The full candidate is assembled under universal_v1/moment_consistency_candidate and is not yet published. After complete physical-prefix audits, retain the latest selected shared neural/auxiliary weights, reuse exactly matching non-active outputs, and score all585 VALID trajectories. Quick81 is recorded only for continuity; physical selection requires both complete585 reports. Exact active/novel-grid repeat checks precede selection. The recorder now logs a candidate as ineligible rather than aborting a watcher if a concurrent physical change makes its full report missing; this does not relax the full-cohort gate. + +## E046b — joint objective result + +The joint objective beats its independent continuation control at all three paired checkpoints, but its best complete-package thermal VRMSE0.45004713 is slightly worse than the selected main-update4000 package at0.45003150. Both branches retained their parameter counts and shared all19 training sources; seed20261109, inherited initializations20261031/20261103. All candidate checkpoint hashes and per-dataset metrics are in E046a. Outcome: no deployment improvement, so retain the selected E044 mean/profile and E033 coarse weights. + +## E048a — all72 thermal TRAIN blend comparison + +Frozen weights and guide forecasts; no new optimizer or randomness. Replay four-frame starts from all72 authorized thermal TRAIN trajectories. Compare nine predeclared mean/profile output blends, preserving coarse blend0.25, momentum pressure0.05 and profile shrinkage0.75. Recompute the complete postprocessing composition for every pair. Float32 GPU training metric uses the unbiased spatial variance and1e-7 epsilon; official float64 scoring is reserved for VALID selection. Source/main/auxiliary hashes and all train memberships are in statistic_blend_v1. + +| Mean blend | Profile blend | Training VRMSE | +|---|---:|---:| +| 0.25 | 0.0 | 0.41026571 | +| 0.25 | 0.25 | 0.40976740 | +| 0.25 | 0.5 | 0.41185432 | +| 0.5 | 0.0 | 0.40688651 | +| 0.5 | 0.25 | 0.40625567 | +| 0.5 | 0.5 | 0.40822653 | +| 0.75 | 0.0 | 0.40535528 | +| 0.75 | 0.25 | 0.40459570 | +| 0.75 | 0.5 | 0.40644322 | + +Outcome: freeze the best TRAIN pair before validation. No dataset VALID metrics from this training comparison; complete-package VALID replay follows. + +## E048 — statistical output-blend calibration preparation + +Hypothesis: the selected E044 statistical objective changes the best fixed output blend, while coefficients are inherited from earlier auxiliary weights. Freeze all weights and replay all72 authorized native thermal TRAIN trajectories from their first four frames. Compare the nine fixed pairs mean blend0.25/0.5/0.75 and profile blend0/0.25/0.5, retaining momentum pressure0.05, coarse interaction0.25 and profile shrinkage0.75. Learned guidance still consumes the same complete statistical forecasts; only output mixing is changed. No new optimization or random parameters. Every candidate uses the same coefficients on every case, while complete physical prefixes stay unchanged. + +Pin parent hashes and train memberships in statistic_blend_v1; select the pair on TRAIN, then replay a complete package on all nine thermal VALID trajectories and use official scoring. All trial TRAIN metrics will be recorded, including regressions. Runtime branches and inherited initialization/sampling seeds remain the selected ones. Validation per dataset pending. + +## E048 — TRAIN-calibrated statistical output blends +- Hypothesis / changed factor: Freeze a mean blend of0.75 and profile blend of0.25 selected from nine predeclared pairs on all72 thermal TRAIN trajectories. No new optimization, parameters or randomness. Apply the same coefficients to every case; complete physical prefixes are unchanged. Preserve the selected shared neural and auxiliary weights, other decoder coefficients and all physical operators. Training postprocessing includes momentum and coarse interactions; guide forecasts remain unchanged. Replay all nine thermal VALID trajectories, reuse72 complete physical outputs, and check exact neural/novel-grid determinism before a winning publication. Inherited seeds and checkpoint hashes remain in the manifest; all TRAIN blends including negative results are in E048a. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `42557d933161b5f8313e9a85910c461b98d0f7671bba8f4d14f50c30ff84d08a`. +- Validation: fixed 81-case subset; reward **0.811229**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453646 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 5000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `be9baea226e3e322affe6cc1ab9240dc6ac8cb0624581989261ca4c01bdb31df`. +- Runtime code SHA256: `2f545d9f81f9b609775cc0073d181a99b24fcfdbd823aaf70c0997a445b0c366`. +- Validation: fixed 81-case subset; reward **0.811460**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450762 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 5000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `be9baea226e3e322affe6cc1ab9240dc6ac8cb0624581989261ca4c01bdb31df`. +- Runtime code SHA256: `121b5a37b6db742769502c11e350f3490338916078a76e88a98511a363f1b6a6`. +- Validation: fixed 81-case subset; reward **0.811340**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.452252 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 5000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `be9baea226e3e322affe6cc1ab9240dc6ac8cb0624581989261ca4c01bdb31df`. +- Runtime code SHA256: `75cf5b74461f47b5b43d9822e62c9cacfb24c8e0bd5dec4431759c9e2257b146`. +- Validation: fixed 81-case subset; reward **0.810686**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.460509 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E048b — stronger TRAIN-selected mean correction did not transfer + +Across all72 TRAIN starts, raising mean blend0.5 to0.75 at profile blend0.25 improves VRMSE0.40625567 to0.40459570. Official VALID replay reverses the ordering:0.45003150 to0.45364634. Outcome: reject the stronger correction and keep mean blend0.5. Do not extend the blend range based on the TRAIN boundary optimum after this generalization failure. Candidate runtime/main/auxiliary hashes and all VALID dataset scores appear in E048. + +## E049 — periodic profile augmentation preparation + +Hypothesis: the statistical branch can overfit the absolute phase of periodic input profiles, even though global means should be translation invariant. Continue the selected E044 field/update6000 weights with independent cyclic16-bin shifts on periodic axes, applying each shift consistently to all context and target profiles. No shifts across nonperiodic boundaries; scalar means, variances and metadata remain unchanged. Compare a matched unaugmented continuation with the same sample sequence and training objective. + +All10115 TRAIN records/all19 datasets,6000 updates, batch512, learning rate0.0001 cosine, sample seed20261112 and separate augmentation seed20261113. Inherited random initialization20261031, no new parameters, covariance projection frozenzero. Two CPU threads per variant, no new GPU trainer. Evaluate1000/3000/6000 checkpoints through the complete shared model at the current fixed blends. Parent weights and manifests are pinned in statistic_equivariance_v1. Validation per dataset pending. This tests the statistical branch; E035’s negative coarse-interaction augmentation result remains recorded separately. + +## E049a — matched periodic profile augmentation comparison + +Continue the selected E044 field-objective shared auxiliary weights with identical samples and seed20261112 for6000 updates. Both variants use reconstructed-field VRMSE, all19 datasets, batch512, learning rate0.0001 cosine and two CPU threads. Periodic variant draws independent cyclic16-bin translations for every sample and permitted axis, applying the same shift to all four context profiles and target profile. Boundary flags prevent nonperiodic translations. Means and variances are invariant. A separate augmentation RNG seed20261113 preserves the control sample sequence. No new parameters; covariance projection remains frozen zero. Inherited initialization20261031. Every checkpoint is replayed through the full shared runtime with fixed output blends. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.44992150 | 0.81152728 | `188e99a71c87407c6f35cb609325ca9807fec7c0d2243afe40b78ca0ccbe8263` | +| control | 3000 | 0.45038680 | 0.81148987 | `2ff5eea5a56d37c54eea94c39a0f860c1999647170fb1209e7daeb3d63bf698a` | +| control | 6000 | 0.44983144 | 0.81153452 | `eb835253e42b001bcf4dfd54029a78393ebf06e129d800868e452ff41447932f` | +| periodic | 1000 | 0.44948206 | 0.81156264 | `7c0852420e06132d6ab7fd50062a31e74d7a64becb1b1509d69967c651a415d4` | +| periodic | 3000 | 0.45248557 | 0.81132161 | `fab8ed8a88e876695f79950e5f33c68c628694e0f9e83f5bccb0396cf4e5515b` | +| periodic | 6000 | 0.45235439 | 0.81133211 | `0e8da17aa1b513005c5e5837d77ad1e71fd3b6bb1a35315ef4a760d9880783f1` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E049 — shared profile translation augmentation +- Hypothesis / changed factor: Controlled continuation of the shared statistical branch, selected modeperiodic update1000. Start from session-trained E044 field/update6000 with no new parameters; compare periodic profile translations against a matched continuation with the same seed20261112 and sample sequence. Augmentation seed20261113 is separate. All19 datasets trained together for6000updates at learning rate0.0001; inherited initialization seed20261031. Only the original four context frames supply means, variances and profiles. The unused covariance projection remains frozen at zero. Same branch evaluated on every case. Replay all nine neural validation cases for each of six checkpoint/variant pairs, with current fixed decoder blends; reuse72 complete physical outputs. All results, seeds and hashes are documented in E049a and the manifest. Exact neural/novel-grid and smoke/diffusion checks precede a winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. +- Runtime code SHA256: `bde67d792a2cc2cd64656ca21dfe1fbfdd653c2560d5376370cc304b090daf51`. +- Validation: fixed 81-case subset; reward **0.811563**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449482 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of periodic_001000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780867**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `03bf730a0e9a89e88f6a8256b9b0b63ce8e4d42aa0e4acf69969ab2e1f8e203d`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.449482 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E050 — synthetic hidden-moment supervision feasibility pilot + +Hypothesis: known latent angular states from independently generated trajectories could supervise a future hidden-state head on the same shared neural body. Begin with a CPU-only feasibility pilot, not model training:32x32 spatial grid,32 even angular modes,128 initial doubled-angle bins, four burn-in steps, four observed context frames and three future frames. Construct fresh smooth positive mixtures of two von Mises angular distributions with concentration and orientation fields from independent filtered random noise. Seeds20261115/20261116, zeta9/alpha-3 and17/-5, L10,dt0.25. Use the independently implemented Fourier Smoluchowski/Stokes solver already developed in this session; do not use benchmark generator code or initial-condition seeds. + +Store complete known angular moments and observed fields, then measure the existing fixed four-frame encoder’s latent error and first forecast error. Source/config/file hashes will be in synthetic_latent_v1. Generated source declaration: synthetic:own Fourier Smoluchowski-Stokes solver; fresh smooth positive doubled-angle mixtures, seeds20261115-20261116,32x32 pilot. No generated trajectory is used by current weights, so the published sources_used list remains unchanged. Any future training use must declare the synthetic source, use the same shared all-dataset model and preserve the no-specialist rule. No new checkpoint or VALID results; outcome pending. + +## E050a — synthetic pilot numerical failure and revised initialization + +The first32x32, broad-orientation pilot became non-finite before producing a complete stored trajectory. No synthetic data entered training or the published package. Preserve its exact script, manifest and log as synthetic_latent_v1/pilot_failed_v0.*. Hypothesis: sharp spatial variations of high angular moments were underresolved on the small grid; this is not yet a confirmed diagnosis. Retry at64x64 with lower spatial bandwidth: phase amplitude0.25 instead of1.2, noise cutoff1.5 instead of2.5, concentration amplitude0.05 instead of0.15 and smoother mixture variation. Preserve the independent seeds, angular discretization and physical parameters. This tests generator feasibility, not surrogate validation. No learned checkpoint or VALID metrics. The revised generated-source description is stored in manifest.json and must be declared if any generated data is later used in training. + +## E050b — isolate synthetic numerical instability + +The64x64 smoother pilot also became non-finite before a complete sample. Archive its exact script/manifest/log as pilot_failed_v1.*. A spatially homogeneous positive-mixture control on16x16 remains finite through eight stored steps, with concentration1 and alignment converging to0.99438. Thus the angular-only CPU evolution works in this control; the spatially varying failure remains unresolved. Real TRAIN metadata confirms dt0.25, so the pilot did not use a mistaken stored interval. Retry the smoother initialization at128x128 with per-frame finite/concentration/alignment diagnostics. No synthetic samples have entered training or the package. No new checkpoint or VALID result. + +## E050c — synthetic pilot rejected for moment-bound violations + +The128x128 smoother pilot remains finite longer, but max|F2|/c grows from0.9954 after burn-in to1.0083,1.0275 and above1.07 in subsequent frames. This violates the necessary positive-distribution moment bound, so finite values alone are not sufficient for a training teacher. Stop the pilot and mark its manifest REJECTED; any generated NPZ files are quarantined in rejected_numerics. No synthetic data entered any optimizer, checkpoint or published sources_used list. + +The homogeneous angular control remains stable, suggesting that the unresolved issue involves spatially varying evolution. Resolution alone has not yet produced an acceptable teacher; do not treat this as a successful synthetic source. Future work would need a converged or positivity-preserving generator and quality checks before dataset expansion. No new learned checkpoint or VALID metrics; published model unchanged. + +## E030 — learned coarse-statistic guidance, update 6000, shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `aecbe81ff548ff47d33b279f85e7f07f1b4f661ba72ecdff211f278994a63956`. +- Runtime code SHA256: `bde67d792a2cc2cd64656ca21dfe1fbfdd653c2560d5376370cc304b090daf51`. +- Validation: fixed 81-case subset; reward **0.811543**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449726 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 6000, no_shrink +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.0; statistical output corrections enabled=True, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `aecbe81ff548ff47d33b279f85e7f07f1b4f661ba72ecdff211f278994a63956`. +- Runtime code SHA256: `b02fbeefa500eb903647f72467e049c24b9d82207ed900d3de67d77b1a5e81de`. +- Validation: fixed 81-case subset; reward **0.811400**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451508 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E030 — learned coarse-statistic guidance, update 6000, no_statistics +- Hypothesis / changed factor: Train one shared guided operator from the validation-selected phase6 parent. Add528 zero-initialized stem weights that consume normalized future coarse mean/profile fields from the frozen shared auxiliary model. All forecasts of these guide fields use only the original four context frames and metadata. Sixteen-step fully backpropagated loss, capped at ten Arena steps; batch6, learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8. Sampling seed20261102; new guidance initialization seed20261101, inherited main initialization20261021 and auxiliary seed20261031. Parent hashes in phase7/lineage.json. Fixed profile shrinkage 0.75; statistical output corrections enabled=False, while learned guidance remains active. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261102. +- Checkpoint SHA256: `aecbe81ff548ff47d33b279f85e7f07f1b4f661ba72ecdff211f278994a63956`. +- Runtime code SHA256: `54e4d7c0dbee2ada00505f657b8d9bc6993085410464ff02a4f98f6c8aac8186`. +- Validation: fixed 81-case subset; reward **0.810944**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.457234 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E045 — late guided checkpoint average +- Hypothesis / changed factor: Equal parameter average of phase7 updates4000/5000/6000, a single predeclared candidate to reduce late-training optimization variation. No new optimization or data fitting; all three checkpoints were trained together across all19 datasets, initialization seed20261021, guidance extension seed20261101 and sampling seed20261102. Accumulate in float64 and return original parameter dtype. Same shared averaged weights apply to every case. Preserve current fixed physical operators and shared auxiliary decoders. Source training checkpoint hashes and equal coefficients are in model_manifest.json and guided_average_v1/lineage.json. Replay nine neural valid cases, reuse72 complete physical outputs, and check exact determinism and novel grids if improved. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `bde67d792a2cc2cd64656ca21dfe1fbfdd653c2560d5376370cc304b090daf51`. +- Validation: fixed 81-case subset; reward **0.811593**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449111 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of late_average + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780897**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.449111 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/quadrature_revert_candidate. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; their576 outputs are independent of neural weights, including any mean-forecast branch. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E038 complete-rollout training launch + +Completed the full learned-guidance schedule after it produced a selected checkpoint. Retired the previous trainer/workers/watcher and verified process exit before launching the next single-GPU trainer. Preserve the latest selected architecture, main parameters, fixed physical operators and auxiliary branches. Train complete available rollout windows:26 corpus,30 Well,10 Arena. Three microbatches of two accumulate to effective batch6; each microbatch graph is freed before the next, fitting one H100. Sampling seed20261106,4000 planned updates, learning rate0.00003 cosine, noise0.01 and group probabilities0.05/0.075/0.075/0.8. Dataset sampling and25% initial-window probability are unchanged. No new architecture parameters; exact inherited seeds and checkpoint hashes are in phase8/lineage.json. Validation pending. + +## E051 — positive angular-density output projection preparation + +Hypothesis motivated by the rejected synthetic pilot: finite angular moments can drift outside the positive-distribution set during numerical evolution, not only during the context encoder. Test a fixed output-step projection on real TRAIN trajectories: reconstruct a128-bin angular density, clip negative values, renormalize to the current predicted concentration, and retain the first33 Fourier moments before decoding. Preserve the entire observed-context encoder and existing initial quadrature. This may add numerical diffusion; positivity alone is not evidence of forecast accuracy. + +Queue after E043 full24/585 validation so only one heavy physical study runs at once. Pin the then-selected physical operator/configuration, and rerun paired baseline/projected30-step forecasts on the second permitted TRAIN trajectories for zeta9/alpha-3 and17/-5. Inherit the selected latent seed and all physical parameters. No learned parameters, checkpoint, synthetic training or inference-time fitting. Only advance if real TRAIN target errors support expansion. Source/provenance/results in positive_step_v1; validation per dataset pending. + +## E052 — spatial error scale diagnosis + +Use the frozen E042 parent (phase7/update3000) and its existing matched nine TRAIN and nine VALID rollouts. Decompose error into the orthogonal blockwise two-dimensional interaction projection at4x16,8x32,16x64,32x128 and64x192. Float64 orthogonality error below1e-10. These target-derived oracle corrections are diagnostic bounds only and never enter a package, optimizer, or model selection as candidate predictions. No new parameters, seed or checkpoint. Full results and parent hash: error_scales_v1/results.json. + +For4x16, TRAIN squared-error fractions by density/pressure/vx/vy are11.5%/10.9%/16.9%/22.8%; for8x32,31.4%/28.3%/35.5%/45.7%. VALID shows the same resolution trend. TRAIN oracle residual VRMSE is0.38930 at4x16 and0.34205 at8x32 versus actual0.41982. VALID diagnostic values0.42018 and0.37229 versus actual0.45046. Outcome: coarse spatial resolution is a measurable capacity constraint; test a modest8x32 branch before larger grids. + +## E053 — matched coarse-resolution continuation preparation + +Compare the existing4x16 interaction forecaster against8x32, with the same width192, all19 TRAIN datasets, initial-context probability0.5, group probabilities0.1/0.1/0.1/0.7, batch512,6000 updates, learning rate0.0001 cosine and seed20261117. Both inherit the session-trained E033/update6000 weights (initialization seed20261103). The fine model lifts input weights by averaging each2x2 group and repeats output weights; its pooled prediction initially matches the parent, while finer input detail persists. Parameters787808 versus2817440. Keep the original interaction loss and fixed0.25 decoder blend to isolate resolution/capacity. Compare updates1000/3000/6000 through the same complete runtime on all nine thermal VALID trajectories. The same branch is applied to every case. + +All10115 TRAIN records were re-audited against split.json before an8x32 float64-accumulated cache was built. No held-out records or synthetic data. The initial strict float32 conversion check failed at normalized max2.88e-5 on active matter before training. An independent float64 check verifies the conversion algebra to5.03e-14; float32 accumulation differences are recorded (normalized RMS at most6.26e-6). The check now requires both float64 algebra parity and bounded float32 roundoff, and passes on four TRAIN groups. No weights were trained before those checks. Checkpoint hashes and all validation outcomes will be appended after training. + +## Full validation of moment_consistency_candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780325**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.088847 | +| well/turbulent_radiative_layer_2D | 0.449111 | + +Replayed all24 active VALID trajectories with the fixed128-pass moment consistency projection; every case uses30/30 physical steps. Reused552 complete corpus/smoke outputs and nine neural outputs from the selected full585 parent after exact unaffected asset/configuration parity. Rebased concurrent shared-neural progress without changing any of the24 complete kinetic outputs. Quick81 outputs are a strict subset with exact input and metadata equality. No held-out cases or inference-time data caches. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E043 — fixed hidden-moment consistency, full-cohort selection +- Hypothesis / changed factor: Add128 fixed Dykstra state-projection passes at128 doubled-angle bins to each initial active quadrature node. Preserve encoded F0,F2,F4 exactly; retain current width1e-4, seed20261026 and all physical parameters. No new learned parameters or inference-time fitting. Six audited TRAIN regimes improved mean VRMSE0.19588287 to0.19229008, with one moderate regression. Replay all24 active VALID trajectories before selection; all30/30 physical steps. Retain current shared-neural progress and score all585 valid trajectories. Compare full585 rewards against the current package, not the smaller active subset. Per-dataset results, complete physical coverage, source hashes and repeat checks are retained. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `287aee6dbc146c522445ebdd14036cb9d00433aa8d7ebe61ef2e2665a73d502a`. +- Validation: fixed 81-case subset; reward **0.812815**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039252 | +| well/turbulent_radiative_layer_2D | 0.449111 | + +Selection basis: full585 physical-operator comparison: candidate0.78032472, current0.78089659 + +Outcome: did not improve validation reward; retained current package. + +## E054 — initial-context sampling preparation + +Hypothesis: statistical training windows should match the scored first-four-frame task more closely. Compare the current50percent initial-context sampling against100percent initial contexts, resuming the selected E049/periodic1000 shared head. Both arms use reconstructed-field VRMSE, no further translation augmentation, all19 TRAIN datasets, seed20261118, batch512,6000updates and learning rate0.0001 cosine. Preserve sample-index/random-number draws; only the rule choosing temporal start changes. No new parameters, frozen-zero covariance projection. CPU training only; full-runtime validation is queued after E053 completes so all six candidates share the then-selected parent. Validate1000/3000/6000 and retain all regressions, hashes and per-dataset results. No synthetic data or held-out targets. + +## E043d — full-cohort projection hypothesis rejected + +The fixed128-pass initial hidden-moment projection improves quick12 active VRMSE0.04060092 to0.03925196 but worsens all24 active VALID trajectories0.08745429 to0.08884714. All30-step physical prefixes complete and exact active/unknown repeat checks pass. Full585 selection correctly rejects the change despite the quick reward improvement. Keep the unprojected narrow1e-4 operator and current averaged neural weights. Full dataset scores and hashes are recorded in E043 and the full-validation entry. E051 positive future-output-step study now runs against this retained parent, TRAIN only. + +## E047b — capacity experiment queued after full-rollout training + +Prepare width48 to96 main-network continuation using the checked E047 conversion. Phase9 waits for phase8/update4000 validation and normal exit of its trainer, workers, launcher and watcher. Then repeat conversion checks on the actually selected parent, freeze its selected auxiliary guides, and run the sole next GPU trainer for2000updates: microbatch1 times6, full30-step rollouts, learning rate0.00003, sampling seed20261119, new zero-sum pair perturbation seed20261110. No new optimizer has started yet. A changed selected auxiliary guide, if any, will be documented as an additional changed factor. All physics/auxiliary inference assets remain inherited; validate1000 and2000 with the existing decoder variants. Files: launch_phase9.py,watch_phase9.py,wider_v1. + +## E055 — synthetic numerical convergence diagnosis + +The rejected E050 spatial pilot does not establish whether spatial bandwidth or angular truncation caused its moment-bound violations. Test a freshly specified, much smoother positive angular-mixture field at spatial32 and64, angular modes32 and64, with the same continuous initial field and fresh independent seed20261120. Phase amplitude0.05, mixture amplitude0.02, concentration amplitude0.01, parameters9/-3 and17/-5, four stored advances of0.25. Monitor finite values, concentration positivity, all retained moment bounds, first9-moment Toeplitz eigenvalues and low-moment convergence against spatial64/angular64. This CPU-only diagnostic writes no training examples, has no learned parameters and uses no real trajectory targets. No outputs may enter any optimizer. Results will determine whether further synthetic teacher work is justified; published sources_used remains unchanged. + +## E053a — matched interaction-grid resolution comparison + +Continue the session-trained E033 interaction branch, comparing its4x16 grid with8x32 lifted initialization. Both variants use identical TRAIN samples and seed20261117,6000 updates, batch512,learning rate0.0001 cosine,all19 datasets and two CPU threads. Keep the original interaction objective and fixed0.25 decoder blend. Fine-grid branch has2817440 parameters versus787808 in control; pooled initialization matches parent in float64 to5.03e-14. No synthetic or held-out data. Every checkpoint is replayed through the complete shared runtime. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.44916459 | 0.81158821 | `7d5c4aab7e7ca6a99be462e961727dc1fbf664e8ef7bcec30368ccfc7573ab9f` | +| control | 3000 | 0.44914566 | 0.81158974 | `0699bced7df85b544d2729aefc358db29ffdbd75213109ab7f67f9fea2cdaafc` | +| control | 6000 | 0.44913370 | 0.81159070 | `a3ee94d8a7784d70e1e5fae13ae8bf9a8ad49558a3c598e012d64c3adebb12e5` | +| fine | 1000 | 0.45124607 | 0.81142089 | `817aa7c53b04e7c5aeedf1540279712131c8ddc84bd2560a9f859f6f0219dd1c` | +| fine | 3000 | 0.45125512 | 0.81142016 | `139bca5da618d59c5cca657194bc4554bcd08c4b7487ecc123cb5f534ab1967d` | +| fine | 6000 | 0.45100125 | 0.81144053 | `89b008c8b5749b16db607dfe3d9c172fc76199bbca21301bc1909013e2c2388a` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E053 — shared interaction-grid resolution +- Hypothesis / changed factor: Matched continuation of the shared interaction forecaster, selected modecontrol update6000. Compare4x16 control against8x32 grid with lifted initialization from session-trained E033/update6000. Both train across all19 datasets with identical sample sequence, seed20261117,6000updates,batch512,learning rate0.0001 cosine. Keep original interaction loss and fixed0.25 correction blend. No new synthetic data or pretrained assets. Same branch runs on every case; correction preserves global means and axis profiles and leaves complete physical prefixes authoritative. Six complete-runtime checkpoint replays on all9 neural VALID trajectories; reuse72 physically complete outputs after unchanged physical assets. All per-dataset metrics and checkpoint hashes are recorded in E053a. Exact neural/unknown and smoke/diffusion repeat checks precede winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `51f53a4b3cb8001dba008fb6ed6df6c76291794f83d0ce0000d9871c35ac0505`. +- Validation: fixed 81-case subset; reward **0.811591**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449134 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E051a — positive future-step projection rejected on TRAIN + +Both paired TRAIN rollouts use all30 physical steps. At9/-3, clipping and renormalizing angular density after every future stored advance worsens VRMSE0.0057188724 to0.0174041912. At17/-5 it worsens0.7328003645 to0.7953209877. The constraint adds bias despite enforcing nonnegative sampled angular density. Outcome: reject without VALID replay; no package change. Protocol/source hashes and step/channel errors are in positive_step_v1. No learned parameters or new random seed; inherit narrow quadrature seed20261026. + +## E055a — synthetic convergence remains inadequate + +All eight short diagnostic runs remain finite. At9/-3, spatial64/angular64 ends with minimum first9-moment Toeplitz eigenvalue about-3.69e-6; spatial64/angular32 differs in low moments by RMS0.000595. At17/-5 even spatial64/angular64 reaches a retained-moment/concentration ratio1.2968 and minimum eigenvalue-1.3238. Spatial32/angular32,32/angular64 and64/angular32 differ from that still-inadequate reference by low-moment RMS0.5498,0.3836 and0.3845. Increasing only angular resolution is insufficient. Outcome: no converged teacher for the strong regime; no synthetic examples were generated or used in training. Do not interpret the highest-resolution run as ground truth. Exact numerical diagnostics/source/seed are retained in synthetic_convergence_v1. + +## E056 — boundary-aware convolutional profile residual preparation + +Hypothesis: shared local profile filters generalize better than position-specific dense weights. Append a width64,three-block1D convolutional residual to the selected16-bin statistical forecaster. Share weights across axes, condition on metadata/horizon and global input statistics, use periodic/replicated padding from boundary flags, and include coordinates only on nonperiodic axes. Center each output profile residual to preserve its mean. Add145675 parameters (statistical branch544587 to690262); zero output head preserves the parent's initial predictions exactly. Nonzero-head periodic-equivariance checks on four TRAIN groups pass below1.5e-9 normalized difference. Only the new residual is claimed equivariant. + +Queue matched base continuation and joint adapter training after E054 selection, using its latest shared statistical weights and inherited initial-context sampling probability. Seed20261121 initializes the new parameters and fixes identical sample draws;6000updates,batch512,learning rate0.0001 cosine,field-VRMSE objective,all19 TRAIN datasets. No new augmentation or synthetic data; unused covariance projection stays frozen zero. Compare1000/3000/6000 in the full shared runtime including guide forecasts, fixed decoder blends, and all9 thermal VALID trajectories. Record all per-dataset results and hashes before selection. + +## E054a — matched initial-context sampling comparison + +Continue the selected E049/periodic1000 shared statistical branch with matched50percent versus100percent initial-context sampling. Both use reconstructed-field VRMSE, identical sample draws and seed20261118,6000 updates,batch512,learning rate0.0001 cosine,all19 datasets and two CPU threads. No translations in either arm. No new parameters; unused covariance projection remains frozen zero. Every checkpoint is replayed through the same complete shared runtime, including its guidance role. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45097503 | 0.81144263 | `0e9088487c92cddb4f19c2ef776da9246bb29789a0d38f0f74b9f32e12f9ba44` | +| control | 3000 | 0.45026510 | 0.81149965 | `e2ef0a745c815bf85b0b5ac6a9ff0003fe50b16f20d43a4e2f0ff700c0ce0cfd` | +| control | 6000 | 0.44998706 | 0.81152200 | `444f5a222242457a02d36ad4e031556e518e24cd3cde3705637bc9972a34b7a6` | +| initial | 1000 | 0.45074985 | 0.81146071 | `a0fda8671f0a7606c0b60ead02a981af730f838500e4ce47b5726da466224fe1` | +| initial | 3000 | 0.45098953 | 0.81144147 | `49d0fb586dc86cc9282d88b2509e9be9cce4c8f0b7821017c80804c2332c6020` | +| initial | 6000 | 0.45044719 | 0.81148502 | `9d00e762d03479761f3aac676b74820d67170615c09d9ed852984146cf5a24f3` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E054 — first-four-frame training distribution +- Hypothesis / changed factor: Matched initial-context sampling comparison, selected modecontrol update6000. Both arms resume E049/periodic1000 with reconstructed-field objective, all19 datasets, seed20261118,6000updates,batch512,learning rate0.0001 cosine. Control chooses frame0 with probability0.5 plus incidental random starts; initial arm always uses the first four frames. No periodic augmentation or new parameters in either arm. Covariance projection remains frozen zero. All six candidates use the same complete runtime and fixed decoder blends; guidance consumes original context only. Replay all nine neural VALID trajectories, reuse72 unchanged complete physical outputs; record all metrics and hashes in E054a. Exact neural/unknown and smoke/diffusion repeats precede winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `47c46b77e41a6426199d670eae5da4f25a6578fbfb5d70ba15bb998919ecf632`. +- Validation: fixed 81-case subset; reward **0.811522**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449987 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E053b and E054b — no deployment improvement + +E053 finer dense8x32 forecasts worsen VALID at every saved update. Best fine result0.45100125 at6000, versus matched4x16 control0.44913370 and selected parent0.44911052. Retain the original E0334x16 branch. A target-error projection shows representational opportunity but does not establish that these fine modes are predictable by a dense forecast head. + +E054 all-first-context training also fails to improve deployment. At updates1000/3000/6000, control VRMSE0.45097503/0.45026510/0.44998706; first-context variant0.45074985/0.45098953/0.45044719. All are worse than selected0.44911052. Retain50percent initial-context sampling and the E049 head. Full per-dataset tables and hashes appear in each experiment's comparison entry; no package weights changed. + +## E057 — shared convolutional interaction forecaster preparation + +Hypothesis: local spatial weight sharing is a better inductive bias than a dense8x32 mapping. Train a159675-parameter shared forecaster from random initialization seed20261122, with width48,five conditioned depthwise blocks,periodic/replicated padding from metadata, coordinates only on nonperiodic axes and a zero output head. Inputs are four8x32 grids plus means/variances/metadata/horizon. All19 TRAIN datasets; batch128,12000updates,learning rate0.0003 cosine,two CPU threads,original interaction loss,50percent initial contexts. Fixed decoder blend0.25. Compare2000/6000/12000 through the complete then-selected shared runtime. Architecture, parameter count, initialization and optimization budget differ from E053; do not attribute differences to convolution alone. + +The first nonzero-head periodic-shift check failed by2.70e-5 normalized error from float32 interaction-centering cancellation on nearly homogeneous fields. Double-precision accumulation in that projection resolves the issue: all periodic checks on four TRAIN groups are now exact, initial predictions equal interaction persistence exactly, and backward checks have finite nonzero output-head gradients. No optimizer ran before these checks. No pretrained or synthetic assets, and no held-out targets. Source/checks in coarse_convolution_v1; pending validation and checkpoint hashes will be recorded. + +## E057b — CPU throughput allocation + +An isolated forward/backward benchmark on artificial tensors, with no optimizer or training data, measured median step times0.411/0.205/0.136seconds at2/4/8 CPU threads. The preliminary E057 run had only a few hundred updates and no saved checkpoint. Retire that CPU process only, preserve its exact configuration/source/log under pilot_two_threads, and restart from the same random initialization seed20261122 with8threads. This remains within the22-CPU allocation alongside the main GPU trainer, its six workers and the two-thread profile experiment. The published model and GPU optimizer are untouched. The final training config explicitly records8threads; no preliminary result enters model selection. Future validation description should use this actual thread count. + +## E058 — fixed context model-defect extrapolation pilot + +Current full corpus VRMSE0.01084193 is dominated by Gray–Scott gamma0.06278, KdV0.02468 and Gray–Scott delta0.02262. Hypothesis: one-step solver discrepancies within the supplied context reveal a persistent numerical-model bias. On the first two authorized TRAIN trajectories of gamma,delta andKdV, independently compute each observed transition's residual x[j+1]-P(x[j]) with the existing selected operator. Average the three residuals with fixed weights0.25/0.25/0.5, and compare the unchanged physical rollout with output correction strength0.25 or1 multiplied by forecast step. No inference parameters are fitted; only the fixed input-state calculation changes. No random seed or learned checkpoint, no synthetic data. Store full26-step TRAIN errors and diagnostic alignment of the context residual with future errors; those target-derived alignments are never inference coefficients. No VALID or package candidate yet. Source hashes/configuration and exact trajectory membership are in context_defect_v1. + +## E058a — fixed model-defect extrapolation rejected + +| TRAIN family | Trajectory | Unchanged | Strength0.25 | Strength1.0 | +|---|---:|---:|---:|---:| +| ape2d/gs_gamma | 0 | 0.029922720 | 0.029934172 | 0.030060413 | +| ape2d/gs_gamma | 1 | 0.085883125 | 0.085890967 | 0.085974602 | +| ape2d/gs_delta | 0 | 0.013151899 | 0.013155781 | 0.013236184 | +| ape2d/gs_delta | 1 | 0.012910574 | 0.012918597 | 0.013015623 | +| ape2d/kdv | 0 | 0.000014622 | 0.000014187 | 0.000014296 | +| ape2d/kdv | 1 | 0.021805965 | 0.023939122 | 0.044701402 | + +Both nonzero strengths worsen all four Gray–Scott cases and the nontrivial KdV case. The nearly exact KdV case changes by less than5e-7. Outcome: reject persistent output-defect extrapolation; do not run validation or alter the package. Context residuals do not supply a useful fixed spatial bias under this formula. Full time/channel errors and target-dependent diagnostic alignments are retained, never used as inference coefficients. No learned checkpoint or seed. + +## E059 — differentiable hidden-state learning feasibility + +The synthetic teacher remains unfit, so investigate whether future TRAIN observations could instead supervise a hidden-state head through the fixed physical dynamics. This is preparation only: no new neural head or optimizer exists yet. Check state derivatives of the independently implemented angular-moment kernel on a small artificial16x16,M8 state, seed20261124. Compare ordinary autograd, substep activation recomputation, and a central finite directional derivative in double precision. No real targets or synthetic training examples. The published TorchScript forward kernel fails under autograd with a complex-to-double overflow before producing a gradient; preserve that failure, then test the identical arithmetic as ordinary Python functions with decorators removed. No published runtime changes. Only if this works should a joint all-dataset neural/state training design be considered; inference would retain fixed parameters and use only four frames. + +## E056a — matched profile-convolution comparison + +Compare matched continuation against a shared1D convolutional residual on16-bin profiles. Both resume the selected parent after E054, share sample sequence and seed20261121,field-VRMSE objective,6000updates,batch512,learning rate0.0001 cosine and all19 TRAIN datasets. Inherit the selected initial-context sampling probability. The residual adds145675 parameters (total statistical branch690262), uses boundary-aware padding and a zero output head. Parent-output parity is exact before training; nonzero residual periodic-equivariance checks pass below1.5e-9 normalized error. The inherited MLP is not claimed equivariant. Every checkpoint uses the complete runtime, including guidance. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.45067359 | 0.81146683 | `eebebf08b139686eb72c688525ab18059b71d54d829ac94f7c62fc0101ce2ecd` | +| control | 3000 | 0.45074070 | 0.81146144 | `bd984e3b48d1f654ae7a6c4a4de0ccc1bda640abb059ced00534603a983f4cb0` | +| control | 6000 | 0.45057447 | 0.81147479 | `b86ccce3ee5a613314ee1fc44c7cf616ea62c48381920b612818b4a276ae4690` | +| convolution | 1000 | 0.45016542 | 0.81150766 | `e35737f2dbde15fcf806e35042b26b2e4e25f70e52b3fc0ae5160682366fd1d2` | +| convolution | 3000 | 0.45075162 | 0.81146057 | `27c73cbbeb60b4b0254a2ebbbf08be5bb0bb9a9d82e21011cea5452591c64a25` | +| convolution | 6000 | 0.45079751 | 0.81145688 | `d121fbad6fe2edf06d186304b086e460da7089f821113d05b0cab10207e55bab` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E056 — shared convolutional profile residual +- Hypothesis / changed factor: Matched shared-profile convolution study, selected modeconvolution update1000. Both arms resume the selected statistical parent after E054 with field-VRMSE objective, identical samples, seed20261121,6000updates,batch512,learning rate0.0001 cosine,all19 datasets and inherited initial-context probability. Convolution variant adds145675 parameters through a width64,three-block residual with metadata/horizon conditioning and boundary-aware padding. Output head starts at zero, preserving the parent exactly; the nonzero residual is equivariant on periodic axes. No new mean-output branch or pretrained assets; unused covariance projection remains frozen zero. All six candidates are tested through the complete shared runtime with fixed decoder blends and guidance from original context only. Replay all9 neural VALID trajectories and reuse72 complete unchanged physical outputs. All metrics and hashes are recorded in E056a. Exact neural/unknown and smoke/diffusion repeat checks precede winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `20f6453e9ad2a9b45a022443f03e8cdc8184b992d395c1d3c69566f69a789f1d`. +- Validation: fixed 81-case subset; reward **0.811508**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.450165 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 1000, shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `d74b5240df075970d9951637f684d5865e953c3cfccd5629de6c2b491307a18c`. +- Runtime code SHA256: `bde67d792a2cc2cd64656ca21dfe1fbfdd653c2560d5376370cc304b090daf51`. +- Validation: fixed 81-case subset; reward **0.810513**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.462711 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 1000, no_shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.0; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `d74b5240df075970d9951637f684d5865e953c3cfccd5629de6c2b491307a18c`. +- Runtime code SHA256: `b02fbeefa500eb903647f72467e049c24b9d82207ed900d3de67d77b1a5e81de`. +- Validation: fixed 81-case subset; reward **0.810489**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.463020 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 1000, mean_profile_off +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=False; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `d74b5240df075970d9951637f684d5865e953c3cfccd5629de6c2b491307a18c`. +- Runtime code SHA256: `54e4d7c0dbee2ada00505f657b8d9bc6993085410464ff02a4f98f6c8aac8186`. +- Validation: fixed 81-case subset; reward **0.809958**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.469864 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E056b — convolutional profile residual did not improve deployment + +The added residual beats its matched control at update1000 (0.45016542 versus0.45067359), but regresses at3000/6000 and never beats the selected0.44911052. Best control0.45057447, best convolution0.45016542. Reject all six candidates; keep the E049 statistical branch. All VALID dataset metrics, seeds and checkpoint hashes remain in E056a. No package weight changes. + +## E059a — physical state gradients and recomputation verified + +Removing only TorchScript decorators resolves the autograd failure while preserving the physical arithmetic. On the16x16,M8,double-precision artificial state, directional autograd and finite differences agree to relative7.17e-8; ordinary and recomputed values/gradients agree exactly. No optimizer or training examples. + +Native256,M32 measurements use authorized TRAIN trajectories only and no optimizer. On9/-3 trajectory1, one stored step (96 substeps) matches the published center-node forecast exactly, yields finite gradient norm9.79e-8, uses2.067GiB peak allocated memory and takes7.48seconds forward/backward. On17/-5 trajectory1, four stored steps (176 substeps each) also match exactly, give finite gradient norm9.51e-6, use7.182GiB and take42.45seconds. Timings share the H100 with main training and are feasibility measurements, not isolated benchmarks. Activation recomputation uses blocks of two internal steps. Seeds20261124/20261125; exact source hashes/reports in differentiable_latent_v1. The kernel operates only on physical state; no inference-time optimizer is proposed. + +## E060 — joint hidden-state head preparation + +Hypothesis: real future TRAIN observations can supervise a missing angular-state correction through differentiable dynamics, avoiding the rejected synthetic teacher. Add a zero-initialized two-channel head to the same shared neural backbone; pool its output to16x16. It is computed on every forward pass for every dataset. The planned physical interpretation is a bounded correction of F6, with all model parameters fixed during inference. Main parameters3066627 to3066725 at width48 (98 added parameters); seed20261126, all added values initially zero. Main-output parity and finite state-head gradients are checked on four TRAIN groups. No optimizer or joint training has started. + +Cache all175 active TRAIN trajectories' first four frames, four future LABEL frames, and the fixed encoded center state/streamfunction. The encoder consumes only frames0..3; future observations are labels and never state inputs. Copy and hash the current independently implemented encoder and Bingham table; audit every trajectory against train pools. These are derived real training data, not synthetic trajectories or pretrained assets. Future joint training must retain the ordinary all19-dataset loss on the same shared backbone; do not train or switch to a separate active-only checkpoint. Any inference implementation must pass zero-head parity and full24-active/full585 selection before publication. Cache and preparation code live in latent_state_v1; physical source changes require rechecking cache compatibility. + +## E057a — convolutional interaction forecaster comparison + +Train a new shared convolutional8x32 interaction forecaster from random initialization seed20261122: width48,five conditioned depthwise blocks,batch128,12000updates,learning rate0.0003 cosine,all19 TRAIN datasets and eight CPU threads. Unlike the dense8x32 branch of E053, spatial weights are shared and periodic axes have exact convolutional translation symmetry up to roundoff. Preserve original interaction loss and0.25 output blend. Interaction centering uses float64 accumulation to avoid cancellation in nearly homogeneous channels. This architecture comparison also changes parameter count and optimization budget; no causal attribution to convolution alone is claimed. Every checkpoint uses the complete then-selected runtime. + +| Variant | Update | Radiative-layer VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| convolution | 2000 | 0.45403915 | 0.81119756 | `98bad765a7bf75c82d06de5833b76fbe30547ee350bc0209b745eace764ca440` | +| convolution | 6000 | 0.45272777 | 0.81130225 | `a2be891cce5aed99e77e404c42f01c59e56aec4b1e51dafc5f93d38fe03c1f68` | +| convolution | 12000 | 0.45190796 | 0.81136784 | `86922705fe4342ed6808e256078e6e399563feba021564b51a06ebe3eefddb4b` | + +The following unchanged dataset metrics apply to every candidate, reusing complete physical prefixes from the documented parent: + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.00641268 | +| ape2d/decay_turb | 0.00038142 | +| ape2d/diff | 0.00000260 | +| ape2d/fisher | 0.00000341 | +| ape2d/gs_alpha | 0.00309470 | +| ape2d/gs_beta | 0.00474027 | +| ape2d/gs_delta | 0.01924967 | +| ape2d/gs_epsilon | 0.00136617 | +| ape2d/gs_gamma | 0.04931292 | +| ape2d/gs_iota | 0.00178479 | +| ape2d/gs_kappa | 0.00485456 | +| ape2d/gs_theta | 0.00467797 | +| ape2d/kdv | 0.01672801 | +| ape2d/kolm_flow | 0.00892557 | +| ape2d/ks | 0.00003913 | +| ape2d/sh | 0.00000371 | +| pdearena/navier_stokes_2d | 0.00008349 | +| well/active_matter | 0.04060092 | + +Outcome: select the best complete-package result for contract checks and comparison against the current published package. All regressions remain recorded above. + +## E057 — shared convolutional interaction forecast +- Hypothesis / changed factor: Shared convolutional8x32 interaction forecaster, selected update12000. Train from random initialization seed20261122 on all19 datasets,12000updates,batch128,learning rate0.0003 cosine. Width48,five boundary-aware depthwise blocks,metadata/horizon conditioning and zero-initialized output. Spatial weights are shared; coordinates are zero on periodic axes. Float64 interaction centering fixes a pre-training cancellation-check failure. Keep original interaction loss and fixed0.25 correction blend; compare2000/6000/12000 through the complete shared runtime. All9 neural VALID trajectories replayed;72 unchanged complete physical outputs reused. No pretrained or synthetic assets; same weights applied to every case. Complete metrics, seeds and hashes are in E057a. Exact neural/unknown and smoke/diffusion repeats precede winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `29efdb1c6552e403f41ff128b035428478610a65682dc88ed4f728e704910aac`. +- Validation: fixed 81-case subset; reward **0.811368**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.451908 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E060a joint state-head preparation and validation dependency audit + +Hypothesis: infer a bounded F6 hidden-moment correction from the same shared backbone, jointly trained on all19 datasets and a differentiable physical loss, to address active-matter hidden-state uncertainty. Added98 zero parameters at width48, seed20261126; unpublished zero prototype main SHA256 `284e981ce16e71afa2ad060ed8189863dd96c60aea0a1bb6a93c00c7861ae10f`. Native nonzero-coefficient parity seed20261128: two-step runtime/training normalized RMS difference 2.6434134e-06, maxabsolute 1.1920929e-07; periodic shift normalized RMS 1.5527538e-06. All six zero-head serving checks are bit exact: TRAIN examples from four groups plus independently constructed2-channel7x11 and13-channel129x177 grids. One joint main/physical dry run used seed20261127,2 main steps and1 physical step, then verified every parameter unchanged; no optimizer step or trained checkpoint. All175 active TRAIN trajectories cached with only first4 frames in encoder and next4 frames as labels. Cache records include membership and SHA256; no synthetic or pretrained data. + +Validation VRMSE per dataset: not evaluated; this was an implementation/gradient check with no trained candidate. Outcome: gradient, runtime, translation, cache and unchanged-parameter checks passed. Fixed-state-head dependencies now include neural/guide weights and serving configuration: changing weights correctly invalidates active-output reuse. New full validation controller replays all24 active plus9 thermal cases, with only552 complete unchanged corpus/smoke outputs reused after exact dependency audit. A deliberate solver mutation is rejected. Queue actual joint2000-update training after phase9 normal exit, with updates1000/2000 compared on all585 VALID trajectories. The selected package remains E045/E049/E033; no model change from this preparation. + +## E061 — fixed spatial interaction filtering + +Hypothesis: small-scale forecast detail may be unreliable. Compare fixed boundary-aware Gaussian filtering of only the two-dimensional interaction, preserving global means and both marginal profiles. Radius grows with sqrt(horizon/30), scales with native short-axis resolution, and uses even extension on nonperiodic axes. Four predeclared radii0/1/2/4 cells on a128-cell short axis. Nine audited TRAIN trajectories, all cooling regimes, same selected E045/E049/E033 package; no learned weights or new seed. Main SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`; filter SHA256 `bdc3ee90c5d898a085456ab95b02e6dbd3a35cb7bd49774655312c3aeb0add00`. + +| Radius | TRAIN VRMSE | +|---|---:| +| 0 | 0.409199345 | +| 1 | 0.409320762 | +| 2 | 0.410080575 | +| 4 | 0.413902060 | + +An initial profile-parity check used a fixed normalized tolerance and failed on a nearly constant pressure field. Preserve the original centered interaction in float64 and test profile changes against the analytic float32 output-rounding bound; all nine pass, along with periodic translation RMS checks. Initial failed log is retained. Outcome: all nonzero filters regress; reject before VALID evaluation. Validation VRMSE by dataset: not evaluated; other TRAIN datasets not evaluated for this fixed-decoder pilot. No package runtime change. + +## E062 — periodic translation averaging diagnostic + +Hypothesis: averaging the same complete neural forecast under periodic translations reduces U-net stride-phase and boundary-alignment errors. Pin the current selected package and replay nine audited thermal TRAIN trajectories for shifts0/2/4/6/32/64/96 native x cells. Compare the single forecast, equal four sub-stride translations and equal four quarter-domain translations after exact inverse shifts. Every member uses identical learned weights, context-only guide forecasts and decoder coefficients. No parameter training, new seed or target-dependent inference transform. Only TRAIN targets select whether a portable complete-model candidate is warranted; VALID results pending. Main checkpoint inherited from E061, with exact parent and memberships in translation_average_v1. + +## E062a — translation averaging TRAIN and thermal VALID result + +Same shared checkpoint SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`; no new parameters or random seed. Nine audited TRAIN trajectories; all future targets remain loss/score inputs only. + +| Fixed rule | TRAIN VRMSE | +|---|---:| +| single | 0.409199345 | +| substride_four | 0.408299870 | +| quarter_domain_four | 0.415312758 | + +Select sub-stride translations0/2/4/6 cells at128-cell x resolution, then scale those fractions to other grids. The complete shared model applies to every member and case; nonperiodic-x boundaries yield identity members with shared computations. Nine thermal VALID trajectories improve **0.449110525 to 0.446880629**, zero failures. Other validation datasets pending. This is an intermediate result, not sufficient for publication: all576 physical VALID cases are being replayed, checking every distinct member’s complete prefix. Full585 comparison, latest-neural rebase and exact repeat checks remain required. Config key translation_ensemble now invalidates physical cache reuse across untranslated/translated packages. Source package and logs are retained under translation_average_v1/candidate. + +## E063 — matched hidden-unit dropout preparation + +Hypothesis: the shared statistical branch overfits during continuation, as multiple recent experiments preferred early checkpoints. Compare control against training-only dropout0.1 after each of the three hidden SiLU activations. Both start from selected E049/periodic1000, packaged auxiliary SHA256 `41c6fb40a84c8bbc5ff6d59967f47edd293f5f3c2f631611f9e732299ba756ca`;544587 parameters, no architecture or inference change. Sample seed20261130, maskseed20261131,6000updates,batch512,LR0.0001 cosine, reconstructed-field VRMSE objective, all10115 TRAIN records/all19 datasets,50percent initial contexts, no new spatial augmentation. Covariance projection stays frozenzero. Each arm uses two CPU threads and identical NumPy sample draws. Forward hooks disappear from the serialized model; pre-training eval parity must be exact on a TRAIN example from each group, and only dropout training values may change. Compare1000/3000/6000 through the same complete serving runtime and nine thermal VALID trajectories. Per-dataset VRMSE/hashes pending; no model publication from preparation. + +## E064 — final nonnegative support preparation + +Hypothesis: later statistical corrections undo the early nonnegative density constraint. Nine audited TRAIN rollouts contain negative density in0.8–8percent of predicted pixels, with minima between-1.1 and-2.7; their corresponding target density is nonnegative everywhere. Project only supplied fields named density or concentration to zero or above, after every correction and the final translation average. Pressure and velocity are unchanged. No new parameters, fitted bounds or random seed; same rule for every case. Main parent SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. MeanTRAIN VRMSE 0.409199345→0.409183765; every step/channel is nonworsening. This is a small consistency gain, not a major accuracy result. + +Queue validation after E062 finishes its full-cohort decision. Use the latest selected parent, apply the identical final unary projection to all585 validated predictions, and verify exact fresh-serving equality on all9 thermal cases. Record per-case source hashes/changed counts, then full585 score plus active/smoke/unknown repeat checks before any winning publication. This derived validation output is a development-only audit; the package applies the projection directly to its own predictions and never reads a cache. Metadata/config key final_nonnegative_fields is a physical dependency, requiring full585 comparison. Any win updates only phase8/9 inference templates and their physical caches; queued phase10 inherits the selected parent at its later launch. Validation per dataset pending. + +## E063a — matched statistical-dropout results + +Continue selected shared mean/profile weights with control versus dropout0.1 after all three hidden SiLU activations, training only. Same architecture/544587 parameters and deterministic inference; hooks are absent from packaged runtime. Seed20261130 fixes identical all19-dataset sample draws, dropoutseed20261131,6000updates,batch512,LR0.0001 cosine,field-VRMSE loss,50percent initial contexts, no periodic augmentation in either arm. Frozen zero covariance projection. Eval parity is exact in both arms before training; dropout changes only training values. + +| Arm | Update | Thermal VALID VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.450102517 | 0.811512720 | `f637840f08e08a3d12d867b4acd9bf59ec443c3bfe3415f0ff0e9983d8bf2b20` | +| control | 3000 | 0.450255515 | 0.811500420 | `d5c71f8fb270a990525db29c62d48ed20ef5946aa6f7087be2c48c8d57d8a928` | +| control | 6000 | 0.450153693 | 0.811508605 | `9875e3b166fe5ae352c897c42bfb709d0c643a859128c5c1d642c2ce4c2f2965` | +| dropout | 1000 | 0.449423886 | 0.811567327 | `6dce1289d5445255f4fbfab170e142d51b2b338372b27ab4790e80244860d140` | +| dropout | 3000 | 0.448526125 | 0.811639695 | `0b60fa6fc224a6c6ba8f0a079d4b21d92a40b40c3201203588c88b03ecf435be` | +| dropout | 6000 | 0.448237033 | 0.811663029 | `f55c11a895d7276342e3e3461bf50d34a5e7d5bddfe3a11db4e1f8946c67ceb2` | + +All candidates reuse the same complete physical outputs from the documented nonlatent parent; the following dataset metrics apply to every row: + +| Dataset | VALID VRMSE | +|---|---:| +| ape2d/burgers | 0.006412678 | +| ape2d/decay_turb | 0.000381422 | +| ape2d/diff | 0.000002603 | +| ape2d/fisher | 0.000003406 | +| ape2d/gs_alpha | 0.003094704 | +| ape2d/gs_beta | 0.004740266 | +| ape2d/gs_delta | 0.019249670 | +| ape2d/gs_epsilon | 0.001366170 | +| ape2d/gs_gamma | 0.049312925 | +| ape2d/gs_iota | 0.001784790 | +| ape2d/gs_kappa | 0.004854561 | +| ape2d/gs_theta | 0.004677970 | +| ape2d/kdv | 0.016728012 | +| ape2d/kolm_flow | 0.008925566 | +| ape2d/ks | 0.000039134 | +| ape2d/sh | 0.000003711 | +| pdearena/navier_stokes_2d | 0.000083487 | +| well/active_matter | 0.040600924 | + +Outcome: best complete-package candidate proceeds to comparison with the current selection; all negative checkpoints remain recorded. + +## E063 — shared statistical-branch regularization +- Hypothesis / changed factor: Matched hidden-unit dropout comparison, selecteddropout update6000. Both arms resume selected shared statistics with all19 TRAIN datasets, sample seed20261130,6000updates,batch512, LR0.0001 cosine and reconstructed-field VRMSE. The changed factor is dropout0.1 after three hidden SiLUs; maskseed20261131. Inference uses the unchanged deterministic architecture and all shared weights on every case. Initialcontext probability0.5, no additional augmentation, covariance projection frozen zero. All six complete-package checkpoints and per-dataset scores/hashes recorded in E063a. Nine thermal VALID trajectories replayed; unchanged complete physical outputs reused from a nonlatent parent. Neural/unknown and smoke/diffusion repeat checks precede a winning publication. +- Seed: 20261102. +- Checkpoint SHA256: `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. +- Runtime code SHA256: `d785c2fb9e59ac1456cb7226d7908cc8c81ba6c9540d72ab84dfe31ccc1e4c9c`. +- Validation: fixed 81-case subset; reward **0.811663**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.448237 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of dropout_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.780967**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261102; checkpoint SHA256 `5969f2d0a3b0b95aadcafc9aeff5cc3f0432348b6ce5518e79b96925b2ad10fe`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.448237 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/phase7/late_average. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed in this model’s quick validation, with exact context equality to full inputs. Reuse these verified outputs and score all585 validation trajectories together. Per-case physical coverage is retained in full_physical_coverage.json. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E038 — complete-rollout autoregressive training, update 2000, shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `d785c2fb9e59ac1456cb7226d7908cc8c81ba6c9540d72ab84dfe31ccc1e4c9c`. +- Validation: fixed 81-case subset; reward **0.811531**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.449874 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 2000, no_shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.0; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `d204d651c7534e1c9a73ab251b0073a771ecf14d28c0d1eaed92180edabe461a`. +- Validation: fixed 81-case subset; reward **0.811936**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.444868 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E038 — complete-rollout autoregressive training, update 2000, mean_profile_off +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=False; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `09804b2a1e385fe55ab3b71dafffee0e269adb29510d3191768963335fa5332a`. +- Validation: fixed 81-case subset; reward **0.811277**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453042 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E065 — exact official-score reuse + +Infrastructure hypothesis: repeated reads of unchanged physical targets dominate CPU validation time. Use official per-case scores only after a separate input/prediction-dependency or byte-equality audit, and extract the unchanged aggregation tail of evaluation/score.py through its Python AST. Every changed case is still scored by the official scorer. Preserve its case order, failures, per-channel/step/window results, dataset means, group means and log-band reward; record scorer, source-report, index and reuse-proof hashes alongside each derived report. No training, parameters, seed or model change. + +Checks in score_report_checks.json reconstruct the existing E045 quick81 and full585 reports with complete dictionary equality, then reproduce the independently scored E063/control1000 quick81 report exactly by combining72 unchanged physical case scores with9 freshly scored thermal cases. Full585 reward remains0.7808965854886851 for E045; every dataset metric is unchanged exactly. Outcome: adopt exact report reuse in future neural full-cohort derivations and the queued E060/E064 controllers. Existing E062 full physical replay still scores its changed physical predictions normally. This is a development efficiency change; no validation artifact is read by inference. + +## E066 — late full-rollout parameter average preparation + +Predeclare one equal parameter average of phase8 updates2000/3000/4000 before3000 and4000 exist. Test whether averaging late optimization variation retains the gain from full30-step training. Use float64 accumulation and restore original dtypes; same resulting single shared network on every case, no new parameters, optimization, seed or runtime ensemble. All three checkpoints train on all19 datasets with samplingseed20261106 and the same frozen guide. At evaluation preserve the then-selected shared auxiliary weights, decoder and fixed operators, and record all source hashes. Score all9 thermal VALID cases officially, reuse72 verified complete physical cases through exact official aggregation, and require neural/unknown repeat checks before a win. Any selected average gets a full585 report. The queued width96 launcher now waits for this experiment after phase8 exits, then widens the latest selected parent. Validation metrics pending. + +## Full validation of eval002000_no_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781240**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.444868 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/phase7/late_average. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E067 — dropout and periodic augmentation preparation + +Hypothesis: periodic profile augmentation may complement the regularization that made E063 improve through6000updates. Continue selected E063/dropout6000 auxiliary SHA256 `4109196a3aea07aa24c1e1c1d0ea958e6ff3c2fb160260672ec242b44fefc740` in two matched arms: dropout0.1 alone versus dropout0.1 plus independent cyclic profile-bin shifts on periodic axes. Same544587 parameters/runtime architecture and frozenzero covariance projection, all10115 TRAIN records/all19 datasets,50percent initial contexts,field-VRMSE objective,batch512,12000updates,LR0.00005 cosine, two CPU threads each. Sample seed20261201, matched dropout masksseed20261202, separate augmentationseed20261203; both arms draw identical data and dropout masks. Training-only hooks vanish from inference, whose initial output parity must be exact on examples from all four groups. Compare2000/6000/12000 using the same selected main/decoder parent, with official scoring of all9 changed thermal VALID cases and exact aggregation of72 audited unchanged physical case scores. Record all per-dataset metrics/hashes; repeat checks and full585 reporting precede completion of any winning model publication. No synthetic or pretrained assets. Validation pending. + +## E067a — matched dropout and periodic-augmentation results + +Continue selected E063/dropout6000 shared mean/profile weights with dropout alone versus dropout plus cyclic profile translations on periodic axes. Same architecture/544587 parameters and deterministic inference; hooks are absent from packaged runtime. Seed20261201 fixes identical all19-dataset sample draws, dropoutseed20261202, augmentationseed20261203,12000updates,batch512,LR0.00005 cosine,field-VRMSE loss,50percent initial contexts. Both arms use dropout0.1 after all three hidden SiLUs. Frozen zero covariance projection. Eval parity is exact before training. Only nine changed thermal cases are freshly scored; the exact official aggregation code combines72 audited unchanged physical case scores. + +| Arm | Update | Thermal VALID VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| dropout | 2000 | 0.444632291 | 0.811955257 | `76f53498712ea942cb053730ed5387873d09cbb1d0099301b74921537f80f5c4` | +| dropout | 6000 | 0.444318958 | 0.811980770 | `8014c4ef2f46b38e0a44715812a9fa08f687419f2fb378a99428a87d87228740` | +| dropout | 12000 | 0.444360818 | 0.811977360 | `ff257d32d992ff7e2834aedd6b3dda8161bc1d0549d8382c132d38995fb4c045` | +| periodic | 2000 | 0.444454130 | 0.811969761 | `12612e8c70024ffb2823011af96271da9daf92338c02e1461ae859214e8a6de7` | +| periodic | 6000 | 0.444025853 | 0.812004652 | `af45d346835da4722967c22962323bc096b4fbbd9171a79d46a49717ce44abe9` | +| periodic | 12000 | 0.444350963 | 0.811978163 | `93d2ce4daa8c2063b90551244670046cb60513d426306ee4523b480b9cf131f7` | + +All candidates reuse the same complete physical outputs from the documented nonlatent parent; the following dataset metrics apply to every row: + +| Dataset | VALID VRMSE | +|---|---:| +| ape2d/burgers | 0.006412678 | +| ape2d/decay_turb | 0.000381422 | +| ape2d/diff | 0.000002603 | +| ape2d/fisher | 0.000003406 | +| ape2d/gs_alpha | 0.003094704 | +| ape2d/gs_beta | 0.004740266 | +| ape2d/gs_delta | 0.019249670 | +| ape2d/gs_epsilon | 0.001366170 | +| ape2d/gs_gamma | 0.049312925 | +| ape2d/gs_iota | 0.001784790 | +| ape2d/gs_kappa | 0.004854561 | +| ape2d/gs_theta | 0.004677970 | +| ape2d/kdv | 0.016728012 | +| ape2d/kolm_flow | 0.008925566 | +| ape2d/ks | 0.000039134 | +| ape2d/sh | 0.000003711 | +| pdearena/navier_stokes_2d | 0.000083487 | +| well/active_matter | 0.040600924 | + +Outcome: best complete-package candidate proceeds to comparison with the current selection; all negative checkpoints remain recorded. + +## E067 — dropout with periodic profile augmentation +- Hypothesis / changed factor: Matched dropout/periodic-augmentation continuation, selectedperiodic update6000. Both arms resume E063/dropout6000 with all19 TRAIN datasets, sample seed20261201,12000updates,batch512, LR0.00005 cosine and reconstructed-field VRMSE. Both use dropout0.1 after three SiLUs, maskseed20261202. Changed factor: independent circular translations of context and target profiles on periodic axes, seed20261203. Initialcontext probability0.5; covariance projection frozenzero. Inference uses the same deterministic shared weights on every case. All six per-dataset scores and checkpoint hashes recorded in E067a. Nine changed thermal VALID cases are freshly scored with the official scorer;72 complete physical case scores are reused after dependency checks and reaggregated with the exact official aggregation code. Neural/unknown and smoke/diffusion repeats precede any winning publication. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `764101bac331d2c72f49db82ff94744d0f1bc6f03d78942af9532914e3b5e3b1`. +- Validation: fixed 81-case subset; reward **0.812005**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.444026 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of periodic_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781309**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.444026 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/phase8/eval002000_no_shrink. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E068 — per-step neural translation averaging rejected on TRAIN + +Hypothesis: average the same neural update across four periodic-X translations at every autoregressive step, sharing the original fixed guide forecasts, to reduce stride artifacts. Changed factor: shifts0,2/128,4/128,6/128 of the working grid; inverse-shift and float64 average before each feedback step. Nonperiodic axes use identity; duplicate shifts are shared. All weights and physical updates are unchanged. This differs from E062 complete-rollout averaging. No new training seed or parameters; diagnostic seed20261206 is used only for a temporary nonzero latent-head implementation check. + +Parent main checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`; shared statistics SHA256 `af0cfd6fa137519bbd329075b23c87c3332e79aba584b7de4027c0b613b2af9a`. Runtime helper SHA256 `f750bfadb4c458bdd5b192615b6fd1365983a44e0f37055ad2632cf445c66f65`. Nine audited thermal TRAIN trajectories, first four context frames and30 forecast steps: VRMSE 0.395806216047 to 0.396866427992. Six complete-package shape/finiteness checks passed; outputs for the TRAIN diffusion/smoke/active cases and two constructed nonperiodic novel cases were unchanged exactly. Unit checks verified nonperiodic identity, canonical unshifted latent output, unchanged parameters and removal of temporary guide hooks. + +VALID VRMSE per dataset: not evaluated; rejected at the TRAIN diagnostic. No publication or changes to active controllers. Full per-case/channel/step results are in step_average_v1/train_results.json. + +## E069 — matched shared coarse-branch dropout preparation + +Hypothesis: the successful E063 statistical-branch regularizer may also improve the coarse spatial interaction forecast. Continue the selected E033/update6000 coarse weights in two matched arms, no dropout versus dropout0.1 after all three hidden SiLUs. Same787808 parameters,4x16 grid and deterministic inference code; no periodic augmentation or changes to mean/profile weights. Train all19 datasets, sample seed20261207, independent dropout seed20261208,6000updates,batch512, LR0.0001 cosine, original normalized coarse-interaction loss,50percent initial contexts, two CPU threads per arm. Evaluate checkpoints1000/3000/6000 through one pinned complete selected package, retain blend0.25; all dataset metrics and hashes will be recorded. Eval identity and train stochasticity are checked before optimization. + +Parent coarse checkpoint SHA256 `ad9a9996912c63eaaed7aa01423bc8598c9a8c966915b277ab64ce1774027dd3`; inherited random initialization seed20261103. No pretrained assets. VALID VRMSE pending. + +## E070 — native active-state precision diagnostic preparation + +Hypothesis: accumulated roundoff in the fixed hidden-state encoder and moment integration contributes to long-horizon phase error. Changed factor: float64/complex128 context, wave numbers, state, coefficient table and integration versus the current float32/complex64 path. Same physics, M32/native256, timestep, six Stokes iterations, observer, fixed narrow1e-4 three-node quadrature and seed20261026; no new learned parameters or optimization. Compare six predeclared TRAIN regimes(1,-1),(3,-2),(5,-4),(9,-3),(15,-4),(17,-5), taking the third authorized TRAIN trajectory in each file,30 future steps from four observations. Return float32 predictions before scoring. + +CPU32-grid implementation checks reproduce the original float32 path exactly and confirm finite float64 dynamics. Full native TRAIN diagnostic is queued behind E062 full validation and E064 final-projection validation on the only H100. No VALID candidate or publication is authorized by this diagnostic alone; if promising, encoder-cache and latent-kernel compatibility must be handled before any later integration. + +Parent kinetic SHA256 `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`; candidate kinetic SHA256 `5db556077cf42a9bcb931ce2cb5aaf51d99c8039b30206712ce3df5e32deeb07`. Checkpoint unchanged from E067; VALID VRMSE pending/not evaluated. + +## E069a — matched coarse-branch dropout results + +Continue selected E033/update6000 coarse interaction weights with no dropout versus dropout0.1 after three hidden SiLUs. Same architecture/787808 parameters and deterministic inference; hooks are absent from packaged runtime. Seed20261207 fixes identical all19-dataset sample draws, dropoutseed20261208,6000updates,batch512,LR0.0001 cosine, original variance-normalized interaction objective,50percent initial contexts. No periodic augmentation. Eval parity is exact before training. Only nine changed thermal cases are freshly scored; the exact official aggregation code combines72 audited unchanged physical case scores. + +| Arm | Update | Thermal VALID VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.444104319 | 0.811998257 | `34d62b9722631e3cd31f88bdf6b7efd7eb6809290dd515128c7456da64da7b5b` | +| control | 3000 | 0.444124794 | 0.811996588 | `c7ced6518bc26fd5ee40809a368a1fa40ccb3a8ebd911f75480daa439718aeec` | +| control | 6000 | 0.444098716 | 0.811998714 | `98141a1a650f0beae40b61d2257fdfc670a24188157af32f07558e9d179f67d4` | +| dropout | 1000 | 0.444112960 | 0.811997553 | `55b16cf95f778d3e7da1d9007abd6ba957ccf84d0935a2a334c3abe2db44164e` | +| dropout | 3000 | 0.444178125 | 0.811992243 | `17e3a8e48e130d6a95d97f768233fd55b843ec15b02480e56164c6792e1f871c` | +| dropout | 6000 | 0.444201102 | 0.811990371 | `6e931f30c9eb43ff2436a8e9bcafbc8c4219139381083f3d5726ab248eeaed74` | + +All candidates reuse the same complete physical outputs from the documented nonlatent parent; the following dataset metrics apply to every row: + +| Dataset | VALID VRMSE | +|---|---:| +| ape2d/burgers | 0.006412678 | +| ape2d/decay_turb | 0.000381422 | +| ape2d/diff | 0.000002603 | +| ape2d/fisher | 0.000003406 | +| ape2d/gs_alpha | 0.003094704 | +| ape2d/gs_beta | 0.004740266 | +| ape2d/gs_delta | 0.019249670 | +| ape2d/gs_epsilon | 0.001366170 | +| ape2d/gs_gamma | 0.049312925 | +| ape2d/gs_iota | 0.001784790 | +| ape2d/gs_kappa | 0.004854561 | +| ape2d/gs_theta | 0.004677970 | +| ape2d/kdv | 0.016728012 | +| ape2d/kolm_flow | 0.008925566 | +| ape2d/ks | 0.000039134 | +| ape2d/sh | 0.000003711 | +| pdearena/navier_stokes_2d | 0.000083487 | +| well/active_matter | 0.040600924 | + +Outcome: best complete-package candidate proceeds to comparison with the current selection; all negative checkpoints remain recorded. + +## E069 — shared coarse-branch dropout +- Hypothesis / changed factor: Matched coarse-branch dropout continuation, selectedcontrol update6000. Both arms resume E033/update6000 with all19 TRAIN datasets, sample seed20261207,6000updates,batch512, LR0.0001 cosine and the original normalized coarse interaction objective. Changed factor: dropout0.1 after three hidden SiLUs, maskseed20261208, versus a control with no dropout. Initialcontext probability0.5, no periodic augmentation. Same787808 parameters, deterministic runtime. All six per-dataset scores and checkpoint hashes recorded in E069a. Nine changed thermal VALID cases are freshly scored;72 complete physical case scores reused only after exact dependency/coverage audit. Neural/unknown and smoke/diffusion repeats precede any winning publication. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `63776e7964b71f5ec9dc1ca059f11e51ace2dfaf79e139d027f9147a1e7dc1eb`. +- Validation: fixed 81-case subset; reward **0.811999**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.444099 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E071 — rollout objective gradient diagnosis + +Hypothesis: the current log1p normalized squared loss and the benchmark VRMSE emphasize channels differently, which may explain remaining neural errors. Fixed phase8/update2000 main checkpoint with its immutable original phase8 TRAIN statistical guides. Compare gradients of the current objective and a smoothed exact VRMSE objective on six audited TRAIN trajectories: third trajectory at thermal cooling times0.03,0.32,3.16, and third Fisher/smoke/active records. Use first four frames, full30/26/10 horizon, no input noise; no optimizer or parameter update. Diagnostic seed20261209. Compute total gradient norm/cosine and, for thermal cases, per-channel gradient norms and mutual cosines. This is a training-objective diagnosis, not a serving-score or physical-operator comparison. VALID per-dataset metrics not evaluated; weights remain unchanged. + +Main checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`; study source SHA256 `45dab7fea0f7848cf870277ac043a3cec19bb656bec80501bdb7998762c0c26a`. Results pending in objective_gradient_v1/results.json. + +## E071a — objective gradient result + +No optimizer ran and all shared-network parameters were verified bitwise unchanged. The following are TRAIN neural-objective diagnostics, not package validation scores; the physical branch is intentionally absent. Same main checkpoint and fixed guide hashes as E071. + +| TRAIN case | Log-loss gradient norm | VRMSE gradient norm | Gradient cosine | +|---|---:|---:|---:| +| well/turbulent_radiative_layer_2D tcool=0.03 | 4.135567 | 4.754782 | 0.972133 | +| well/turbulent_radiative_layer_2D tcool=0.32 | 8.674965 | 10.694219 | 0.984043 | +| well/turbulent_radiative_layer_2D tcool=3.16 | 10.710047 | 12.229944 | 0.984117 | +| ape2d/fisher | 43.078826 | 65.607155 | 0.999756 | +| pdearena/navier_stokes_2d | 3.224614 | 3.841176 | 0.987786 | +| well/active_matter | 25.649524 | 34.562863 | 0.997298 | + +Thermal channel gradients mostly have small or positive mutual cosines; pressure dominates at low/mid cooling time, while the second velocity component dominates the high-cooling-time example. VRMSE increases density/first-velocity gradient magnitudes, but its total direction is strongly aligned with the existing loss on these examples. Fisher and active examples have larger gradient norms despite moderate mean loss; this motivates investigating gradient balance separately if capacity/latent-state trials plateau. Outcome: no objective change queued from this six-case diagnostic; current capacity and physical-state experiments remain the next main training runs. Peak allocated GPU memory30.30GiB for the diagnostic due to two consecutive retained graphs, still within the singleH100 budget alongside running work. No VALID evaluation, new checkpoint or publication. + +## E072 — physical reflection augmentation preparation + +Hypothesis: reflection symmetry can regularize shared mean/profile forecasts beyond periodic shifts and dropout. Continue E067/periodic6000 in matched control/reflection arms. All19 TRAIN datasets,544587 parameters, frozen-zero covariance projection,12000updates,batch512,LR0.00005 cosine, reconstructed-field VRMSE,50percent initial contexts, two CPU threads each. Both arms use dropout0.1 after three SiLUs and periodic profile-bin shifts. Sample seed20261210, dropoutseed20261211, reflectionseed20261212, shiftseed20261213. Only changed training factor: with probability0.5, reflect the first spatial axis and apply polar-vector, rank-two tensor and vorticity-pseudoscalar signs. Means/profile components change sign consistently; variances are invariant. Apply only to audited reflection-symmetric collocated PDEs with periodicX. KdV odd derivatives, forced Kolmogorov forcing and staggered-grid Arena are excluded; they remain part of shared training without this augmentation. + +Native-field versus profile-transform checks on all19 TRAIN dataset examples give normalized discrepancies below3e-16; applying the transformation twice is exact. Independent corpus update checks give relative discrepancies at most4e-7. The active-moment RHS symmetry holds to4.3e-16 in float64 for the resolved2/3 band. The raw32-grid downsampling had a2.37e-5 error in both float32 and float64, identifying a Nyquist-derivative ambiguity rather than roundoff; filtering resolved modes removes it. These diagnostic filters do not change training arrays or serving. Initial CPU JIT optimization failure is retained and resolved using the runtime’s existing optimized_execution(False) convention. All failed diagnostic logs and gated trainer startups are retained; no optimizer ran before the physics-check gate passed. No architecture or inference-transformation changes. Evaluate updates2000/6000/12000 through one pinned complete package, recording every dataset score and checkpoint hash. + +Parent shared statistics SHA256 `af0cfd6fa137519bbd329075b23c87c3332e79aba584b7de4027c0b613b2af9a`; inherited initialization seed20261031. No pretrained assets. VALID VRMSE pending. + +## E073 — reflection-average TRAIN diagnostic preparation + +Hypothesis: projecting the reflection-augmented statistics forecaster onto reflection-equivariant outputs may reduce residual asymmetry. Fix the E072 reflection/update12000 checkpoint before its VALID results; average original and inverse-reflected mean/profile predictions using the same branch weights. No new parameters, seed or training. Evaluate the additive-field VRMSE criterion on all72 thermal TRAIN trajectories, context0..3 and horizons1..30. This criterion includes unresolved target interaction variance and is not a complete surrogate score. If it improves, require a complete TRAIN rollout comparison before any VALID candidate; otherwise stop this line. No inference implementation or publication is part of this initial diagnostic. Checkpoint hash and outcome pending. + +## E072a — matched horizontal-reflection results + +Continue selected E067/periodic6000 shared mean/profile weights with the same dropout and periodic shifts in both arms; add horizontal reflection only in the experimental arm. All19 TRAIN datasets,544587 parameters, covariance projection frozen zero. Sample seed20261210, maskseed20261211, reflectionseed20261212, periodicshiftseed20261213,12000updates,batch512,LR0.00005 cosine, field-VRMSE objective,50percent initial contexts. Reflection probability0.5 for eligible collocated equations with periodicX; transform scalar/vector/tensor fields and spatial profiles consistently. Exclude KdV, forced Kolmogorov and staggered Arena. Same deterministic runtime and decoder blends. Field/profile and independent fixed-PDE symmetry checks passed before training. Nine changed VALID cases freshly scored plus72 audited unchanged physical case scores. + +| Arm | Update | Thermal VALID VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 2000 | 0.444315900 | 0.811981019 | `912a88895dda9a7d7f17f40eae2fc29eb600ae0f79bb87e1d9c2265c99bef4aa` | +| control | 6000 | 0.445454090 | 0.811888427 | `e10240f8bf97e528f6638be278ae056b6e4979d875bddf86e1e88a6ce1586d97` | +| control | 12000 | 0.444812734 | 0.811940572 | `8b93e822b99478d90fe325ce1735f6cfefe8296bc46e98b0df6850790463db94` | +| reflection | 2000 | 0.443882243 | 0.812016359 | `22a571ef85c56dd0afc88c58affb0a413e5b0faeffea004d69d487f8927d1e8a` | +| reflection | 6000 | 0.445341933 | 0.811897541 | `ee0ea35e8737073b61b0b770e6ca2ab349e0e0f22931e7406bf860f14ea9900f` | +| reflection | 12000 | 0.444827306 | 0.811939387 | `34be550fe47dc5a305604113f10a42ebd923119f55c01b7396d1882a0de6f1c9` | + +All candidates reuse the same complete physical outputs from the documented nonlatent parent; the following dataset metrics apply to every row: + +| Dataset | VALID VRMSE | +|---|---:| +| ape2d/burgers | 0.006412678 | +| ape2d/decay_turb | 0.000381422 | +| ape2d/diff | 0.000002603 | +| ape2d/fisher | 0.000003406 | +| ape2d/gs_alpha | 0.003094704 | +| ape2d/gs_beta | 0.004740266 | +| ape2d/gs_delta | 0.019249670 | +| ape2d/gs_epsilon | 0.001366170 | +| ape2d/gs_gamma | 0.049312925 | +| ape2d/gs_iota | 0.001784790 | +| ape2d/gs_kappa | 0.004854561 | +| ape2d/gs_theta | 0.004677970 | +| ape2d/kdv | 0.016728012 | +| ape2d/kolm_flow | 0.008925566 | +| ape2d/ks | 0.000039134 | +| ape2d/sh | 0.000003711 | +| pdearena/navier_stokes_2d | 0.000083487 | +| well/active_matter | 0.040600924 | + +Outcome: best complete-package candidate proceeds to comparison with the current selection; all negative checkpoints remain recorded. + +## E072 — shared statistics reflection augmentation +- Hypothesis / changed factor: Matched physical-reflection augmentation, selectedreflection update2000. Continue E067/periodic6000 shared mean/profile weights, all19 TRAIN datasets, sample seed20261210, 12000updates,batch512,LR0.00005 cosine, field-VRMSE and50percent initial contexts. Both arms use dropout0.1 after three SiLUs(maskseed20261211) and periodic profile translations(shiftseed20261213). Changed factor: horizontal reflection with probability0.5(seed20261212), reversing X-profile bins and polar/tensor component signs, for eligible reflection-symmetric collocated equations. Exclude KdV, forced Kolmogorov and staggered Arena. Same544587 parameters and deterministic runtime. All six candidate per-dataset scores/checkpoint hashes recorded in E072a; official9-case scores plus audited72-case reuse; neural/unknown and smoke/diffusion repeats precede any winning publication. +- Seed: 20261106. +- Checkpoint SHA256: `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. +- Runtime code SHA256: `8c28c764c772da156dc303635aee26bf2bbfc6d555ad2c846c69a8754f0c72c7`. +- Validation: fixed 81-case subset; reward **0.812016**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.443882 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## Full validation of reflection_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781320**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.443882 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/dropout_augmentation_v1/validation/periodic_006000. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E073a — reflection averaging rejected on TRAIN + +Fixed E072/reflection12000 checkpoint, SHA256 `34be550fe47dc5a305604113f10a42ebd923119f55c01b7396d1882a0de6f1c9`. On all72 audited thermal TRAIN trajectories, the reconstructed additive-field criterion worsens from0.5565570787423186 to0.558363237314754. Same weights, no new seed or parameters. Outcome: reject at the TRAIN screen; no full-rollout candidate, VALID scores or publication. Full per-case/channel results retained in reflection_statistics_v1/equivariance/results.json. + +## E074 — complete neural translation average with one physical branch + +Hypothesis: complete translated neural forecasts may retain E062’s thermal benefit while avoiding repeated physical computation or physical-output changes. Use the same four periodic-X shifts0,2/128,4/128,6/128 and shared neural/auxiliary weights for every case; average inverse-translated complete neural forecasts. Compute the physical branch once from the canonical supplied context and retain its returned prefix exactly after averaging. Identity duplicates share computation. No new parameters, seed or optimizer. This differs from E068, which averaged each autoregressive neural step, and E062, which recomputes and averages full physical forecasts. + +Assemble an unpublished portable package from the current selection, check canonical physical equality on TRAIN diffusion/smoke/active examples and unknown-schema behavior, then replay nine audited thermal TRAIN trajectories. Save all case/channel/step metrics. If promising, defer any validation or publication until E062/E064 decisions complete, avoiding composition with their running installers. VALID per-dataset VRMSE not yet evaluated; parent/checkpoint hashes will be pinned by the study. + +## E074a — canonical-physics neural translation average rejected + +Complete-package parity checks preserve the TRAIN diffusion, smoke and active physical predictions exactly; both constructed nonperiodic novel schemas are also unchanged, finite and correctly shaped. On nine full thermal TRAIN rollouts using the latest E072 guides, parentVRMSE0.3969341901511231 becomes0.39740124346328004. No new parameters or seed. Main checkpoint SHA256 `3bc98551d692db80aef859172eb31ef406e68b1b07e0d0e31d6a17ccbb2b9b67`; shared statistics SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Outcome: reject at the TRAIN screen, no VALID candidate or publication. The implementation remains an unpublished efficiency alternative; no running controllers or physical caches changed. E062’s already-running full-physical ensemble study continues because it has a distinct physical effect. + +## E075 — periodic shared-statistics projection diagnostic + +Hypothesis: the periodic-augmentation-trained statistical branch still depends on an arbitrary profile-bin origin. Average the same branch over all16 cyclic shifts of its periodicX context profile, undoing the shift of each predicted X profile before averaging. Global means, variances and Y profiles retain their original geometric meaning. No new weights, seed or tuned coefficient. Pin current selected shared-statistics weights and evaluate the additive-field criterion on all72 authorized thermal TRAIN trajectories, first four frames and horizons1..30. This is distinct from whole-neural and whole-physical translation ensembles: only the statistical forecasts change. If promising, require a complete TRAIN rollout comparison before any VALID candidate. No inference package or publication in this diagnostic. + +Pinned branch SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. VALID per-dataset scores not evaluated. + +## E038 — complete-rollout autoregressive training, update 3000, shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`. +- Runtime code SHA256: `ce7a2dd1166720a647fcb4df3ca2c263e4268a99d95afdb6a4ced5bf79c0be75`. +- Validation: fixed 81-case subset; reward **0.811955**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.444636 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 3000, no_shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.0; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`. +- Runtime code SHA256: `8c28c764c772da156dc303635aee26bf2bbfc6d555ad2c846c69a8754f0c72c7`. +- Validation: fixed 81-case subset; reward **0.812259**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.440919 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E038 — complete-rollout autoregressive training, update 3000, mean_profile_off +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=False; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`. +- Runtime code SHA256: `78e4e4f6e621bb62774c334b02d6c5cd2ba6f87b70fb017ecc097896fc112369`. +- Validation: fixed 81-case subset; reward **0.811775**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.446849 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E075a — periodic statistical projection rejected on TRAIN + +On all72 thermal TRAIN trajectories, the selected branch’s additive-field criterion worsens from0.5598388719889853 to0.5639645279281669 after averaging all16 cyclic X-profile shifts. Pinned mean checkpoint SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. No new parameters or seed. Outcome: reject at this TRAIN screen; no full surrogate rollout candidate, VALID results or publication. All per-case/channel results retained in periodic_statistics_v1/results.json. + +## E076 — main-network reflection sampler preparation + +Prepared a training-only whole-window version of the E072 reflection augmentation for a later shared-backbone experiment. Same audited PDE eligibility and scalar/vector/tensor signs. A separate reflection RNG preserves the original group, trajectory, start-frame and periodic-shift sample sequence exactly. Optional flag defaults off. Sampler test seed20261215, reflectionseed20261216:80 TRAIN windows across all four groups,31 reflected; disabled path, metadata/sample sequence, and whole-window transform/involution checks are exact. Every one of10115 source records passes TRAIN membership audit. No GPU trainer, new checkpoint, VALID candidate or publication was launched. + +Potential next experiment after the queued capacity and state-head trials: matched control and reflection continuations from the same selected parent. Preserve physical state-head supervision if that parent has a trained latent head. No launcher or fixed training budget has been committed; choose those after the current trials report. Implementation and checks are in reflection_main_v1. + +## Full validation of eval003000_no_shrink + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781563**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.440919 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/reflection_statistics_v1/validation/reflection_002000. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E077 — compiled-backbone throughput diagnostic + +Hypothesis: compiling the shared autoregressive backbone may reduce training time on the single H100. This is a no-optimizer, no-weight-update diagnostic on one audited TRAIN trajectory (thermal tcool0.32, trajectory2), batch1 and two autoregressive steps, BF16 autocast. Seed20261217. Pinned main checkpoint SHA256 `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`, guide SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Compare eager and default fullgraph torch.compile outputs, gradients, and interleaved CUDA-event timings. All model parameters must remain unchanged. No VALID results or publication candidate; one shape/two steps cannot establish suitability for training. + +Initial attempt failed before compiled timing due to an invalid cross-device cache rename in Triton. Failure log retained as compile_main_v1/initial_cross_device_failure.log. Retried with both Inductor and Triton caches on local /tmp storage. This changes diagnostic cache placement only; ongoing training and serving remain eager. + +## E077a — full compilation unsupported; mixed compilation diagnostic + +The second cache placement attempt also failed: this environment rejects cross-directory file rename even under /tmp. A development-only Triton cache class now writes a unique temporary file adjacent to its destination, preserving atomic same-directory replacement. Cache bytes/group checks passed. This does not modify installed libraries or the serving package. Both filesystem failure logs are retained. + +With the cache fixed, full-graph compilation fails during backward with a ComplexFloat-to-Float stride error in the global Fourier mixer. No optimizer or new weights were created. A third computational variant compiles the surrounding network while explicitly retaining the global Fourier mixer in eager mode (fullgraph=False); all other protocol choices, TRAIN example and selected checkpoints remain fixed. Numerical and speed results pending in compile_main_v1/mixed. Validation VRMSE: not evaluated; no inference/training adoption. + +## Paired analysis of E038 updates 2,000 and 3,000 + +Using identical E072 auxiliary weights and the no-shrink decoder, the nine thermal VALID trajectories improve by a mean VRMSE -0.0029633274994067538; five of nine improve. The gain concentrates in middle rollout steps; the last six steps regress on average. Descriptive paired bootstrap seed20261218,20,000 resamples gives a 95percent interval [-0.009792993346356565, 0.0033186238294739545] for the mean difference. With one trajectory per cooling regime, this is empirical variation within the observed pool, not established uncertainty for unseen regimes. No selection rule or model change. Official reports and hashes, per-channel/step differences and standalone plots are retained in phase8/analysis003000. E066 equal checkpoint averaging was already declared before update3,000. + +## E078 — non-Gaussian statistic feature screen + +Hypothesis: per-channel skewness and kurtosis reveal thermal mixing information absent from separate means, variances and axis profiles. CPU-only diagnostic on all72 audited thermal TRAIN trajectories, contextframes0..3 and mean/profile targets4..33. Pinned current mean checkpoint SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Compare matched linear residual probes using existing normalized statistical features versus those features plus clipped standardized third/fourth spatial moments. Fixed ridge penalty100, eight stratified folds (one trajectory per cooling regime per fold), seed20261219; feature standardization uses fold-training rows only. No hyperparameter search. + +The fixed parent already trained on these trajectories; only probe corrections are cross-fitted. This is a feature screen, not an estimate of validation performance or a deployable specialist. No full surrogate training, new shared checkpoint, VALID evaluation or publication. A positive result would require training one shared branch across all19 datasets and full-package validation. Protocol/source and results retained in moment_features_v1. + +## E078a — moment feature screen rejected + +The parent additive-field TRAIN criterion is0.5598388813998767. Cross-fitted residual probes worsen it to0.5697928408169022 with existing features and0.5697874929134444 with added skewness/kurtosis. Normalized mean-only error is0.0930144262474265 for the parent,0.11150019876098338 for control and0.11139854203664112 for moments. The added moments have negligible benefit over control and neither probe improves the parent. Reject this feature screen; no full shared-model training, new checkpoint, VALID evaluation or inference integration. All72 records pass TRAIN membership; parent parameters remain exactly unchanged. Seed20261219, fixed penalty100, parent mean SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Full per-record/channel values and cross-fitted predictions retained in moment_features_v1/results.json and arrays.npz. + +## E077b — mixed compilation scheduler failure + +Excluding the Fourier mixer from compilation passes the earlier complex-operation boundary but fails in the compiler fusion scheduler with AssertionError4 versus512. Failure log retained as compile_main_v1/mixed.log. A conservative diagnostic disables the compiler’s mixed-order reduction fusion option while retaining the same eager Fourier boundary. This changes compiler optimization only, with identical inputs, checkpoints and no optimizer. Results pending in compile_main_v1/conservative. No change to ongoing training or serving. + +## E077c — conservative compilation throughput result + +With the global Fourier mixer eager and mixed-order reduction fusion disabled, the no-update two-step thermal TRAIN test completes. Eager loss0.0855544135 versus compiled0.0855036378; normalized output RMS difference0.00456819, relative gradient difference0.03222236, all finite, every parameter unchanged. Compilation99.23seconds; six interleaved timing pairs give median1.6251times speedup (roughly75–80ms eager versus46–49ms compiled). Peak allocated1.16GiB. This is one width48 shape/two steps on a shared H100, not sufficient for adoption. All failures and results retained. + +## E079 — full-rollout width96 compilation screen + +Extend E077’s conservative compiler setup to the queued width96 architecture and all19 TRAIN datasets. Pin current main SHA256 `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03` and guide `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`; use the existing function-preserving channel duplication with seed20261110/noise0.01. No optimizer steps. One third authorized TRAIN record per dataset, initial four frames, batch1, full26/10/30 corpus/Arena/Well horizons, BF16, input noise0.01 seed20261220. Compare full predictions, gradients, losses and two interleaved timing pairs. Predeclare numerical screen: relative loss difference<1percent, normalized output RMS<0.05, relative gradient norm difference<0.1, gradient cosine>0.99, finite outputs/gradients and unchanged parameters. Record all datasets regardless of screen outcome. Training/serving remain unchanged while this runs; any later adoption is a separate logged execution change and requires actual-parent checks. No VALID scores or publication candidate. + +## E080 — preserve eager low-precision casts in compilation + +E079’s first full30-step thermal case has1.7237times throughput, relative loss difference0.0017755, normalized output RMS0.0158855, gradient cosine0.9962419, but relative gradient difference0.1027377 exceeds its predeclared0.1 limit. This fails that numerical-equivalence screen; it is not a validation-performance result. Leave E079 running for all19 outcomes. + +Hypothesis: Inductor’s documented elimination of intermediate BF16 downcast/upcast pairs accounts for much of the accumulated difference. Repeat the same full19-dataset width96 no-update suite with emulate_precision_casts=True, which preserves those eager rounding points and associated floating-point fusion behavior. All seeds, inputs, checkpoint values, other compiler settings and numerical thresholds remain fixed. Assert widened parameter tensors exactly match E079. Queue after E079 completes to stay within one H100; no optimizer, training adoption, VALID score or package change. Both source versions and cache helper hashes retained. + +## E081 — neural inference precision TRAIN screen + +Hypothesis: accumulated BF16 rounding contributes to neural rollout error. Compare the same complete selected surrogate with neural autocast enabled versus disabled (float32 inputs and weights under existing backend defaults). No parameter, guide, decoder or physical-operator changes; no new seed. Pinned main SHA256 `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`, guide `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Require exact unchanged outputs for the small TRAIN diffusion/Arena/active physical cases, finite outputs on thermal and two novel schemas, and exact repeated candidate outputs. Then compare the same nine audited thermal TRAIN continuations across all30 steps. Reject without VALID evaluation if their mean does not improve. No compiler inference, test-time fitting or extra model member. Validation VRMSE per dataset: pending/not evaluated. Source and checks retained in neural_precision_v1. + +## E079a — full-horizon compiler screen rejects adoption + +All19 TRAIN cases complete, outputs/gradients finite, every parameter unchanged. Median throughput ratio2.1216438618488094. Seven cases fail the predeclared numerical screen; Gray–Scott delta has relative gradient difference0.7038 and cosine0.8733. No optimizer, VALID score or training/serving adoption. Widened diagnostic checkpoint SHA256 `d6256e2b679c7237aa5751cc1ec8c18f278fb8e87b50b8dba225e0e4ff7064c0`. Seeds and source hashes remain in the E079 protocol. + +| Dataset | Normalized output RMS difference | Relative gradient difference | Relative loss difference | Speedup | Screen | +|---|---:|---:|---:|---:|---| +| well/turbulent_radiative_layer_2D | 0.015885 | 0.102738 | 0.001776 | 1.724 | fail | +| ape2d/diff | 0.003026 | 0.138812 | 0.028866 | 2.223 | fail | +| ape2d/burgers | 0.004920 | 0.053860 | 0.003987 | 1.914 | pass | +| pdearena/navier_stokes_2d | 0.002232 | 0.016548 | 0.000549 | 1.533 | pass | +| well/active_matter | 0.221067 | 0.054861 | 0.003199 | 1.985 | fail | +| ape2d/decay_turb | 0.002556 | 0.011180 | 0.004082 | 1.988 | pass | +| ape2d/fisher | 0.011552 | 0.083745 | 0.035292 | 2.196 | fail | +| ape2d/gs_alpha | 0.011394 | 0.037654 | 0.000057 | 2.122 | pass | +| ape2d/gs_beta | 0.008843 | 0.013001 | 0.001582 | 2.009 | pass | +| ape2d/gs_delta | 0.054145 | 0.703820 | 0.002044 | 2.091 | fail | +| ape2d/gs_epsilon | 0.005628 | 0.035144 | 0.000983 | 2.137 | pass | +| ape2d/gs_gamma | 0.004126 | 0.014093 | 0.000090 | 2.138 | pass | +| ape2d/gs_iota | 0.025078 | 0.046624 | 0.004819 | 2.119 | pass | +| ape2d/gs_kappa | 0.017761 | 0.103080 | 0.001799 | 2.221 | fail | +| ape2d/gs_theta | 0.039929 | 0.166120 | 0.004075 | 2.107 | fail | +| ape2d/kdv | 0.004689 | 0.085845 | 0.000537 | 2.157 | pass | +| ape2d/kolm_flow | 0.008747 | 0.034871 | 0.000087 | 2.175 | pass | +| ape2d/ks | 0.007982 | 0.032241 | 0.001332 | 2.219 | pass | +| ape2d/sh | 0.048194 | 0.018184 | 0.001603 | 2.203 | pass | + +Diagnostic detail: float32 reduction of the long gradient vectors yielded a cosine slightly above1 for one case. This does not affect the seven failures, which also violate other limits. E080 computes gradient norms/dots in CPU float64 after backward to remove this measurement roundoff; model arithmetic is unaffected. E079’s original measurements/source are retained. The compiler’s own graph compilation/autotuning preserves RNG state; E080 uses the same input-noise seed and record sequence. + +## E081 — shared neural inference precision +- Hypothesis / changed factor: Disable neural BF16 autocast for the same shared weights and original four-frame inputs, retaining float32 inputs/weights under existing backend defaults. No new parameters or seed; all physical and auxiliary arithmetic unchanged. Nine audited thermal TRAIN rollouts improve0.3858202574 to0.3850654613. Six TRAIN/novel schema smoke cases are finite and exactly repeatable; diffusion, Arena and active physical outputs remain exactly unchanged. Freshly score all9 thermal VALID cases; reuse72 audited complete physical predictions for quick81. If selected, derive full585 from the same complete576 physical prefixes, with all inputs/metadata and dependencies audited. No compiler inference, specialist checkpoint, pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `7556bb5478ee29099305209db7baebc83fc388b44a726f5a769f3152b1c54a03`. +- Runtime code SHA256: `37fcf15c33b676f512d9ac6a0b051787d5ca4a3e92f4f30cc065acf409af0516`. +- Validation: fixed 81-case subset; reward **0.812211**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.441505 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E081a — inference precision improvement does not transfer to VALID + +The nine thermal TRAIN rollouts improve0.3858202573790942 to0.38506546133427394 with autocast disabled. All six TRAIN/novel schema smoke outputs are finite and exactly repeatable; complete diffusion, Arena and active physical predictions are bitwise unchanged, and all three learned checkpoints are identical. However, all nine thermal VALID trajectories worsen0.4409189158885829 to0.441504742943192. Quick81 reward0.8122107255146385 versus selected0.8122587790567208, zero failed cases. The preceding E081 record contains all per-dataset VALID scores and checkpoint/runtime hashes. Reject; no inference-template, training or selected-package precision change. No full585 derivation needed for this negative neural-only candidate. + +Compatibility preparation generalized the queued latent-head installer to recognize either the original autocast expression or the optional model flag. Its source-normalization check matches each parent exactly after removing only the latent-head plumbing; existing behavior is unchanged. This optional flag remains absent from the selected package. No latent training code or physical-cache dependency rule changed. + +## E080a — rounding preservation helps but does not pass globally + +All19 no-update TRAIN cases finish, every output/gradient finite, parameters unchanged. Median speedup2.061026698990928. Sixteen pass the unchanged numerical screen. Active fails normalized output RMS; Gray–Scott delta and theta fail gradient limits. Global compiler adoption is rejected. Thermal relative gradient difference falls from0.10274 to0.05615, and diffusion from0.13881 to0.00296. CPU float64 gradient reductions remove the earlier cosine roundoff. The unavailable CUDA toolkit libdevice caused a documented fallback to bundled Triton libdevice; warnings retained. + +| Dataset | Output RMS difference | Relative gradient difference | Relative loss difference | Speedup | Screen | +|---|---:|---:|---:|---:|---| +| well/turbulent_radiative_layer_2D | 0.009919 | 0.056153 | 0.000303 | 1.929 | pass | +| ape2d/diff | 0.000848 | 0.002965 | 0.000774 | 2.130 | pass | +| ape2d/burgers | 0.001804 | 0.005764 | 0.000360 | 2.115 | pass | +| pdearena/navier_stokes_2d | 0.001830 | 0.029056 | 0.000454 | 1.666 | pass | +| well/active_matter | 0.117604 | 0.006451 | 0.000787 | 2.061 | fail | +| ape2d/decay_turb | 0.001009 | 0.004750 | 0.000505 | 2.023 | pass | +| ape2d/fisher | 0.002030 | 0.005293 | 0.001402 | 2.000 | pass | +| ape2d/gs_alpha | 0.010585 | 0.028301 | 0.000003 | 2.047 | pass | +| ape2d/gs_beta | 0.004780 | 0.005609 | 0.000096 | 1.882 | pass | +| ape2d/gs_delta | 0.036557 | 0.333432 | 0.000618 | 1.940 | fail | +| ape2d/gs_epsilon | 0.004138 | 0.013351 | 0.000090 | 2.039 | pass | +| ape2d/gs_gamma | 0.002936 | 0.006737 | 0.000069 | 2.103 | pass | +| ape2d/gs_iota | 0.024890 | 0.036554 | 0.000194 | 2.075 | pass | +| ape2d/gs_kappa | 0.012319 | 0.042181 | 0.000277 | 2.244 | pass | +| ape2d/gs_theta | 0.048089 | 0.349601 | 0.002550 | 2.223 | fail | +| ape2d/kdv | 0.001916 | 0.004853 | 0.000018 | 1.999 | pass | +| ape2d/kolm_flow | 0.002239 | 0.007663 | 0.000143 | 2.095 | pass | +| ape2d/ks | 0.004062 | 0.008440 | 0.000067 | 2.141 | pass | +| ape2d/sh | 0.028627 | 0.010698 | 0.001178 | 2.179 | pass | + +## E082 — geometry-restricted compilation screen + +Hypothesis: the expensive128x384 rectangular geometry can be accelerated while unsupported tensor geometries retain eager execution of the same shared parameter set. No dataset-name routing or separate checkpoint. This is a narrower execution scope, not a relaxation of E079/E080 thresholds. Test all nine pre-existing audited thermal TRAIN diagnostic trajectories, one per cooling regime, batch1/full30 steps, width96, the same3k main and E072 guide, BF16 with eager Fourier mixer and preserved precision casts. Use a dedicated CUDA input-noise generator seed20261220 and hash every noisy context. All nine must satisfy the original output/loss/gradient thresholds before considering an actual-parent preflight. All other geometries would remain eager; inference remains unchanged. No optimizer, VALID score, new trained model or current training change. Source/protocol/weights/hashes under compile_main_v1/thermal_suite. + +## E082a — rectangular compiler scope also rejected + +All nine full30-step thermal TRAIN numerical screens complete, finite outputs/gradients and unchanged parameters; median speedup1.6945201049195608. Seven pass. Cooling0.06 has relative gradient difference0.21023584;0.32 has0.12336355, both above the unchanged0.1 limit. Reject this restricted execution scope. No compiler wrapper, wider-training flag, actual-parent preflight or serving change was installed. All queued trainers remain eager. Source, context hashes and complete per-case metrics retained in compile_main_v1/thermal_suite. + +## E083 — offline TRAIN hidden-state fitting diagnostic + +Hypothesis: the planned boundedF6 correction can improve active-matter continuation when given a useful state estimate. Before judging whether the shared head can learn that estimate, fit its512 coefficient values directly offline for one audited TRAIN trajectory (zeta9,alpha−3,trajectory2). Initialize at zero;20 Adam updates atLR0.1, seed20261222, using only the first4 future TRAIN frames in the same differentiable physical loss as the queued head. Keep scale0.01,alignment cubic factor and16x16 grid fixed. Evaluate the fixed update20 on a full30-step TRAIN physical rollout with the original three-node quadrature, and report fitsteps1–4 and unusedsteps5–30 separately. No checkpoint/hyperparameter search, no neural weight updates. + +These coefficients use TRAIN targets and are diagnostic only: they cannot be packaged or used for inference cases. This is a local fitting study on one trajectory, not a proved optimum or evidence that four observed frames predict the correction. No VALID/held-out data, synthetic training data, deployment candidate or benchmark reward. Encoder/split/membership and the selected cache-record hash are audited before reading labels; protocol, sources, hashes and coefficients stay under latent_oracle_v1. + +## E038 — complete-rollout autoregressive training, update 4000, shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `e98efa15fa411007efbc795a658a28a5074f1f2b818913d00c43a42299afebb5`. +- Runtime code SHA256: `ce7a2dd1166720a647fcb4df3ca2c263e4268a99d95afdb6a4ced5bf79c0be75`. +- Validation: fixed 81-case subset; reward **0.811685**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.447969 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E083a — offline hidden-state fit result + +Fixed update20 reduces the four-step fitting loss from 0.000517013482749 to 0.000413423113059. Both native30-step physical rollouts complete. The target-fitted state regresses full30 TRAIN VRMSE: 0.00892846228338 to 0.0142502669091. Fit steps1–4: 0.000504987527507 to 0.000399531632118; unused steps5–30: 0.0102243814766 to 0.0163811492594 (delta +0.00615676778282). The fit uses the center state; these forecasts retain the original three-node quadrature. + +Only512 direct coefficients were optimized, seed20261222,20fixed Adam updates atLR0.1. No shared surrogate or learned state-head parameters changed. Diagnostic coefficient SHA256 `bbdfe8fe89ccb014b019e6778ad6b462fa5f7e0f0d39a34f0447e938d19c0d1a`. This one-trajectory local fit is not a proved optimum and does not show that four observed frames predict the correction. No VALID score for any dataset, no held-out access, no deployment candidate or publication of coefficients. Full results, per-step/channel scores, sources and figures are in latent_oracle_v1. The queued shared-head protocol is unchanged. + +## E038 — complete-rollout autoregressive training, update 4000, no_shrink +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.0; mean/profile output corrections enabled=True; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `e98efa15fa411007efbc795a658a28a5074f1f2b818913d00c43a42299afebb5`. +- Runtime code SHA256: `8c28c764c772da156dc303635aee26bf2bbfc6d555ad2c846c69a8754f0c72c7`. +- Validation: fixed 81-case subset; reward **0.811915**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.445123 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E038 — complete-rollout autoregressive training, update 4000, mean_profile_off +- Hypothesis / changed factor: Train the selected shared architecture on full rollout windows, capped at26 corpus,30 Well and10 Arena steps. Three microbatches of two accumulate to effective batch6, limiting memory on the single H100. Learning rate0.00003 cosine, native thermal grids, noise0.01, group probabilities0.05/0.075/0.075/0.8; sampling seed20261106. Exact selected parent, inherited seeds and fixed auxiliary hashes are in phase8/lineage.json. If the parent has learned guidance, its fixed statistics use only the original four context frames and metadata. Profile shrinkage 0.75; mean/profile output corrections enabled=False; the selected coarse interaction correction is retained. Rerun nine neural validation cases and reuse72 complete physical outputs documented in physical_cache_source.json. No pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `e98efa15fa411007efbc795a658a28a5074f1f2b818913d00c43a42299afebb5`. +- Runtime code SHA256: `78e4e4f6e621bb62774c334b02d6c5cd2ba6f87b70fb017ecc097896fc112369`. +- Validation: fixed 81-case subset; reward **0.811251**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.453363 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E066 — late full-rollout parameter average +- Hypothesis / changed factor: Equal parameter average of phase8 updates2000/3000/4000, predeclared before3000/4000 exist. All three checkpoints share full30-step autoregressive training on all19 datasets, sample seed20261106, initialization inherited from this session’s scratch-trained lineage. Float64 accumulation restored to original dtypes. No new training, parameters, random seed or inference ensemble. Preserve the selected shared auxiliary guides, decoder settings and physical operators at evaluation time. Source hashes and equal coefficients in lineage.json and the manifest. Replay and officially score all9 thermal VALID cases; reuse72 complete physical cases after dependency/coverage checks, aggregate using the exact official code, and require neural/unknown repeats on a win. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `8c28c764c772da156dc303635aee26bf2bbfc6d555ad2c846c69a8754f0c72c7`. +- Validation: fixed 81-case subset; reward **0.812311**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.440282 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: improvement; published atomically. + +## E084 — short versus long hidden-state gradient diagnosis + +E083 improves its four fitted TRAIN steps but regresses its full30 rollout. On the same audited TRAIN9/−3 trajectory2, compute exact autograd derivatives at zero2x16x16 correction for the original stabilized VRMSE atsteps1–4 and5–30. Same native256/M32/6Stokes dynamics,scale0.01,center state,seed20261223; no optimizer or shared-network change. Check exactforecastprefix, finite gradients and unchanged coefficients. Report float64 gradient cosine/norms and local derivatives along the already fixed E083update20. This single-trajectory local diagnosis is not a VALID score, finite-update guarantee or shared-head learnability test. No deployment candidate; queuedheadprotocol unchanged. Sources/protocol pinned in latent_horizon_v1. + +## Full validation of late_average + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781615**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.440282 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/phase7/late_average. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E047 wider main-model training launch + +Grow the selected shared main network from width48 to96, preserving the initial function within documented float32 roundoff. Initialize only from session-trained weights, duplicate channels and split outgoing weights, with zero-sum symmetry-breaking noise0.01 seed20261110. Main parameters3066627 to9526707. Six microbatches of one accumulate effectivebatch6, full30-step unroll with26/10 corpus/Arena caps, learning rate0.00003 cosine, noise0.01, group probabilities0.05/0.075/0.075/0.8, sampling seed20261119,2000updates. Start only after phase8 trainer/workers/watcher exit. Freeze selected auxiliary guides; exact parent and hashes in phase9/lineage.json. Validation pending. + +## E084a — early and late state-gradient result + +The two objectives have nearly orthogonal gradients at zero correction: cosine0.0044190691758825215; late/early gradient norm ratio439.77179892533485. Thus this example does not show opposed local gradients. Along the fixed E083update20, the early and late local directional derivatives are both negative (-0.0002630996148348695,-0.026207977598443843), despite the finite-update long forecast regression. A local derivative does not predict such a finite correction through these sensitive dynamics; the differentiable center-state objective also differs from three-node serving. Exactprefix and unchangedcoefficients checks pass, gradientsfinite; native30backwardpeak25.916GiB,~209.6s onsharedGPU. GradientSHA256 `d4447147baf8c3887331525bce2bcc9892d433aadbea8af781de9e7d54c54b96`. No newweights/VALIDmetrics/servingchange; shared-headprotocolremainsunchanged. + +## E085 — initial versus later TRAIN windows + +Hypothesis: the current0.25 initial-window sampling probability may misallocate updates when all evaluation contexts start at frame0. Pin selected late-average mainSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61` and complete sharedpackage, with no training. Replay the same nine thermal TRAIN trajectories, one per cooling regime, from start0,midpoint33,and lastpossible67 with30-step horizons. Fresh TRAIN reads must exactly equal nativecached frames; compute the official channel/stepVRMSE formula, clamp10, and summarize positions/channels/horizons. Fixedservingseed0, nooptimizer,newparameters,VALIDorheldoutaccess. This descriptive comparison does not by itself identify an optimal sampling distribution. Results pending in window_position_v1. + +## E085a — later TRAIN windows are harder + +The fixed selected complete model scores initial/middle/late TRAIN windowVRMSE 0.383153984741939/0.5268689325498044/0.5673285633357836. Both later windows are worse than initial in all nine cooling regimes. Everychannelmean error increases; full per-case/channel/step and targetvariance diagnostics retained. Fresh reads equal nativecache; weights/runtime unchanged. This does not show that later training is harmful or identify an optimal sampling fraction, but motivates a matched initial-context sampling trial. No VALID scores, optimizer or deployment change. + +## E086 — shared-statistics initial-context sampling + +Continue the selected E072reflection2000 mean/profile weights in matched CPUarms: initialcontext probability0.5control versus0.9experimental. All19TRAINdatasets,544587parameters,6000updates,batch512,LR0.00005cosine. Both retain fieldVRMSE, dropout0.1, periodicprofiletranslations, eligiblehorizontalreflection0.5 and frozenzero covarianceprojection. Seeds sampling20261224/dropout20261225/periodic20261226/reflection20261227. Inherited scratchseed20261031; parentmeanSHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Record and random-number draws are matched; the threshold changes starts and can change actualhorizons near trajectoryends. Save1000/3000/6000 and evaluate allsix complete sharedpackages with unchanged decoder coefficients; no separately trained dataset model. No synthetic/pretrained assets or inference-time fitting. Existing model/reflection checks inherited byte-for-byte; matched-mask/rate sanitycheck passes beforelaunch. Full protocol/source hashes in initial_statistics_v1. Validation pending. + +## E085b — main-network initial-context sampler preparation + +Prepared an independent initial-window override for a future matched shared-backbone trial. Probability2/3 raises the explicit initialpolicy0.25→0.75 while preserving the original group, trajectory, baseline-start and periodic-shift draw sequence exactly; defaultdisabled path is bitwise identical. Seeds sampling20261228/override20261229. On80 real TRAIN windows acrossall4groups, 20 starts change; initial counts [48, 68]. Every input/target window equals the corresponding source slice and periodic shift, all10115records pass TRAINmembership audit. Sourcehashes andchecks in initial_main_v1. No trainer, checkpoint, VALID candidate, inferencechange or newtrainingbudget launched. Any future continuation must retain selected learned-state supervision if present. + +## E086a — matched initial-context sampling results + +Same selected E072reflection2000 auxiliary parent; all19 TRAIN datasets,6000 updates, batch512,LR0.00005 cosine,544587 parameters, fieldVRMSE, dropout0.1, periodic shifts and eligible reflection0.5 in both arms. Initialcontext probability0.5control versus0.9experimental. Sampling/dropout/periodic/reflection seeds20261224/25/26/27; all initialization descends from session scratchseed20261031. No runtime or decoder change. All changed nine thermalVALID cases freshly scored,72 complete physical cases reused after audit. + +| Arm | Update | Thermal VALID VRMSE | Quick81 reward | Training checkpoint SHA256 | +|---|---:|---:|---:|---| +| control | 1000 | 0.441333924574 | 0.812224730617 | `e9ff2ada10e23fb88676116911caa31f67ac4b9d3ed3c77b424562061c12bfff` | +| control | 3000 | 0.440927499736 | 0.81225807449 | `812686f8fe10216a5be371ce873cbfd8b2d9fb899da002cda6bb8f79a41a29cd` | +| control | 6000 | 0.440568632754 | 0.812287542186 | `9ee5bfc4d2ac7614d9c583e68f85a77334069381a1a671fdb2713b6b1b558ed8` | +| initial | 1000 | 0.441076006654 | 0.812245887136 | `132fa6ba8ac417fb72d0698711b0610a47c046aecdefd45fec7aca090450c549` | +| initial | 3000 | 0.441032874694 | 0.812249426376 | `0aadf69afa3921665de281060066478689123b718d07a58bbc791e0ed12d57b0` | +| initial | 6000 | 0.441308067539 | 0.812226851063 | `47e9174f904ae06d491b9496104bb4dc17a80ea9c89fa64bddc3111152baa876` | + +Unchanged dataset metrics for every candidate: + +| Dataset | VALID VRMSE | +|---|---:| +| ape2d/burgers | 0.00641267771349 | +| ape2d/decay_turb | 0.000381421944384 | +| ape2d/diff | 2.60304291901e-06 | +| ape2d/fisher | 3.40589902092e-06 | +| ape2d/gs_alpha | 0.00309470353535 | +| ape2d/gs_beta | 0.0047402657863 | +| ape2d/gs_delta | 0.0192496697308 | +| ape2d/gs_epsilon | 0.00136617003054 | +| ape2d/gs_gamma | 0.0493129245648 | +| ape2d/gs_iota | 0.00178478978776 | +| ape2d/gs_kappa | 0.00485456077448 | +| ape2d/gs_theta | 0.00467797007751 | +| ape2d/kdv | 0.0167280115129 | +| ape2d/kolm_flow | 0.00892556562073 | +| ape2d/ks | 3.91337424002e-05 | +| ape2d/sh | 3.71134325297e-06 | +| pdearena/navier_stokes_2d | 8.34867076885e-05 | +| well/active_matter | 0.0406009242797 | + +Compare the best complete-package candidate with the current selection; retain every negative result. + +## Full validation of control_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781592**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087454 | +| well/turbulent_radiative_layer_2D | 0.440569 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/universal_v1/phase8/late_average. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E086 — initial-context statistical training +- Hypothesis / changed factor: Matched shared-statistics initial-context sampling, selectedcontrol update6000. Continue selected E072reflection2000 weights across all19 TRAIN datasets, 6000updates,batch512,LR0.00005 cosine; sample/dropout/periodic/reflection seeds20261224/25/26/27. Changed factorinitial probability0.5 versus0.9; both arms retain dropout0.1, periodic shifts, eligible reflection0.5 and reconstructed-field VRMSE. Same544587 parameters, frozenzero covariance projection and deterministic inference on every case. Full per-dataset results and hashes for allsix candidates are in E086a. Nine fresh officialVALID scores and audited unchanged physical outputs; full585 report and neural/unknown/smoke/diffusion repeat checks precede publication. No pretrained or synthetic assets, no inference fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `73bd86b52f8a18af91dc0a0e2752c7076c9bfae1bb414a471e5ff65d2f0a53c0`. +- Validation: fixed 81-case subset; reward **0.812288**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000039 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040601 | +| well/turbulent_radiative_layer_2D | 0.440569 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E086b — initial-context statistical weighting rejected + +Allsix complete-package candidates regress relative to selected thermalVRMSE0.4402820249420574. The best is control6000 at0.4405686327535333, full585reward0.7815915571085751,zero failures. Initial90percent improves its matchedcontrol at1000 only, and loses at3000/6000; finalcontrol0.440568632754 versusinitial0.441308067539. Reject the increased initial-context weighting for this auxiliary branch. No inference-template, running-training guide or selectedweight change. This does not decide the separate, untrained main-network sampling hypothesis. All per-dataset scores/checkpoint hashes and negative fullreport remain retained. + +## E070a — native CUDA precision preflight + +On one audited TRAIN context at native256/M32, two physical steps with the original three-node quadrature, the new float32 path exactly matches the pinned original physical operator onGPU. Float64 captures its CUDAgraph, returns finite float64 forecasts fromcomplex128 state and differs fromfloat32 by maximumabsolute4.2149330314877886e-07. All three complete2/2steps. No targets were read and no optimizer ran; this is implementation evidence, not an accuracy result. Both precision modes generate the same float32 white quadrature noise from the existing fixedseed, then filter/integrate at their selected precision. No kinetic-source modification. The six-regime/full30 TRAIN diagnostic remains queued afterE062/E064; original E067 physical parent/configuration stays pinned. Nativechecks/audit in precision_active_v1/cuda_checks.json. + +## E087 — conservative density transport feasibility + +Hypothesis: use the continuity equation to improve the shared thermal forecast density. Independent finite-volume MC reconstruction/HeunRK2,CFL0.4,native128x384,supplieddomainextent,periodicX/openYreplicatedghosts,linear velocity interpolation. Compare selected modeldensity against transport driven by its forecast velocities and an oracle driven by knownTRAINfuture velocities on the same nine audited initial TRAIN windows fromE085. The oracle is diagnostic only and must never be packaged or evaluated onVALID/heldout cases. No blend/grid/CFLsearch, no optimizer/newlearnedparameters; parentmainSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Report density and full4channelcompositeVRMSE. Manufactured periodic advection errors32/64/128 are0.000354954/0.000101952/0.000030071; periodicmass preservedto1e-13, constantfield exact withopenboundaries, positivity and CUDAgraph/eager/repeat exactness pass (checkseed20261230). These manufactured checks are not synthetic trainingdata. Source/protocol/hashes in density_transport_v1; no VALID candidate or runtime change. + +## E087a — native continuity replacement rejected + +Allnine TRAIN pilots complete. Original densityVRMSE0.14801000442843196 versusforecast-velocity transport0.4597758131612161 andTRAINoraclevelocity transport0.2766399596436549. Fourchannelcomposite0.383153984741939→0.46109543692513505 (oracle0.41531147354574477). Reject direct replacement beforeVALID. No pretrainedassets,newweights or servingchanges; oraclepredictions remain TRAIN-only artifacts. The oracle itself is worse than the learned density, so velocity forecast error alone does not explain failure. + +A follow-up convention/conservation diagnosis computes centered continuity balances on the same TRAINframes0–33, with fixed4x4/8x8 pooling, identity versus swapped velocity assignments and diagnostic least-squares flux coefficients. Also inspect globalmass change versus open-boundary massflux. No fitted coefficient will be deployed from this diagnostic, and coarse temporal sampling can bias it. Protocol/sourcehash in balance_protocol.json; results pending. + +## E087b — TRAIN continuity convention diagnosis + +Fixed identity velocity assignment has relativebalanceRMS residual0.86244 at4x4 pooling and0.48944 at8x8, versus swappedcomponents3.26722/3.00647. Thus a component swap is not supported. Fitted identity coefficients change from[0.62582,0.52606] at4x4 to[0.95757,0.83357] at8x8; the fitted8x8 residual remains0.47774. This scale dependence does not identify a uniform time correction. Globalmass/boundaryflux coefficient varies0.17–0.96 across coolingregimes, with weak late-cooling masschange signals. Keep the suppliedconventions and reject the current directdensitytransport candidate. Temporal undersampling, unresolvedflux correlations and spatialdiscretization remain plausible causes; this diagnostic cannot separate them. No coefficient, optimizer, VALIDscore, checkpoint or servingchange. Fullcase metrics and exactmethod in balance_results.json. + +## E066a — paired average description + +The selected equal2k/3k/4k average improves six of nine thermalVALID cases relative to update3000; overallVRMSE0.440918915889→0.440282024942. Its final6step mean improves0.00658671154 while middlehorizons regress. All four comparisons have byte-identical configuration, meanweights and coarseweights; reporthashes and standalonePNG/SVG in full_average_v1/paired_analysis.json andpaired_average. This is descriptive on nineVALID trajectories, with no fitting or alteredselectionrule. + +## E088 — wider checkpoint average predeclaration + +Predeclare one equal parameter average of phase9 updates1000/2000 before either sourcecheckpoint exists. Float64 accumulation restored tooriginaldtypes; samewidth96/all19TRAIN lineage andseed20261119, no newparameters/training/seed/inferenceensemble. Compare the same three already-established decoder settings as individualphase9 checkpoints: shrink0.75,no_shrink0,and mean/profile correctionsdisabled. Lockedphase9inference template/completephysicalcache; ninefreshthermalVALID scores plus72auditedphysicalcases,full585derivation andexactneural/unfamiliarrepeatchecks beforewinningpublication. Thewaiting phase9watcher is restarted before anycheckpoint exists toappend thiscomparison after itsoriginal2kvalidation; the alreadyqueued phase10launcher waitsforwatchernormalexit, so it will use the finalselectedparent. Maintrainer, source, optimizer, frozen guides andphase10trainingprotocol unchanged. Sources/protocol in full_average_v1/wider_protocol.json andwider_sources.json. Validation pending. + +## E089 — native-grid residual preparation and full-rollout resource check + +Hypothesis: the inherited strided U-net limits local spatial dynamics; add a native-grid width64 residual with eight metadata-conditioned depthwise5x5 blocks, dilation1/2/4/8 repeated twice, and zero final projection. Same normalized four-frame features before stride padding and guide injection; same weights on every case. Adds372363 randomly initialized parameters, initialization/check seed20261231. Only this new branch is checkpointed during training. Inherited complete model is not claimed translation equivariant. No optimizer updates, trained candidate or VALID comparison yet. + +Four audited TRAIN contexts (all four groups, resized32x48) preserve inherited outputs exactly. An optional nonzero latent-state head also preserves output and state coefficients exactly. Nonzero branch periodic shifts are exact in double precision on7x11/31x47/9x13 grids; arbitrary small periodic and replicated boundaries remain finite. Checkpoint recomputation matches eager outputs and all gradients exactly; gradient reaches stem and final depthwise block. Sources and context/checkpoint hashes in native_residual_v1/checks.json. Parent mainSHA1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61. + +Native128x384 TRAIN tcool0.03 trajectory0 full30 forward/backward at width96 with phase9initialSHA81445375f8d162239675f5b5e061b2bcd6678f81eae9c468e2de04d7544f9ea8, frozenE072guides and checkseed20270101: original and extended outputs/loss are bit-exact (loss0.172014117241), all gradients finite, nonzero newheadgradient0.208984375, inherited parameters unchanged. Peakallocated28.3834→28.9053GiB; peakreserved29.3164→29.8320GiB. SharedGPU timings7.39/7.22seconds are not an isolated speed comparison. Zero newhead means newinteriorgradients are correctlyzero at initialization. No VALID/held-out targets, optimizer, publication or queued training. Actual continuation must retain physical-head supervision if that architecture is selected. + +## E089a — spatial error bands limit the fine-detail hypothesis + +Descriptive decomposition of the nine previously audited E085 initial TRAIN rollouts, fixedE066 mainSHA1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61. PeriodicX Fourier and even-extension cosineY, bands in maxabsolute cycles/cell; ParsevalnormalizedMSE agrees within1e-12. No optimization, new randomness, VALID or prediction transformation. Errorenergy: mean8.97percent,wavelength>32cells48.84percent,16–32cells31.25percent,8–16cells7.94percent,4–8cells1.75percent,2–4cells1.26percent. Thus only3.01percent lies at wavelengths8cells or below. Oracle removal of that TRAINtarget-dependent error reducesVRMSE0.383153985→0.374927869; removing all wavelengths16orbelow gives0.351265132. These are descriptive oracles, never inference operations. Bands add in normalizedMSE, notVRMSE. The result weakens a fine-detail-only explanation; any native residual benefit would need broader scales too. Its receptivefield123cells permits that but this does not show learnability. No training protocol or queue added. Per-channel/step energies, source/prediction/targethashes andPNG/SVG figure in native_residual_v1/error_bands.*. + +## Full validation of candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781818**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437624 | + +Replayed all576 physical VALID cases under the same four-member translation rule; all distinct member prefixes are complete, audited in translation_member_coverage.json. Member weights are identical and every member applies to every case; identity duplicates share computation. Replayed all9 thermal VALID cases after retaining the latest selected shared neural/guide weights. Physical modules/configuration are identical to the original ensemble replay and outputs are independent of neural weights only because every member has a complete physical prefix. Quick81 inputs/metadata match their full585 counterparts exactly. Compare full585 rewards, with active/neural/smoke/unknown repeat checks before any winning publication. No held-out data or test-time fitting. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E062 — fixed shared-model translation average, full cohort +- Hypothesis / changed factor: Same complete surrogate averaged over four fixed periodic-x translations, fractions0/2/4/6 of128 cells; identity for nonperiodic-x boundaries and shared computation for duplicate members. No new parameters or seed; same learned weights applied to all four members and every case. Select the fixed shift rule on nine audited TRAIN trajectories:0.40919935 to0.40829987; quarter-domain translations regress to0.41531276. Initial nine-case thermal VALID improves0.44911052 to0.44688063. Replay all576 physical VALID cases, requiring every distinct member to complete its physical prefix; preserve later selected neural improvements and rerun all9 thermal cases before full585 comparison. Exact member coverage, source package and rebases retained. No pretrained assets, synthetic data, dataset specialist checkpoint or inference-time training. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `5aefab0263a8789ffb9c221d50ba13fc27c59ceceb5d284e6fd58f0a04d54301`. +- Validation: fixed 81-case subset; reward **0.812595**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437624 | + +Selection basis: full585 physical-operator comparison: candidate0.78181788, current0.78161511 + +Outcome: improvement; published atomically. + +## E087c — small complementary continuity blend + +A poor standalone transport forecast may still have complementary error. On the same nine initial TRAIN cases and fixedE066 checkpoint, fit a single convex density blend in[0,1] using only E087forecast-velocity predictions (never oraclefuturevelocities). All endpointchannel-stepVRMSEs are below10, so the meanVRMSE is convex; derivative bisection givesalpha0.01344492073972926. DensityVRMSE0.148010004428→0.147888832664; leave-one-cooling-regime-out fits/evaluations give0.147916682547. The derivative atzero is−0.0180187542072. No newnetworkparameters or randomseed. This is a very small TRAINgain, not a validation result. Coefficient, quadraticstatistics, allsource/target/predictionhashes and fullcrossfitresults in density_transport_v1/blend_results.json; no oracleprediction files read by this study. + +## E087d — additional TRAIN confirmation predeclaration + +Freeze the nine-case fittedalpha exactly, without refitting, and evaluate the other63 authorized thermalTRAINtrajectories (all72 minus the nine calibrationrecords). Retain pinnedE066 complete model and guides from window_position_v1/package; no translationensemble or positivityrebase during this diagnostic. Read nativeframes0–33 from authorizedTRAIN sources, require byteidentity with cachednativearrays, supply onlyframes0–3 to the model, and transport density with its own forecast velocities. Source, parenthashes, and fullmembershipaudit in density_transport_v1/additional_train/protocol.json and audit.json. No optimizer/newseed/VALID data. All63 are already in backboneTRAIN, so this only checks the blend beyond its calibration subset. Results pending; no servingcandidate or queue modification. + +## E087d result — fixed blend confirms on63 additional TRAIN trajectories + +Alpha remains0.01344492073972926 from the originalnine calibrationrecords. On the other63 initialTRAIN trajectories, densityVRMSE0.154377421716→0.154178949505, fullfour-channelVRMSE0.379773997608→0.379724379556;52/63casesimprove. Freshnativeinputs/targets matchcachedTRAINwindows exactly. No coefficient refitting, optimizer, newseed, oraclevelocities or VALID data. Allrollouts/sources/targethashaudit under density_transport_v1/additional_train. This remains a very small effect. + +## E087e — fixed density-blend validation predeclaration + +Assemble the fixed blend after the complete current shared forecast and any final support projection. Eligibility depends only on density/pressure/velocity0/velocity1 fields, periodic-X/open-Y boundaries and validspacing/time metadata; notdatasetname. Same independentMC/Heuncontinuity/CFL0.4 with a1024-substep workcap and deterministic unchanged-forecast fallback. It consumes only suppliedcontext and modelvelocities. Packagedoperator matches savedTRAINforecast-velocity arithmetic exactly, leavesotherchannels andineligible/empty/workcappedcasesexact, and preservesconstantdensity/determinism on7x11CPUgrid. Initialpreflight log/source archived after removal of a redundant always-true assertion; cleanedcheck rerun passes, finalsourcehashes in density_blend_v1/checks.json. No change to solverarithmetic. + +Queue one full585comparison afterE064normalcompletion: ninefresh eligibleVALID forecasts; reuse576parentpredictions only after eachinputmetadatafailseligibility and allinheritedassets/config/source matchexcepttheisolated finaltransform call. Exactofficialaggregation and neural/novelrepeatchecksbeforewin. Newoperatorasset/config invalidate genericphysical-cache reuse; this explicitmetadataaudit supplies the proof for this comparison. Ifselected, updatephase8/9templates/cachesunderlocks. Optionalasset equality added tofuturelatentnonactiveaudit; physicalkernel, encodedcache, maintrainers andphase10protocol unchanged. Sources/protocol in density_blend_v1/protocol.json. Validation pending. + +## Full validation of candidate + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781820**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437602 | + +Applied the exact final unary nonnegativity projection to all585 previously validated parent predictions, using only supplied field names. The projection occurs after the full model, all corrections and any translation average, so no dynamics are approximated or recomputed differently. Source prediction hashes and per-case changed counts are in projection_audit.json. Freshly score every changed case with the official scorer, reuse per-case scores only for byte-identical copied predictions, and aggregate all585 using the exact official aggregation code; provenance and scorer/report hashes are retained. Replayed all9 thermal VALID cases through the assembled package and verified bit-exact equality to projected outputs; active/smoke/diffusion and unfamiliar-schema repeat checks precede a winning publication. Quick81 input/metadata equality verified. The model itself reads only supplied context and package assets. No held-out trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E064 — final physical-support projection +- Hypothesis / changed factor: Final nonnegativity projection for density and concentration, selected by supplied field names, after complete forecasts and any translation average. No new parameters or seed. On nine audited TRAIN rollouts, statistical corrections had introduced negative density in0.8–8percent of pixels despite an earlier clamp; all corresponding targets are nonnegative. Fixed zero projection reduces meanTRAIN VRMSE0.40919935 to0.40918377. Transform all585 parent VALID predictions exactly, record changed counts/source hashes, replay all9 thermal cases and require bit-exact transformed/serving parity. Full585 selection plus active/smoke/unknown repeats. No tuning of the zero bound, pretrained assets or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `992e25d838bbd7bfe4db95013b7790eca11b4fab518ca1b24b637b276a33c0da`. +- Validation: fixed 81-case subset; reward **0.812597**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437602 | + +Selection basis: full585 physical-operator comparison: candidate0.78181972, current0.78181788 + +Outcome: improvement; published atomically. + +## E089b — portable native-residual prototype + +Assemble the zero-output branch on the selectedE062 four-translation package; no inference plumbing or inheritedweights changed. Fullmodel parameters4771385 (372363added), newbranchinitializationseed20261231, prototypecheckpointSHA28060c09084349af3d6cfb9c189732445b28f0a411bfde620b4209267564857a. Allruntimecode is embedded inmodel.py; no research-directory dependency at inference. Prototypeonly, no optimizer or publication. Repeat complete-package parent/zero parity onfour freshly audited TRAINcontexts (allfourgroups, includingstrongactive17/-5) andtwo independent unfamiliar2/13-channel grids, contextonly/no targets. Existingnovelsseed20261126. Resultpending undernative_residual_v1/prototype/parity; prototype notselected. + +## E090 — extend immutable state-encoder TRAIN labels to30 steps + +E083short-horizon hidden-stateoracle regression andE084 near-orthogonal early/lategradients motivate preparing longer supervision. For all175 authorizedactiveTRAIN records, readnativeframes0–33 and require byte-exactagreement with existingfourcontextframes andfirstfourlabels, matchingmetadata and completeTRAINmembership. Keep all existing encodedstates, originalcache, physicalkernel andqueuedphase10protocol untouched. Store30 futureframes only as labels in separate latent_full_targets_v1; no recomputation of state fromtargets. Recordeverybase/target/contexthash andmarkcomplete onlyafterall175. Nonewseed, modelparameters, optimizer, VALID data or inferenceasset. This enables a later separatelydeclared full/sampledhorizonloss experiment; it does notqueueone. Preparationrunning; protocol/sourcehashes in latent_full_targets_v1/protocol.json. + +## Full validation of attempt0 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781821**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437585 | + +Replayed all9 eligible thermal VALID trajectories through the complete assembled package. Reused576 parent predictions and official per-case scores only after identical parent weights, all existing runtime assets/configuration, exact serving-source equality after removing the single new transform call, and false metadata eligibility for every reused case. This is a final transform; it cannot alter earlier state evolution. Input/prediction hashes and eligibility proof are retained. The exact official aggregation combines585 cases; quick81 input/meta equality is audited. A single coefficient fit on nine TRAIN trajectories was confirmed unchanged on63 more TRAIN cases. Only forecast velocities are used; no oracle velocities, held-out data or inference-time fitting. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E087e — small fixed continuity blend +- Hypothesis / changed factor: Fixed density continuity blendalpha0.01344492073972926, independently implemented MC finite-volume reconstruction/HeunRK2/CFL0.4, applied after the complete forecast and any nonnegative projection. Supplied field/boundary metadata determines eligibility. Uses only last context density/velocity and model forecast velocities. Fit one coefficient on nine TRAIN cases (densityVRMSE0.148010004428 to0.147888832664), leave-one-regime-out0.147916682547, then freeze on other63TRAIN:0.154377421716 to0.154178949505,52/63improve. No new network parameters or seed. Fresh nine VALID rollouts plus576 exactly audited ineligible parent outputs and full585 comparison; exact neural/novel-grid repeats before any improvement. No oracle future velocities, pretrained weights or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `7ff7479a422972cabedba1416898450cb254160c6ccdd66d7ec789be9d2e517c`. +- Validation: fixed 81-case subset; reward **0.812598**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437585 | + +Selection basis: full585 physical-operator comparison: candidate0.78182111, current0.78181972 + +Outcome: improvement; published atomically. + +## E089b result — full portable prototype parity + +Allsix complete-package checks are exact: four TRAINcontexts spanningallfourgroups (strongactive17/-5 included), plusindependent unfamiliar2-channel7x11 and13-channel129x177 inputs. Threeforecaststeps forTRAIN/twofornovels; allfinitefloat32 andexpectedshapes. Compare against E062four-translation parent withsameall-case memberrule. ProtoSHA28060c09084349af3d6cfb9c189732445b28f0a411bfde620b4209267564857a; branchseed20261231. No newtargets, training, VALID or publication. Exactinputs/read-onlymembershipaudit andchecks retained undernative_residual_v1/prototype. + +## E090 result — all175 longer TRAIN targets prepared + +Completed175 native30-step label files,15138838400bytes,82.55seconds CPU/wall preparation. Everyrecord matches fullTRAINmembership; freshfourcontextframes andfirstfourtargets matchimmutableexistingcache exactly, metadataequal,allfinitefloat32. Originalencodedstates andcachemetadata unchanged; eachbase/context/targetSHA recorded. No newencoderstate, GPUoptimizer, VALID data or inferredhiddenstate fromfuturetargets. Completion/manifest inlatent_full_targets_v1. No queuedtrainingprotocol change. + +## E091 — nested physical activation recomputation + +Hypothesis: wrappingeachfullstoredframe in an outer activationcheckpoint, while retaining originaltwo-substep checkpoints inside, can reduce full30-gradientmemory without changing mathematics. Copyunchangedphysicalkernel andloss source; only insertouterrecomputation. No optimizer, pretrainedassets, inferencechange, main/state-headtrainingchange or VALID data. Seed20270103, zero2x16x16F6coefficients, fixedscale.01 andsamecenterstateobjective. Stages: strong17/-5 TRAINtrajectory2 fourstepsoriginalvsnested, requireexactoutputs/loss andgradientrelativeL2<1e-5; moderate9/-3 trajectory2 nested30 vsE084stored weightedfullgradient (relativeL2<1e-4,lossdifference<5e-8); thenstrong30finite-gradient/prefix/memorycheck. Use auditedE090targets andoriginalencodedstates, nofitting. SharedH100timings notisolatedthroughput. Source/protocol undernested_physics_v1; results pending. + +Initialpreparation looked for the Binghamtable at the wrongdirectory; thefirststudy exited beforephysicalevaluation becauseits sourcehashmanifest was absent. Bothfailedsource/log retained. Correctedpreparation copies theexistingzero-prototype table and verifiesits SHA againsttheimmutableencoder manifest. Numericalloss/kernel/targets unchanged. + +## E089c — isolated matched continuation plumbing + +Prepare isolated control/native-residual training sources froma suppliedparent, retaining originalall-dataset windows, guidance, objectiveandoptimizer; ifa futureparenthas a latenthead, retainitsoriginalphysical supervision source. Do notchange thecurrentphase9/phase10 trainer/source. Reseedmainnoise afterarchitectureconstruction; deterministicno-update dryrun only. Runone fullpipeline backwarditeration ineach nonlatentarm, seed20270104,two microbatchesofone,twosteps,noise.01,uniformgroups. Bothdrawsame active/thermal datasets andhorizons andhaveexactloss0.05352762434631586. Everyparameter unchanged, nooptimizerstep/checkpoint. Nativebranchheadcontributesnewgradients as expected; gradnormcontrol.324292868376/native.335362970829. Peak1.119/1.679GiB; timingsare sharedGPU. All10115indexrecords freshlyverifiedTRAIN. Sources/checks/audit in native_residual_v1/matched_preflight. ParentSHA1f24abcc..., prototypeSHA28060c09... . Noactualtrainingprotocol/queue/VALID comparison; latentpath stillneedsitsownactual-parent numericalgate beforefuturetraining. + +## E091 result — long physical gradients fit with nested recomputation + +Allstagespass. StrongTRAIN17/-5trajectory2 fourstep outputs/loss/gradients arebit-exact;peak7.0738716→2.7079763GiB. Moderate9/-3trajectory2 full30 nestedgradient relativeL2 versusE084weightedearly/late reference3.6428751e-6;lossdifference5.5879354e-10. Nested30peak2.6898851GiB,loss.00904015265405,gradnorm.248568729. Strong17/-5full30finite,prefixexact,loss.637867689133,gradnorm.042030938,peak3.3442583GiB. SharedGPU timesstrong4old53.66s/nested67.59s,moderate30nested278.12s,strong30nested512.83s; notisolatedspeedcomparison. Parametersunchanged; nooptimizer/training/servingchange. Outputtensorhashes andchecks in nested_physics_v1/results.json. Phase10physicalsource andfour-step protocol remainunchanged. + +## E092 — full30 TRAIN hidden-state oracle predeclaration + +Testwhether fullhorizon supervision can improvea boundedF6correction instrongTRAIN17/-5trajectory2. Optimize512zero2x16x16statecoefficients for exactlythreeAdamupdates,LR.01,seed20270105,usingall30futureTRAINframes andthe unchangedcenterphysicalVRMSE objective/floor/clamp. Fixedupdate3, no LRsearch/checkpointselection. Firstupdate reusesthe verified E091zero-state adjoint atidentical kernel/loss/context/state/target/precision, withsourceandtensorhashaudit; subsequentgradients arefreshnestedrecomputation. Evaluatezeroandfixedupdate3usingcanonicalthree-node1e-4quadratureseed20261026,full30; missingphysicalframes score10. This is offline statefittingonly: futureTRAINlabels enterloss, notthecontextencoder. Coefficients MUST NEVERbe deployed or usedonVALID/held-out/inferencecases. Sharedmodelparameters andpublishedpackage unchanged; nolearnabilityclaim fromoneoracle. E090longtargets usedonlyforTRAINloss. Sources/protocol in latent_full_oracle_v1. Results pending. + +## E047 — shared main-model capacity growth, update 1000, shrink +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.75,mean/profile corrections=True; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8a2c2218ffa70f7d1e24000d95311d6930caf4f901bee88a01fab2c2f0911f65`. +- Runtime code SHA256: `23d0dedfbed9f1689b7ef7afe1f96fa0c848612f4f748e1c3c6c3148263dc39e`. +- Validation: fixed 81-case subset; reward **0.811817**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.447136 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E070 result — numerical precision has a small, mixed TRAIN effect + +All six predeclared native256/M32 TRAIN pairs complete all30 steps. Mean VRMSE float32 0.213677023714 versus float64 0.213597271491, a 0.0373% relative reduction; five improve, zeta5/alpha-4 worsens from0.140238355616 to0.140473912097. Strong17/-5 changes0.630049141475 to0.629725532087. Summed shared-GPU elapsed times604.37/892.89 seconds are descriptive and not an isolated throughput comparison. Identical float32 quadrature draws, seed20261026, original E067 physical parent/config, no optimizer or new learned checkpoint. New kernelSHA5db556077cf42a9bcb931ce2cb5aaf51d99c8039b30206712ce3df5e32deeb07; all paired predictions and hashes are retained in precision_active_v1. Outcome: small TRAIN-only gain, insufficient to prioritize a costly full active VALID replay and encoder-cache migration. Do not integrate or silently replace the physical kernel; selected serving and immutable state cache remain unchanged. No validation VRMSE or publication candidate from this diagnostic. + +## E047 — shared main-model capacity growth, update 1000, no_shrink +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.0,mean/profile corrections=True; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8a2c2218ffa70f7d1e24000d95311d6930caf4f901bee88a01fab2c2f0911f65`. +- Runtime code SHA256: `4e136bbd12f7c8a1eafb85b580f07be76ce2f7e3d9d76cb8223d745f780e4d7f`. +- Validation: fixed 81-case subset; reward **0.812315**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.441021 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E093 — shared-head projection of early and full physical adjoints + +Before choosing longer shared-head supervision, compare how four-step and30-step coefficient derivatives project into the actual zero neural head. Reuse audited E084/E091 state adjoints on moderate9/-3 and strong17/-5 TRAIN trajectory2; selected E066 width48 mainSHA1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61 and E072 guide, seed20270106. Same native context to128 pooled neural input, original BF16/guidance, no noise or shift. No solver rerun, optimizer, new labels or VALID. Check input/state/adjoint hashes, zero coefficients, finite gradients, exactly zero backbone state-loss gradient at the zero head and unchanged parameters. Report gradient norms/cosines before and after the98-parameter head projection. This is a local TRAIN derivative diagnostic, not a learnability or finite-update claim. Phase10 protocol unchanged. Sources/protocol under head_gradient_v1; results pending. + +## E047 — shared main-model capacity growth, update 1000, mean_profile_off +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.75,mean/profile corrections=False; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8a2c2218ffa70f7d1e24000d95311d6930caf4f901bee88a01fab2c2f0911f65`. +- Runtime code SHA256: `0065691a81aba7508e657e8c8e368ff199594f9ae5937d21fd8f23795c529c5d`. +- Validation: fixed 81-case subset; reward **0.811825**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.447035 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E093 result — long-horizon differences survive the shared-head projection + + +No optimizer or parameters changed. Both TRAIN examples have finite98-parameter head gradients and exactly zero backbone state-loss gradient at the zero head. + +moderate: coefficient early/full cosine0.004768404, full/early normratio381.1359; projected shared-head cosine0.085773849, normratio271.6535, early/full norms0.011110632/3.018242141. + +strong: coefficient early/full cosine0.354147787, full/early normratio23.1967; projected shared-head cosine0.303358347, normratio26.6474, early/full norms0.019149241/0.510276769. + +Seed20270106; selected E066 mainSHA1f24abcc..., fixed E072 guide. Projected-gradientSHA697c494a297f3dc1cc4b6a04df53f6ed34ee6b43b0411e4c84f590bc51f0da3b. The moderate example remains weakly aligned, so the short objective does not become a good local proxy merely by restricting updates to the head. Both examples remain positively aligned; do not call them conflicting/opposed. These are two unshifted TRAIN states at zero correction, not a finite-update or learnability result. No VALID metrics, deployable checkpoint, or phase10 protocol change. Full audit in head_gradient_v1/results.json. + +## E094 — isolated longer physical-supervision plumbing + +Prepare a separate trainer supporting30-step TRAIN labels and E091 nested activation recomputation, with configurable replacement of some scheduled short physical losses by longer losses. Original phase10 source, objective and queue remain unchanged; no new optimizer is scheduled. Seed20270107 for one no-update matched preflight: selected E066 backbone plus zero head, original versus nested source, two main microbatches with two forecast steps and one two-step physical loss, same records/noise/shifts and optimizer disabled. Recheck all175 base and30-label hashes/TRAIN memberships, exact old-label prefixes, unchanged context/state objects, and every8/every64 cadence arithmetic. Audit all10115 main TRAIN index records. Require exact main/physical losses and gradient-norm relative difference below1e-6. Actual-parent full30 neural backward/resource gate remains required before any future optimizer. This prepares capability only; no chosen long-horizon continuation protocol or VALID comparison. Sources/protocol in long_state_training_v1. + +## E094 preflight — loss parity passes, gradient-norm gate fails + +Both one-iteration dryruns have exact mainloss0.037365127354860306 and physicalloss0.0031871255487203598, same records/horizons/shifts and unchanged parameters. Gradientnorm control0.50008505582809448, nested0.5000419020652771; relative difference8.62928462e-05 exceeds the predeclared1e-6 gate. Preserve failed source/logs and do not adopt the trainer. Diagnose with two repeat original runs under original training backend flags, followed by original/nested runs under deterministic convolution flags; save full gradients before clipping. No optimizer/phase10 changes. Do not relax the gate or attribute this to backend nondeterminism without the comparisons. Sources/protocol under long_state_training_v1. + +## E095 — CUDA graph training feasibility without compiler fusion + +Test fixed-shape full30 neural forward/backward capture as a possible way to reduce launch overhead. This is separate from previously rejected torch.compile studies and does not modify any running trainer. Selected width48 E066 main, E072 guides; two authorized native thermal TRAIN initialwindows (tcool.03/.32 trajectory2), seed20270108, inputnoise.01 generated once outside capture, deterministic convolution flags. PyTorch make_graphed_callables requires AMP caching disabled: first measure default cache-enabled versus disabled eager losses/forecasts/gradients explicitly, then compare graph with cache-disabled eager. Require exact final forecasts and gradient relativeL2<1e-6 for both contexts and a fixed live head-bias perturbation; restore weights afterward. No optimizer, trained checkpoint, VALID or serving change. Capture warmup3; any shared timing is not an isolated throughput claim. Launch after E092 and E094 diagnosis complete. Sources/protocol in cuda_graph_training_v1. + +## E094 diagnosis — controlled recomputation gradients are exact + +Under cudnn.benchmark=False and cudnn.deterministic=True, original/nested sources have bit-exact mainloss0.037385668605566025, physicalloss0.00318712554872036, every parameter gradient, and gradientnorm0.5000669360160828. Peak memory2.258983 to1.850681GiB. Two repeated original runs under the original training flags differ in their full gradients by relativeL2 5.88977e-6 (primarily stem weights), even though their reported gradientnorms agree. The deterministic convolution choice also shifts the main forward loss relative to the original flags, so these are explicit numerical factors. Initial nondeterministic preflight failure remains recorded; controlled comparison isolates nested recomputation successfully. All175 long targets pass hash/membership/old-prefix/context-state checks and all10115 main records are TRAIN. No optimizer, VALID, serving or queued phase10 change. Preparation complete, but actual-parent full30 backward/resource gate remains required. Actual source hashes for all isolated arms are in source_provenance.json; stale inherited manifests in copied control/instrumented folders are preserved separately and replaced with truthful local source manifests. + +## E092 result — fixed three-update full-horizon TRAIN state fit + +All three declared Adam updates complete (LR.01, seed20270105, exactly512 direct coefficients; no checkpoint/LR search). Center-state objective 0.637867689133 to 0.628882646561. The fixed update3 improves the canonical three-node full30 TRAIN rollout: 0.630049141475 to 0.621326571054; completed physical steps zero/update3=30/30 out of30. Missing frames retain their predeclared score10. Coefficient SHA256 `2cf669cefd9adffd25dbebd582ef2a7fe40bb7ccf0e62c7274090cc1316ac05e`. No shared neural weights changed. The immutable E091 initial zero adjoint was reused exactly for update1; later gradients were fresh nested recomputation. Recorded peak3.183126GiB, cumulative three-update shared time775.33s; not isolated throughput. Full per-step/channel scores, source/context/target hashes and PNG/SVG are in latent_full_oracle_v1. + +This uses all30 future labels of one authorized TRAIN trajectory (zeta17/alpha-5, trajectory2), so the fitted coefficients are an oracle diagnostic and MUST NEVER enter inference or VALID/held-out cases. Unlike E083's short-horizon fit on a different moderate trajectory, the objective matches the full horizon here; that cross-case contrast is not a controlled causal comparison. It motivates testing longer shared supervision but does not establish four-frame learnability or generalization. No validation metrics, deployable checkpoint, selected-package change or automatic new training queue. + +## E095a result and E095b numerical follow-up + +Original capture fails before replay because the model constructs a CPU boolean coordinate mask and copies it to CUDA inside forward, which CUDA graph capture forbids for unpinned CPU memory. No optimizer or training/serving change. Eager full30 AMP cache enabled/disabled forecasts and losses are exact on both TRAIN cases, but full-gradient relativeL2 differences are0.007809954731 and0.004939142117. Thus disabling AMP caching is a real numerical training factor, not silently interchangeable execution. Source/results retained in cuda_graph_training_v1. + +Follow up in isolated cuda_graph_training_v2: replace only the boundary-coordinate boolean tensor copy with the same GPU scalar multiplications. Require both full30 eager forecasts and every parameter gradient to be bit-exact to the original source before attempting capture again. Same inputs/noise/seed20270108, deterministic convolution flags, cache-disabled graph/eager comparison and live parameter change gate. No compiler, optimizer, validation or running-source modifications. Any eventual acceleration adoption requires a separately evaluated training experiment because of the measured AMP-cache difference. + +## E095b result and E095c constant-buffer follow-up + +The scalar-multiplication rewrite preserves the first full30 TRAIN final forecast exactly but fails the stricter all-gradient bit-equality check before capture. The assertion-only check did not save a difference magnitude, so none is inferred. No optimizer or capture; source/log/failure record remain in cuda_graph_training_v2. + +Try the narrower transformation in cuda_graph_training_v3: preallocate the four exact boolean boundary masks as a nonpersistent model buffer outside forward, then retain the original tensor multiplication and views. No new trainable parameters or random draws. Same two TRAIN inputs, seed20270108 and unchanged bit-exact eager source gate; save comparison magnitudes before assertions. Graph parity still compares cache-disabled eager only, with the separate AMP-cache training factor disclosed. Running trainers/serving unchanged. + +## E095c result and E095d repeatability diagnosis + +Preallocating the coordinate mask also preserves the full30 final TRAIN forecast exactly, but the first all-gradient comparison differs by relativeL2 0.008851873054 and maxabsolute0.00634765625, so capture is not attempted. Record the failure without weakening the exact source gate. + +Before attributing this to coordinate arithmetic, test repeatability in cuda_graph_training_v4 with CUBLAS_WORKSPACE_CONFIG=:4096:8 and torch.use_deterministic_algorithms(True), in addition to deterministic convolution flags. Compare three repetitions of one original model instance and a second original instance on the same TRAIN input, then retain the original constant-buffer/capture gates. Same inputs/noise/seed20270108, no optimizer or running-source changes. If an operation lacks a deterministic implementation, record that failure. The original AMP-cache difference remains a separate training factor. + +## E095d result — full deterministic training capture is exact + +With deterministic algorithms and fixed cuBLAS workspace, all same-instance repetitions, a new original instance, and the constant-buffer source comparison are bit-exact in forecasts and every gradient. CUDA capture succeeds (8.04s shared setup); both full30 TRAIN inputs and a live head-bias perturbation match cache-disabled eager forecasts/gradients exactly. Weights restored, no optimizer. Capture sourceSHA35ca8df418fb99daf68b83de6695a205289e56c1419e5cc3738142cc6398505d. AMP-cache enabled/disabled gradient differences remain0.00418496/0.00472504 despite exact forecasts. Strict eager calls about1.89–2.07s and graph1.48–1.54s shared, but these backend settings are slower than the earlier ordinary eager measurements; no speed/adoption claim. The reported~0.21GiB replay allocated metric excludes retained private-pool memory and MUST NOT be represented as the training memory footprint. + +## E095e — ordinary-backend capture characterization + +Use the source rewrite already proved in E095d, with the original benchmark-enabled, nondeterministic backend settings. Record default-backend gradient variability rather than requiring it to be bit-exact before capture. Keep original exact graph-gate result as a separate reported metric; no relaxed claim of equivalence. Same two TRAIN inputs/noise/seed20270108, cache-disabled graph and eager paths, live parameter check, then six alternating eager/graph timing pairs. Record capture peak and reserved memory to include graph-pool retention. Shared timings are not isolated throughput. No optimizer, running trainer, serving, VALID or queue change. Sources/protocol under cuda_graph_training_v5. + +## E095e input-alias diagnosis and E095f correction + +Ordinary-backend capture runs, but the paired timing check revealed an input alias: PyTorch uses supplied sample tensors as mutable static graph storage, so replaying case1 overwrote the stored argument0 tensors. The initial graph comparisons each used the correct case, but the later timing pairs inadvertently used case1 for both positions. Do not report them as a two-context throughput test. This is an isolated diagnostic; input files and running trainers are untouched. + +In cuda_graph_training_v6, clone dedicated capture samples before capture, retain CPU snapshots of every original argument tensor, require all originals unchanged, and assert each paired timing loss equals that case's original eager reference. Same data/noise/seed20270108, numerical and live-parameter checks, ordinary backend, cache-disabled paths, six alternating pairs. No optimizer or adoption. Preserve prior source/results and the alias diagnosis. + +## E095f result — input-safe capture and measured default-backend variability + +Dedicated capture samples preserve every original input tensor exactly. Each of six alternating timing pairs uses its intended TRAIN context and reproduces the corresponding eager loss exactly. Mean shared times eager0.867567504s versus graph0.646542220s (25.48% less); this is a paired exploratory shared-GPU result, not isolated throughput. Capture setup3.8835s, peakallocated15.48786GiB, peakreserved31.88086GiB, retainedreserved16.03320GiB after capture. The private pool is explicitly accounted for. + +Forecasts remain bit-exact. Graph/eager gradient relativeL2 differences case0=2.1885e-10,case1=0.00713708,live-bias-change=0.00125385, so the strict original1e-6 gradient gate FAILS under ordinary nondeterministic backend settings. This is not numerical equivalence to the current trainer. E095d's fully deterministic exact proof remains valid, and AMP-cache plus backend changes remain explicit training factors. No optimizer, trained checkpoint, VALID, serving or running-source changes. Source/result hashes and all negative iterations are retained. Future training needs a separate accumulation/optimizer/unused-head preflight and validation. + +## E096 — mixed eager/graph joint-optimizer preflight + +Before any future graph training, implement a training-only scalar rollout wrapper with dedicated capture inputs and allow_unused_input=True for the zero state head. Compare exactly two disposable optimizer updates in an eager arm and an arm using graph thermal30 plus eager fallback for other schemas. Selected E066 width48 main plus zero98-parameter latent head, seed20270109, fully deterministic backend and AMP caching disabled in both arms. Each update accumulates four TRAIN groups (thermal30; corpus/Arena/active2), then strong17/-5 trajectory2 physical2 loss at native256 with nested recomputation. Step2 reverses main microbatch order and uses the second thermal context. AdamW LR3e-5/head3e-4,beta.9/.95,weightdecay1e-4,clip1. Require exact corresponding losses and gradient/updated-parameter relativeL2<1e-6, finite gradients, no main-loss gradient for the unused state head, and original inputs unchanged. Record all context membership/source and disposable checkpoint hashes. This is a numerical pipeline check, not a model-selection run; NEVER publish these two-update diagnostic checkpoints. No running source, queued phase10 or serving change. Sources/protocol in training_graph_v1. + +## E096 initial failure — unsupported deterministic pooling backward + +The first eager arm stops at its first physical backward, before any optimizer update: PyTorch reports that adaptive_avg_pool2d_backward_cuda has no deterministic implementation. Preserve the original helper, generated model and log. This does not affect the running phase9 or queued phase10 source. + +For divisible-by16 state-head grids, test an explicit disjoint-block derivative while calling the exact original adaptive pooling in forward. Each input belongs to one output bin, so the backward needs only replication and division, with no atomic accumulation. Other shapes retain the original implementation. Predeclare exact forward and gradient comparisons on random nonzero1x2x128x128 and1x2x128x384 inputs/adjoints, seed20270110. Only if both pass, rerun the original two-update E096 protocol with seed20270109 and every original gate. No numerical threshold relaxation, changed trainable parameters, serving update or new training queue. + +## E096 pooling derivative check + +The square128x128 case is exact initially, but rectangular gradients differ by one rounding step (maximum9.3132257e-10 in the declared check). A sequential scalar-division attempt also fails exactness; preserve both sources and results. Elementwise probes identify the original arithmetic: device-tensor division by block width, then block height. Python scalar division can instead multiply by a rounded reciprocal. Match the original order and device division explicitly. With seed20270110, original forwards and gradients now match bit-exactly on128x128,128x384 and additional160x240/112x176 grids, including non-power-of-two block sizes. No tolerance change. Restart the original E096 two-update joint-optimizer preflight; no running trainer, serving source or queue changed. + +## E097 — shared coarse attention preparation + +Thermal TRAIN error-band analysis places most normalized error at large spatial scales, while earlier local convolutional auxiliary forecasts did not improve deployment. Test content-dependent attention between all64 sites of the existing4x16 coarse field. The proposed shared branch has four blocks of width128 with four attention heads, metadata/time conditioning, boundary-aware relative position bias, absolute coordinates only on nonperiodic axes, and a zero output head. It predicts the same interaction component and retains the existing mean/profile-preserving decoder. New random seed20270111; all19 datasets use the existing audited10115 TRAIN statistic/grid records, no pretrained assets. + +Preparation only: check zero-head persistence, padded/unpadded feature equality with a nonzero head, periodic translation equivariance, finite gradients, and CPU batch resource requirements before declaring the training schedule. No baseline retraining, new optimizer, VALID evaluation or selected-package change yet. Architecture and pinned metadata source are in coarse_attention_v1. + +## E096 result — joint optimizer and accumulation checks pass + +Both declared disposable updates pass with bit-exact corresponding losses, every gradient, and every updated parameter under the strict cache-disabled backend. Each update combines all four TRAIN groups, captured thermal30 and eager fallbacks, then the physical state loss; the second changes microbatch order and uses a nonzero learned head. All original inputs remain unchanged. Physical loss0.00743609434 then0.00741805416; head maximum weight0.000299999 then0.000577064. Eager/graph saved diagnostic SHA256 `5ba4b7d7acb45c788d1f6e5dcd0f17c2d9072d8c45012eb90f071015bfe24895` / `4a5339b585aae679d59dec43a44634087b37ff0dba0b792efc1ad369a872bf4b`; serialized files differ by name but their parameter tensors are exactly equal. These are test artifacts, NEVER deployment candidates. Final sources, initial failures and exact pooling-derivative checks are retained in training_graph_v1. No running trainer, phase10 protocol or serving change. + +## E097 CPU training launch + +The shared coarse attention model passes zero-head persistence and exact padded/unpadded feature checks on four TRAIN groups. With nonzero output weights and relative-position biases, periodic translation errors are below2.1e-15 in double precision; all tested gradients are finite. Parameter count846459. On four CPU threads, batch512 forward/backward takes0.79–0.97s in the preflight, with peak process RSS2.20GiB. All10115 TRAIN records and both cache index hashes are audited. + +Launch12000 updates from random initialization seed20270111, batch512, four CPU threads, AdamW LR0.0003 with100-update warmup and cosine decay to10%, weight decay0.0001, gradient clipping1. Preserve the original coarse interaction objective, TRAIN groups0.1/0.1/0.1/0.7, initial-context probability0.5 and horizon sampling. Check the first100-update pipeline artifact; select only among updates2000/6000/12000 using the fixed existing output blend0.25. This architecture trial also changes parameter count, seed and optimization settings; no isolated causal claim about attention alone. Replay the complete runtime on all9 thermal VALID cases; physical reuse must pass dependency/input/coverage audits, with active replay if required by a selected latent-state parent. Full585 comparison and exact repeated predictions precede any publication. No pretrained or synthetic training assets, held-out data, baseline retraining or changes to the running main trainer. Training source/protocol in coarse_attention_v1; validation pending. + +## E097 pipeline result + +The fixed100-update artifact (training SHA256 `7a941365c389515dc1436523a2abd083a1d961fec6cc4cb51c7a14cf3bc08098`) passes complete-package shape, float32, finiteness and exact repeated-output checks on four authorized TRAIN contexts and two unfamiliar shapes/channels. Corpus diffusion, Arena and active predictions are exactly equal to the parent on these checks. The thermal and unfamiliar predictions change as expected. Source/input hashes and exact TRAIN membership are retained in coarse_attention_v1/prototype. The isolated physical-reuse audit accepts this coarse-only change and rejects a deliberately altered mean-branch weight. This artifact is a plumbing check only, never a selection candidate; no VALID targets were read. The controller is running and waits for declared updates2000/6000/12000. Frozen source hashes cover package assembly, proof and validation; no running main or state-head protocol changes. + +## E098 — temporal hidden-moment observation TRAIN pilot + +Hypothesis: the existing fixed Stokes observer’s F4 innovations across the three supplied context transitions reveal part of the missing F6 state. Use the analytic frozen-flow coupling d(rhs_F4)/dF6=2i(A-iB), multiplied by metadata dt, in a fixed pointwise ridge inverse. Smooth the inferred correction with Gaussian lengthL/32; compare fixed gains0,0.1,1, with increment magnitude bounded by0.01c and final |F6|<=c. Apply after each of the existing three context observations. This is an approximate local interval model: F4 innovation can also reflect closure, integration and observation error, so improvement is not assumed. No model parameters are fitted or changed during inference. + +Before future-label diagnostics, require exact zero-gain native encoder states/streamfunctions at all three transitions and agreement of the analytic coupling with the full physical RHS in double precision. Preflight random seed20270112; original quadrature seed20261026. Then compare canonical three-node, full30 TRAIN rollouts on the third authorized trajectories of moderate9/-3 and strong17/-5. The two nonzero gains are declared before results. Shared neural weights, all other physical settings, and ongoing training protocols remain unchanged. No VALID/held-out data or training optimizer. Source/protocol hashes and pinned parent are in temporal_moment_v1; this diagnostic is not automatically deployed. + +## E098 preflight result + +Zero-gain physical state and streamfunction are bit-exact to the pinned parent at all three context observations. The analytic F6-to-F4 frozen-flow coupling agrees with the original RHS expressions to relativeL24.03e-14 in double precision. The first double check encountered a TorchScript complex128 conversion overflow; its source/log are retained. The corrected check executes the exact original flow/RHS AST with only JIT decorators removed, leaving the candidate/runtime float32 scripted kernel and scientific threshold unchanged. No future frames were read during this preflight. Proceed with the declared two-regime, three-gain, full30 TRAIN diagnostic; no optimizer or deployment. + +## E099 — grouped physical CUDA graph feasibility + +Hypothesis: grouping four existing internal active-fluid steps into one graph launch reduces host overhead while preserving the same ordered operations. Isolated kernel copy retains the single-step graph for remainder steps; no compiler, floating-point precision change or learned parameters. Queue after phase9’s final checkpoint. On audited cached TRAIN9/-3 and17/-5 trajectory2 states, run three alternating-order pairs of30 stored intervals with graph groups1/4. Require bit-exact complex physical states and streamfunctions at every interval. Report shared-H100 timing plus capture and reserved-memory costs; no isolated throughput claim. No new seed, optimizer, future-label reads or validation. This diagnostic does not change the immutable encoder, any running/queued trainer, or the package. Source hashes and pinned parent in kinetic_graph_batch_v1. + +## E047 — shared main-model capacity growth, update 2000, shrink +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.75,mean/profile corrections=True; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. +- Runtime code SHA256: `23d0dedfbed9f1689b7ef7afe1f96fa0c848612f4f748e1c3c6c3148263dc39e`. +- Validation: fixed 81-case subset; reward **0.811890**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.446230 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of attention_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781820**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437601 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/density_blend_v1/attempt0. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E097 — shared coarse attention, update 2000 +- Hypothesis / changed factor: Shared4x16 coarse attention, update2000, auxiliary initialization/sampling seed20270111. Four blocks, width128, four heads, metadata/time conditioning and boundary-aware relative bias;846459 parameters. Train from scratch on all19 datasets and10115 audited TRAIN records,12000 updates,batch512,four CPU threads, AdamW LR0.0003,100-update warmup and cosine decay, original interaction objective and fixed blend0.25. Training checkpointSHA256 9eff14b2e22ebc44d0548fee2f0ad828434a58b7d1a73dd6be4561fa0c945cb3. Every case evaluates the same shared branch. Candidate inherits the documented current parent; all nine thermal VALID rollouts are fresh, physical reuse follows dependency/input/coverage audits, and latent parents trigger fresh24-active replay with audited552 nonactive reuse. Full585 reports and exact neural/active/Arena/unfamiliar repeat checks precede a win. This architecture comparison also changes parameter count, seed and optimization settings. No pretrained assets, held-out data or inference-time fitting. Active training jobs retain their declared templates. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `2aa5cd215e197a3ea997568a47b7f745d61a928522938f7f7607bbe220a6f289`. +- Validation: fixed 81-case subset; reward **0.812597**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437601 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E047 — shared main-model capacity growth, update 2000, no_shrink +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.0,mean/profile corrections=True; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. +- Runtime code SHA256: `4e136bbd12f7c8a1eafb85b580f07be76ce2f7e3d9d76cb8223d745f780e4d7f`. +- Validation: fixed 81-case subset; reward **0.812364**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.440425 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E100 — cross-group neural gradient diagnosis + +E071 found Fisher/active neural gradient norms substantially larger than thermal examples, but did not measure cross-group alignment. Hypothesis: nominal group probabilities need not reflect actual update influence, and nonthermal gradients may interfere with thermal prediction. Pin selected main SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61` and guide `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Inspect37 deterministic TRAIN examples: fourth trajectories for all16 corpus families, Arena, active17/-5, and all9 thermal cooling regimes at start0; repeat the10 Well cases at start34. Full26/10/30 horizons, current log1p normalized squared loss, batch1, no input noise, seed20270113, controlled convolution flags. No physical branch or optimizer. + +Save every flattened gradient, per-example norms/cosines, sample group means, and projections of the existing0.05/0.075/0.075/0.8 mixture onto the thermal mean gradient. Compare only diagnostically with unit-normalized per-example gradients. Require finite gradients and unchanged parameters. This finite sample does not estimate the whole training distribution or establish a balancing method’s validation benefit. Queue behind the E098 TRAIN pilot; no running/queued trainer changes, new model candidate, VALID or held-out use. Source/protocol and TRAIN memberships in group_gradient_v1. + +## E047 — shared main-model capacity growth, update 2000, mean_profile_off +- Hypothesis / changed factor: Grow the selected shared main network from width48 to96 using function-preserving channel duplication and zero-sum outgoing-pair perturbations, seed20261110. Main parameters3066627 to9526707. Sampling seed20261119,2000updates, six microbatches of one,effectivebatch6,full30-step rollout with corpus26/Arena10 caps. Learning rate0.00003 cosine,noise0.01,all19 TRAIN datasets and unchanged group sampling. Exact parent and frozen guide hashes are in phase9/lineage.json. Profile shrinkage0.75,mean/profile corrections=False; retain selected coarse branch and fixed physical operators. Rerun all9 neural valid cases and reuse72 complete physical outputs from the documented cache. No pretrained assets or inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. +- Runtime code SHA256: `0065691a81aba7508e657e8c8e368ff199594f9ae5937d21fd8f23795c529c5d`. +- Validation: fixed 81-case subset; reward **0.811554**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.450399 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E088 — wider checkpoint average, shrink +- Hypothesis / changed factor: Predeclared equal parameter average of phase9 updates1000/2000, variantshrink. Declared before either checkpoint existed. Same width96 shared model and all19 TRAIN datasets, samplingseed20261119, inherited scratch-trained initialization; float64 averaging restored to original dtypes. No new parameters, training, random seed or inference ensemble. Three decoder settings match the existing phase9 comparison: profile shrinkage0.75,mean/profile output corrections=True. Use the locked phase9 inference template and audited complete physical cache; freshly replay/officially score nine thermalVALID cases, reuse72 unchanged physical case scores, derive full585 and require exact neural/unfamiliar repeats on a win. No pretrained or synthetic assets and no inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8e2ba25ce733a307eeb5d8c3dc397ac32ebaac64218ce63c1d84e8c0739a2d14`. +- Runtime code SHA256: `23d0dedfbed9f1689b7ef7afe1f96fa0c848612f4f748e1c3c6c3148263dc39e`. +- Validation: fixed 81-case subset; reward **0.811971**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.445238 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E047 final individual checkpoint analysis + +Width96/update2000 with the unchanged no-shrink decoder gives thermalVALID0.440425400628 versus selected width48/E0660.437585061919. 4/9 cases improve. Mean per-channel changes (candidate minus selected): {'density': -0.00183693175437736, 'pressure': 0.012341972627327748, 'velocity_0': 0.00020684648425119598, 'velocity_1': 0.0006494674825490401}. The final6-step mean worsens by0.012671631; the middle7-step window improves0.003426289. All configuration values except width, serving model source, guide/coarse weights, physical kernel and density adapter are byte-identical. This describes the full wider-training protocol, not an isolated causal effect of capacity. All three2k decoder variants failed selection; the already-declared1k/2k parameter average remains pending. Paired report hashes, all per-case/time/channel values and standalonePNG/SVG are in phase9/analysis002000. No new fitting or selection rule. + +## E088 — wider checkpoint average, no_shrink +- Hypothesis / changed factor: Predeclared equal parameter average of phase9 updates1000/2000, variantno_shrink. Declared before either checkpoint existed. Same width96 shared model and all19 TRAIN datasets, samplingseed20261119, inherited scratch-trained initialization; float64 averaging restored to original dtypes. No new parameters, training, random seed or inference ensemble. Three decoder settings match the existing phase9 comparison: profile shrinkage0.0,mean/profile output corrections=True. Use the locked phase9 inference template and audited complete physical cache; freshly replay/officially score nine thermalVALID cases, reuse72 unchanged physical case scores, derive full585 and require exact neural/unfamiliar repeats on a win. No pretrained or synthetic assets and no inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8e2ba25ce733a307eeb5d8c3dc397ac32ebaac64218ce63c1d84e8c0739a2d14`. +- Runtime code SHA256: `4e136bbd12f7c8a1eafb85b580f07be76ce2f7e3d9d76cb8223d745f780e4d7f`. +- Validation: fixed 81-case subset; reward **0.812537**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.438327 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E088 — wider checkpoint average, mean_profile_off +- Hypothesis / changed factor: Predeclared equal parameter average of phase9 updates1000/2000, variantmean_profile_off. Declared before either checkpoint existed. Same width96 shared model and all19 TRAIN datasets, samplingseed20261119, inherited scratch-trained initialization; float64 averaging restored to original dtypes. No new parameters, training, random seed or inference ensemble. Three decoder settings match the existing phase9 comparison: profile shrinkage0.75,mean/profile output corrections=False. Use the locked phase9 inference template and audited complete physical cache; freshly replay/officially score nine thermalVALID cases, reuse72 unchanged physical case scores, derive full585 and require exact neural/unfamiliar repeats on a win. No pretrained or synthetic assets and no inference-time fitting. +- Seed: 20261119. +- Checkpoint SHA256: `8e2ba25ce733a307eeb5d8c3dc397ac32ebaac64218ce63c1d84e8c0739a2d14`. +- Runtime code SHA256: `0065691a81aba7508e657e8c8e368ff199594f9ae5937d21fd8f23795c529c5d`. +- Validation: fixed 81-case subset; reward **0.811844**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.446795 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E101 — fixed shared-model output-average TRAIN pilot + +The wider2k model has mixed per-case/channel errors despite a worse mean validation score. Hypothesis: its forecast errors may complement the selected width48 model. Before any validation ensemble comparison or portable implementation, replay both complete pinned packages on the nine initial thermal TRAIN windows used in E085. Compare each member and the fixed equal0.5 output average, computed in float64 then cast float32 and scored with the official scorer. No alpha fitting or sweep, new weights or random seed. All configuration except width and all serving/auxiliary/physical sources match exactly. Parent SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`; wider SHA256 `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. + +Queue behind E100; source/protocol and frozen package copies in output_ensemble_v1. A positive pilot would still require a portable implementation in which every member applies to every case, strict checks for any shared identical physical work, complete VALID comparison and repeat checks. No per-dataset checkpoint switching, incomplete-prefix shortcut, current package change or alteration to queued state-head training. Validation metrics not evaluated for this hypothesis yet. + +## E098 TRAIN pilot result + +Canonical three-node, full30 forecasts; same pinned physical parent, fixed observer gains and quadrature seed20261026. No shared parameters or training optimizer. + +| TRAIN regime | Gain0 | Gain0.1 | Gain1 | +|---|---:|---:|---:| +| 9.0/-3.0, trajectory2 | 0.008928462283 | 0.008840420770 | 0.011601164090 | +| 17.0/-5.0, trajectory2 | 0.630049141475 | 0.629604972778 | 0.627294747476 | + +Mean TRAIN VRMSE: gain0.0=0.319488801879, gain0.1=0.319222696774, gain1.0=0.319447955783. Complete per-step/channel errors, context-innovation magnitudes, prediction hashes and standalone PNG/SVG are retained in temporal_moment_v1. These two trajectories do not establish generalization; no VALID comparison, complete translated-runtime claim or publication candidate. Result SHA256 `59e9f3e2a600cd4fe9eefe67d22a2993e8b2e1fbd48ff211677ea56bfa2bed48`. + +## E098b — frozen temporal-observer TRAIN confirmation + +Gain0.1 improves both declared pilot cases and reduces their mean0.319488801879 to0.319222696774 (0.0833percent relative). Gain1 worsens the moderate case and only narrowly improves the two-case mean. Freeze gain0.1 with all other observer settings unchanged, and predeclare six fresh fourth TRAIN trajectories at zeta/alpha1/-1,3/-2,5/-4,9/-3,15/-4,17/-5. Compare gain0 and0.1 using full30 canonical three-node rollouts; no refitting or new settings. Source files/table are byte-identical to the checked E098 implementation. Original and corrected preflight hashes are now explicitly distinguished in E098 assessment.json; its earlier version is preserved. + +Queue behind E099. No new seed, shared checkpoint, optimizer, VALID evaluation or automatic deployment; this small pilot effect needs confirmation across regimes. Pinned original parent and source proof in temporal_moment_confirm_v1. The current/queued state-head encoder and training protocol remain unchanged. + +## E100 result — unit normalization is not supported by this sample + +All37 full-horizon TRAIN gradients are finite and all shared parameters remain bit-identical. Initial-window mean gradient norms for corpus/Arena/active/thermal are10.9542/2.7600/14.4719/3.1536. Their cosines with the thermal mean are-0.08161/+0.08644/+0.03962/1. The weighted signed projections of nonthermal groups onto the thermal direction are-0.04470/+0.01789/+0.04301, versus+2.52291 from thermal. Large nonthermal magnitudes therefore do not imply large opposition in this sample. + +The nominal weighted-mixture cosine to the thermal mean is0.903095; unit-normalizing each example gives0.899671. On the start34 Well-only sample, those values are0.921436 and0.728220. The latter renormalizes the two available group probabilities and is not an all-group estimate. Outcome: no balancing change queued from this diagnostic. Small finite batch1/no-noise TRAIN samples do not establish population behavior or validation impact. Raw gradient SHA256 `0870d2be686c114ae9c79b94a8ee9b8761100688bed09397b980ecd5b2ca20fa`; all rows,37x37 cosine matrix and source/main/guide hashes retained in group_gradient_v1. Seed20270113, no optimizer, VALID, new checkpoint or inference change. + +## E060 joint state-head training launch + +Initialize a two-channel16x16 F6 state head at zero on the selected shared width48 backbone, adding98 parameters. Apply it to every case and use only its first four-frame pass for physical hidden-state inference. Jointly optimize the all19-dataset autoregressive loss and a four-step differentiable physical VRMSE loss every8 updates; one parameter set, no specialist checkpoint. Draw physical TRAIN examples uniformly by parameter file, then trajectory, with even periodic shifts; all175 cache records have verified TRAIN membership and SHA256. Fixed correction0.01*c*(abs(F2)/c)^3*tanh(head), unchanged three-node quadrature. Train2000 updates, effectivebatch6, full30-step main rollout, LR0.00003 cosine, headLR10x, physicalweight1, samplingseed20261127/physicalseed20261128, zero-initseed20261126. No synthetic or pretrained assets. Selected parent, frozen guide, source and cache hashes are in phase10/lineage.json. Repeated six-case zero-head serving parity passed on the actual launch parent. Validate updates1000/2000 using all585 VALID cases, replaying all24 active and9 thermal cases. Validation pending. + +## E098b membership correction before data reads + +The original fourth-trajectory selector failed its TRAIN membership assertion before reading any frames. Inspection of TRAIN pools shows the zeta1/-1 file has only three authorized trajectories. A subsequent attempt to form a fourth entry from the union of TRAIN intervals also failed before any frame read; the provisional disjoint-interval hypothesis was incorrect. Both sources/logs are retained. Pin and audit all six choices before relaunch: {'1/-1': 1, '3/-2': 3, '5/-4': 3, '9/-3': 3, '15/-4': 3, '17/-5': 3}. Use the fourth authorized trajectory where available and the second for zeta1/-1. All six are new to this observer hypothesis; no excluded targets, predictions or scores were accessed. The only changed factor is the declared confirmation cohort selector, made before outcomes; observer gain0.1 and numerical settings stay fixed. Selection audit/protocol are in temporal_moment_confirm_v1. + +## E099 result — exact grouping, no demonstrated throughput gain + +Every physical state and streamfunction is bit-exact over all180 paired stored intervals across two TRAIN states and three alternating-order30-step repetitions. Mean shared integration times (single/group4) are33.1102/37.5483s at96 internal substeps, 62.4672/59.8771s at176 internal substeps. Five of six pairs favor single-step launches; the final strong-case pair reverses direction as concurrent work changes. Outcome: numerical feasibility passes, but no consistent speed gain is established, so no adoption. + +These timings include state clones and omit output decoding; they are not isolated or complete-rollout throughput. Reserved allocator memory after constructing both operators is0.682GiB, versus0.340GiB after the first; this is process-local and not total device use. The cached TRAIN files also deserialize their old target tensors, which the check never accesses or uses—clarifying the preparation’s shorthand “no future-label reads.” Source/protocol/raw timings and result SHA256 `35236a6c01d14f37ac50f9997c4a153457150857d6242843b9b54bb6ce2f4d90` are retained in kinetic_graph_batch_v1. No optimizer, VALID, encoder, training or serving change. + +## E101 TRAIN result and portable ensemble preparation + +Fresh nine-case TRAIN thermal VRMSE: selectedwidth48=0.381542788307, width96/2k=0.376278064086, fixed0.5 outputaverage=0.375775414370. Both individual members and the average are scored with the official scorer, zero failures. The copied diagnostic truth files initially omitted the scorer-required fields attribute; scoring stopped before the first case. Copy that attribute from each matching supplied context into local truth copies, verify every target value unchanged and the stored average exactly equal to the declared formula, then rescore the already-generated predictions. Failed logs/source and scorer_metadata_correction.json are retained; no prediction or scientific setting changed. + +Proceed to portable implementation with exactly these two fixed checkpoints and equal coefficients. Both complete shared models execute on every case, every distinct translation and every channel chunk. Each keeps its independent neural recurrence, statistical corrections, final support projection and continuity blend before output averaging. Share only identical fixed solver step results and kinetic continuations scoped to one case/translation/chunk. On a non-finite physical step each member retains its own neural fallback; partial prefixes are not treated as complete. Guides/coarse weights are shared by both members. No latent-head models are accepted by this assembler. + +Require exact equality to the average of independently executed original packages on four TRAIN groups and two unfamiliar channel/grid cases, exact repeats, and an explicit partial-solver/channel-chunk check. Source-controlled transform and cache helper hashes in output_ensemble_v1/package_source_hashes.json. Fresh thermal VALID comparison and complete physical-reuse audit remain required before any selection; no VALID ensemble scores have been computed yet. Current state-head training remains on its declared single-model template. Future training builders must explicitly handle an ensemble parent before use. + +## E101 validation gate preparation + +The generic physical-dependency checker now treats physical_cache.py and model_ensemble configuration as physical dependencies; an ensemble cannot silently use the ordinary neural-only reuse path. A focused check confirms both changes are detected and identical packages still pass. For any latent-head runtime, extra ensemble weights are also conservatively included. The record helper hashes both added runtime assets and describes two complete shared members in README; unchanged single-model packages retain their prior behavior. + +The E101 controller waits for exact six-context package parity and then uses an isolated pinned-source proof: first-member weights/config match the reference, the entire serving source equals the reviewed deterministic transform, the second checkpoint/helper hashes match, no latent head exists, and only576 fully covered physical outputs may be reused. A deliberately changed ensemble coefficient must be rejected. Replay all9 thermal VALID cases through the actual package and also verify exact equality to the fixed independent-member average. Reaggregate official scores for585cases, preserve all predictions/provenance, and require exact neural/active/Arena/unfamiliar repeats before a full-score win. No ordinary audit is weakened and no partial physical prefix is reused. Controller/source hashes retained in output_ensemble_v1. + +## E101 portable parity result + +All four TRAIN-group contexts and both unfamiliar7x11/129x177 contexts (including13channels) are float32, finite, correctly shaped, exactly equal to the float64-accumulated average of independently executed original packages, and exactly repeatable. The partial-solver check with1/13channels also preserves each member’s distinct fallback exactly while halving physical step calls. Total learned parameters13,925,729, including one shared1,332,395-parameter guide/coarse set; both main backbones run on every case. Final predictSHA256 `a90fb248d84e8e1b27fb0d629cc6447509a8c233201f541fdf2a8b8d8a629ba7` and physical-cacheSHA256 `f5caf34dcfc52c37bd155420114054924200196833e6431e21ef62599cc37b24`. Six-context audit and source hashes retained in output_ensemble_v1/prototype. Fresh thermal/full585 validation is now running; no model publication from this correctness check alone. + +## Full validation of validation + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781879**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.436891 | + +Nine fresh complete thermal VALID rollouts, exactly equal to the declared half-and-half independent member average. Reuse576 complete physical outputs only under the pinned-source ensemble proof in full_reuse_audit.json: identical fixed physical inputs, no latent head, exact recording/replay, all translations complete, identical final transforms and equal average. No incomplete physical prefix reused. Official nine-case scores plus unchanged576 official case scores are reaggregated by exact scorer code; all585 prediction files retained. Full585 comparison and exact complete-package repeats precede publication. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E098b confirmation result + +Fixed gain0.1 versus unchanged gain0, six audited TRAIN trajectories, original canonical three-node physical settings and seed20261026. + +| TRAIN regime | Trajectory | Gain0 | Gain0.1 | +|---|---:|---:|---:| +| 1.0/-1.0 | 1 | 0.000686656862 | 0.000739129935 | +| 3.0/-2.0 | 3 | 0.005778065039 | 0.007371151402 | +| 5.0/-4.0 | 3 | 0.237885609992 | 0.239844936852 | +| 9.0/-3.0 | 3 | 0.007429069662 | 0.007360796030 | +| 15.0/-4.0 | 3 | 0.215661111139 | 0.215589949334 | +| 17.0/-5.0 | 3 | 0.572184448172 | 0.572194334429 | + +Mean0.173270826811 to0.173850049664, 2/6 improve. Frozen gain does not improve this confirmation mean. Do not deploy or validate this unmodified setting. All12 physical rollouts complete30steps. Low-coupling cases produce much larger inferred F6 increments and frequent amplitude limiting, suggesting insufficient regularization of weak observations. No model parameters, held-out data, VALID or inference package changed. Raw result SHA256 `ac54ba99f39d4bb4bae5f7a01e42402fa24fe70efaecf34ae806759e4d06590f`; full per-step/channel errors, innovation magnitudes and prediction hashes retained in temporal_moment_confirm_v1. + +## E102 — absolute observability ridge for the fixed context observer + +E098b’s frozen gain0.1 worsens the six-regime mean0.173270826811 to0.173850049664, with2/6 improvements. In weak-coupling regimes, the relative ridge yields large inferred F6 changes and frequent clipping. Change one factor: raise the fixed ridge floor from1e-8 to0.01, keeping gain0.1, relative ridge0.01, smoothing and all bounds unchanged. For dimensionless interval couplingH and lambda>=0.01, |H|/(|H|²+lambda)<=5, so gain0.1 limits pointwise inverse noise amplification to0.5 before smoothing/state bounds. No dataset-name or parameter-regime gating. + +Test exactly the same six audited TRAIN confirmation records with this one declared floor; no sweep, new seed, learned weights, optimizer, VALID or automatic deployment. Preserve original baselines. For a case whose three original ridge values already exceed0.01, the only changed clamp is inactive throughout encoding; verify unchanged kernel, exact source substitution, context/prediction hashes and recomputed score before reusing that identical prediction. Otherwise replay the complete30-step canonical three-node operator. Queue after the E101 ensemble validation/repeat/publication gate, prioritizing completion of that package. Source/protocol in regularized_moment_v1. + +## E101 — fixed two-member shared-model output average +- Hypothesis / changed factor: Equal output average of selected width48/E066 and width96/phase9 update2000 no-shrink, fixed0.5 before TRAIN pilot and VALID comparison. Both complete shared models execute every case, translation and channel chunk, with independent neural recurrence and per-member final corrections. Identical guides and physical computations are shared only within a case. Complete TRAIN/unfamiliar averages, repeatability and partial-prefix/chunk fallback are exact. Main seeds20261021 and extension20261110/training20261119; both checkpoint hashes in manifest. No new optimization, pretrained assets or inference-time fitting. All nine thermal VALID outputs fresh and exact to independently generated member averages;576 complete physical outputs reused only after pinned source/configuration/coverage/input proof. Full585 comparison plus exact repeat checks precede a win. The active state-head trainer retains its original single-model template. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `087b1a80f5bdd055700c84312cca54cc5773ad4f9a4d5fb39a9d3b5f98a1c633`. +- Validation: fixed 81-case subset; reward **0.812656**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.436891 | + +Selection basis: full585 physical-operator comparison: candidate0.78187855, current0.78182111 + +Outcome: improvement; published atomically. + +## E103 — shared ensemble weight calibration + +Hypothesis: one TRAIN-calibrated global output coefficient may improve the fixed E101 half-and-half ensemble. Pin both original scratch-trained members and all guides/corrections, width48 main SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, width96 main SHA256 `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. Fit one scalar across all30 steps and four channels on the original nine initial thermal TRAIN trajectories using exact VRMSE derivative bisection in [0,1]; first verify neither endpoint reaches the cap. Leave one cooling regime out to check scalar generalization against fixed0.5. Only if fitted and crossfit means improve, freeze the nine-case coefficient and confirm on all other63 thermal TRAIN trajectories, with no refitting. These cases were available to backbone training but are independent of the scalar fit. Every model still executes every case; no dataset-specific coefficient or inference-time fitting. No new seed, network weights, synthetic or pretrained assets. No VALID or submission change yet. Protocol/source hashes in ensemble_blend_v1; outcome pending. + +## Full validation of attention_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781861**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437105 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E097 — shared coarse attention, update 6000 +- Hypothesis / changed factor: Shared4x16 coarse attention, update6000, auxiliary initialization/sampling seed20270111. Four blocks, width128, four heads, metadata/time conditioning and boundary-aware relative bias;846459 parameters. Train from scratch on all19 datasets and10115 audited TRAIN records,12000 updates,batch512,four CPU threads, AdamW LR0.0003,100-update warmup and cosine decay, original interaction objective and fixed blend0.25. Training checkpointSHA256 5ed620a41088acd76b0b8ed619ad7656d975a6a44ab6f643070fb9278c69a07c. Every case evaluates the same shared branch. Candidate inherits the documented current parent; all nine thermal VALID rollouts are fresh, physical reuse follows dependency/input/coverage audits, and latent parents trigger fresh24-active replay with audited552 nonactive reuse. Full585 reports and exact neural/active/Arena/unfamiliar repeat checks precede a win. This architecture comparison also changes parameter count, seed and optimization settings. No pretrained assets, held-out data or inference-time fitting. Active training jobs retain their declared templates. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `3fc609f00c3eb08eb740026a939322330b55b21a404c8b49c7eb4a0c202a9762`. +- Validation: fixed 81-case subset; reward **0.812638**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437105 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E102 fixed-floor TRAIN result + +Only the observer ridge floor changes from 1e-8 to 0.01. Gain 0.1, relative ridge 0.01, smoothing, bounds and original quadrature seed 20261026 remain fixed. No learned checkpoint or optimizer. + +| TRAIN regime | Trajectory | Unchanged | Previous observer | Floor 0.01 | Exact reuse | +|---|---:|---:|---:|---:|---| +| 1.0/-1.0 | 1 | 0.000686656862 | 0.000739129935 | 0.000688046529 | False | +| 3.0/-2.0 | 3 | 0.005778065039 | 0.007371151402 | 0.005766271618 | False | +| 5.0/-4.0 | 3 | 0.237885609992 | 0.239844936852 | 0.241874032464 | False | +| 9.0/-3.0 | 3 | 0.007429069662 | 0.007360796030 | 0.007371246335 | False | +| 15.0/-4.0 | 3 | 0.215661111139 | 0.215589949334 | 0.215589949334 | True | +| 17.0/-5.0 | 3 | 0.572184448172 | 0.572194334429 | 0.572194334429 | True | + +Mean unchanged 0.173270826811, previous observer 0.173850049664, regularized observer 0.173913980118; 3/6 improve versus unchanged. Does not improve this TRAIN cohort mean; reject the fixed setting. These six TRAIN cases motivated the regularization change. They are not independent confirmation or VALID. No optimizer or new checkpoint. All forecasts complete 30 steps. Reuse is allowed only where all three original ridge values exceed the new floor, with source, context, prediction and exact recomputed score checks. Result SHA256 `1b7893058762c9e217596119e1b4b3055af7af10564019c5b116e2c66db34304`. No VALID scores or inference change. + +## E103 calibration result and frozen confirmation + +Nine TRAIN trajectories: fixed0.5 VRMSE 0.375775414587, fitted weight 0.7107390279678207 on the width96 member gives 0.375209351537, leave-one-cooling-regime-out mean 0.375388709017. Both declared gates pass. Scalar coefficient is now frozen; queued confirmation generates fresh complete predictions from both original E101 members on all other63 thermal TRAIN trajectories, then scores equal and fitted averages using the official float32 contract. No future values enter inference and no coefficient is refitted. Fixed inherited checkpoint hashes and fit SHA256 `7fa357fde954825d121410ea6ff340126e7438e75fb53a922bd0f69830089c74` retained. Confirmation is queued after E102 to limit GPU contention; main training remains unchanged. No VALID or deployment change. + +## E104 — complete-output reflection pilot + +Hypothesis: fixed horizontal reflection averaging may reduce orientation-dependent errors in the complete shared forecast. Pin E101 two-member package and its equal output coefficients, all runtime weights and corrections. For collocated [density,pressure,velocity_0,velocity_1] with periodicX/openY, reverse X and horizontal velocity, forecast from the four transformed context frames, then undo the reflection and average with the original full forecast at fixed0.5. Other metadata schemas would use identity; no dataset-name checkpoint routing. Each branch retains the complete original four-translation recurrence and final corrections. No new parameters, random seed or coefficient fitting. First use the original nine audited initial TRAIN trajectories, checking exact reflection involution and all source/input hashes. Positive pilot requires fixed-rule confirmation on the other63 TRAIN trajectories before VALID. Queue after E103 confirmation; no deployed model changes. Runtime/source hashes and protocol in reflection_output_v1. + +## E103 frozen-weight confirmation result + +All63 other initial thermal TRAIN trajectories, freshly audited against split membership and native caches, with complete float32 predictions from both pinned E101 members. Original training initialization/extension seeds retained; no new seed or neural checkpoint. Width96 weight 0.7107390279678207, frozen before confirmation. Official mean VRMSE: width48 0.376180351468, width96 0.374600858275, equal average 0.373317513623, fitted average 0.373352777188; 30/63 improve over equal averaging. Frozen coefficient fails confirmation; reject this weight without a VALID trial. These63 trajectories were available to backbone training, but excluded from the scalar fit. No refitting, VALID scores or deployment change. Full per-case results/predictions and hashes retained; result SHA256 `7d0ffbdd7f08a509474fd0999971c6d7f8f2a9870965cf752b8778604e387472`. + +## E104 reflection pilot result + +Original nine audited initial thermal TRAIN trajectories, pinned E101 two-member package and fixed reflection weight0.5, no fitted coefficients or new seeds/weights. Complete forecast original VRMSE 0.375775414370, reflected-and-transformed-back forecast 0.759461122154, average 0.516054711734; 0/9 improve. Reflection average fails the TRAIN pilot; reject this setting. Exact transform involution, component signs, input membership and runtime hashes verified. No VALID or serving change. Result SHA256 `9d0905283e6a52677a59fe2be5ee148afc11fe1c77e15d6ec795fa76e2a78795`, all predictions and official reports retained in reflection_output_v1. + +## E105 — full-horizon unresolved Gray–Scott context state + +Earlier fine-grid context assimilation measured only three forecast steps and found near-identical errors. Test whether retaining the unresolved state matters over the full26-step continuation. Use eight TRAIN trajectories, third/fourth authorized records for gamma,delta,alpha,theta. Keep the exact current double-precision2x-grid, ETDRK2,dtmax0.25, observation filter and solver coefficients. Variant initializes from supplied context0, advances each of three observed intervals, and replaces only the resolved deconvolved modes with context1/2/3, preserving unresolved modes. Baseline resets from context3 as currently deployed. No fitting, clipping, new parameters or seed. Fixed projections must reconstruct each supplied observation within1e-11 times field scale and leave inputs unchanged. Score all26steps of both canonical physical variants; this pilot omits translation averaging and neural work. Positive mean is necessary before broader TRAIN confirmation and a complete VALID package. No future frames enter initialization, no held-out or VALID reads, no inference change. Protocol, exact TRAIN cohort and original source hashes in gs_context_long_v1. + +## E105 full-horizon result + +Eight TRAIN physical comparisons: reset mean 0.03417093699225, retained-context mean 0.03417048656099; 4/8 improve. All16 rollouts complete26steps, all inputs unchanged, maximum observation reconstruction error 1.67e-15. Mixed and negligible full26-step effect:4/8 improve and the mean improves by0.00132%. Keep the simpler current operator; no further confirmation or VALID candidate. No new weights or seed, VALID scores or inference change. Result SHA256 `505482a10dda203085b57f7ded8f951cf8ab6c1fefd208160103f3023247de2e`; per-family/per-step results and predictions retained in gs_context_long_v1. + +## E106 — matched main reflection training + +E104 shows severe orientation sensitivity: original TRAIN9 VRMSE0.375775 versus reflected0.759461, fixed averaging0.516055. Test training augmentation, not the rejected inference average. Pin the current no-latent E101 parent for two matched first-member continuations, even if an independent experiment later becomes selected. Both begin from identical E066 width48 weights; keep width96 second member, E072 guide, E033 coarse branch, physical operators, four translations, equal output averaging and final corrections fixed. Both members execute every inference case. Train all19 datasets for1000 updates per arm, sequential control then reflection after the phase10 trainer exits normally. Batch2x3/full30(corpus26/Arena10),LR3e-5 originalAdamW/cosine,noise0.01,groups.05/.075/.075/.8. Same sampling/noise seed20270118 and independent reflection RNG20270119; only eligible whole-window horizontal reflection0.5 differs, using the already verified E076 sampler. No inference reflection is added. Inherited scratch initialization seed20261021; no pretrained or synthetic assets. + +Before queue activation require actual-parent full30 neural backward checks for both arms with finite gradients, bit-unchanged parameters and zero optimizer updates, plus exact initial-state/second-member package identity and correct13925729parameter/member-hash accounting. A private ensemble-aware installer avoids the known generic replacement-helper limitation. Freshly audit all10115 TRAIN records before optimization. Evaluate all four declared checkpoints at500/1000 through the complete pinned two-member package, replaying nine thermal VALID cases and auditing576 unchanged complete physical outputs. Full585comparison is against the actual current selected package, including any newer learned-state winner. Winning neural changes require exact three thermal and two novel-schema repeats; physical source/configuration/results remain proven unchanged. Phase10 is untouched. Sources/protocol in reflection_training_v1; preparation and validation pending. + +## E106 full-rollout preflight result + +Actual E101 first-member weights pass full30-step native thermal forward/backward in both control and reflection arms. Control loss0.22673711, gradient norm13.13635; reflection loss0.62737024, gradient norm18.01826. All gradients finite and every model parameter bit-unchanged; zero optimizer updates. Each process peaks at30.0204GiB, and the shared H100 remains within80GB. The private package installer preserves every initial parameter and the second-member bytes, reports13925729parameters and updates member0hash correctly; physical dependency audit reports no changes. Source snapshots and preflight checks retained. Controller and validator are queued behind phase10 normal trainer completion; no new main optimizer has started. + +## E107 — late spatial-convergence screen + +Hypothesis: earlier three-step Gray–Scott convergence tests may miss high frequencies generated later. Change only the internal grid from2x to4x in the current double ETDRK2/dtmax0.25 operator; reset from context3 as currently deployed. Use all eight E105 TRAIN trajectories and all26 forecast steps. Reuse2x controls only after exact source, context, target, prediction hashes and recomputed scores agree; fresh4x forecasts and late Parseval-weighted high-frequency energy are recorded. No new weights, seed, parameter fit, context-retention layer, neural/translation averaging, VALID or serving change. Any positive effect must justify expanded TRAIN confirmation and runtime cost before integration. Concurrent H100 timings are descriptive, not an isolated speed or inference-budget claim. Protocol/source hashes in gs_resolution_long_v1. + +## E107 full-horizon resolution result + +Eight TRAIN trajectories:2x VRMSE 0.03417093699225,4x 0.03416989803292; 5/8 improve. Five of eight improve, but mean benefit0.00304% is negligible relative to added cost. Retain2x; no expanded confirmation or VALID candidate. At step26, the largest fluctuation-energy fraction above the2x nonlinear cutoff is 8.022e-13. This indicates negligible unresolved spatial energy in these predicted trajectories, not agreement of the PDE model with truth. All26 forecasts complete, exact prior-control reuse audit passes. No new seed/checkpoint, VALID or serving change. Result SHA256 `1caeee2cf93686ceebdc63b0d020cd2ae443f0cb874da6fc8ef582d828b6dcdd`. + +## E108 — frozen-density acoustic profile + +Hypothesis: unresolved temporal sampling of vertical acoustic motion contributes to pressure-profile error. From the supplied last two frames, form X-mean pressure and vertical-velocity increments, freeze the last density profile, and propagate linear acoustics p_t=-K*v_y, (rho0*v)_t=-p_y with K=(5/3)*last mean pressure. Use nativeY Rusanov finite volumes, SSPRK2/CFL0.4 and zero incoming acoustic characteristics at both open boundaries. Sum propagated increments onto the last pressure profile; no future fields enter this calculation. Fixed gamma, numerics, initialization and boundary rule; no sweep. + +Test one scalar convex correction to only the centeredY pressure profile of E101, preserving pressuremean, interaction and allotherchannels. Fit on the originalnine TRAIN trajectories with the exact clamped pressure VRMSE objective, predeclared1001-point scalar grid then48 local ternary refinements. Leave-one-cooling-regime-out check must also improve before coefficient freeze and confirmation on the other63TRAIN trajectories. All are authorizedTRAIN; no VALID used. First require manufactured-wave convergence/open-boundary escape and exactzero-increment persistence; these are checks, not synthetic training. CPU1thread, no new randomseed or networkweights, no separately trained dataset model, no inference-time fitting. Prototype is not deployed. Protocol/source hashes in acoustic_profile_v1. + +## E108 acoustic-profile result + +Manufactured-wave relativeL2 errors0.117684 at128cells and0.064057 at256cells; outgoing pulse relative norm6.97e-26 after escape, zero-increment persistence exact. These numerical checks pass without training data or optimizer. On allnine authorized TRAIN trajectories, the global scalar optimum is exactly0. Baseline and fitted pressure VRMSE both0.795080431354; leave-one-cooling-regime-out0.795080431354. Global TRAIN scalar optimum is zero: reject this acoustic profile correction. No confirmation or VALID. No further coefficient, gamma or boundary sweep, checkpoint or inference change. Raw full-profile replacement results, directions and protocol are retained, SHA256 `42385fb19741494ea8d7c222e21a3f60b646c80756d0214bcf1497715593e212`. + +## E109 — zero-residual bottleneck attention preparation + +Hypothesis: content-dependent global mixing at the coarsest shared U-net grid may improve long-range pressure/field interactions beyond the existing input-only diagonal Fourier mixer. Add one optional four-head attention residual at width288 (native thermal16x48tokens), LayerNorm/QKV/SDPA with no dropout, and an exactly zero output projection.332352newparameters, initializationseed20270126. Existing boundary-aware coordinates/features are retained; no extra position encoding. Only the first E101member is extended; the frozen96member retains its exact original architecture via optional constructorflagFalse. All members execute every case. + +Use a forked CPU generator for the newparameters, preserving original CPU and CUDA sampling/noise RNG sequences. Pin the same no-latent E101 parent as E106. If zero-function, realTRAIN/novel-shape, repeatability, full30backward/resource and complete-physical-reuse checks pass, train one1000-update candidate with E106control’s exact no-augmentation seed20270118/sampling/optimizer schedule and reuse its500/1000 control checkpoints, avoiding redundant control training. Proposed candidate runs after bothE106arms; full585 comparisons againstactualcurrentbest, with freshthermal predictions and rigorously audited physical reuse. No new optimizer or queue yet; no pretrained/synthetic assets, VALID or deployment change. Protocol and sources in bottleneck_attention_v1. + +## E109 conditional training queue + +The optional branch is prepared with332352newparameters (firstmember3398979,total14258081). Original base parameter values and CPUconstructorRNG state are exact; all new output-projection weights are zero. The second member retains its original architecture and checkpoint. The private source audit permits only the reviewed neural constructor/optional residual changes and unchanged complete physical work, never partial prefixes. Actual six-caseTRAIN/unknown parity, repeated forecasts and full30backward checks are still running. + +Queue one1000-update no-augmentation attention continuation only after those gates pass and bothE106arms finish. Reuse E106control at500/1000, with exact same parent, data sampler, seed20270118, batch2x3/full30, learning rate, noise, optimizer and schedule. At launch freshly verify all10115TRAIN memberships, immutable source hashes, actual E106control configuration, identical guides and second-member bytes. Independently initialized attention parameters useCPU seed20270126 without consuming CUDA noise RNG. No new optimizer is running yet. All9thermalVALID forecasts will be fresh per checkpoint,576 complete physical outputs require the restricted source proof plus input/coverage audit, and full585selection compares the actual current best. No changes to phase10, E106 or deployed inference. Launch/validation/source snapshots in bottleneck_attention_v1. + +## E109 zero-residual preflight result + +All six complete package predictions (four audited TRAIN groups plus two unfamiliar schemas, including13channels and odd grids) exactly equal the E101parent and an independent repeat. Original CPU and CUDA constructor RNG states are preserved; the frozen96member loads its exact original architecture/state. Firstmember3398979parameters, fullensemble14258081. Full30native thermal backward has identical initial forward loss0.2267371118068695 to E106control, finite gradients, bit-unchanged parameters and zero optimizer updates; peak30.190510GiB. Total gradient norm13.13464737 is not a gradient-parity test because the parameter set differs and ordinary convolution backward is not strictly deterministic. No optimizer-equivalence claim. The reviewed constructor-only serving change and unchanged physical assets/configuration pass the restricted complete-prefix reuse proof. Preflight result SHA256 `fffb9b83a0ab2844641175ea79be36d725c18de28fb2e19d608c32b662f4c8f0`. Controller/validator remain queued behind E106 completion; no new optimizer or VALID trial yet. + +## E110 — full-horizon shared-state backward resource check +- Hypothesis / changed factor: Nested recomputation permits a full30 physical loss to backpropagate through the actual jointly trained shared neural state head alongside a full30 native thermal objective on one H100. This isolated no-update check combines one full30 native thermal batch2 backward with one full30 nested physical backward through the actual phase10/update1000 state head. No source, optimizer or cadence in the active run changes. +- Seed: 20270118. All10115 TRAIN memberships and all175 physical base/target hashes checked before loading the selected TRAIN labels; physical stress case is zeta17/alpha-5 trajectory2. +- Checkpoint SHA256: `0fc425a0cf92a0d25cc8beae5b2cb291cef63f56f6c17c7f9ebd84342e06848d`. This is an existing session-trained candidate used for feasibility, not a new selected checkpoint. +- Validation VRMSE per dataset: not run; no VALID or held-out data used. +- Acceptance: finite combined/head gradients, nonzero head gradient, peak allocation below40GiB, all parameters bitwise unchanged. Zero optimizer updates; no training queue created. A future parent and matched training comparison remain undecided. +- Outcome: prepared; resource/gradient check pending. Sources and protocol retained in `long_joint_preflight_v1`. + +## E111 — fixed short-horizon tangent decoder TRAIN screen +- Hypothesis / changed factor: A fixed short-horizon tangent decoder from the four input frames may reduce initial forecast error before the learned long-horizon forecast takes over. Only complete neural-output decoder: fixed four-frame slope(-3*x0-x1+x2+3*x3)/10, last-frame anchor, convex weight .5*exp(-(k-1)/3) at forecast step k=1..K. Tangent density/concentration projected nonnegative. No fitted coefficients or sweeps. +- Pin E101 main hashes `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61` and `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`; all runtime assets hashed. No new seed or weights; inherited scratch initialization lineage. +- Original9 TRAIN pilot, then only if positive confirm the unchanged rule on other63 TRAIN trajectories, all with exact cache/input/target audits and complete float32 official-formula VRMSE. No coefficient fit, sweep, VALID or held-out data. +- A positive screen is not deployable by itself: it requires a generic neural-output decoder preserving physical prefixes, complete runtime parity, full585/current-best comparison and repeated predictions. Both shared ensemble members continue to execute every case. +- Validation VRMSE per dataset: not run. Outcome: screen pending. Decoder SHA256 `01dd101db8582c1b1a8fbb88e2e5cc5c0acc729c2aa7b99949c8dfd85f6846d6`; source/protocol retained in tangent_decoder_v1. + +## E111 fixed tangent TRAIN result + +Nine initial TRAIN trajectories, no fitted coefficients or new weights/seeds. Complete E101 mean VRMSE 0.37577541437035 → 0.41652274850307 with the fixed four-frame tangent decoder; 0/9 improve. First-step mean error rises0.214869→0.428722, and all four channel means worsen. The TRAIN pilot fails; reject this rule without63-case confirmation, portable implementation or VALID. This does not establish that all learned temporal corrections fail. All inputs/targets exactly match authorized native TRAIN caches; repeated outputs exact and baseline official-formula difference<1e-15. Parent checkpoint hashes unchanged. Result SHA256 `2ca8df7b09cd549f5499f411cf257e1f30d8adc688964bd30b8efffe1e231c01`; no submission behavior change. + +## Full validation of attention_012000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781839**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437374 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E097 — shared coarse attention, update 12000 +- Hypothesis / changed factor: Shared4x16 coarse attention, update12000, auxiliary initialization/sampling seed20270111. Four blocks, width128, four heads, metadata/time conditioning and boundary-aware relative bias;846459 parameters. Train from scratch on all19 datasets and10115 audited TRAIN records,12000 updates,batch512,four CPU threads, AdamW LR0.0003,100-update warmup and cosine decay, original interaction objective and fixed blend0.25. Training checkpointSHA256 3dc787950f3010c2e2ba43802d28cbb79d06d8155598347f01883aeb1a8ef675. Every case evaluates the same shared branch. Candidate inherits the documented current parent; all nine thermal VALID rollouts are fresh, physical reuse follows dependency/input/coverage audits, and latent parents trigger fresh24-active replay with audited552 nonactive reuse. Full585 reports and exact neural/active/Arena/unfamiliar repeat checks precede a win. This architecture comparison also changes parameter count, seed and optimization settings. No pretrained assets, held-out data or inference-time fitting. Active training jobs retain their declared templates. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `e747ba2901a6388c86c04a5883c6ecbb37d403058989cc125802e1bb2e1b263a`. +- Validation: fixed 81-case subset; reward **0.812616**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437374 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E112 — matched shared profile attention + +Content-dependent global coupling between profile locations may improve broad thermal structure beyond the flat shared mean/profile MLP. Pin E101 for matched control and attention continuations of its common E072 mean/profile checkpoint SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`. Both complete neural ensemble members keep their weights and share this auxiliary on every case. Add two width96/four-head attention blocks over32 profile tokens, four normalized context profiles/allchannels per token, axis identity/nonperiodic positions, conditioning on the existing192-wide context/metadata/time hidden feature. Mean/profile output projections start at zero. New230710parameters; auxiliary775297, complete ensemble14156439 for the candidate. Control disables/freezes the same zero branch during training and strips it at packaging. + +Sequential CPU-only6000updates per arm,4threads,batch512,AdamW5e-5 cosine to10%,original reconstructed-field VRMSE, groups.1/.1/.1/.7,initialwindow.5,existing dropout.1/periodic augmentation/eligible reflection.5 for both. Sampling/dropout/reflection/periodic seeds20270202/03/04/05; newattention initialization20270206 in a CPU-only RNG fork. Inherited random initialization20261031; no pretrained or synthetic assets. All10115TRAIN memberships freshly audited. Only the attention residual differs between the matched arms. + +CPU and CUDA preflight outputs exactly match the parent across four groups and unknown IDs with1/4/11channels; constructor CPU RNG unchanged, all original weights exact, finite nonzero adapter gradients(.3136894/.3136899), all parameters bitwise unchanged, zero optimizer updates. Fixed decoder blends remain. Select only2k/6k: fresh9thermal VALID plus576 complete unchanged physical outputs under explicit source/config/input/coverage proof; full585 comparison against actual current best and exact neural/unknown repeats before a win. Pinned no-latent parent even if independent experiments later win. Validation VRMSE per dataset pending; training now authorized by the passed preflight. Sources/protocol/checks retained in profile_attention_v1. + +## E110 full-horizon shared-state backward result + +Actual phase10/update1000 shared weights, seed20270118, one full30native thermal batch2 and strong17/-5 TRAINtrajectory2 physical30 loss. All10115TRAIN memberships and175physical base/target hashes pass. Main loss0.23039406538, physical loss0.638578653336; main gradient norm17.561602, physical norm0.672068321 (head0.66953244), cosine0.000317889077. All gradients finite, head nonzero, every parameter bitwise unchanged; zero optimizer updates. Peak30.0204GiB. Main/physical elapsed24.88/574.36s while sharing the H100, not an isolated throughput measurement. + +Full30 joint backward is feasible at these trained weights. This is not a validation/generalization result and does not select a model or queue a longer-supervision trainer. A future parent/cadence/matched objective comparison still needs declaration. Active phase10 protocol unchanged. No VALID or new checkpoint. Result SHA256 `bbbec107d32b948b098314003ec684a1253ad0743713ab24be668eae43fadc39`; source/model/target provenance and all checks retained. + +## Full validation of control_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781821**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437589 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E112 — shared profile attention control, update 2000 +- Hypothesis / changed factor: Matched shared mean/profile continuation control,update2000/6000, pinned E101 complete ensemble. All19 datasets/10115TRAIN records, CPU4threads,batch512,AdamW5e-5cosine,original reconstructed-field VRMSE, groups.1/.1/.1/.7,initialwindow.5,dropout.1,periodic augmentation and eligible reflection.5 in both arms. Sampling/dropout/reflection/periodic seeds20270202/20270203/20270204/20270205; new attention seed20270206. Only the candidate trains two width96/four-head blocks over32 profile tokens, conditioned on the existing context/metadata/time hidden feature, with zero initial mean/profile projections. Control freezes/disables that residual and strips it from the runtime. Both ensemble members share this same statistical model on every case; main checkpoints, all corrections and output blends stay fixed. Training checkpointSHA256 8bb06dc5e23e41950fc4cd630151154ad4d932fa27b8c0902a5dc62e62dc7018, packaged auxiliarySHA256 1a116d5fe60af4b21b1a63b9c447127a42b3d90ab7ed69af6afcf910523c9b2e. Fresh9thermal VALID predictions plus576 audited complete unchanged physical outputs; full585 compare to actual current best. Exact neural/unfamiliar repeats before a win. No pretrained/synthetic assets, held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `dc47ce38f0a90e6b4ba86fef0b281b9865673558f14ca90ec9bda1565c6d9caa`. +- Validation: fixed 81-case subset; reward **0.812598**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437589 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of control_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781829**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437486 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E112 — shared profile attention control, update 6000 +- Hypothesis / changed factor: Matched shared mean/profile continuation control,update6000/6000, pinned E101 complete ensemble. All19 datasets/10115TRAIN records, CPU4threads,batch512,AdamW5e-5cosine,original reconstructed-field VRMSE, groups.1/.1/.1/.7,initialwindow.5,dropout.1,periodic augmentation and eligible reflection.5 in both arms. Sampling/dropout/reflection/periodic seeds20270202/20270203/20270204/20270205; new attention seed20270206. Only the candidate trains two width96/four-head blocks over32 profile tokens, conditioned on the existing context/metadata/time hidden feature, with zero initial mean/profile projections. Control freezes/disables that residual and strips it from the runtime. Both ensemble members share this same statistical model on every case; main checkpoints, all corrections and output blends stay fixed. Training checkpointSHA256 de4c90fb94df0ab9882b4a083fbecab081665d3870ceca6f29efa20ffd9b764b, packaged auxiliarySHA256 e02e6de6b0f086b34b6837711acdd6e401e398d44df465e7558f22d41f893ffc. Fresh9thermal VALID predictions plus576 audited complete unchanged physical outputs; full585 compare to actual current best. Exact neural/unfamiliar repeats before a win. No pretrained/synthetic assets, held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `2c7dcde60e8480183ccc816be93899dfd8a0b7cceecbdc6a7d2721757d7b2547`. +- Validation: fixed 81-case subset; reward **0.812606**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437486 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of attention_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781846**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437289 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E112 — shared profile attention attention, update 2000 +- Hypothesis / changed factor: Matched shared mean/profile continuation attention,update2000/6000, pinned E101 complete ensemble. All19 datasets/10115TRAIN records, CPU4threads,batch512,AdamW5e-5cosine,original reconstructed-field VRMSE, groups.1/.1/.1/.7,initialwindow.5,dropout.1,periodic augmentation and eligible reflection.5 in both arms. Sampling/dropout/reflection/periodic seeds20270202/20270203/20270204/20270205; new attention seed20270206. Only the candidate trains two width96/four-head blocks over32 profile tokens, conditioned on the existing context/metadata/time hidden feature, with zero initial mean/profile projections. Control freezes/disables that residual and strips it from the runtime. Both ensemble members share this same statistical model on every case; main checkpoints, all corrections and output blends stay fixed. Training checkpointSHA256 fb30b4dede1335c91a8611ef9022496f5ae62499a15096a7d8d455731d908db3, packaged auxiliarySHA256 550d3ce23a8ad45d9b1ae88118056ba2d2e0a5d31074ada1d59989102f671ef4. Fresh9thermal VALID predictions plus576 audited complete unchanged physical outputs; full585 compare to actual current best. Exact neural/unfamiliar repeats before a win. No pretrained/synthetic assets, held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `2c1ef4e9ec67af7732f53a15b339ec40386c518854a83027c5f84f6ccb73f2a0`. +- Validation: fixed 81-case subset; reward **0.812623**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437289 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E112 portable zero-residual result + +Both control/candidate auxiliary installers preserve every original weight and both main checkpoint bytes, with correct total counts13925729/14156439 and no physical dependency change. Complete zero-initialized attention package exactly matches the pinned E101 reference and independent repeated execution on all six TRAIN/unfamiliar cases, covering four groups and unfamiliar13-channel input. All original TRAIN input hashes and memberships freshly checked. Only zero-step diagnostic auxiliary artifacts were created; they are not selection candidates. New initialization seed20270206; inherited main/mean hashes unchanged. No VALID used by this check. Result SHA256 `462ae4538bee1375ce74d469d6e220aa7d04d62ad9a6fde559fc4d1ed9845098`; portable_preflight retains source/input/install/forecast evidence. + +## Full validation of attention_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781839**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437366 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E112 — shared profile attention attention, update 6000 +- Hypothesis / changed factor: Matched shared mean/profile continuation attention,update6000/6000, pinned E101 complete ensemble. All19 datasets/10115TRAIN records, CPU4threads,batch512,AdamW5e-5cosine,original reconstructed-field VRMSE, groups.1/.1/.1/.7,initialwindow.5,dropout.1,periodic augmentation and eligible reflection.5 in both arms. Sampling/dropout/reflection/periodic seeds20270202/20270203/20270204/20270205; new attention seed20270206. Only the candidate trains two width96/four-head blocks over32 profile tokens, conditioned on the existing context/metadata/time hidden feature, with zero initial mean/profile projections. Control freezes/disables that residual and strips it from the runtime. Both ensemble members share this same statistical model on every case; main checkpoints, all corrections and output blends stay fixed. Training checkpointSHA256 ac7e91bb740cf31ef3ef08ce6ee18cb3337fd2b1d0612f23bfa88d8121440702, packaged auxiliarySHA256 4d8a39551be249c0f0ba21986f01186904729238b31807e3a345378a04b3339d. Fresh9thermal VALID predictions plus576 audited complete unchanged physical outputs; full585 compare to actual current best. Exact neural/unfamiliar repeats before a win. No pretrained/synthetic assets, held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `9736877bbdec75c0beedc9495ef18bac9633770f9cd7c10f3605d251053dce4e`. +- Validation: fixed 81-case subset; reward **0.812616**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437366 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E112 descriptive auxiliary versus complete-rollout analysis + +All9 thermal VALID cases, no fitting or model selection in this analysis. The auxiliary-only field uses canonical unshifted context statistics and its predicted mean/axis profiles; complete rollout scores include both neural members, translations and all corrections. These are distinct computations. Exact native target interaction floor for axis-additive reconstruction is0.45076384; this diagnostic target projection is never used at inference. + +| Candidate | Auxiliary field VRMSE | Mean normalized absolute error | Profile normalized RMS error | Complete thermal VRMSE | +|---|---:|---:|---:|---:| +| parent | 0.5616377 | 0.12494179 | 0.25973191 | 0.43689118 | +| control_002000 | 0.56244575 | 0.12616602 | 0.25977492 | 0.4375891 | +| control_006000 | 0.56214908 | 0.1256348 | 0.26012361 | 0.4374858 | +| attention_002000 | 0.56039406 | 0.12682445 | 0.25474187 | 0.43728877 | +| attention_006000 | 0.55920671 | 0.12373633 | 0.25365633 | 0.43736645 | + +All checkpoint hashes, per-case/per-step values and standalone PNG/SVG retained in profile_attention_v1/analysis. This comparison does not establish a causal mechanism for the complete rollout. + +## E113 — separate recurrent guidance from final statistics + +E112 final attention improves standalone VALID profiles(.2597319→.2536563 normalizedRMS) and auxiliary field reconstruction(.5616377→.5592067), but its complete thermal rollout regresses(.4368912→.4373665). Hypothesis: changing the guide seen by the frozen recurrent backbone can offset decoder gains. Use exactly E112 attention6000 for final mean/profile corrections while restoring original E072 guidance inside every autoregressive step. No checkpoint/blend sweep or new optimizer. Both shared statistics models and both main members execute every case/translation/channel chunk, with no dataset-name checkpoint routing. All physical operators, main weights and remaining corrections/blends stay fixed. + +Guide SHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`, decoder SHA256 `4d8a39551be249c0f0ba21986f01186904729238b31807e3a345378a04b3339d`; inherited E112 initialization20270206/training20270202 and original session-trained main/E072 lineage, no new random weights or pretrained assets. Total14701026parameters. Exact source transform and a zero-decoder complete TRAIN/unfamiliar parent-parity check precede new inference. Expanded latent dependency guard now detects either guidance_model.py or guidance_weights.pt changes; equal/deliberately changed temporary fixtures pass. This adds conservative dependencies and does not permit any new reuse. + +First replay joint E112 and decoupled E113 complete packages on original9 initial thermal TRAIN windows against exact cached E101, all membership/input/target/baseline-average audits fresh. Only if the fixed candidate mean improves E101, confirm it unchanged on the other63TRAIN trajectories; no refitting. These trajectories were available to the model/auxiliary training. No VALID before both gates pass; a later win requires restricted complete-physics reuse proof, full585/current-best comparison and exact neural/unfamiliar repeats. No zero-step diagnostic checkpoint may be deployed. Validation VRMSE per dataset pending; preparation complete and preflight/TRAIN study pending. Sources/protocol retained in decoupled_statistics_v1. + +## E114 — late Gray-Scott temporal convergence + +Small temporal errors may accumulate over a complete26-step Gray-Scott forecast despite earlier short-horizon convergence. Change only the fixed internal maximumtime step.25→.125 of the existing double ETDRK2/2xgrid operator. Reuse audited E105 reset controls on the same eight TRAIN trajectories and all26steps; exact source/context/target/prediction/recomputed-score proof precedes reuse. No context retention, space-grid change, neural/translation averaging, new seed/weights, parameter fit or VALID. Gate for broaderTRAINconfirmation: at least0.1%mean relative reduction and5/8improving cases; no timestep sweep. Parent main/solver hashes unchanged, with runtime config factor declared above. Shared-H100 timings are descriptive; any complete-package cost still needs assessment before adoption. Validation VRMSE per dataset not run. Outcome pending, protocol/sources retained in gs_time_long_v1. + +## E114 full-horizon temporal result + +All8TRAIN cases worsen when the fixed maxstep is halved: mean VRMSE 0.03417093699225→0.040213565006868 (17.684%higher),0/8 improve. The declared gate fails; retain maxdt.25, no timestep sweep, broader confirmation or VALID. Finite full26outputs, exact source/context/target/prediction/recomputed-score control reuse checks pass. No new weights or seed. Initial launch had a Python keyword SyntaxError before any data/model reads; original source/log archived, syntaxonlyfix recompiled, scientificprotocol/gatesunchanged. Result SHA256 `5b99f51cf8c45ebb97e7053cdad70c8cfc85b3d385c5326706139bde24581c82`; all predictions and provenance retained in gs_time_long_v1. + +## E113 conditional validation controller + +The fixed candidate may enter VALID only after exact zero-decoder parent parity and both original9/other63TRAIN mean-improvement gates. The controller exits without reading VALID if either gate fails. A passing candidate gets9fresh thermal forecasts,576completephysicaloutputs only under exact source/config/input/coverage proof, full585 comparison to current best and exact trained neural/unfamiliar repeats before publication. Guides/decoder/mainweights/blends remain exactly frozen from preparation; no new search or optimization. Controller source hashes retained. + +## E115 — occasional full-horizon shared-state supervision + +Hypothesis: full30 physical supervision may improve late active dynamics unconstrained by four-step state loss. Match the existing phase10/E060 control exactly at initialization, guide, sampler/noise/physical draws and optimizer: initialSHA256 `1a082bab4c1db8eae9a5c0fb6fa8cec159de67ba2e7ec9b9b7abe800844c8989`, guideSHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`, zeroheadseed20261126, trainingseed20261127/physical20261128,2000updates,batch2x3,full30,LR3e-5/head10x,noise.01,groups.05/.075/.075/.8. Change only physical objective horizon every64th main update:31full30 losses replace short4, other219physical losses stay4, every8 cadence/weight1 unchanged. Reuse the existing short-loss control; do not train it again. No synthetic/pretrained assets. + +Nested activation recomputation has already passed controlled original-short gradient parity and full30 joint backward through the trained shared model(E110,30.0204GiB). Initial parameters exactly match both originalphase10 and earlier E094zero-head checks; model/loss/kernel source is identical to E110. No new GPU preflight was needed for unchanged graph shapes/architecture. Fresh10115TRAIN memberships and all175long/base hashes pass; every existing phase10 physical file/trajectory/even-shift draw matches the deterministic candidate schedule. The31 long draws span21parameter files. All runtime/template assets and fixed guide remain the original single-model state-head architecture; no ensemble or selected-parent rebasing. + +Queue only after E106/E109 main training and phase10 full validation finish normally; one main GPU optimizer at a time. Several additional training hours expected. Save1k for recovery/audit only; select only final2k with fresh24active+9thermal VALID and552strictly audited unchanged corpus/Arena outputs. Compare complete585 against actual current best and matched phase10/update2000; exact active/neural/unfamiliar repeats before a win. Validation VRMSE per dataset and trained checkpoint hash pending. Protocol/source/lineage/long-target audits and physical schedule retained in long_state_supervision_v1. + +## E100b — initial versus later thermal gradient alignment + +Reuse the immutable37-gradient E100 artifact; verify its SHA256, parent/guide hashes and allTRAIN memberships. Compare matching9thermal trajectories atstarts0/34, full30batch1/no noise, seed20270113; no new gradient evaluation or optimizer. Hypothesis: later windows may locally oppose the initial-window objective. Initial/later mean gradient norms3.153638/3.8079253, cosine0.05323798, later projection on the initial mean divided by initial norm²=0.064283298. 3/9 paired initial/later cosines are negative. This local smallTRAINsample does not determine an optimal sampling probability or finite-update/VALID effect. No new sampler training queued by this analysis. All paired losses/norms/cosines and artifact hashes retained in group_gradient_v1/temporal_alignment. No VALID or new checkpoint. + +## E113 fixed guidance/decoder TRAIN result + +Exact zero-decoder complete parent parity passes on all six TRAIN/unfamiliar cases, with strict reviewed source and runtime-asset proofs. Pilot9 official complete VRMSE: baseline 0.37577541437035, joint 0.37480059414003, candidate 0.37483030273231; 6/9 improve over E101. Frozen other63TRAIN confirmation: baseline 0.37331751362258, candidate 0.37219334454229; 49/63 improve. Both TRAIN gates pass; candidate may enter the declared full585 VALID comparison. No new seeds/weights or coefficient fitting, no held-out access. All inputs and targets exactly match authorized native TRAIN caches; cached E101 predictions exactly equal the fixed independent-member average and reproduce official scores. Pilot result SHA256 `e79eb05c3f9d572bbda805728cf27d9eac8c637d89833fbf9b6ecb4cde6f6a20`. Confirmation SHA256 `ac56d3fdfdbb92d70b7faa179d7cc383a8a2b01ddc470e7295d967b2578bec8a`. All source/input/prediction/audit artifacts retained in decoupled_statistics_v1. + +## Full validation of validation + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781845**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437292 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E113 — original recurrent guide with attention final decoder +- Hypothesis / changed factor: Keep the original E072 shared statistical guidance inside each autoregressive step, while using exactly E112attention6000 for the final mean/profile decoder. Both shared statistical models and both main members execute every case/translation/channel chunk. All main checkpoints, physical operators, output coefficients and remaining corrections are fixed. Inherited E112newparameterseed20270206/training20270202; no new optimizer, coefficient fit, synthetic or pretrained asset. GuideSHA256 3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c, decoderSHA256 4d8a39551be249c0f0ba21986f01186904729238b31807e3a345378a04b3339d, total14701026parameters. Frozen complete architecture passed original9TRAIN and other63TRAIN mean-VRMSE gates without refitting. Exact zero-decoder six-case TRAIN/unfamiliar parent parity and restricted source/asset proof passed. Fresh all9thermal VALID predictions plus576 fully covered unchanged physical outputs with input/source/config/coverage audits; full585 compare against actual current best and exact trained neural/unfamiliar repeats before a win. No held-out access or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `a7cd48983ca9f17bdbb7333cad44f4791838fe77a2316eca374be234bcbbd7e2`. +- Validation: fixed 81-case subset; reward **0.812623**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437292 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E116 — locked hidden-feature dropout preparation + +Hypothesis: training regularization may improve the shared autoregressive member without injecting different random forcing at each forecast step. Only changed factor: dropout0.1 on the hidden GELU features in all14 CNext blocks, independent per example/channel, shared spatially and locked for every step of each microbatch. New independent mask seed20270211; sampling/noise20270118 and unused reflection RNG20270119 exactly match E106 control. Reuse that control; train only one1000-update no-reflection candidate after E115, full30/batch2x3/AdamW LR3e-5/noise.01/groups.05/.075/.075/.8. Pinned E101 width48 first member is trained; width96 and E072/E033 guides remain fixed. Inference has no masks, hooks, new parameters or code changes. Parent first SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`; same initial artifact as E106 `956092907d8d7d163de708e1be1c6fa7e5b369bbdc6954dc48178acf351efbc0`. Validate500/1000 on all9 thermal cases fresh, audit576 complete unchanged physical predictions, and compare full585 against current selection and matched control; repeat neural/unfamiliar inference before a win. Preflight used zero optimizer updates. Import-order failure before model/data access was fixed and archived. The original cross-process equality-to-historical-loss gate failed. An untouched E106 check-only rerun also differs from its historical loss under unchanged cuDNN benchmarkTrue. Direct constructor/mask global RNG checks and same-process complete30 forecasts with disabled hooks versus removed hooks are exact. Thus control parity is established directly without assuming cross-process kernel-choice reproducibility. No training hyperparameter, data, mask rule, inference arithmetic or selected checkpoint changed. The original failed preflight and untouched control rerun are retained; no optimizer updates occurred. Native B2/full30 control loss0.22610411047935486, grad13.41086483001709, peak30.020408153533936GiB; all gradients finite and parameters bit unchanged. Native B2/full30 dropout loss0.42838016152381897, grad19.64864158630371, peak29.972065925598145GiB; all gradients finite and parameters bit unchanged. All14 masks preserve identity across30 uses, global CPU/CUDA RNG is unchanged, eval/disabled paths exact, and new windows receive new masks. Validation per-dataset VRMSE: not yet applicable; training queued only after successful preflight. No pretrained/synthetic assets or held-out access. Preflight result SHA256 `0b3e68be1e806ba10c6da1a74423cd6ac4f2c72dc8d709f5eb1234a399a1e207`. + +## E117 — current ensemble error diagnosis + +All72 initial thermal TRAIN and9 initial thermal VALID trajectories, with fresh pool/native-input/native-target/official-score audits and exact TRAIN equal-member identity. Reuse cached E101 complete forecasts; no new inference, seed, coefficient fit, optimizer or selection. First member SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. Exact float64 squared-error decomposition into mean, X profile, Y profile and interaction sums within1e-10. Overall TRAIN VRMSE 0.373624751216; VALID 0.436891175623. Other dataset metrics remain the selected E101 values. + +| Channel | TRAIN VRMSE | VALID VRMSE | TRAIN mean / X / Y / interaction shares | VALID shares | +|---|---:|---:|---|---| +| density | 0.15033264 | 0.17460174 | 0.0029 / 0.0265 / 0.0823 / 0.8883 | 0.0025 / 0.0302 / 0.0812 / 0.8862 | +| pressure | 0.76348992 | 0.88917316 | 0.1004 / 0.0243 / 0.1886 / 0.6868 | 0.2213 / 0.0206 / 0.2372 / 0.5209 | +| velocity_0 | 0.17302387 | 0.20265762 | 0.0128 / 0.0122 / 0.2134 / 0.7615 | 0.0152 / 0.0140 / 0.2498 / 0.7210 | +| velocity_1 | 0.40765257 | 0.48113218 | 0.0330 / 0.0783 / 0.0506 / 0.8381 | 0.0460 / 0.0852 / 0.0647 / 0.8041 | + +First-step errors are close across cohorts. Pressure VRMSE in steps13–30 is0.8539627 TRAIN versus1.0294060 VALID; the gap emerges during rollout. Mean plus Y-profile errors account for45.8473% of VALID pressure squared error versus28.8944% TRAIN. Mean pressure spatial-std ratio is0.5930 TRAIN /0.5866 VALID; centered spatial correlation0.6518 /0.5544. Reduced predicted variance is not by itself evidence that an amplitude increase would improve accuracy. All nine cooling regimes occur in both pools, and VALID is worse than its corresponding TRAIN mean in each regime; there is only one VALID trajectory per regime. This descriptive comparison supports investigating long-rollout generalization; it does not identify a causal mechanism or an optimal new parameter. Per-step/per-case metrics, hashes and viewed standalone PNG/SVG retained in ensemble_error_v1. Result SHA256 `5ac7251b77e04cdcfac28041cdad68d0e9a99b1eb79c3a72283a49aa0c7627db`. + +## E118 — persistent coarse feature memory preparation + +Hypothesis: a persistent feature memory may improve later rollout generalization beyond the four-frame state window. The E117 diagnosis motivates the late forecast question but does not prove missing memory is its cause. Only changed factor: a96-wide GRUCell per coarsest-grid location in first E101member, with LayerNorm288/inputLinear288to96/SiLU/GRU/outputLinear96to288. Output projection starts at zero;111840newparameters, initializationseed20270212 in an isolated CPU RNG fork. Original parameter values/order and CPU/CUDA noise RNG must remain unchanged. Memory is explicit local forecast state, zero at every case/translation/channelchunk/trainingmicrobatch and carried without detaching through all autoregressive steps. There is no inference-time parameter fitting. Both complete sharedmembers execute everycase; second96member andE072/E033guides stayfixed. Reuse E106 no-reflection control with identical seed20270118/sampler/noise/full30batch2x3/AdamW LR3e-5/schedule. If all preflight gates pass, train only one1000-update candidate after E116; select500/1000 throughfresh9thermal+strict576completephysicsreuse/full585currentbest+matchedcontrol/repeats. Required gates: original/secondweightidentity, constructorRNG, real temporal sensitivity/earlier-featuregradients, sixTRAIN/unfamiliar completezero-parent/repeatparity, full30B2finitebackward/nonzerooutputgrad/unchangedparameters/peakunder40GiB, strictsourceproof and correctpackageaccounting. Initial artifactSHA256 `28df2cb464ee1bacbe28dfe39a3097dc37104dccf64d6dcc19ec8111c700657b`, initialpackagedmain `adb5c975f045099677a9e8dbe7694d1e6c7d2a9c73edc18792360426550857dc`, completeensemble14037569parameters. These initial weights are a diagnostic prototype, never a selected candidate. Validation per-dataset VRMSE pending; no optimizer/VALID/queue until gatespass. No pretrained/synthetic assets. Protocol/source retained in temporal_memory_v1. + +## E119 — fixed temporal forecast filter TRAIN screen + +Hypothesis: centered temporal averaging may remove model-generated variation complementary to existing spatial/member averages. Pin E101 both members/guides/corrections; firstSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. Onlychangedfactor: fixed0.25previous+0.5current+0.25next after the complete forecast, using last supplied context as previous forstep1, leaving the finalstep unchanged. All future values in this filter are the model’s own predictions. No fitted coefficients, extra inference, optimizer, newseed or sweep. Metadata eligibility is the four density/pressure/velocityfields withperiodicX/openY, independent ofdatasetname. Pilot original9initialTRAIN; require>=0.1%relative meanVRMSE gain and>=5/9improved. Ifpass, freeze and confirmother63TRAIN withsame>=0.1%/32of63gates, nofit. Only afterbothgatesmayportable/full585VALID/repeatchecksbeconsidered. Baselinecache/memberidentity/nativecontext/target/membership/officialscoreaudited. Validation per-dataset VRMSE not applicable yet; no model deployment. No synthetic/pretrained assets. Source/protocol in temporal_filter_v1. + +## E119 fixed temporal filter TRAIN result + +pilot 9TRAIN: officialVRMSE 0.37577541437035->0.40033387295036, 0/9improve, relativegain-6.53541%; gate fails. Reject this fixedfilter; no VALID or deploymentchange. No refitting, alternativekernel/sweep, newcheckpoint/seed/optimizer. Exactconstant/lineartrajectory checks and allTRAINaudits pass. ResultSHA256 `2d7e31e6e40b60073cb39b51f2a2e8952105b351a98d5085ba6f89673300fb3f`. + +## E118 temporal memory preflight result + +All six complete TRAIN/unfamiliar forecasts (including13channels and odd grids) equal E101 and an independent repeat exactly. Constructor CPU/CUDA RNG and strict second-member architecture/state are unchanged. A nonzero diagnostic memory output has real temporal sensitivity and nonzero gradient1.2679772964929725e-07 to earlier input features; diagnostic weights are not saved or deployed. Native thermalB2/full30 backward loss0.22639204561710358, totalgrad13.770532608032227, clipped memoryoutputgrad0.002742402022704482; all gradients finite/parameters bitunchanged/zerooptimizerupdates, peak30.22698402404785GiB. Original eager training may differ across processes under cuDNN benchmarkTrue; no historical trainingloss equality claim. Serving parity/repeats are exact. Strict complete-physical dependency/source proof and all10115TRAIN membership/indexhash checks pass. Privateinstaller preserves allinitialweights and secondmemberbytes, updatesmemberhash and14037569parametercount correctly. One1000-update memory arm may now queue after E116, with unchanged declared hyperparameters and reused E106control. No optimizer has started yet. Preflight SHA256 `6dd67de1ec87681b40bcfea9e103f32273c64090befb772f9c676e4ccf09e86d`. + +## E120 — Fourier horizon features in the shared guide + +Hypothesis: four sine/cosine frequency pairs may represent oscillatory future means/profiles better than the current six slowly varying time features. Onlychangedfactor:1536zero-initialized parameters project sin/cos(2*pi*[1,2,4,8]*horizon/30) into the first192-wide guide hidden feature. No new RNG draws or initializationseed; original parameters and constructorRNG remainexact. Same E101/E072 guideparentSHA256 `3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c`; bothmainmembers and allotherauxiliaries/corrections/blends fixed. Reuse completed E112control at2k/6k. Trainonlyone6kCPUcandidate withsame originalweights,seed20270202/dropout03/reflection04/periodic05,4threads,batch512,AdamW5e-5cosine,groups.1/.1/.1/.7,initial.5,dropout.1,periodicshift/eligible reflection.5 andfieldVRMSEobjective acrossall19datasets. Gate optimization on CPU/CUDA4group+unknown1/4/11zerooutput/oldweight/RNG/finitegradient/no-update checks and portablezero sixTRAIN/unknownparent/repeatparity+correctinstallation. Then select2k/6k usingfresh9thermalVALID,576completephysical only understrictsource/config/input/coverageproof,full585currentbest+matchedcontrolcomparison,neural/unfamiliar repeatsbeforewin. Auxiliary546123/fullensemble13927265parameters. No new optimizer yet; validation per-datasetVRMSE pending. No pretrained/synthetic/held-out assets or inference-time fitting. Protocol/source in fourier_time_v1. + +## E120 Fourier-time preflight result + +All10115TRAINmemberships audited. CPU/CUDA outputs exactly match the parent acrossfourgroups andunknown1/4/11channels, originalweights andconstructorRNG exact, finite nonzero newprojectiongradients and parameters unchanged withzerooptimizerupdates. Adaptergradientnorms 0.03553203221735819, 0.035531076354147706. Complete sixTRAIN/unfamiliar forecasts, including13channels/oddgrids, equal the E101parent and independentrepeat exactly. Installer preserves bothmainweightfiles and alloriginalauxiliaryparameters, reports546123auxiliary/13927265totalparameters and correcthashes. One6kCPUcandidate may now start under the declared protocol, reusingE112control; no inference or deploymentchange from this preflight. Portable resultSHA256 `02c86614018de53ddc37597fbbcea3e072f200105a899648c4f8e1e7d2a2dba7`. + +## Full validation of fourier_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781833**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437447 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E120 — shared Fourier horizon features, update 2000 +- Hypothesis / changed factor: Fourier horizon features in shared mean/profile guide,update2000/6000,pinned E101 ensemble. Only new1536zero-initialized parameters project sin/cos(2*pi*[1,2,4,8]*horizon/30) into first192-wide hidden feature. No RNG draws for new initialization; original six time features and all output blends fixed. Reuse E112control: identical original guide weights,all19datasets/10115TRAINrecords,CPU4threads,batch512,AdamW5e-5cosine, reconstructed-field VRMSE,groups.1/.1/.1/.7,initial.5,dropout.1,periodic augmentation/reflection.5. Seeds sampling/dropout/reflection/periodic20270202/20270203/20270204/20270205; no redundant control training. Both complete mainmembers and everycase share the same statistical branch; other weights/corrections fixed. Training checkpointSHA256 a35b1e6adaf7b6163a43e57ab539e5e8adcf0ceb6630f70c85e9350ff9cae2eb, auxiliarySHA256 7271cfa191cf089c3dd2782362842cd69ff9262d53a01e39312ff47f0f602f33. Matched control full585reward 0.7818207793704589; candidate 0.7818325706365746. Fresh9thermalVALID+576audited complete unchanged physical outputs; full585 compared to actual current best. Exact neural/unfamiliar repeats beforewin. No pretrained/synthetic assets,held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `eaf972cd97209ad2db2ef63638d41a054adbe37540ce2cfc95134d8dc9b717d1`. +- Validation: fixed 81-case subset; reward **0.812610**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437447 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of fourier_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781851**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437226 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E120 — shared Fourier horizon features, update 6000 +- Hypothesis / changed factor: Fourier horizon features in shared mean/profile guide,update6000/6000,pinned E101 ensemble. Only new1536zero-initialized parameters project sin/cos(2*pi*[1,2,4,8]*horizon/30) into first192-wide hidden feature. No RNG draws for new initialization; original six time features and all output blends fixed. Reuse E112control: identical original guide weights,all19datasets/10115TRAINrecords,CPU4threads,batch512,AdamW5e-5cosine, reconstructed-field VRMSE,groups.1/.1/.1/.7,initial.5,dropout.1,periodic augmentation/reflection.5. Seeds sampling/dropout/reflection/periodic20270202/20270203/20270204/20270205; no redundant control training. Both complete mainmembers and everycase share the same statistical branch; other weights/corrections fixed. Training checkpointSHA256 877cfd8ec306851d96688e857f6f3fef936c1bb3149b80ca11ef05ab89d86a3b, auxiliarySHA256 4c2095c6079a9fb0314412fb91d36d3582f4e2eec8f4bcb081829c8a55c501b7. Matched control full585reward 0.7818293241563412; candidate 0.7818508202363947. Fresh9thermalVALID+576audited complete unchanged physical outputs; full585 compared to actual current best. Exact neural/unfamiliar repeats beforewin. No pretrained/synthetic assets,held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `3b927b764ec5f2b644596bc4e8f190000abe902934223131e5fe1a9b3999df85`. +- Validation: fixed 81-case subset; reward **0.812628**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437226 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E120 descriptive auxiliary versus complete-rollout analysis + +All9 thermal VALID cases, no fitting or model selection in this analysis. The auxiliary-only field uses canonical unshifted context statistics and its predicted mean/axis profiles; complete rollout scores include both neural members, translations and all corrections. These are distinct computations. Exact native target interaction floor for axis-additive reconstruction is0.45076384; this diagnostic target projection is never used at inference. + +| Candidate | Auxiliary field VRMSE | Mean normalized absolute error | Profile normalized RMS error | Complete thermal VRMSE | +|---|---:|---:|---:|---:| +| parent | 0.5616377 | 0.12494179 | 0.25973191 | 0.43689118 | +| control_002000 | 0.56244575 | 0.12616602 | 0.25977492 | 0.4375891 | +| control_006000 | 0.56214908 | 0.1256348 | 0.26012361 | 0.4374858 | +| fourier_002000 | 0.56245266 | 0.12564692 | 0.26028531 | 0.43744656 | +| fourier_006000 | 0.56229385 | 0.1249025 | 0.26081413 | 0.43722603 | + +All checkpoint hashes, per-case/per-step values and standalone PNG/SVG retained in fourier_time_v1/analysis. This comparison does not establish a causal mechanism for the complete rollout. + +## E121 — extra neural grid-phase TRAIN screen + +Hypothesis: an extra outer periodic translation may reduce grid-phase error left by the current even-phase neural ensemble. Pin E101 firstSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`,allguides/corrections. Fixed candidate: average original complete forecast with inverse-shifted forecast from context shiftedround(X/128) alongperiodicX. No coefficient/shift sweep, optimizer, newseed or weights. Original innerfourtranslations unchanged. Pilot9initialthermalTRAIN,require>=0.1%relativegain and5/9improved; onlyifpassconfirmother63samefixedrule,>=0.1%/32of63gates,no refit. Thermalcaseshavezerophysicalcoverage, explicitly checked fromsuppliedmetadata. A futureportablecandidate mustpreserve the original E101physicalprefix exactly (includingitsfourphysicaltranslations) andaddonlyanotherneuralview, with everysharedmember appliedeverycase. No extra physicalintegration intended; partialprefix/chunk/source/determinism checks andfull585VALID remain mandatorybeforedeployment. Validation per-datasetVRMSE notapplicableyet. No pretrained/synthetic assets orheld-out access. Protocol/source in extra_phase_v1. + +## E121 extra neural phase TRAIN result + +pilot 9TRAIN: officialVRMSE 0.37577541437035->0.37544769497592, 6/9improve, relativegain0.0872115%; gate fails. Reject this fixed extra view; no VALID or deploymentchange. No refit, shift/kernel sweep,newweights/seed/optimizer. Pool/native/member/score/inverse-transform/no-physical-eligibility audits pass. ResultSHA256 `fb6c1a55e64582bad2c01b271aa7d165b297603ff9b96529abe88d8d9d87f10c`. + +## E122 — batched physical quadrature feasibility + +Hypothesis: integrate the three independent angular-state quadrature nodes together to reduce GPU launches. Only a leading node axis and per-node indexing/reductions change; original arithmetic, six flow iterations, substeps, decoder, seed20261026 and quadrature amplitude0.0001 remain fixed. Pin E101 mainSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`, kernel `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`. + +Use only immutable audited TRAIN encoded states at zeta9/alpha-3 and zeta17/alpha-5, trajectory2 each. Require bit-exact initial transforms, flow and RHS, then all node states/streamfunctions/decoded outputs/guards/averages at every step of four intervals on both cases. Only if both pass, compare three alternating-order paired30-step runs on each. Reject this throughput-only change on any numerical mismatch; do not tune tolerances. For further development require >=10% median speed gain on both cases and faster batching in5/6 pairs. Shared H100 timings include state clones and decoding, not full-package throughput. No optimizer, new seed, new weights, VALID per-dataset scores or deployment. Cached files deserialize old TRAIN targets that this diagnostic does not access. Protocol and source hashes in quadrature_batch_v1. + +## E122 quadrature batching result + +Numerical gate fails; timing gate not passed. TRAIN zeta9/alpha-3/trajectory2 primitive differences {}; completed 1 paired runs. Median speed gains [], faster pairs 0/6 (only meaningful after numerical gate). Do not adopt or expand this throughput-only change. No tolerance adjustment, optimizer, new weights, VALID per-dataset scores or deployment. ResultSHA256 `bbbf542087fbe01ca9fb16985aa2dc6ea93b20c483abc234df17c787a5c7fa08`. + +## E123 — matched initial-window sampling preparation + +Hypothesis: allocating more main-network updates to initial windows may improve frame-zero forecast generalization. E085/E100b establish different window difficulty and nearly orthogonal initial/later gradients, not an optimal probability. Only changed factor is an independent2/3 frame-zero override after the original baseline start draw, increasing the explicit initial policy0.25 to0.75; uniform starts can also equal0. All original group/trajectory/baseline-start/periodic-shift/reflection RNG streams and noise RNG stay unchanged. Override seed20270213. Reuse E106 no-reflection control500/1000, same pinned E101 first48/fixed96/E072/E033, sampling/noise20270118, unusedreflection20270119, full30(corpus26/Arena10)batch2x3,LR3e-5/AdamW/schedule/noise0.01/groups.05/.075/.075/.8. No new parameters or inference code. Both complete members run every case. One1000-update arm queues after E118 only after80-window sampler/raw-source/RNG checks, all10115TRAIN memberships, unchanged installer/runtime and actual-parent nativeB2/full30 finite backward with unchanged weights/zero updates. Validate500/1000 with fresh9thermal+strict576completephysicsreuse/full585currentbest+matchedcontrol/repeats. Initial artifactSHA256 `956092907d8d7d163de708e1be1c6fa7e5b369bbdc6954dc48178acf351efbc0`, first `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. No optimizer or VALID result yet, no pretrained/synthetic assets or held-out access. Protocol/source in initial_sampling_v1; earlier E085b preparation remains untrained. + +E122 numerical detail: primitive FFT/flow/RHS checks all pass on the first TRAIN state, but decoded node output differs by7.7486038208e-7 at step1 and3.5464763641e-6 at step4; fixed average differs by4.1723251343e-7 and1.1622905731e-6 respectively. All values are finite and guards agree. This fails the predeclared exactness gate; the second case and30-step/timing expansion are not run. The four-step shared-device timings15.4593s sequential/10.4418s batch are not used to accept a numerically changed implementation. No source/tolerance/serving changes follow. + +## E123 initial-window sampling preflight result + +All10115TRAIN memberships pass. On80real windows across all four groups, disabled override equals untouched E106 exactly; candidate inputs/targets equal native slices/shifts and preserve every original group/trajectory/start/shift/reflection draw. Initial counts[31, 68], changed starts37; global Python/NumPy/Torch RNG unchanged. Native thermalB2/full30 backward loss0.19188666343688965, grad8.966861724853516, peak30.020387172698975GiB; all gradients finite, all parameters bit-unchanged, zero optimizer updates. Installer/runtime/second-member bytes and first state remain exact, memberhash correct,13925729 totalparameters. One1000-update arm now eligible to queue after E118, fixed protocol and reused E106control. No new trainer has started. Preflight resultSHA256 `44c23f25e5ef57b163ef7c86f86d18a3b3225d0641b85ac80c700adb34c92312`. + +## E124 — matched main VRMSE objective preparation + +Hypothesis: replacing the main log1p normalized squared error with the scored normalized RMSE may improve rollout reward because these objectives weight channel/step errors differently. Only changed factor: sqrt(MSE/(unbiased spatial targetvariance+1e-7)+1e-12), clamped at10 per channel/step before equal averaging. The fixed smoothing differs from exact VRMSE by at most1e-6 and gives finite derivatives at zero; no smoothing/loss sweep. Reuse E106 no-reflection control500/1000, same pinned E101 first48/fixed96/E072/E033, sample/noise20270118,unusedreflection20270119,initialpolicy0.25,full30(corpus26/Arena10)batch2x3,LR3e-5/AdamW/schedule/noise0.01/groups.05/.075/.075/.8. No new parameters, sampling changes, RNG draws or inference code. Both complete members run every case. One1000-update arm queues after E123 only after9auditedTRAIN official-score/native-target checks, zero/saturated finite gradients, unchanged disabled loss/RNG/model/sampler,all10115TRAIN memberships, correct ensemble installer and actual-parent nativeB2/full30 finite backward with unchanged weights/zero updates/peakunder40GiB. Validate500/1000 fresh9thermal+strict576completephysicsreuse/full585currentbest+matchedcontrol/repeats. Initial artifactSHA256 `956092907d8d7d163de708e1be1c6fa7e5b369bbdc6954dc48178acf351efbc0`, first `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`. No optimizer or VALID result yet, no pretrained/synthetic assets or held-out access. Protocol/source in score_objective_v1. + +## E124 main VRMSE objective preflight result + +All10115TRAIN memberships pass;9audited native TRAIN target/context pairs match source slices. Training objective versus official cached forecast VRMSE differs by at most2.639444218743847e-12; disabled objective equals original exactly. Zero/saturated gradients are finite and zero, ordinary gradient positive; loss consumes no RNG. Native thermalB2/full30 backward loss0.44385552406311035, grad14.14299201965332, peak30.020387172698975GiB; all gradients finite, parameters bit-unchanged, zero optimizer updates. Original sampler/model/guidance sources, serving runtime, second member and initial main state remain exact. Privateinstaller memberhash/count13925729 correct. One1000-update arm may now queue after E123, same declared protocol and reused E106control. Preflight resultSHA256 `6cc37d1dac0c1e3eb7af73c33e05ac2a1d0408403aa373b70ab05386dae9c8b1`. + +## E125 — independent-stream physical quadrature feasibility + +Hypothesis: three independent CUDA streams can overlap original single-node physical kernels without the changed FFT/reduction shapes of E122. Only scheduling changes: three private original operators, unchanged kernelSHA256 `cf75e10045f053d2774594b2423dfbf1c11fcbcbf9852849ab7f6c700bc18f7f`, fixed quadrature1e-4/seed20261026/config. Pin E101 first `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`; no new weights. + +Audit immutable TRAIN9/-3/trajectory2 and17/-5/trajectory2 encoded states. Require every node state, streamfunction, decoded field, guard decision and fixed-order average bit-exact/finite after each of four stored steps on both. Stop on mismatch, no tolerance change. Only if both pass, compare three alternating-order paired30-step runs per case with the same exactness gate. Further development requires>=10% median speed gain on each case and concurrent faster in5/6 pairs. Shared H100 timings include clones/decoding, exclude setup/context/neural work, and are not isolated throughput. Guards are compared after timing; a portable guarded rollout may synchronize more often. No optimizer, VALID per-dataset scores, held-out reads or deployment. Cached TRAIN targets deserialize but are not accessed. Protocol/source in quadrature_streams_v1. + +## E126 — symmetric moving-frame TRAIN screen + +Hypothesis: averaging small moving-coordinate-frame views may reduce shared neural forecast sensitivity. Pin E101 firstSHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`, second `0c9c49ae6a7da1fbae8d3f21d94b26311467ba1fd991b59877c872973f1a6c23`, allguides/corrections/innertranslations. For supplied density/pressure/velocity fields with periodicX/openY, use sign±1: contextframe k rolls bysign*k Xcells and velocity_0 gains sign*Lx/(X*dt); inverse predictions use k4..33. Average original,+1,-1 completeforecasts equally in float64 thenfloat32, no coefficient fitting. + +Pilot9initialTRAIN requires>=0.1%relative gain and5/9improved; onlyifpassconfirmother63samefixedrule with>=0.1%/32of63. Check source/native/member/officialscore/metadata eligibility and exact integer inversion; velocity float32 roundtrip tolerance2eps*max(1,abs velocity,abs boost). Manufactured time-index/sign check is not training data. WaitE125 completion before fresh inference. No newweights/seed/optimizer/VALID. A futureportable candidate wouldpreserve original completephysicalprefix andaddonlyneural views, with unchanged sharedmodels on everycase and full585/repeat/source checks; noneisimplementedyet. Protocol/source in moving_frame_v1. + +E126 implementation correction before inference: audit JSON serialization rejected a NumPy float32 tolerance. Preserve the failed source/log/hash manifest and all prepared inputs/package under failed_serialization names; cast only the already computed tolerance to Python float. Its numerical value, transformed inputs, model, seed, gates and weights are unchanged. No forecast or score was produced before failure. Restart the same declared pilot, still waiting for E125 before GPU inference. + +## Full validation of control_000500 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781841**, failed cases 0. Neural initialization seed 20261021; sampling seed 20270118; checkpoint SHA256 `a086fbfc38f5f997f3a6ffa7b8a8f89206d1958a07290f1c9b43e16fb6b10aa5`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437347 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E106 — main reflection control, update 500 +- Hypothesis / changed factor: Matched first-member continuation arm control, update500/1000, sampling/noise seed20270118 and independent reflection RNG20270119. Pinned E101 parent for both arms, width48 main trained on all19 datasets, fixed width96 member and E072/E033 guides. Only whole-window eligible horizontal reflection0.5 differs between arms; control consumes the same independent RNG without transforming. Full30 rollout (corpus26/Arena10),batch2x3,AdamW LR3e-5 cosine,noise0.01,groups.05/.075/.075/.8. Both complete members run every case with unchanged equal output weights, physics, translations and final corrections. No latent head in this pinned-parent experiment; no changes to phase10. Training checkpointSHA256 32e69aa36f35f8fc0dc51519c4719a213330b14b23e32e35f0006ce46abb7429, packaged first memberSHA256 a086fbfc38f5f997f3a6ffa7b8a8f89206d1958a07290f1c9b43e16fb6b10aa5. All9 changed thermal VALID forecasts fresh;576 complete physical predictions reused after unchanged source/config/input/coverage audit. Compare full585 against the actual current selected package; exact neural/unknown repeat checks precede a win. No pretrained or synthetic assets, held-out data or parameter fitting at inference. +- Seed: 20261106. +- Checkpoint SHA256: `a086fbfc38f5f997f3a6ffa7b8a8f89206d1958a07290f1c9b43e16fb6b10aa5`. +- Runtime code SHA256: `087b1a80f5bdd055700c84312cca54cc5773ad4f9a4d5fb39a9d3b5f98a1c633`. +- Validation: fixed 81-case subset; reward **0.812618**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437347 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E106 control descriptive paired analysis, update500 + +Thermal VRMSE: E101 parent0.4368911756234454, matched control0.43734695742255775, candidate0.43734695742255775. Candidate improves0/9 versus control and4/9 versus parent. Late steps13–30: control0.5258797683704173, candidate0.5258797683704173. Other576 official case results match E101 exactly. Descriptive only after full585 scoring/selection; no fit/new optimizer or model change. Nine thermal trajectories limit generalization claims. Per-case/channel/horizon results and unviewed standalone figures in matched_main_analysis_v1/reflection_training_v1_control_000500. CheckpointSHA256 `a086fbfc38f5f997f3a6ffa7b8a8f89206d1958a07290f1c9b43e16fb6b10aa5`, analysisSHA256 `23d63916472b8cb4a983524178891fb469ee699872f32ef6ee98a31c056b6605`. + +E106 control500 figure inspected:4/9 thermal cases improve versus E101; densityVRMSE0.17460174->0.17439273, pressure0.88917316->0.88966110, vx0.20265762->0.20312290, vy0.48113218->0.48221110. Steps1–5 rise0.26740730->0.26821558,6–12 rise0.32764430->0.33049928,13–30 fall0.52645492->0.52587977. Overall thermal0.4368911756->0.4373469574 andfullreward0.7818785482->0.7818408118, zero failures; E101 retained. Descriptive only; these small mixed changes do not establish why continuation failed. PNG/SVG now viewed. + +## E127 — context-conditioned damped modes in the shared guide + +Hypothesis: context-dependent frequencies and damping may predict broad temporal evolution better than fixed horizon features. E120 fixed-frequency results were negative; this tests the complete conditional decoder, not an isolated capacity-matched frequency claim. Add41760parameters: reuse original first guide layer with six zero time inputs for context; Linear192to96 yields32amplitudes/frequencyshifts/decays. Amplitudes tanh, basefrequencies2**linspace(-2,3,32) cycles/30steps modulatedbyexp(0.5*tanh(f)), positive decaysoftplus(d)+0.05. Dampedcosine-minus-one andsine features feedzero64to363output(no bias), added tooriginal normalizedmean/profile delta before its existinghorizon/30factor. Isolated CPU default-generator seed20270214 preserves originalCPU/CUDARNG/weights. Extra module has no dropout; originalthreeSiLU dropoutdraws unchanged. Auxiliary586347/fullensemble13967489parameters. Reuse E112control2k/6k andE072parentSHA3b423a0afbb46bafc914d62daaf8611a1b30ae424b8c517d846cbb93d4d6578c. One6kCPUarm,4threads,batch512,AdamW5e-5cosine,all19datasets/groups.1/.1/.1/.7,initial.5,fieldVRMSE,dropout.1/reflection.5/periodicaugmentation; seeds20270202/03/04/05. Both E101 mainmembers andotherweights/blendsfixed,guidecomputed everycase/member. Before optimizer require CPU/CUDA4group+unknown1/4/11zero-parent/RNG/oldweight/nonzerograd/unchangedparameter checks; an unsaved nonzero probe must showcontext/time dependence andhorizon0identity; sixcompleteTRAIN/unknown13channel/oddgridparent/repeatparity andcorrectinstaller. Select2k/6k viafresh9thermal+strict576completephysicsreuse/full585currentbest+matchedcontrol/repeats. No optimizer/VALID yet, no pretrained/synthetic assets orheld-out access. Protocol/source in context_modes_v1. + +## E127 conditional-modal preflight result + +All10115TRAIN memberships pass. CPU/CUDA zerooutputs equalparent acrossfourgroups andunknown1/4/11channels; originalweights andCPU/CUDAconstructorRNG exact. Nonzero unsavedprobe hascontext/horizondependence andexactzeroathorizon0. Newoutputgradientnorms 0.06172472836282498, 0.06172468270231952; allgradientsfinite,parametersunchanged,0optimizerupdates. Six completeTRAIN/unfamiliar forecasts including13channels/oddgrids equalE101parent andindependentrepeat exactly. Installerpreservesbothmainweights andalloriginalauxiliaryparameters,586347auxiliary/13967489totalparameters/correcthashes. One6kCPUarm is now eligible underfixedprotocol/reusedE112control. No selectedruntimechange. Portable resultSHA256 `5642a75cdc4630d0d5cb9dbdf174f5239c0ebc4bca24dc291627a75ffd553db5`. + +## Full validation of modal_002000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781848**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437264 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E127 — context-conditioned guide modes, update 2000 +- Hypothesis / changed factor: Context-conditioned damped modes in shared guide,update2000/6000,pinned E101 ensemble. Add41760 parameters:32 context-conditioned amplitudes/frequencies/decays and a zero-initialized64-to363 output decoder. Isolated CPU seed20270214 preserves original CPU/CUDA RNG; reuse the first feature layer at zero time for conditioning. Original six time features/blends fixed. Reuse E112control: identical original guide weights,all19datasets/10115TRAINrecords,CPU4threads,batch512,AdamW5e-5cosine, reconstructed-field VRMSE,groups.1/.1/.1/.7,initial.5,dropout.1,periodic augmentation/reflection.5. Seeds sampling/dropout/reflection/periodic20270202/20270203/20270204/20270205; no redundant control training. Both complete mainmembers and everycase share the same statistical branch; other weights/corrections fixed. Training checkpointSHA256 13720a28f5783e427e78d190e4cfbad9cbe0c16a7bd2f657636565236df1f09d, auxiliarySHA256 5a6fc2a917512a417c900dcaa3cab61dcbeefa0702a5d62f5c4f6cbfd6779643. Matched control full585reward 0.7818207793704589; candidate 0.7818476782097061. Fresh9thermalVALID+576audited complete unchanged physical outputs; full585 compared to actual current best. Exact neural/unfamiliar repeats beforewin. No pretrained/synthetic assets,held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `fab2828eb9909050da2250ef346ef5dacf8814a4df00057f52dde118c7083bd7`. +- Validation: fixed 81-case subset; reward **0.812625**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437264 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## Full validation of modal_006000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.781843**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261106; checkpoint SHA256 `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.087494 | +| well/turbulent_radiative_layer_2D | 0.437317 | + +Exact reuse protocol for a shared-neural change: all physical assets/configuration match /run-state/workspace/research/output_ensemble_v1/validation. The source per-case log audit verifies552 complete corpus/smoke physical rollouts and24 complete active rollouts; all576 outputs have unchanged dependencies. For a shared latent-state head, those dependencies include neural weights, guide assets, postprocessing and serving configuration; neural-weight changes cannot reuse these active predictions. The nine neural radiative-layer outputs were replayed with exact context/metadata equality to full inputs and freshly scored with the official scorer. Reaggregate their official per-case scores and the576 audited unchanged per-case scores using the exact aggregation code extracted from the official scorer. Report provenance includes scorer/source-report hashes and the reuse audit. All585 prediction files are retained. This avoids rereading unchanged targets without approximating the metric. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E127 — context-conditioned guide modes, update 6000 +- Hypothesis / changed factor: Context-conditioned damped modes in shared guide,update6000/6000,pinned E101 ensemble. Add41760 parameters:32 context-conditioned amplitudes/frequencies/decays and a zero-initialized64-to363 output decoder. Isolated CPU seed20270214 preserves original CPU/CUDA RNG; reuse the first feature layer at zero time for conditioning. Original six time features/blends fixed. Reuse E112control: identical original guide weights,all19datasets/10115TRAINrecords,CPU4threads,batch512,AdamW5e-5cosine, reconstructed-field VRMSE,groups.1/.1/.1/.7,initial.5,dropout.1,periodic augmentation/reflection.5. Seeds sampling/dropout/reflection/periodic20270202/20270203/20270204/20270205; no redundant control training. Both complete mainmembers and everycase share the same statistical branch; other weights/corrections fixed. Training checkpointSHA256 ceeb818fc48f9cd173f6b6a15e11050e214bde49c182f27d28621242ebc5ff93, auxiliarySHA256 833d46bd76767f424bada214848b1f525c55fb38137d21ac02d24fa40a7044f4. Matched control full585reward 0.7818293241563412; candidate 0.7818432887454752. Fresh9thermalVALID+576audited complete unchanged physical outputs; full585 compared to actual current best. Exact neural/unfamiliar repeats beforewin. No pretrained/synthetic assets,held-out data or inference-time fitting. +- Seed: 20261106. +- Checkpoint SHA256: `1f24abcc814d6b2fff9a530b350ee4d8cffc8c80cb9374bfee6509c02ae0bd61`. +- Runtime code SHA256: `4fed78da43711bed8edfda60eecfdf9e01dd65ae6ce4c9cf3e1fde924c65e35f`. +- Validation: fixed 81-case subset; reward **0.812620**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.040528 | +| well/turbulent_radiative_layer_2D | 0.437317 | + +Selection basis: same physical operators; fixed81-case neural comparison + +Outcome: did not improve validation reward; retained current package. + +## E127 descriptive auxiliary versus complete-rollout analysis + +All9 thermal VALID cases, no fitting or model selection in this analysis. The auxiliary-only field uses canonical unshifted context statistics and its predicted mean/axis profiles; complete rollout scores include both neural members, translations and all corrections. These are distinct computations. Exact native target interaction floor for axis-additive reconstruction is0.45076384; this diagnostic target projection is never used at inference. + +| Candidate | Auxiliary field VRMSE | Mean normalized absolute error | Profile normalized RMS error | Complete thermal VRMSE | +|---|---:|---:|---:|---:| +| parent | 0.5616377 | 0.12494179 | 0.25973191 | 0.43689118 | +| control_002000 | 0.56244575 | 0.12616602 | 0.25977492 | 0.4375891 | +| control_006000 | 0.56214908 | 0.1256348 | 0.26012361 | 0.4374858 | +| modal_002000 | 0.56177743 | 0.12524239 | 0.25910797 | 0.43726399 | +| modal_006000 | 0.56154999 | 0.12426852 | 0.25932932 | 0.43731703 | + +All checkpoint hashes, per-case/per-step values and standalone PNG/SVG retained in context_modes_v1/analysis. This comparison does not establish a causal mechanism for the complete rollout. + +## E125 independent-stream quadrature result + +Numerical gate passes, timing gate passes. TRAIN{'zeta': 9, 'alpha': -3, 'trajectory': 2}: 4paired runs, maximum differences{'state': 0.0, 'psi': 0.0, 'decoded': 0.0, 'average': 0.0}. TRAIN{'zeta': 17, 'alpha': -5, 'trajectory': 2}: 4paired runs, maximum differences{'state': 0.0, 'psi': 0.0, 'decoded': 0.0, 'average': 0.0}. Median speed gains[0.2869366333359454, 0.20713581208897716], faster pairs6/6, shared H100/incomplete-package timing. Portable/full585/repeat/source checks still required before any selection. No optimizer, new weights, VALID scores or serving changes. ResultSHA256 `02e80c9d2b52946d2f75dc3a141a3361dc630c095b6e220668b49876858510ce`. + +## E128 — shared initial context encoder with fixed two-member rollouts + +A common initial context encoder from the all19 joint-state model may retain its active improvement while the fixed E101 members retain thermal accuracy. Changed factor: Add fixed context-encoder component evaluated every case, distinct translation and channel chunk; common state consumed by both fixed E101 members. Fixed encoder checkpoint SHA256 `0fc425a0cf92a0d25cc8beae5b2cb291cef63f56f6c17c7f9ebd84342e06848d`, seed20261127/physical20261128, update1000; both E101 main members and E072/E033 guides/coefficients unchanged. Total16992454 parameters. No new training, seed, synthetic or pretrained assets. Initial source/config audit restricts potential reuse to552 complete corpus/smoke cases; new context model/weight assets are included in the conservative latent dependency guard with mutation fixtures passing. Preflight and fresh33/full585 validation pending; no score or runtime claim. Protocol in shared_state_ensemble_v1/protocol.json. + +## E126 moving-frame TRAIN result + +pilot 9TRAIN: officialVRMSE 0.3757754143703486->0.43018241452106953, 0/9improve, relativegain-14.478594945304124%; gate fails. Reject this fixed view ensemble; no VALID or deployment change. Native/member/score/metadata/inverse-transform checks pass. No refit, transform/weight sweep, optimizer, newweights/seed or synthetic training. ResultSHA256 `4172a00636e713052d47f6cf0e9abdc25248ee43f15cea725dbc20d516b34865`. + +## E129 — guarded portable stream scheduling prototype + +E125 exact independent-stream scheduling may retain its benefit with real per-step guards, context encoding and learned shared initial state. Schedule original3single-node operators on3privateCUDAstreams; check original guard for each node each step and return common valid prefix. Original arithmetic, encoding, quadrature seed/scale, models and coefficients fixed. Fixed E128 common encoder SHA 0fc425a0cf92a0d25cc8beae5b2cb291cef63f56f6c17c7f9ebd84342e06848d; no new optimizer or seed. Gate: Fourstep bothTRAINcases then3alternating-order30step pairs percase; exact complete guarded operator outputs/prefixes, finite, including original learned context coefficients. Atleast10percent median gain each case and5of6wins including constructor/context/guards/decoding, sharedGPU timings only; exact portable package/repeat checks still required before deployment. No validation or deployment yet; sources/protocol in guarded_streams_v1. + +## E128 complete-package preflight + +Strict load and parameter counts pass: main48=3066627, main96=9526707, common context encoder=3066725, total16992454 including fixed guides. Six TRAIN/unfamiliar cases including13channels/oddgrids match the fixed E101 outputs off active matter and fixed phase10/1k outputs on active matter, with exact independent repeats. Partial spectral and kinetic prefixes preserve each member’s separate neural fallback exactly; physical work and initial context encoding are shared once per case/distinct shift/channel chunk, and case resets pass. No optimizer updates. Preflight SHA256 `d29ce9346a6a3c9fc5370d3f6596c5f0d8dd690080957fda25215f34274afdcf`. Fresh33/full585 validation starts under the fixed E128 protocol. + +## Full validation of eval001000 + +All 585 validation trajectories, starting at frame 0; no held-out trajectories. Reward **0.782451**, failed cases 0. Neural initialization seed 20261021; sampling seed 20261127; checkpoint SHA256 `0fc425a0cf92a0d25cc8beae5b2cb291cef63f56f6c17c7f9ebd84342e06848d`. Same model as the corresponding quick-validation experiment; changed factor: expanded validation sample only. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.007568 | +| ape2d/decay_turb | 0.000289 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000005 | +| ape2d/gs_alpha | 0.003239 | +| ape2d/gs_beta | 0.004901 | +| ape2d/gs_delta | 0.022624 | +| ape2d/gs_epsilon | 0.000889 | +| ape2d/gs_gamma | 0.062780 | +| ape2d/gs_iota | 0.006619 | +| ape2d/gs_kappa | 0.008764 | +| ape2d/gs_theta | 0.019424 | +| ape2d/kdv | 0.024675 | +| ape2d/kolm_flow | 0.011662 | +| ape2d/ks | 0.000024 | +| ape2d/sh | 0.000005 | +| pdearena/navier_stokes_2d | 0.000081 | +| well/active_matter | 0.085238 | +| well/turbulent_radiative_layer_2D | 0.441418 | + +Replayed and officially scored all24 active and all9 thermal VALID trajectories with this checkpoint. 24/24 active cases have a complete30-step physical prefix; any remainder is the same model’s neural fallback. Reused552 complete corpus/smoke outputs and their official per-case scores only after exact unaffected module/configuration comparison and serving-source normalization of the isolated latent-head plumbing. Per-case input/output hashes and solver coverage are in reuse_audit.json. The exact official aggregation code reassembles all585 case scores, with source-report/scorer hashes beside each report. Neural weights and guides affect active hidden-state inference, so no active output was reused. The quick81 report is an exact input/metadata-matched subset of these585 predictions. Selection requires full585 reports for candidate and current package. No held-out trajectories. + +Outcome: recorded full validation for audit and comparison; no automatic comparison against the distinct quick-validation sample. + +## E060 — joint hidden-state learning, update 1000 +- Hypothesis / changed factor: Joint shared-backbone/state-head training update1000/2000. Zero-initseed20261126, main samplingseed20261127 and physicalsamplingseed20261128. All19-dataset autoregressive objective, effectivebatch6/full30-step rollout plus native four-step differentiable moment-dynamics VRMSE every8 updates. Physical examples uniformly sample TRAIN parameter files then trajectories with even periodic shifts; all175 cache memberships and hashes audited. Same shared head computed for every case; bounded F6 correction0.01*c*(abs(F2)/c)^3*tanh(head), no parameter fitting at inference. HeadLR10x mainLR0.00003 cosine; fixed selected guides and unchanged three-node quadrature. Replay all24 active and9 thermal VALID trajectories for this checkpoint; reuse552 complete corpus/smoke predictions only after unaffected module/configuration/serving-source audit. Full585 comparison is mandatory; quick81 scores are logged only as the exact subset. No synthetic or pretrained assets. Exact active, neural and unfamiliar-schema repeat checks precede any winning publication. +- Seed: 20261127. +- Checkpoint SHA256: `0fc425a0cf92a0d25cc8beae5b2cb291cef63f56f6c17c7f9ebd84342e06848d`. +- Runtime code SHA256: `9616372e912d369c5aafd4fdf9d22e0b41d14da1de950142164f6d27ea064c39`. +- Validation: fixed 81-case subset; reward **0.813120**, failed cases 0. + +| Dataset | VRMSE | +|---|---:| +| ape2d/burgers | 0.006413 | +| ape2d/decay_turb | 0.000381 | +| ape2d/diff | 0.000003 | +| ape2d/fisher | 0.000003 | +| ape2d/gs_alpha | 0.003095 | +| ape2d/gs_beta | 0.004740 | +| ape2d/gs_delta | 0.019250 | +| ape2d/gs_epsilon | 0.001366 | +| ape2d/gs_gamma | 0.049313 | +| ape2d/gs_iota | 0.001785 | +| ape2d/gs_kappa | 0.004855 | +| ape2d/gs_theta | 0.004678 | +| ape2d/kdv | 0.016728 | +| ape2d/kolm_flow | 0.008926 | +| ape2d/ks | 0.000033 | +| ape2d/sh | 0.000004 | +| pdearena/navier_stokes_2d | 0.000083 | +| well/active_matter | 0.039602 | +| well/turbulent_radiative_layer_2D | 0.441418 | + +Selection basis: full585 physical-operator comparison: candidate0.78245076, current0.78187855 + +Outcome: improvement; published atomically. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/README.md b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/README.md new file mode 100644 index 0000000000000000000000000000000000000000..29bc95ce70f85999fc080e4aaa17755aa99e5ae2 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/README.md @@ -0,0 +1,43 @@ +# Shared neural and spectral surrogate + +A single shared neural operator is evaluated for every case and every autoregressive step. A deterministic spectral branch conditions on PDE metadata and uses independently implemented differential operators where their form was verified on training data. Unsupported equations use the shared neural update. There are no separately trained specialist checkpoints, no pretrained weights and no inference-time fitting. All runtime assets are included. + +Selection uses validation rollouts only. The neural model was trained from random initialization on all train pools, with sampling probabilities documented in EXPERIMENTS.md. See EXPERIMENTS.md for per-dataset results and negative experiments. Run `bash infer.sh INPUT_DIR OUTPUT_DIR`. + +Active matter also uses a fixed Fourier angular-moment Smoluchowski/Stokes operator. A recurrent context encoder inserts the observed low moments into its physical state; no parameter optimization or training occurs at inference. The Bingham initialization table contains mathematical Bessel ratios generated from scratch. Neural forecasts remain available as a fallback if a physical update is non-finite or violates positivity/moment bounds. The entry point pins the package directory through an open file descriptor so atomic publication cannot mix runtime assets. + +The active context encoder transports hidden angular moments with the supplied concentration and orientation, then applies a fixed linear Stokes observation layer and a necessary moment-positivity projection. It consumes only the four supplied frames. This is a fixed state computation; model and solver parameters do not change at inference. + +The active hidden state uses three fixed uncertainty quadrature nodes: its encoded center and symmetric small angular perturbations. Each node uses the same physical parameters and shared neural model; identical computations are shared. Outputs average the common valid physical prefix. A fixed local random generator makes the quadrature reproducible for every input, independent of case IDs and ordering. No separate specialist checkpoints or inference-time training are used. + +A two-channel head on the same shared neural backbone predicts a bounded correction to the sixth angular moment from the supplied four-frame context and metadata. The head is computed for every case. It was trained jointly with the all-dataset autoregressive objective and a differentiable physical continuation loss on TRAIN trajectories; no separate active-matter checkpoint is selected. Only the first neural pass enters the physical initial state. Neural and guide weights therefore affect active-matter predictions. Model selection replays all24 active and all9 thermal validation trajectories and compares full585-case rewards; only complete corpus/smoke predictions with verified unchanged dependencies may be reused. + +Four fixed translations of the same complete surrogate are averaged after undoing each shift. The shifts are0,2/128,4/128,6/128 of the first spatial axis when its boundaries are periodic, otherwise identity. The same shared weights and four-member rule apply to every case; identical member computations are shared. The shift rule was selected on TRAIN rollouts and checked on all585 VALID trajectories, including all distinct physical member continuations. No new learned weights or inference-time fitting. + +The final output projects density and concentration to nonnegative values, after all learned corrections and any translation average. Applicability follows supplied field names. This prevents later statistical corrections from reintroducing negative density; the fixed zero bound was checked on TRAIN trajectories and the complete package compared on all585 VALID cases. + +A small fixed continuity blend refines density after the complete shared forecast and any final support projection. It uses only the supplied last density/velocity and the model’s forecast velocities, with metadata-based field and boundary eligibility. One coefficient was fit on nine TRAIN trajectories and confirmed on63 additional TRAIN trajectories without refitting. The independently implemented conservative update has fixed CFL control and a bounded-work fallback. No future observations or inference-time fitting are used. + +Buoyant smoke uses a fixed staggered-grid finite-difference fluid operator with semi-Lagrangian advection, explicit viscosity, and a pressure projection. Numerical spacing and scheme were identified using training trajectories only, with no generator code or initial-condition seeds. The implementation uses only PyTorch. + +Neural forecasts use a fixed, horizon-dependent shrinkage toward profiles averaged over periodic directions, with nonnegative density and concentration. The coefficient was selected on nine training rollouts before validation. The physical operators are unaffected. + +The shared neural operator evaluates long rectangular grids at native resolution. The checkpoint manifest and experiment ledger record training grid sizes, rollout horizon, sampling probabilities and checkpoint lineage. + +Reaction-diffusion updates retain a finer internal Fourier grid and use double-precision second-order exponential integration. The fixed observation operator maps cell averages to and from that state. Numerical choices were selected on training trajectories and tested on validation rollouts. CUDA graphs accelerate the fixed updates without changing their arithmetic. + +The Burgers and forced-vorticity updates also retain a finer internal Fourier grid in double precision. A fixed cell-averaging observation map returns the native output grid. Training-only resolution studies selected a twofold grid increase; smaller time steps and fourfold grids added little. + +The shared neural network includes a metadata-conditioned global Fourier feature layer. It was added with a zero output projection, preserving the parent model at initialization, then trained jointly across all four groups. New parameters were initialized from scratch in this session. + +A shared mean-forecast branch predicts channel spatial means from the four supplied frames, metadata, and requested horizon. It was initialized randomly and trained across all19 datasets. This same branch is evaluated on every case, including cases handled by fixed physical updates; no separately trained dataset specialists are selected. For neural forecasts, a fixed validation-selected blend adjusts the spatial means. Physical updates remain authoritative. The auxiliary checkpoint, parameter count, seed, and source hashes are recorded in model_manifest.json. + +The shared statistical branch also consumes channel profiles with 16 bins along each spatial axis, derived only from the supplied context. These profiles describe broad spatial structure that global means and variances omit. + +The same statistical branch jointly forecasts global means and coarse spatial profiles. A fixed blend corrects the autoregressive forecast’s broad spatial structure, using interpolation consistent with the supplied boundary conditions. Every case uses the same learned statistical weights. + +For density/pressure/velocity fields on a periodic-x and open-or-wall-y grid, a small fixed pressure correction uses the forecast momentum balance. A mixed-boundary gradient projection preserves the predicted global pressure mean. Its blend was selected on training trajectories and checked on validation; it uses only context and the model’s own forecast, with no parameter fitting at inference. + +The autoregressive network consumes fixed future mean/profile forecasts through a jointly trained guidance adapter. Those guide fields are computed only from the original four supplied frames. The auxiliary parameters remain fixed during inference. + +A shared coarse-field branch forecasts two-dimensional spatial interactions on a 4 by 16 grid from the four input frames, metadata and horizon. It was initialized randomly and trained across all19 datasets. Its fixed output blend preserves the existing global means and axis profiles, and complete physical prefixes remain authoritative. The same weights are evaluated on every case; no inference-time fitting occurs. diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/arena.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/arena.py new file mode 100644 index 0000000000000000000000000000000000000000..370e9af2b204b0326adc20cb9b86a8874a1c8dbe --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/arena.py @@ -0,0 +1,39 @@ +"""Fixed MAC-grid buoyant-fluid update, independently implemented in PyTorch.""" +import math +import torch +import torch.nn.functional as F + +def eligible(meta,channels): + return channels==3 and str(meta.get('pde',''))=='Navier-Stokes: buoyant smoke (PDEArena)' and list(meta.get('fields',[]))==['u','vx','vy'] and 'buoyancy_y' in meta.get('constants',{}) + +class ArenaOperator: + def __init__(self,meta,x): + self.n,self.m=x.shape[-2:];self.device=x.device;self.dtype=torch.float64;x=x.to(self.dtype) + self.hx=float(meta['domain_extent'][0])*(self.n-1)/self.n**2;self.hy=float(meta['domain_extent'][1])*(self.m-1)/self.m**2 + self.interval=float(meta['dt'])*13/14;self.substeps=max(1,math.ceil(self.interval/1.5000001));self.dt=self.interval/self.substeps;self.nu=.01;self.buoyancy=float(meta['constants']['buoyancy_y']) + self.c=x[0].clone();self.u=F.pad(x[1],(0,0,0,1));self.v=F.pad(x[2],(0,1,0,0));device=self.device;dtype=self.dtype + ix=torch.arange(self.n,device=device,dtype=dtype);iy=torch.arange(self.m,device=device,dtype=dtype) + self.eig=(2*torch.cos(math.pi*ix/self.n)-2)[:,None]/self.hx**2+(2*torch.cos(math.pi*iy/self.m)-2)[None,:]/self.hy**2;self.eig[0,0]=1 + self.cx=(2/self.n)**.5*torch.cos(math.pi/self.n*ix[:,None]*(ix[None,:]+.5));self.cy=(2/self.m)**.5*torch.cos(math.pi/self.m*iy[:,None]*(iy[None,:]+.5));self.cx[0]*=2**-.5;self.cy[0]*=2**-.5 + self.coords={} + for name,shape,offset in [('c',(self.n,self.m),(.5,.5)),('u',(self.n+1,self.m),(0.,.5)),('v',(self.n,self.m+1),(.5,0.))]: + xx,yy=torch.meshgrid((torch.arange(shape[0],device=device,dtype=dtype)+offset[0])*self.hx,(torch.arange(shape[1],device=device,dtype=dtype)+offset[1])*self.hy,indexing='ij');self.coords[name]=(xx,yy,offset) + def sample(self,f,xx,yy,offset,mode='border'): + grid=torch.stack([2*(yy/self.hy-offset[1])/(f.shape[1]-1)-1,2*(xx/self.hx-offset[0])/(f.shape[0]-1)-1],-1) + return F.grid_sample(f[None,None],grid[None],mode='bilinear',padding_mode=mode,align_corners=True)[0,0] + def advect(self,f,name): + xx,yy,off=self.coords[name];u=self.sample(self.u,xx,yy,(0,.5));v=self.sample(self.v,xx,yy,(.5,0));return self.sample(f,xx-self.dt*u,yy-self.dt*v,off,'border' if name=='c' else 'zeros') + def diffuse(self,f): + z=F.pad(f,(1,1,1,1));return f+self.dt*self.nu*((z[2:,1:-1]+z[:-2,1:-1]-2*f)/self.hx**2+(z[1:-1,2:]+z[1:-1,:-2]-2*f)/self.hy**2) + def project(self,u,v): + u[0].zero_();u[-1].zero_();v[:,0].zero_();v[:,-1].zero_();div=(u[1:]-u[:-1])/self.hx+(v[:,1:]-v[:,:-1])/self.hy + ph=(self.cx@div@self.cy.T)/self.eig;ph[0,0].zero_();p=self.cx.T@ph@self.cy + u[1:-1]-=(p[1:]-p[:-1])/self.hx;v[:,1:-1]-=(p[:,1:]-p[:,:-1])/self.hy;return u,v + @torch.no_grad() + def advance(self): + for _ in range(self.substeps): + c=self.advect(self.c,'c');u=self.advect(self.u,'u');v=self.advect(self.v,'v');v[:,1:-1]+=self.dt*self.buoyancy*.5*(c[:,1:]+c[:,:-1]);u=self.diffuse(u);v=self.diffuse(v);self.u,self.v=self.project(u,v);self.c=c + return torch.stack([self.c,self.u[:-1],self.v[:,:-1]]).float() + + def step(self,x): + return self.advance() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/coarse_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/coarse_model.py new file mode 100644 index 0000000000000000000000000000000000000000..fc93ff2927589c947bd9a2ee18251f3a3eb5d26d --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/coarse_model.py @@ -0,0 +1,86 @@ +"""A universal coarse spatial interaction forecast, learned from scratch.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + +GRID = (4, 16) + + +def interaction(x): + return x - x.mean(-1, keepdim=True) - x.mean(-2, keepdim=True) + x.mean((-2, -1), keepdim=True) + + +class CoarseForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4 * GRID[0] * GRID[1]) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, MAXC * GRID[0] * GRID[1]), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, grids=None): + b, t, c = means.shape + assert t == 4 and c <= MAXC and grids is not None + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = ((means - last[:, None]) / scale[:, None]).clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + centered = (grids - means[:, :, :, None, None]) / scale[:, None, :, None, None] + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale) * mask / 5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + F.pad(centered.clamp(-100, 100), (0, 0, 0, 0, 0, MAXC-c)).flatten(1), + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + delta = self.net(torch.cat([features, md, self.ds(ids), time], 1)).reshape(b, MAXC, *GRID)[:, :c] + future = grids[:, -1] + scale[:, :, None, None] * delta * h[:, :, None, None] + return interaction(future) + + +@torch.no_grad() +def forecast_coarse(model, context, md, ids, steps): + x = context.double() + mean = x.mean((-2, -1)).float() + variance = x.var((-2, -1), unbiased=False).float() + grid = F.adaptive_avg_pool2d(x, GRID).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + grids=grid[None].expand(steps, -1, -1, -1, -1)) + + +def resize_axis(x, size, axis, periodic): + old = x.shape[axis] + position = (torch.arange(size, device=x.device, dtype=x.dtype) + .5) * old / size - .5 + left = position.floor().long() + weight = position - left + a, b = (left.remainder(old), (left+1).remainder(old)) if periodic else (left.clamp(0, old-1), (left+1).clamp(0, old-1)) + shape = [1] * x.ndim + shape[axis] = size + return x.index_select(axis, a) * (1-weight.reshape(shape)) + x.index_select(axis, b) * weight.reshape(shape) + + +@torch.no_grad() +def correct_coarse(prediction, forecast, weight, bc): + raw = prediction.double() + current = interaction(F.adaptive_avg_pool2d(raw, GRID)) + future = torch.where(torch.isfinite(forecast), forecast.double(), current) + difference = interaction(future - current) + difference = resize_axis(difference, raw.shape[-2], -2, bc[0]) + difference = resize_axis(difference, raw.shape[-1], -1, bc[1]) + # Preserve all existing global means and axis profiles exactly in real arithmetic. + return (raw + weight * interaction(difference)).float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/config.json b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5398ab144f29eb33d7847b432ea7d848f8f96722 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/config.json @@ -0,0 +1,85 @@ +{ + "width": 48, + "depth": 2, + "detail_decay": 0.8, + "physics": true, + "active_tau": 10.0, + "kinetic": { + "modes": 32, + "resolution": 256, + "iterations": 6, + "cuda_graph": true, + "transport": "phase", + "observer_weight": 1.0, + "observer_ridge": 0.01, + "observer_iterations": 12, + "quadrature_scale": 0.0001, + "quadrature_seed": 20261026, + "latent_scale": 0.01 + }, + "arena": true, + "neural_regularization": { + "profile_mix": 0.0, + "horizon": 30, + "pressure_mix": 0.0, + "pressure_start": 10, + "pressure_end": 30 + }, + "spectral_options": { + "gs_scale": 2, + "gs_dt": 0.25, + "gs_order": 2, + "gs_double": true, + "cuda_graph": true, + "advection_scale": 2, + "advection_double": true + }, + "native_rectangular": true, + "global_mixing": true, + "mean_forecast": { + "weight": 0.5, + "profile_weight": 0.25, + "width": 192, + "covariance_features": false + }, + "momentum_pressure": { + "weight": 0.05 + }, + "statistic_guidance": true, + "coarse_forecast": { + "width": 192, + "weight": 0.25 + }, + "translation_ensemble": { + "axis": 0, + "fractions": [ + 0.0, + 0.015625, + 0.03125, + 0.046875 + ], + "eligibility": "periodic boundary pair; identity for other boundaries", + "same_shared_weights_for_every_member_and_case": true, + "duplicate_computations_shared": true + }, + "final_nonnegative_fields": [ + "density", + "concentration" + ], + "density_transport": { + "weight": 0.01344492073972926, + "cfl": 0.4, + "max_substeps": 1024 + }, + "latent_state": { + "moment": "F6", + "coefficient_grid": [ + 16, + 16 + ], + "scale": 0.01, + "alignment_power": 3, + "input": "first shared neural pass on the supplied four frames and metadata", + "parameters_fixed_at_inference": true + } +} \ No newline at end of file diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/constraints.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/constraints.py new file mode 100644 index 0000000000000000000000000000000000000000..5a191ecbd935e3c2619942fe7e59b00c7e991565 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/constraints.py @@ -0,0 +1,29 @@ +"""Fixed physical projection and forecast-uncertainty regularization.""" +import math +import torch + +@torch.no_grad() +def project_active(context,prediction,meta,tau=10.0): + # This is a metadata-conditioned part of one shared operator, not a separate checkpoint. + fields=list(meta.get('fields',[])) + if str(meta.get('pde',''))!='active_matter' or context.shape[1]!=11 or fields!=['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3']: + return prediction + x=context;y=prediction.clone();ksteps,c,nx,ny=y.shape;device=y.device + L=float(meta.get('constants',{}).get('L',meta['domain_extent'][0]));dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=L/nx,device=device)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=L/ny,device=device)[None,:];k2=kx*kx+ky*ky + t=torch.arange(1,ksteps+1,device=device,dtype=torch.float32) + cf=torch.fft.rfft2(x[-1,0]);y[:,0]=torch.fft.irfft2(cf[None]*torch.exp(-0.05*dt*k2[None]*t[:,None,None]),s=(nx,ny)) + err=(y[:,0]-y[:,3]-y[:,6])/2;y[:,3]+=err;y[:,6]+=err + sym=(y[:,4]+y[:,5])/2;y[:,4]=sym;y[:,5]=sym + y[:,1:3]-=y[:,1:3].mean((-2,-1),keepdim=True);y[:,7:]-=y[:,7:].mean((-2,-1),keepdim=True) + e=(y[:,7]-y[:,10])/2;y[:,7]=e;y[:,10]=-e;sym=(y[:,8]+y[:,9])/2;y[:,8]=sym;y[:,9]=sym + kk=k2.sqrt().clamp_min(1e-20);ax=-ky/kk;ay=kx/kk + A=torch.stack(torch.broadcast_tensors(ax,ay,1j*kx*ax,.5j*(kx*ay+ky*ax),.5j*(kx*ay+ky*ax),1j*ky*ay)) + ids=[1,2,7,8,9,10];std=x[:,ids].std((0,2,3),unbiased=False).clamp_min(1e-4);weight=std.pow(-2) + yy=torch.fft.rfft2(y[:,ids]);den=(A.abs().square()*weight[:,None,None]).sum(0).clamp_min(1e-20) + aa=(A.conj()[None]*yy*weight[None,:,None,None]).sum(1)/den + y[:,ids]=torch.fft.irfft2(A[None]*aa[:,None],s=(nx,ny)) + if tau>0: + prior=x[-1].mean((-2,-1),keepdim=True);w=torch.exp(-t/tau) + y[:,1:]=prior[None,1:]+(y[:,1:]-prior[None,1:])*w[:,None,None,None] + return y diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/density_transport.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/density_transport.py new file mode 100644 index 0000000000000000000000000000000000000000..fe2339073e86f02d282df378e6ecc9acb1097e31 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/density_transport.py @@ -0,0 +1,144 @@ +"""Independent conservative continuity update with MC reconstruction and RK2.""" +import math +import torch + + +def neighbor(value, direction, axis, periodic): + if periodic: + return torch.roll(value, -direction, axis) + first = value.select(axis, 0).unsqueeze(axis) + last = value.select(axis, value.shape[axis]-1).unsqueeze(axis) + if direction == 1: + return torch.cat([value.narrow(axis, 1, value.shape[axis]-1), last], axis) + return torch.cat([first, value.narrow(axis, 0, value.shape[axis]-1)], axis) + + +def minmod3(a, b, c): + positive = (a > 0) & (b > 0) & (c > 0) + negative = (a < 0) & (b < 0) & (c < 0) + magnitude = torch.minimum(torch.minimum(a.abs(), b.abs()), c.abs()) + return torch.where(positive, magnitude, torch.where(negative, -magnitude, 0.)) + + +def rhs(density, velocity, spacing, periodic): + result = torch.zeros_like(density) + for j, (dx, wrap) in enumerate(zip(spacing, periodic)): + axis = j-2 + left = neighbor(density, -1, axis, wrap) + right = neighbor(density, 1, axis, wrap) + slope = minmod3(2*(density-left), .5*(right-left), 2*(right-density)) + vn = velocity[j] + face_velocity = .5*(vn+neighbor(vn, 1, axis, wrap)) + positive = density+.5*slope + negative = right-.5*neighbor(slope, 1, axis, wrap) + flux = face_velocity*torch.where(face_velocity >= 0, positive, negative) + previous_flux = neighbor(flux, -1, axis, wrap) + if not wrap: + # Replicated primitive ghost cells: the incoming lower face uses + # the boundary density and normal velocity directly. + previous_flux = previous_flux.clone() + previous_flux.select(axis, 0).copy_(density.select(axis, 0)*vn.select(axis, 0)) + result -= (flux-previous_flux)/dx + return result + + +@torch.no_grad() +def forecast(context_density, velocities, dt, extent, periodic=(True,False), cfl=.4, cuda_graph=True): + """velocities[0] is observed; subsequent entries are chosen forecast velocities.""" + assert velocities.shape[1] == 2 and velocities.shape[-2:] == context_density.shape + spacing = [float(l)/n for l,n in zip(extent, context_density.shape)] + maximum = (velocities[:,0].abs()/spacing[0]+velocities[:,1].abs()/spacing[1]).amax().item() + substeps = max(1, math.ceil(float(dt)*maximum/cfl)) + small_dt = float(dt)/substeps + density = context_density.clone() + start_velocity = velocities[0].clone() + end_velocity = velocities[1].clone() + fraction = density.new_zeros(()) + + def body(): + velocity0 = start_velocity+(end_velocity-start_velocity)*fraction + velocity1 = start_velocity+(end_velocity-start_velocity)*(fraction+1/substeps) + stage = density+small_dt*rhs(density, velocity0, spacing, periodic) + updated = .5*(density+stage+small_dt*rhs(stage, velocity1, spacing, periodic)) + density.copy_(updated) + fraction.add_(1/substeps) + + graph = None + if cuda_graph and density.is_cuda: + initial = density.clone() + stream = torch.cuda.Stream() + stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3): + body() + torch.cuda.current_stream().wait_stream(stream) + density.copy_(initial); fraction.zero_(); torch.cuda.synchronize() + graph = torch.cuda.CUDAGraph() + with torch.cuda.graph(graph): + body() + density.copy_(initial); fraction.zero_() + output = [] + minimum = float(density.min()) + for k in range(len(velocities)-1): + start_velocity.copy_(velocities[k]); end_velocity.copy_(velocities[k+1]); fraction.zero_() + for _ in range(substeps): + graph.replay() if graph is not None else body() + assert torch.isfinite(density).all() + minimum = min(minimum, float(density.min())) + output.append(density.clone()) + return torch.stack(output), {'substeps':substeps, 'minimum_density':minimum, + 'spacing':spacing, 'cfl':cfl, 'cuda_graph':graph is not None} + + +"""A fixed continuity blend selected on TRAIN; observations remain context-only.""" +import math +import numpy as np +import torch + + +def density_eligible(meta, shape): + fields = list(meta.get('fields', [])) + bc = list(meta.get('boundary_conditions', [])) + extent = meta.get('domain_extent', []) + return (len(shape) == 3 and shape[0] == 4 and min(shape[-2:]) > 1 + and fields == ['density', 'pressure', 'velocity_0', 'velocity_1'] + and bc == ['periodic', 'periodic', 'open', 'open'] + and len(extent) == 2 and all(math.isfinite(float(v)) and float(v) > 0 for v in extent) + and math.isfinite(float(meta.get('dt', 0))) and float(meta.get('dt', 0)) > 0) + + +@torch.no_grad() +def correct_density(context, prediction, meta, weight, cfl=.4, max_substeps=1024, device='cuda'): + if len(prediction) == 0 or weight == 0 or not density_eligible(meta, prediction.shape[1:]): + return prediction + x = torch.as_tensor(context, device=device, dtype=torch.float32) + p = torch.as_tensor(prediction, device=device, dtype=torch.float32) + velocity = torch.cat([x[-1:, 2:4], p[:, 2:4]]) + if not torch.isfinite(velocity).all() or not torch.isfinite(x[-1, 0]).all() or x[-1, 0].min() < 0: + return prediction + spacing = [float(v)/n for v, n in zip(meta['domain_extent'], prediction.shape[-2:])] + maximum = (velocity[:, 0].abs()/spacing[0]+velocity[:, 1].abs()/spacing[1]).amax().item() + substeps = max(1, math.ceil(float(meta['dt'])*maximum/cfl)) + if substeps > max_substeps: + return prediction + try: + transported, settings = forecast(x[-1, 0], velocity, float(meta['dt']), + meta['domain_extent'], (True, False), cfl, device == 'cuda') + except RuntimeError: + # The eager implementation has the same arithmetic if graph capture fails. + try: + transported, settings = forecast(x[-1, 0], velocity, float(meta['dt']), + meta['domain_extent'], (True, False), cfl, False) + except (RuntimeError, AssertionError): + return prediction + except AssertionError: + return prediction + if not torch.isfinite(transported).all() or settings['minimum_density'] < -1e-7: + return prediction + output = prediction.copy() + output[:, 0] = ((1-weight)*p[:, 0].double()+weight*transported.double()).float().cpu().numpy() + # Preserve the final physical support projection when the parent has one. + if np.all(prediction[:, 0] >= 0): + output[:, 0] = np.maximum(output[:, 0], 0.) + print('density transport blend:', weight, 'substeps:', substeps, flush=True) + return output diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/launch.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..199f186c5c4a10c772dbdc88d98fc9a7582f3ae9 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/launch.py @@ -0,0 +1,11 @@ +"""Pin one complete package directory for the duration of inference.""" +import os,sys,runpy +from pathlib import Path +if len(sys.argv)!=3:raise SystemExit('usage: bash infer.sh INPUT_DIR OUTPUT_DIR') +inputs,outputs=[str(Path(p).resolve()) for p in sys.argv[1:]] +package_fd=os.open(Path(__file__).parent,os.O_RDONLY|os.O_DIRECTORY) +package_path=f'/proc/self/fd/{package_fd}' +os.environ['PDE_SURROGATE_PACKAGE_DIR']=package_path +sys.path.insert(0,package_path) +sys.argv=[package_path+'/predict.py',inputs,outputs] +runpy.run_path(sys.argv[0],run_name='__main__') diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/make_bingham_table.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/make_bingham_table.py new file mode 100644 index 0000000000000000000000000000000000000000..3cba92fb5f0503306c910a27c10b6e42c7099aa6 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/make_bingham_table.py @@ -0,0 +1,9 @@ +import numpy as np +from scipy.special import ive +from pathlib import Path +s=np.linspace(0,12,8193);mu=-np.expm1(-s);r=mu[1:];k=np.where(r<.53,2*r+r**3+5*r**5/6,np.where(r<.85,-.4+1.39*r+.43/(1-r),1/(r**3-4*r**2+3*r))) +for _ in range(6): + z=ive(1,k)/ive(0,k);k=np.maximum(k+(r-z)/np.maximum(1-z*z-z/k,1e-14),1e-10) +t=np.zeros((65,len(s)),np.float32);t[0]=1 +for j in range(1,65):t[j,1:]=ive(j,k)/ive(0,k) +np.save(Path(__file__).with_name('bingham_table.npy'),t) diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/mean_model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/mean_model.py new file mode 100644 index 0000000000000000000000000000000000000000..264dc34336192462ea2ddb804fa45875a4129b01 --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/mean_model.py @@ -0,0 +1,107 @@ +"""Shared context-statistic forecast; all parameters learned from random initialization.""" +import torch +from torch import nn +import torch.nn.functional as F +from model import MAXC, METADIM, DATASETS + + +class MeanForecast(nn.Module): + def __init__(self, width=192): + super().__init__() + self.covariates = nn.Linear(4*MAXC*MAXC,width,bias=False) + nn.init.zeros_(self.covariates.weight) + self.ds = nn.Embedding(len(DATASETS) + 1, 24) + self.net = nn.Sequential( + nn.Linear(METADIM + 24 + MAXC * (11 + 4*2*16) + 6, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), + nn.Linear(width, width), nn.SiLU(), nn.Linear(width, MAXC*33), + ) + nn.init.zeros_(self.net[-1].weight) + nn.init.zeros_(self.net[-1].bias) + + def forward(self, means, variances, md, ids, horizon, mask=None, profiles=None, covariances=None): + # One independently specified horizon per row. All supplied statistics + # describe only the four context frames; no inference-time fitting. + b, t, c = means.shape + assert t == 4 and c <= MAXC + last = means[:, -1] + scale = (variances.mean(1) + means.var(1, unbiased=False)).sqrt().clamp_min(1e-3) + relative = (means - last[:, None]) / scale[:, None] + relative = relative.clamp(-100, 100) + logvar = torch.log1p(variances.clamp_min(0) / scale[:, None].square()) + if mask is None: + mask = torch.ones_like(last) + assert profiles is not None + last_profiles=profiles[:,-1] + profiles=(profiles-means[:,:,:,None,None])/scale[:,None,:,None,None] + profiles=F.pad(profiles.clamp(-100,100),(0,0,0,0,0,MAXC-c)).flatten(1) + features = torch.cat([ + F.pad(relative, (0, MAXC-c)).flatten(1), + F.pad(logvar, (0, MAXC-c)).flatten(1), + F.pad(torch.asinh(last), (0, MAXC-c)), + F.pad(torch.log(scale)*mask/5, (0, MAXC-c)), + F.pad(mask, (0, MAXC-c)), + profiles, + ], 1) + h = horizon.float().reshape(-1, 1) / 30 + time = torch.cat([h, h.square(), torch.log1p(h*30)/4, + torch.sin(h), torch.cos(h), torch.exp(-h)], 1) + hidden=self.net[0](torch.cat([features,md,self.ds(ids),time],1)) + if covariances is not None: + covariance=covariances/scale[:,None,:,None]/scale[:,None,None,:] + covariance=F.pad(covariance.clamp(-100,100),(0,MAXC-c,0,MAXC-c)).flatten(1) + hidden=hidden+self.covariates(covariance) + delta=self.net[1:](hidden).reshape(b,MAXC,33)[:,:c] + future_mean=last+scale*delta[:,:,0]*h + future_profiles=last_profiles+scale[:,:,None,None]*delta[:,:,1:].reshape(b,c,2,16)*h[:,:,None,None] + future_profiles=future_profiles-future_profiles.mean(-1,keepdim=True)+future_mean[:,:,None,None] + return future_mean,future_profiles + + +@torch.no_grad() +def forecast_statistics(model, context, md, ids, steps): + mean = context.double().mean((-2, -1)).float() + variance = context.double().var((-2, -1), unbiased=False).float() + profx=F.adaptive_avg_pool2d(context.double(),(16,1)).squeeze(-1) + profy=F.adaptive_avg_pool2d(context.double(),(1,16)).squeeze(-2) + profiles=torch.stack([profx,profy],-2).float() + flattened=context.double().flatten(-2) + centered=flattened-flattened.mean(-1,keepdim=True) + covariance=(centered@centered.transpose(-2,-1)/centered.shape[-1]).float() + horizons = torch.arange(1, steps+1, device=context.device) + return model(mean[None].expand(steps, -1, -1), + variance[None].expand(steps, -1, -1), + md.expand(steps, -1), ids.expand(steps), horizons, + profiles=profiles[None].expand(steps,-1,-1,-1,-1), + covariances=covariance[None].expand(steps,-1,-1,-1)) + + +@torch.no_grad() +def forecast_means(model,context,md,ids,steps): + return forecast_statistics(model,context,md,ids,steps)[0] + +def interpolate_profile(profile,size,periodic): + position=(torch.arange(size,device=profile.device,dtype=profile.dtype)+.5)*16/size-.5 + left=position.floor().long();weight=position-left + if periodic: + i=left.remainder(16);j=(left+1).remainder(16) + else: + i=left.clamp(0,15);j=(left+1).clamp(0,15) + output=profile[...,i]*(1-weight)+profile[...,j]*weight + return output-output.mean(-1,keepdim=True) + +@torch.no_grad() +def correct_statistics(prediction,means,profiles,mean_weight,profile_weight,bc): + raw=prediction.double();center=raw.mean((-2,-1)) + means=torch.where(torch.isfinite(means),means.double(),center) + output=raw+mean_weight*(means-center)[:,:,None,None] + if profile_weight: + current_x=F.adaptive_avg_pool2d(raw,(16,1)).squeeze(-1) + current_y=F.adaptive_avg_pool2d(raw,(1,16)).squeeze(-2) + current=torch.stack([current_x,current_y],-2) + forecast=torch.where(torch.isfinite(profiles),profiles.double(),current) + difference=(forecast-forecast.mean(-1,keepdim=True))-(current-current.mean(-1,keepdim=True)) + dx=interpolate_profile(difference[:,:,0],raw.shape[-2],bc[0]) + dy=interpolate_profile(difference[:,:,1],raw.shape[-1],bc[1]) + output=output+profile_weight*(dx[:,:,:,None]+dy[:,:,None,:]) + return output.float() diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/model.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/model.py new file mode 100644 index 0000000000000000000000000000000000000000..66d11509dc3f8d4031ff6913c1077cc32b5e0e6f --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/model.py @@ -0,0 +1,119 @@ +"""Shared metadata-conditioned convolutional operator, trained from scratch.""" +import math +import numpy as np +import torch +from torch import nn +import torch.nn.functional as F + +DATASETS=['ape2d/burgers','ape2d/decay_turb','ape2d/diff','ape2d/fisher','ape2d/gs_alpha','ape2d/gs_beta','ape2d/gs_delta','ape2d/gs_epsilon','ape2d/gs_gamma','ape2d/gs_iota','ape2d/gs_kappa','ape2d/gs_theta','ape2d/kdv','ape2d/kolm_flow','ape2d/ks','ape2d/sh','pdearena/navier_stokes_2d','well/active_matter','well/turbulent_radiative_layer_2D'] +CONSTANTS=['Viscosity','Viscosity X','Viscosity Y','Diffusivity','Reactivity','Feed Rate','Kill Rate','Critical Number','Domain Extent','tcool','L','zeta','alpha','buoyancy_y'] +MAXC=11 + +def metadata(meta,dataset): + vals=[math.log(max(float(meta.get('dt',1)),1e-12))/5] + vals += [math.log(max(float(v),1e-12))/5 for v in meta.get('domain_extent',[1,1])] + bc=meta.get('boundary_conditions',['periodic']*4) + vals += [float(b=='periodic') for b in bc]+[float(b=='wall') for b in bc] + const=meta.get('constants',{}) + for k in CONSTANTS: + v=float(const.get(k,0)); vals += [math.copysign(math.log1p(abs(v)*100),v)/8,float(k in const)] + return np.array(vals,np.float32),DATASETS.index(dataset)+1 if dataset in DATASETS else 0,(bc[0]=='periodic',bc[2]=='periodic') + +METADIM=3+8+2*len(CONSTANTS) + +def pad(x,n,bc): + x=F.pad(x,(n,n,0,0),mode='circular' if bc[1] else 'replicate') + return F.pad(x,(0,0,n,n),mode='circular' if bc[0] else 'replicate') + +class Block(nn.Module): + def __init__(self,c,cond): + super().__init__(); self.dw=nn.Conv2d(c,c,5,groups=c); self.norm=nn.LayerNorm(c); self.fc1=nn.Linear(c,2*c);self.fc2=nn.Linear(2*c,c) + self.film=nn.Linear(cond,2*c); nn.init.zeros_(self.film.weight);nn.init.zeros_(self.film.bias) + self.scale=nn.Parameter(torch.ones(c)*0.1) + def forward(self,x,e,bc): + h=self.dw(pad(x,2,bc)).permute(0,2,3,1);h=self.norm(h) + g,b=self.film(e).chunk(2,-1);h=h*(1+g[:,None,None,:])+b[:,None,None,:] + h=self.fc2(F.gelu(self.fc1(h)))*self.scale + return x+h.permute(0,3,1,2) + +class GlobalMix(nn.Module): + def __init__(self,width,cond=192,latent=24,modes_x=32,modes_y=96): + super().__init__();self.latent=latent;self.mx=modes_x;self.my=modes_y + self.input=nn.Conv2d(width,latent,1);self.output=nn.Conv2d(latent,width,1) + self.weight=nn.Parameter(torch.randn(2,latent,modes_x,modes_y,2)*0.5) + self.condition=nn.Linear(cond,latent*2);nn.init.zeros_(self.condition.weight);nn.init.zeros_(self.condition.bias) + nn.init.zeros_(self.output.weight);nn.init.zeros_(self.output.bias) + def forward(self,x,e,bc): + h,w=x.shape[-2:];q=self.input(x) + # Even extensions impose the same zero-normal-gradient convention as replicated padding. + if not bc[0]:q=torch.cat([q,q.flip(-2)],-2) + if not bc[1]:q=torch.cat([q,q.flip(-1)],-1) + xf=torch.fft.rfft2(q.float(),norm='ortho');yf=torch.zeros_like(xf) + nx=min(self.mx,q.shape[-2]//2);ny=min(self.my,xf.shape[-1]) + weight=torch.view_as_complex(self.weight.contiguous());gate=torch.view_as_complex(self.condition(e).float().reshape(len(e),self.latent,2).contiguous())[:, :, None, None] + yf[:,:,:nx,:ny]=xf[:,:,:nx,:ny]*weight[0,:,:nx,:ny]*(1+gate) + yf[:,:,-nx:,:ny]=xf[:,:,-nx:,:ny]*weight[1,:,-nx:,:ny]*(1+gate) + z=torch.fft.irfft2(yf,s=q.shape[-2:],norm='ortho')[:,:,:h,:w] + return self.output(F.gelu(z)) + +class Operator(nn.Module): + def __init__(self,width=48,depth=2): + super().__init__(); self.width=width;self.depth=depth;d=192;ws=[width,width*2,width*4,width*6] + self.ds=nn.Embedding(len(DATASETS)+1,32) + self.meta=nn.Sequential(nn.Linear(METADIM+32+MAXC*3,d),nn.SiLU(),nn.Linear(d,d),nn.SiLU()) + self.stem=nn.Conv2d(4*MAXC+2,ws[0],3);self.global_mix=GlobalMix(ws[0],d) + self.guide_stem=nn.Conv2d(MAXC,ws[0],1,bias=False);nn.init.zeros_(self.guide_stem.weight) + self.latent_head=nn.Conv2d(ws[0],2,1);nn.init.zeros_(self.latent_head.weight);nn.init.zeros_(self.latent_head.bias) + self.enc=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws]) + self.down=nn.ModuleList([nn.Conv2d(ws[i],ws[i+1],2,stride=2) for i in range(3)]) + self.up=nn.ModuleList([nn.Conv2d(ws[i+1]+ws[i],ws[i],1) for i in range(3)]) + self.dec=nn.ModuleList([nn.ModuleList([Block(c,d) for _ in range(depth)]) for c in ws[:-1]]) + self.head=nn.Conv2d(ws[0],MAXC,1);nn.init.zeros_(self.head.weight);nn.init.zeros_(self.head.bias) + def forward(self,x,md,ids,bc=(True,True),guidance=None,return_latent=False): + # Reversible feature normalization is computed from supplied states, with fixed learned parameters. + b,t,c,h,w=x.shape + mean=x[:,-1].mean((-2,-1),keepdim=True) + std=x.std((1,3,4),unbiased=False).clamp_min(1e-3)[:,:,None,None] + z=(x-mean[:,None])/std[:,None] + mask=torch.ones(b,c,device=x.device) + sm=torch.cat([torch.asinh(mean.flatten(1)),torch.log(std.flatten(1))/5,mask],1) + if c0 for v in [dt,lx,ly]):return prediction + state=torch.cat([context,prediction]).double();density,pressure,vx,vy=[fields.index(name) for name in names] + rho=state[:,density];u=state[:,vx];v=state[:,vy];nx,ny=rho.shape[-2:];dx=lx/nx;dy=ly/ny + ddx=lambda value:(torch.roll(value,-1,-2)-torch.roll(value,1,-2))/(2*dx) + force_x=-gradient(rho*u,dt,0)-ddx(rho*u*u)-gradient(rho*u*v,dy,-1) + force_y=-gradient(rho*v,dt,0)-ddx(rho*u*v)-gradient(rho*v*v,dy,-1) + faces_x=.5*(force_x+torch.roll(force_x,-1,-2));faces_y=.5*(force_y[...,:-1]+force_y[...,1:]) + estimate=project_faces(faces_x,faces_y,dx,dy)[len(context):] + original=prediction[:,pressure].double();center=original.mean((-2,-1),keepdim=True) + updated=(1-weight)*original+weight*(estimate+center) + result=prediction.clone();result[:,pressure]=torch.where(torch.isfinite(updated),updated,original).float() + return result diff --git a/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/numerics.py b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/numerics.py new file mode 100644 index 0000000000000000000000000000000000000000..3c45e36262df2d382b1be3a53691ec656a37ca9a --- /dev/null +++ b/tasks/pde-surrogate/pde-surrogate-astra-xhigh-1h100-48h-20261020b/submissions/S303/numerics.py @@ -0,0 +1,21 @@ +"""Portable periodic Fourier resampling with even-Nyquist handling.""" +import torch +def resample_axis(x,n,axis): + old=x.shape[axis] + if old==n:return x + real=not x.is_complex();xf=torch.fft.fft(x,dim=axis);shape=list(x.shape);shape[axis]=n;y=torch.zeros(shape,device=x.device,dtype=xf.dtype);nn=min(old,n);nyq=nn//2+1 + sl=[slice(None)]*x.ndim;sl[axis]=slice(0,nyq);y[tuple(sl)]=xf[tuple(sl)] + neg=nn-nyq + if neg: + sl[axis]=slice(-neg,None);y[tuple(sl)]=xf[tuple(sl)] + if nn%2==0: + sl[axis]=nn//2 + if nMAXC: + return np.concatenate([predict_member(model,x[:,i:i+MAXC].cpu().numpy(),meta,ds,steps,device,detail_decay,physics,active_tau,kinetic_cfg,arena,neural_regularization,native_rectangular,spectral_options) for i in range(0,c,MAXC)],1) + factor=2 if min(h,w)>=192 or min(h,w)>=128 and max(h,w)/min(h,w)>2 and not native_rectangular else 1 + z=F.avg_pool2d(x,factor) if factor>1 else x + md,di,bc=metadata(meta,ds);md=torch.tensor(md,device=device)[None];ids=torch.tensor([di],device=device) + future_means,future_profiles=forecast_statistics(model.mean_forecast,x,md,ids,steps) if hasattr(model,'mean_forecast') else (None,None) + future_coarse=forecast_coarse(model.coarse_forecast,x,md,ids,steps) if hasattr(model,'coarse_forecast') else None + kinetic_steps=0 + detail=x[-1]-F.interpolate(z[-1,None],size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else None + ctx=z[None];out=[] + solver=SpectralUpdate(meta,(h,w),device,**(spectral_options or {})) if physics and eligible(meta,c) else None + if arena and arena_eligible(meta,c) and min(h,w)>1:solver=ArenaOperator(meta,x[-1]) + fullstate=x[-1];physical_steps=0;physical_name=type(solver).__name__ if solver is not None else None + for k in range(steps): + with torch.autocast(device_type=device,dtype=torch.bfloat16,enabled=device=='cuda'):y,current_latent=model(ctx,md,ids,bc,guidance=(future_means[k:k+1],future_profiles[k:k+1]) if future_means is not None else None,return_latent=True) + if k==0:latent=current_latent[0] + if not torch.isfinite(y).all():y=ctx[:,-1].clone() + if solver is not None: + candidate=solver.step(fullstate) + if torch.isfinite(candidate).all(): + physical_steps+=1 + fullstate=candidate;y=(F.avg_pool2d(candidate,factor) if factor>1 else candidate)[None] + else:solver=None + ctx=torch.cat([ctx[:,1:],y[:,None]],1) + yh=F.interpolate(y,size=(h,w),mode='bilinear',align_corners=False)[0] if factor>1 else y[0] + if detail is not None:yh=yh+detail*(detail_decay**(k+1)) + if solver is not None:yh=fullstate + yh=torch.where(torch.isfinite(yh),yh,x[-1]) + out.append(yh.cpu().numpy()) + pred=np.stack(out).astype(np.float32) + if physical_name:print(f'physical steps: {physical_name} {physical_steps}/{steps}',flush=True) + if neural_regularization is not None and solver is None: + raw=torch.as_tensor(pred,device=device);pred=regularize(raw,meta,**neural_regularization).cpu().numpy() + if future_means is not None and solver is None: + pred=correct_statistics(torch.as_tensor(pred,device=device),future_means,future_profiles,model.mean_weight,model.profile_weight,bc).cpu().numpy() + if active_tau is not None: + raw=torch.as_tensor(pred,device=device);projected=project_active(x,raw,meta,active_tau) + pred=torch.where(torch.isfinite(projected),projected,raw).cpu().numpy() + if kinetic_cfg and str(meta.get('pde',''))=='active_matter' and c==11 and list(meta.get('fields',[]))==['concentration','velocity_0','velocity_1','D_0','D_1','D_2','D_3','E_0','E_1','E_2','E_3'] and all(k in meta.get('constants',{}) for k in ['L','alpha','zeta']): + from kinetic import KineticOperator + cfg=dict(kinetic_cfg);cfg['latent_coefficients']=latent + try: + op=KineticOperator(meta,x,**cfg) + except RuntimeError: + if not cfg.get('cuda_graph',False):raise + cfg['cuda_graph']=False;op=KineticOperator(meta,x,**cfg) + physical=op.rollout(steps,guard=True);used=len(physical);kinetic_steps=used + if used:pred[:used]=physical.cpu().numpy() + print(f'kinetic frames: {used}/{steps}',flush=True) + if getattr(model,'momentum_pressure_weight',0): + pred=pressure_projection(x,torch.as_tensor(pred,device=device),meta,model.momentum_pressure_weight).cpu().numpy() + if future_coarse is not None and solver is None: + prefix=max(physical_steps,kinetic_steps) + if prefix1;self.initialized=False + self.double=gs_double if meta['pde']=='Gray Scott' else advection_double if advective else False + self.real_dtype=torch.float64 if self.double else torch.float32 + self.complex_dtype=torch.complex128 if self.double else torch.complex64 + if self.fine:shape=tuple(int(scale*s) for s in shape) + self.name=name=str(meta['pde']);self.shape=shape;self.p=p=meta['constants'];nx,ny=shape;lx,ly=meta['domain_extent'];dt=float(meta['dt']) + kx=2*math.pi*torch.fft.fftfreq(nx,d=lx/nx,device=device,dtype=torch.float64)[:,None];ky=2*math.pi*torch.fft.rfftfreq(ny,d=ly/ny,device=device,dtype=torch.float64)[None,:] + self.kx=kx.to(self.real_dtype);self.ky=ky.to(self.real_dtype);k2=kx*kx+ky*ky;self.k2=k2.to(self.real_dtype) + self.mask=(torch.fft.fftfreq(nx,device=device).abs()[:,None]<1/3)&(torch.fft.rfftfreq(ny,device=device)[None,:].abs()<1/3) + fx=torch.fft.fftfreq(nx,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(ny,device=device,dtype=torch.float64)[None,:] + self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if self.fine: + ox,oy=self.original_shape;fx=torch.fft.fftfreq(ox,device=device,dtype=torch.float64)[:,None];fy=torch.fft.rfftfreq(oy,device=device,dtype=torch.float64)[None,:];self.observation=(torch.sinc(fx)/torch.sinc(fx/8)*torch.sinc(fy)/torch.sinc(fy/8)).to(self.real_dtype) + if name=='Diffusion':L=-p['Viscosity X']*kx*kx-p['Viscosity Y']*ky*ky;dm=dt + elif name=='Fisher-KPP':L=-p['Diffusivity']*k2+p['Reactivity'];dm=dt;self.order=2 + elif name=='Gray Scott':L=torch.stack([-2e-5*k2-p['Feed Rate'],-1e-5*k2-p['Feed Rate']-p['Kill Rate']]);dm=gs_dt;self.order=gs_order + elif name=='Swift-Hohenberg':L=p['Reactivity']-(p['Critical Number']-k2)**2;dm=.1;self.order=2 + elif name=='Kuramoto-Sivashinsky':L=k2-kx**4-ky**4;dm=.1;self.order=2 + elif name=='Korteweg-de-Vries':L=-p['Viscosity']*k2+1j*(kx**3+ky**3);dm=.01;self.order=2 + elif name=='Burgers':L=-p['Viscosity']*k2;dm=.0005 + elif 'Navier-Stokes' in name:L=-p['Viscosity']*k2-(0.2 if 'Kolmogorov' in name else 0);dm=dt/128 + else:raise ValueError(name) + self.force=0 + if 'Kolmogorov' in name: + yy=torch.arange(ny,device=device,dtype=self.real_dtype)*ly/ny+7/16*ly/self.original_shape[1] if self.fine else (torch.arange(ny,device=device)+7/16)*ly/ny + self.force=torch.fft.rfft2((-4*torch.cos(4*yy)).expand(nx,ny)) + self.substeps=max(1,math.ceil(dt/dm));h=dt/self.substeps;z=L*h + cast=lambda v:v.to(self.complex_dtype if v.is_complex() else self.real_dtype) + self.e=cast(torch.exp(z));self.e2=cast(torch.exp(z/2));self.q=cast(h/2*phi(z/2,1));ph1,ph2,ph3=phi(z,1),phi(z,2),phi(z,3) + self.etd1=cast(h*ph1);self.etd2=cast(h*ph2) + self.f1=cast(h*(ph1-3*ph2+4*ph3));self.f2=cast(h*(ph2-2*ph3));self.f3=cast(h*(-ph2+4*ph3)) + self.graph=None + if cuda_graph and torch.device(device).type=='cuda' and name!='Diffusion': + nc=2 if name in {'Burgers','Korteweg-de-Vries','Gray Scott'} else 1 + self.buffer=torch.zeros((nc,nx,ny//2+1),device=device,dtype=self.complex_dtype) + try: + def body():self.buffer.copy_(self.substep(self.buffer)) + stream=torch.cuda.Stream();stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(stream): + for _ in range(3):body() + torch.cuda.current_stream().wait_stream(stream);torch.cuda.synchronize();graph=torch.cuda.CUDAGraph() + with torch.cuda.graph(graph):body() + self.graph=graph + except RuntimeError: + self.graph=None + def fft(self,x):return torch.fft.rfft2(x) + def ifft(self,v):return torch.fft.irfft2(v,s=self.shape) + def nonlinear(self,v): + name=self.name;u=self.ifft(v) + if name=='Fisher-KPP':n=-self.p['Reactivity']*u*u + elif name=='Gray Scott': + uvv=u[0]*u[1].square();n=torch.stack([self.p['Feed Rate']-uvv,uvv]) + elif name=='Swift-Hohenberg':n=u*u-u*u*u + elif name=='Kuramoto-Sivashinsky': + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=-.5*(dx*dx+dy*dy);n=n-n.mean((-2,-1),keepdim=True) + elif name in {'Burgers','Korteweg-de-Vries'}: + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=(6 if name=='Korteweg-de-Vries' else -1)*(u[0]*dx+u[1]*dy) + elif 'Navier-Stokes' in name: + inv=torch.where(self.k2==0,0,1/self.k2.clamp_min(1e-20));vx=self.ifft(1j*self.ky*inv*v);vy=self.ifft(-1j*self.kx*inv*v) + dx=self.ifft(1j*self.kx*v);dy=self.ifft(1j*self.ky*v);n=vx*dx+vy*dy + else:n=torch.zeros_like(u) + return self.fft(n)*self.mask+self.force + @torch.no_grad() + def step(self,u): + if self.fine: + if not self.initialized: + latent=torch.fft.irfft2(torch.fft.rfft2(u.to(self.real_dtype))/self.observation,s=self.original_shape);v=self.fft(resample(latent,self.shape));self.initialized=True + else:v=self.internal + else:v=self.fft(u.to(self.real_dtype))/self.observation + if self.name=='Diffusion':return self.ifft(self.e*v*self.observation) + if self.graph is not None: + self.buffer.copy_(v) + for _ in range(self.substeps):self.graph.replay() + v=self.buffer + else: + for _ in range(self.substeps):v=self.substep(v) + if self.fine: + self.internal=v + low=resample(self.ifft(v),self.original_shape) + return torch.fft.irfft2(torch.fft.rfft2(low)*self.observation,s=self.original_shape).float() + return self.ifft(v*self.observation).float() + def substep(self,v): + if self.order==2: + nv=self.nonlinear(v);stage=self.e*v+self.etd1*nv;return stage+self.etd2*(self.nonlinear(stage)-nv) + nv=self.nonlinear(v);a=self.e2*v+self.q*nv;na=self.nonlinear(a);b=self.e2*v+self.q*na;nb=self.nonlinear(b);c=self.e2*a+self.q*(2*nb-nv) + return self.e*v+self.f1*nv+2*self.f2*(na+nb)+self.f3*self.nonlinear(c)